NREL/TP-210-4567 . UC Category: 270 . DE92001154
NREL/TP--210-4567
DE92
__----
NREL Preprint
001154
_---
r the 22nd IEEE
lists Conference
Materials Science and Engineering
Division
E. Nelsen, Editor
National Renewable Energy Laboratory
(formerly the Solar Energy ResearchInstitute)
1617 Cole Boulevard
Golden, Colorado 80401-3393
A Division of Midwest Research Institute
Operated for the U.S. Department of Energy
under Contract No. DE-AC02-83CH10093
Prepared under Task No. PVllll
October 1991
On September 16,1991, the Solar Energy Research lnstltute was designated a national laboratory, and its name was changed
to the Natlonal Renewable Energy Laboratory.
NOTICE
This report was prepared as an account of work sponsored by an agency of the United States government. Neither the United States government nor any
agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights.
Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer. or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the’U&ted States government or any agency thereof. The views and opinions of authors
axpmssed herein do not necessarily state or reflect those of the United States government or any agency thereof.
Printed in the United States of America
Available from:
NationalTechnical information Service
U.S. Department of Commerce
5285 Port Royal Road
SprIngfIeld, VA 22161
Price: Microfiche A01
Printed Copy A09
codes ar8 u.S6d for pricing all publiitions. The code is determined by the number of pages in the publiition. Information pertaining to the pricing codes
Can be found in the current issue of the following publications which are generally available in most libraries: Energy Research Abstracts (ERA); Governm8nt&porTsAnnouncements
and Index (GRAatul I); Scientificand Tech&a/Abstract
Reports (STAR); and publication NTIS-PR-360 available from NTIS
at the above address.
.
.
DISCLAIMER
Portions of this document may be iilegible
electronic image products. Images are
produced from the best available original
document.
PREFACE
This document contains preprints of papers prepared by photovoltaic scientists at the National Renewable
Energy Laboratory (NREL) and collaborating researchers for the 22nd IEEE Photovoltaic Specialists
Conference. The conference was held on October 7-10, 1991, in Las Vegas, Nevada. The NREL work
described here was funded by the U.S. Department of Energy under Contract No. DE-AC%!-83CHlOO93.
Approved for
National Renewable Energy Laboratory
Materials Science and Engineering Division
. ..
111
CONTENTS
Papers in this document are listed below according to the order in which they were presented. Late news
papers appear after the scheduled presentations. Some papers including NREL coauthors are not available
in this publication; those that are included appear after the papers with NREL primary authors.
y-i&
Pas No.
Cost-Reduction Technology for High-Efficiency Photovoltaics:
Research Issues and Progress
J.PBenner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . l-1
The U.S. DOE/NREL Polycrystalline Thin Film Photovoltaic Project
K. Zweibel, H.S. Ullal, R.L. Mitchell, and R. Noufi . . . . . . . , . . . . . . . , . . . . . . . . . . . . . . 2-l
Physical, Chemical and Structural Modifications to Thin-Film
CuInSe, Based Photovoltaic Devices
J.R. Tuttle, M. Contreras, D.S. Albin, and R. Noufi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-l
A Study of ITO/CdS/CuInGaSe, Thin Film Solar Cells
K. Ramanathan, R.G. Dhere, and T.J. Coutts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-l
Optical Characterization of CuInSe, Solar Cells Obtained
by the Selenization Method
R.G. Dhere, K. Ramanathan, and T.J. Coutts . . . . . . . . . . . . . . , . . . . . . . . . . . . . . . . . . . . 5- 1
Optical Band Gap and Lattice Constant of Cd,Zn,,SeyS,-YThin-Film
Alloys for Heterojunction Photovoltaic Cells
R, Noufi, J. Tuttle, D. Albin, M. Contreras, J. Carapclla,
A. Mason, and A. Tennant . . . . . . . . . . . . . . .:. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-1
Preparation and Characterization of Polycrystalline, r.f.
Sputtered, CdTe Thin Films for PV Application
F. Abou-Elfotouh, M. Soliman, A.E. Riad, M. Al-Jassim,
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-l
andT.Coutts
Advanced High-Efficiency Concentrator Tandem Solar Cells
M.W. Wanlass, T.J. Coutts, J.S. Ward, K.A. Emery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-l
Status of the Photovoltaic Manufacturing Technology (PVMaT) Project
C.E. Witt, L.O. Herwig, R.L. Mitchell, and G.D. Mooney . . . . . . . . . . . . . . . . . . . . . . . . . 9-1
The Effect of Microstructure and Strain in In/Cu/Mo/Glass Precursors
on CdS/CulnSe, Photovoltaic Device Fabrication by Selenization
D.Albin,J.Carapella,J.Tuttle,andR.Noufi
.,...............................
- iv
10-l
CONTENTS (Continued)
Me No.
Fundamental Research in Crystalline Silicon Photovoltaic
Materials: Program Perspective
B.L. Sopori and J.P. Benner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-I
Hydrogen in Silicon: Diffusion and Defect Passivation
B.L. Sopori, K.M. Jones, X. Deng, R. Matson, M.M. Al-Jassim,
S. Tsuo, A. Doolittle, and A. Rohatgi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-1
Hot Spot Susceptibility and Testing of PV Modules
E.C. Molenbroek, D.W. Waddington, and K.A. Emery . ‘. . . . . . . . . . . . . . . . . . . . . . . . . . 13-1
Weathering Degradation of EVA Encapsulant and the Effect of Its
Yellowing on Solar Cell Efficiency
F.J. Pem, A.W. Czandema, K.A. Emery, and R.G. Dhere . . . . . . . . . . . . . . . . . . . . . . . . . 14-1
Minority-Carrier Lifetime of Polycrystalline CdTe in
CdS/CdTe Solar Cells
R.K. Ahrenkiel, B. Keyes, L. Wang, and S. Albright . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15-1
High-Efficie:cy Heteroepitaxial InP Solar Cells
M.W. Wanlass,T.J. Coutts, J.S. Ward, andK.A.Emery
. . . . . . . . . . . . . . . . . . . . . . . . . . 16-1
ASTM Photovoltaic Standards Development Status
C.R.Ostenvald . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17-1
Design of a Fiber Optic Based Solar Simulator
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-1
B.L.SoporiandC.Marshall
Interim Qualification Tests and Procedures for Terrestrial
Photovoltaic Thin-Film Flat-Plate Modules
R. DeBlasio, L. Mrig, and D. Waddington . . . . . . . . . . , . . . . . . . . . . . . . . . . . . . . . . . . 19-1
The U.S. DOE/SERI Amorphous Silicon Photovoltaics Program
B.L. Stafford, W. Luft, B. von Roedem, R. Crandall, and W. Wallace . . . . . . . . . . . . . . . . 20-l
Effects of Helium Dilution on Glow Discharge Depositions of
a-Si,,Ge,:H Alloys
Y.S. Tsuo, Y. Xu, I. Balberg, and R.S. Crandall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-1
Pilot Production of 4 cm* ITO/InP Photovoltaic Solar Cells
T.A.Gessert,X.Li,T.J.
Coutts,andN. Tzafaras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22-1
Modeled Performance of Monolithic, 3-Terminal InP/Ga&q,,,As
Concentrator Solar Cells as a Function of Temperature and
Concentration Ratio
C.R. Osterwald, M.W. Wanlass, J.S. Ward, B.M. Keyes,
K.A. Emery, T.J. Coutts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23-1
CONTENTS (Concluded)
Title
Effect of Base Doping on Radiation Damage in GaAs Single-Junction
Solar Cells
K.A. Bertness, B.T. Cavicchi, S.R. Kurtz, J.M. Olson,
A.E. Kibbler, and C. Kramer.. . . . . . . . , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24-l
InP Concentrator Solar Cells
J.S. Ward, M.W. Wanlass, T.J. Coutts, K.A. Emery, and C.R. Osterwald . . . . . . . . . . . . . . 25-l
Back Surface Fields for GaInP, Solar Cells
D.J. Friedman, S.R. Kurtz, A.E. Kibbler, and J.M. Olson . . . . . . . . . . . . . . . . . . . . . . . . . 26-l
Controlled Light-Soaking Experiment for Amorphous Silicon Modules
W. Luft, B. von Roedem, B. Stafford, D. Waddington, and L. Mrig . . . . . . . . . . . . . . . . . 27-l
A Numerical Analysis of PV Rating Methods
K. Emery and S. Nann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28-1
.
vi
COST REDUCflON TECHNOLOGY FOR HIGH-EFFICIENCY
PHOTOVOLTAICS: RESEARCH ISSUES AND PROGRESS
J. I’. Benncr
National Renewable Energy Laboratory
(formerly the Solar Energy ResearchInstitute)
Golden, Colorado
.
arsenide(&As). indium phosphide (InP), and combinations of
thesematerials and related alloys. (These alloys are generally
called III-V alloys becausetheir constituent elementsare from
columns III and V of the periodic table.) The common feature
in all high-efficiency solar cells is the crystalline perfection of
the active semiconductor materials.
ABSTRACT
In the past two years, the long-awaited performance of
advanced III-V basedphotovoltaic (PV) technologies has been
realized. Concentrator cells in several configurations have
exceeded 30% conversion efficiency, and single-junction
devices have exceeded25% efficiency under standardone-sun
conditions. Thin-film submodules have surpassed 20%
efficiency. These accomplishmentsare due in part to NREL’s
(formerly SERI) consistent, long-term support of research to
understand and control the preparation of materials for these
cells: However, the focus of this research is now shifting.
Industry’s decision to invest in the development of production
processestests not only on its perceptions of the market, but
also on its confidence in the expected performance of the
factory. This paper will provide an overview of the
Department of Energy (DOE) PV program’s researchon III-V
solar cells directed toward establishing the baseof technology
necessary to design lower cost production processes and
improve the reliability of manufacturing projections.
While most of the program’s work is focussed on
manufacturing issues and cost reduction for lower-efficiency
technologies, the crystalline technologies’ path to high
efficiency serves two important purposes. First, high
efficiency is the program’s hedge should the targets for the
balance-of-system(BOS) costs prove unattainable. Second,it
is generally accepted that utilities seek high-efficiency
technologies in all power generation options and will therefore
prefer, possibly even demand,high-efficiency modulesfor their
PV applications.
In megawatt-scale PV systems, high-efficiency PV
modules can command a premium price. This tends to create
some confusion. It is convenient to assume that the
higher-efficiency modules must necessarily be significantly
more expensive to produce. In this paper, we shall examine
the possibility that known prc~~ssesthat have achieved highly
efftcient PV devices need not be significantly more expensive
than their lower-efficiency counterparts. Through this
examination, the critical researchareasand programobjectives
of NREL’s Crystalline Materials and Advanced Concepts
Program will becomeevident.
INTRODUCTION
A key strategy within DOE’s National Photovoltaics
Program Five Year Plan (1) is to minimize the risk of program
failure by maintaining, despite budget fluctuations and other
influences, multiple paths to the ultimate goal of providing
cost-competitive solar-generatedelectric power. As a result,
the targets for module performance are very broad. Within
this decade, technology improvements are intended to yield
modules of 10%~20%efficiency with the cost, stability, and
other characteristics needed to build systems generating
electricity for less than $0.2O/kWhr. This target range
encompassesall current PV technologies. from amorphousand
polycrystalline thin films. through crystalline silicon, and
concentrator modules. With a little more time, ongoing
research is expected to increase the upper end of the range to
better than 25% efftciency.
HIGH EFFICIENCY,
THE DlRECT
LOWER COST
PATH TO
Them are a number of models for growth in PV industry,
applications and annual sales.The ultimate goal of all of these
is to reach “energy significance”, a point at which them are
many gigawatts of PV generating capacity with much of it
located in large installations. All analysesof large installations
point out that high-efficiency modules can support a much
higher cost. The module cost per square meter of a 20%
efficient module can obviously be at least twice that of a 10%
efficient module and still produce the sameprice per watt. In
large systems,the high-efficiency module carries an additional
premium value becausethe samefield can produce much more.
energy. The resulting savings in BOS cost add directly to the
allowable module cost. A simplified expression, derived from
DeMeo’s equation (2). is shown as follows:
Experimental and theoretical evidence indicates that
crystalline PV materials will be needed for advanced solar
celts designed to reach the upper end of the efficiency target
range. Many types of solar cells have demonstrated
high-efficiency performance, namely flat-plate cells of at least
20% efftciency and concentrator cells of 30% or higher
efficiency. These have been fabricated in silicon, gallium
l-l
COE =
tc,+qJ
-------_-Efficiency
known that the cost of the active layer of semiconductor
needed for effective solar cells is less than 10% of the cost
target for any thin-film module, including GaAs (4).
Crystalline GaAs technology has demonstrated thin-film,
submodule efficiencies of more than 20% (5); still, the
raw-materials costs are not significantly higher than those
projectedfor other thin-film options and are far less than those
encountered in wafered crystalline silicon modules. This
technology, called cleavage of lateral epitaxial films for
transfer (CLEW, also demonstratesthat relatively large areas
of single-crystalline, thin-ftlm semiconductorscan be applied
to glass and that cells incorporating these films can be
interconnected with module-level cell interconnection
processing, also a key cost-reduction step for thin-film
modules relative to wafered silicon photovoltaic technology.
(l+i)FCR
-------- + (~.Ol(,
S.
where
COE= Cost of electricity in $/kWh
C,= PV module cost per unit area
C,= BOS cost per unit area
i= indirect capital cost factor
FCR= Fixed charge rate
S= Annual incident energy
Efficiency is the systemefiiciency, including not only the
module performanceunder standardconditions but also module
heating, wiring, and mismatch losses. The value of 0.016 is
calculated from assumed costs of power conditioning,
operation, and maintenance.
The challenge is to develop technologies that can produce
crystalline films without consuming a crystalline GaAs wafer.
Heteroepitaxy of GaAs on germanium is an important first
step. Solar cells in this system routinely yield efficiencies as
high as those of the best homoepitaxial devices. Many
methodshave been identified for preparing thin, single-crystal
semiconductor films on glass (6,7,8,9). We also have
established that high-efficiency does not absolutely Rquire
defect-free single-crystal materials. Experiments in specially
designedGaAs sampleshaving controlled dislocation densities
in the range from ld crns2 to more than IO8 cmm2
demonstratedvery little loss in cell performancefor materials
better than 16 cms2(lo), as shown in Figure 1. Continuing
progressin GaAs/Si heteroepitaxial technology may reach the
needed quality level.
With advances in zone melt
recrystallization (ZMR) silicon, near-single-crystal, <loo>
oriented films can be prepared by lateral growth without
seeding (11). Such films might eventually serve as the low
cost substrate for high-efficiency, thin-film silicon or GaAs
modules.
One of the first observationsstemming from thesesystems
studies is that the value of high-efficiency modules is resilient
under changing economic assumptions. Table 1 shows
changesin cost/performancetargets(calculated by the Electric
Power ResearchInstitute) that have occufied just in the last
year.
Another observation is that the area-relatedBOS cost used
to estimate the allowable module costs in Table 1 is $50/m2.
This value is much less than the BOS costs today. Thus,
while $50/m2 is a widely excepted target, it is a target
nonetheless. If area-related BOS costs actually only fall to
$70/m’. then another $20/m2 must be shavedoff the allowable
module costs.
“High efficiency is the most direct path to lower cost” was
one of the major conclusions of DOE’s Energy Research
Advisory Board several years ago. Actually achieving high
efficiency in experimental devices is the first stepon that path.
Table 2 lists the highest-efficiency, compound semiconductor
solar cells measured by NREL or Sandia National
Laboratories.
REDUCING
OF SUBSTRATE
Several theoretical analyses (12,13) indicate that even
semi-crystalline materials can reach efficiencies in excess of
20% provided the density of mid-gap electronic statesis held
to less than 1013cmJ. This correspondsto a grain diameter
of about 1 cm assuming high-quality intragrain properties and
COSTS
High-efficiency PV like all other approaches, needs
significant advances in cost-reducing technologies for
production. The basecost of any product is establishedby the
value of the materials used to produce it. It has long been
Table 1. Cost/PerformanceTargets For Fixed Flat Plate
Systemsat $O.O8/kWhCOE. 1989dollars. Southwest U.S. site
‘Module
Efticiency
(%I
10
15
20
25
May 1990 (3) April 1991 (2)
(%lm2)
40
85
130
175
Wm2)
27
66
105
143
103
10)
5
6
7
108
LXsldbtion deo,sity (:Om-“)
Figure 1 GaAs Cell Efficiency versus Dislocation Density
l-2
109
Table 2. High-Efficiency, Cornpound Semiconductor
Cell Type
Awn V,,
Solar
J,,
Cells*
FF Efficiency Source
(cm2) (mV) (mh/cm2) (%)
ml
4
16
4
1
1011 27.6
83.8
4034 6.6
79.6
four rerminnl
822 19.7
62.2
23.3
21.0
25.8
10.1
Kopin
Kopin
Boeing/Kopin
S. Chu( 1985)
4
4
0.2
0.3
4
1022
1035
1038
891
25.1
24.3
28.7
25.5
77.7
878
29.3
85.4
17.6
21.9
Kopin
ASEC
SERI
Spire
Spire
28.7**
27.5
31.8
27.5
27.6
34.2,’
31.0*+
30.2
Spire
SERI
SERI
SERI
Varian
Boeing
Sandia
SERI
Thin Film:
GaAs cell
GaAs submodule
Gahs stackedon CuInSe2
poly-GaAs/poly-Ge
Epitaxial
Single Junction:
GaAs/ with AlGahs window
GaAs/Ge with AlGaAs window
GaAs/GaAs with GaInP window
GaAs/Si
InP/lnP homojunction
Concentrator
28.2
27.6
87.1
85.3
86.4
25.7
Cells:**
single-junction Gabs
200
single-junction GaInAsP (1.l S eV) 171 899
three terminal
monolithic InP/GaInAs (0.75 eV)
50
monolithic GaInP/GaAs
1 2292 13.6 182.5
monolithic AIGaAs/GaAs
1 2403 13.96 83.4
100
stackedGaAs on GaSb
stackedGaAs on Si
500
stackedGaAs on GaInAsP (0.95eV) 40
:* V, = open-circuit voltage, J,, = short-circuit current, FF = fill factor
Re-calibration at Sandia will likely result in down-grading of theseefficiencies by 4 %
Of course,this minimum entry level of production corresponds
to enough cells to supply more than 50 MW of concentrator
modules annually.
no grain boundary passivation. Each order-of-magnitude
reduction in grain boundary recombination center density
achieved by passivation treatment reduces the grain diameter
requirement by the sameamount. Even with the advancesin
hydrogen passivation in silicon and sulphide surface
passivation in III-V materials, grain diameters of at least 100
pm appearto be the minimum acceptabletarget for materials
with crystalline properties.
REDUCING
CELL FABRICATION
Simplifying the device structures needed to achieve high
efficiency is also an important area for development. A major
drawback to the mechanically stacked cells is that most
production stepsmust be repeatedfor each of the cells in the
stack. Arguments have been that overall yields will be
improved because only the subcells passing electrical
inspection will proceed on to final assembly, However, this
seemsto imply that the yield in the subcell lines must be too
low lo be viable in cost effective manufacturing. The
monolithic cascade devices represent a significant
simplification in device design needed for manufacturability.
The epitaxial growth step is more complex than that needed
for single-junction, but the balance of the process is nearly
identical. Advanced concepts for further improving the
performance of single-junction cells may. lead to further
simplification with no performance penalty.
COST
One of the first terrestrial products using high-efficiency
technology is likely to be ultra-high-efficiency concentrator
cells. It has been shown that advancing the technology to
larger wafer sizes, while maintaining a wafer throughput of
more than 1000 per week, results in a significant reduction in
production cost (14). Costs are added for each piece going
through each processing step. Many of the steps are
performed at the wafer level. Larger wafers yield a larger
divisor, the number of solar cells per-wafer, to reduce the
per-cell cost of all of the processing steps before the wafers
are diced. At this scale of production, 30% efficient GaAs
cells for operation in 1000x concentratorscould cost less than
half the target price (15) neededto reach DOE mid-term goals.
Thin cell designs, using back-surfaceoptical reflection OT
optical confinement, are in initial stagesof evaluation for III-V
solar cells. As in advanced silicon cell designs, if all other
loss mechanismhave been minimized, reducing the volume of
l-3
the cell to reduce bulk recombination can give higher open
circuit voltages and efficiencies. The reflector can bc madeby
growing alternating layers of semiconductors with different
indices of refraction, known as a Bragg reflector (16.17). Or,
in thin films such as CLEW, a simple coating is all that is
needed. Spire has already obtained more than a 20.tnV
increase in open-circuit voltage by using a Bragg reflector in
a l-pm-thick GaAs cell. A more speculative, but potentially
more substantial, improvement might be realized through the
nature of the recombination event in high quality GaAs.
Recombinationis not necessarilya toss in GaAs, provided that
the material quality is sufficient that radiative recombination
is the dominant mechanism. If the photon emitted by this
process can be confined and re-absorbed, a process called
photon recycling, then the effective minority carrier lifetime of
the material becomesmuch longer. Photon recycling might
open a new range for high- efficiency device performance.
A final area for device development is in designing
structures that retain high- efficiency performance when
produced in less-than-ideal material. The thin structures
possessthis feature to someextent. Other structuresare under
evaluation that may be even more tolerant (18).
REDUCING
FI!,M DEI’OSITION
COST
The largest component of the production cost of
high-efficiency solar cells today is the cost of epitaxy. There
are many reasons for this. The largest MOCVD reactor is a
batch systemcapable of coating substrateswith a total areaof
nearly a third of a square meter. This is more than adequate
for large-scale concentrator cell manufacturing, but a
continuous deposition systemwit1 eventually be neededif costeffective flat-plate modules are to be produced. Eleven dollars
buys only one gram of trimethylgallium (TMGa) containing
about 25 cents worth of pure gallium. If demand was
measuredin metric tons, insteadof kilograms as neededtoday,
suppliers indicate that TMGa cost could drop to 70 cents per
gram. Equipment and expenses needed to provide a safe
production environment for the use of large quantities of
compressed toxic gases and pyrophoric sources add to the
burden. Large-scale PV plants may need to generate the
reactants on site to improve safety, minimize material costs,
and potentially benefit from improved control of epitaxy.
These needs are being addressedby the research program.
Colorado State University, for example, is investigating the
remote-plasmagenerationof arsine and arsine radicals and the
use of these species for GaAs crystal growth (19). As an
added benefit of this research,they have obtained epitaxy at
growth temperaturesunder 3OO“C,which may be of use for
heteroepitaxy on siticon or other substrates with a large
mismatch in their thermal expansion coefficients.
Ultimately, for large-area crystalline thin films, chemical
vapor deposition (CVD) apparatus such as that used in
amorphous silicon production may be needed for III-V
materials. Perhapsmuch of the substratemanipulation and gas
handling technology could be adapted to III-V production if
the program has provided an adequate technology base of
reaction chemistry, surface interactions and other basic
mechanisms. It is important to recognize that CVD is a
current PV production process. It can be cost-effective for
future PV products as well. Table 3 provides a summary of
CVD costs estimated using the Solar Array Manufacturing
Industry Costing Standards (SAMICS) and Interim Price
Estimation Guidelines (IPEG) techniques developed for the
DOE program a decadeago (20). While the absolutenumbers
are open to debate,the major difference is that the consumable
sourcescost more for thin-film GaAs than for silicon.
CONCLUSIONS
The approachin high-efficiency researchwill continue to
be to develop technologies needed for the cost-effective
production of crystalline-film-based products. White the vision
of the project is focussedon advanced flat-ptate modules, at
any time the market warrants corporate entry, much of the
technology is directly
applicable to producing
ultra-high-efficiency concentratorcells. Cost reduction in the
high-efficiency technologies requires a three-prongedattack.
First;increasing the efficiency is still the most direct path to
lowering the cost of PV modules provided that the additional
device sophistication doesn’t add cost. The focus of
high-efficiency cell research will be to simplify the &vice
structures and to design structures which a more tolerant of
less-than-idealmaterials. The secondprong is to improve the
technologiesfor crystal growth. Production cost factorsrelated
to large-area uniformity, yield, process safety and source
utilization efficiency are. in fact, inseparable from the more
esoteric scientific issues such as thermal geometry, reaction
chemistry, crystallographic orientation effects, spinodal
decomposition, and interdiffusion. Finally, an improved
technology and new conceptsare neededto eliminate the cost
of consuming a crystalline substrate. Greatersophistication in
crystal growth than is apptied today underlies all threeof these
paths.
This work was performed under Contract No.
DE-ACO2-83CH10093to the U. S. Departmentof Energy
Table 3. IPEG/SAMICS Analysis of Chemical Vapor’
deposition Costs* ($/m2 estimate for 1 p deposited
thickness, 2x106 m2/yr production rate)
PECVD
a-S
MOCVD
TCO
MOCVD
CaAs
20
40
40
Equipment
Factory area
Labor
Consumables
Utility
2.76
0.15
0.26
0.44
1.50
1.57
0.11
0.17
0.28
0.20
1.57
0.11
0.17
12.86
0.20
Total cost
5.11
2.32
14.90
Growth rate (as)
l
PECVD=plasma-enhanced
CVD, MOCVD = MetalorganicCVD
REFERENCES
7.
8.
9.
U.S. Department of Energy, May 1987, National
Photovoltaics Promam Five Year ResearchPlan
1987-1991,DOE/CH10093-7,Washington, D.C.
E. A, DeMeo, Proceedingsof the Tenth Photovoltaic
Solar Energy Conference (Kluwer, Dordrecht, 1991).
p. 1269.
E. A. DeMeo and others, Conference Record of the
21st IEEE Photovoltaic Specialists Cotiference. (Il:El!,
New York, 1990), p. 16.
J. Stone and others, Conference Record of the 171h
IEEE Pbotovoltaic Specialists Conference (IFEE, New
York, 1984), p. 1178.
R. W. McClelland, et al., Conference Record of the
21st IEEE Photovoltaic Specialists Conference. (1FF.E.
New York, 1990), p. 168.
C. S.Bax and others, Comparison of Thin Film
Transistor and SOI Technologies, H. W. Lam and M.
J. Thompson (Ed.), Materials ResearchSociety
Symposia Proceeding Vol 33,(Elsevier, New
York,1984),p. 215.
McClelland, R. W., C. 0. Botler and J. C. C. Fan
(1980), Appl. Phys. I&t. 37(6), 560.
Smith, H. I.. M. W. Geis, C. V. Thompson and 11.A.
Atwater (1983), J. Crystal Growth 63, 527-546.
10.
1-l.
12.
13.
14.
15.
16.
17.
18.
19.
20.
l-5
Yablonovitch, E., T. Gmitter, J. P. Harbison and R.
Bhat (1987) Appl. Phys. Lett. 51(26). 2222.
S. P. Tobin, Proceedingsof the Fourth International
Photovoltaic Science and engineering Conference,
(1989) p.47.
J. A. Knapp, L. R. Thompson, G. J. Collins, (1990) J.
Mater. Res.,Vol.S, No. 5. p.998.
A. K. Ghosh and otherq(l980) J. Appl. Phys., 51(I),
p.446.
M. Yamaguchi and Y. Itoh, (1986)J. hppl. Phys.,
60( 1). p.413.
J. P. Benner and others, Proceedingsof the Biennial
Congressof the International Solar Energy Society
Vol. 1 (Permagon,New York, 1991). p. 46.
Chamberlin, J. L., and D. L.King, Conference Record
of the 21st IEEE Photovoltaic Specialists Conference.
j
(IEEE, New York, 1990), p. 870.
P. D. Dapkus private communication, 1987.
S. P. Tobin, S.-M. Vernon. M. M. Sanfacon, A.
Mastrovito, this conference record.
S. M. Durbin and J. L. Gray, this conference record.
B. G. Pihlstrom, L. R. Thompson, G. J. Collins,(l991),
Solar Cells, Vol. 30, p. 415.
Jackson, B. and others (1983). Advanced Photovoltaic
Module Costing Manual, SERI/TR-214-1965.
DE84G00029.
THE U.S. DOE’NREL POLYCRYSTALLME THIN FILM PHOTOVOLTAICS PROJECT
Kenneth Zweibel. Harin S. Ullal,
L. Witchell. Rommel Noufi
Richard
National RenewableEnergy Laboratory
Golden, CO 80401,USA
Tei: (303) 231 7141. Fax: (303) 231 1030
ABSTRACT
Theoreticalefficiencies are 23.5% for CIS and 275% for CdTe (1).
In a similar way, module stability also appearsto be excellent (se-e
results below), with two SiemensSolar IndustriesCIS modules.for
example,measured at 99.8% and 97.5% of their original efficiencies
after almost tOO0days of outdoor testing at NREL.
During the last eighteen months, pmgress in polycrystalline thin
films facilitated by the U.S. Department of Energy/National
Renewable Energy Laboratory (U.S. DOE&XL) program has
included (1) acceleratedgroti.of the U.S. industrial infrasuucture
suppotting CuInSq (CIS), CdTe and Si-film; (2) achievementof a
record thin-film CIS power module (4 ft’t apertureareaefficiency
of 9.7% verified by NREL: (3) improved apertureareaefficiency of
8.1% verified by NFUZLfor a thin Nm CdTe module of areanear
1 ft$ (4) progressin improved CIS/Mo adhesion:(5) a breaktbmugh
total area thin filq CdTe solar cell efficiency of 13.4%verified by
NOEL; and (6) continuedsuccessof multiyear outdoorstability tests
of prototype CIS and CdTe modules. These and other technical
resultsare summarizedin this paper.In addition. *wewill describe a
U.S. DOE/NREL-hmded competition for university research
subcotmacts designed to strengthen the technical base of
polycrystallIinethin flllns.
16
U.S. OOE Module Eilictency Goal I 15%1
14 212
5 ARC0 Solar (CIS)
-
*
0 Solarex(a-Si. not light-soaked)
c] Photon Energy (CdTe)
$10
.a,
&’ 8
iTI
2 6
G
24
2
-l980
1985
1990
Calendar Year
INTRODUCTION
Notes: All mcuules 1000 cm2 areaor more: aSi elliclenaes pmr 10llgnt-mauoeadegraaaton
All awnwe areas
Figure 1
Substantialtechnicalprogresshasbeenmadeduring the last eighteen
monthsin the researchand developmentof polycrystalline thin film
solar cells and solar power modules, facilitated by the U.S.
DOE/NREL program.The major technical achievementsinclude ( 1)
acceleratedgrowth of rhe U.S. industrial infrastructure supporting
CuInSe, (CIS). CdTe xtd Si-film; (2) achievementof a record thinfilm CIS power module (4 fi?) aperture area efficiency of 9.7%
verified by NREL; (3) improved apemrreareaefficiency of 8.1%
verified by NREL for a thin film CdTe moduleof areanear 1 ft$ (4)
pqress in improvedCiS/?vfoadhesion:(5) a breakthroughtotal area
thin film CdTe solar cell efficiency of 13.4%verified by NREL; and
(6) continuedsuccessof multiyearoutdoor stability testsof prototype
CIS and CdTe modules. These remarkable technical results are
summarized in this paper. Additionally. we will describe a U.S.
DOWEL-funded competition for university researchsubconnacts
designedto sttengthenthe technical baseof polycrystalline thin film
projects.
Progress
in thin film photovoltaic module
efficiencies
However, the strengths of the polycrystalllne thin films remain
somewhatobscure becauseprogressin the laboratory and at the
prototypemodule level still needsto be translatedinto successwith
manufactured products. We will complete our paper with a
discussionof the status and issues associatedwith this necessary
uansition.
U.S. INFRASTRUCTURE
Progress in developing CIS. CdTe and Si-films continuesto be very
stung. ~Manyof the technical goals (module efficiency, stability,
cost) associatedwith attaining truly low cost PV (under 6 cents&WI?
system cost) appearto be achievable by these technologiesgiven
existing researchtrends.Figure 1 shows the relative progressof the
thin films towards high-efficiency, 1-e modules. Such progress
should continue, since cell efficiencies are still rising. and the
ultimate pmcticul efficiencies of these materials are 16%-18%.
2-l
Technical achievementsin the late 1980ssuch as the 11% l-h! CIS
module (2) made by the then r\RCO Solar (now Siemens Solar
Industries)led to an increasein interestin polycrystailine thin films.
Oneof the critical barriers to the progress of thesetechnologieswas
the narmwnessof their industrial base.In CIS. only the then ARC0
SolarmadeCIS modulesand few other corporateentities wereeven
capableof making efficient CIS cells (i.e., only International Solar
EIectric Technology (ISET) and Boeing Aerospace).The situation
was similar for CdTe: one company was ending its participation
(AMETEK) and another, Photon Energy, was extremely small and
resource-timited.In the case oi CdTe. but Mt CIS. there were nonU.S.companiesof significance: BP Solar in England and Matsushita
Battery in Japan.In the caseof Si-fii, there was only one small
company with a serious commitment. AstroPower. Thus, a major
focus of increased U.S. DOE/NREL funding to support
polycrystaUine thin films was to strengthen the corporate
infras~cnUe Of thesetechnologies.This was doneby a competitive
procurement during 1990. mainly through adding hinds to
underfunded. resource-limited efforts such as those at ISET,
AstmPower and Photon Energy, and by allowing the entry of new
participants such as Solar Cells Inc. (CdTe), Solarex (CIS) and
Martin &Marietta(CIS). Table 1 providesa summaryof t,heindustrial
“parmers”tit U.S.DOB/NRELis funding to acceleratethe progress
in polycrystallIne thin IiIm modules.as weI3 as the goals of their
three-yearsubcontracts.
areas.However, from the standpointof overaIl technical progress,it
is hard to argue charin most ways the SIemensSolar CIS power
moduie is the most significant achievementin thin fiIm PV. No
other moduleof anything like its size comescloseto ir in efficiency.
Figure 2 is the NREL-measuredI-V curve of this module,showing
its apemne area (3883 cm*) efficiency of 9.7%. The relevant
parameterswere: maximum power,37.7W; ocen-circuit voltaPe.24
v; short-circuit current. 2.44 amps:and fill f&, 0.644. -
3.01
TABLE 1
Module Development Goals for
Industrial Partners
iSET
Siamens Sofar
Solarex
Martin Manetta
ClJltlSc2
Cu. In SounerlngSelenlratlon. 119’0ettnxncy- 900 cm’
Cu. In Sounenn~Selen~zat~on. t2.N eftiuency-3900 cm2
Cu. In. Se Elemental Sourtermg. 12% efficiency-900 cm:
Rotatmg Cylindrical Magnetron Spultermg. 8% etficlency-900 cm’
Photon Energy
Solar Cells Inc.
Sotaymg. :2.5% etliaency-3900 cm2
Close Spaced Subbmatlon. 10% elfrlency-i200
AstroPower
Silicon Films
Thtn S&con 6lms on Ceramic Subsrrales.
12?6 effrlency-1200 cm-
ceTe
Cd
P
sE 1.0
a
k
= 0.5
0
The government/industrypannershipsof Table 1 can be viewed as
complementaryto presentor potential U.S. DOE/NREL-fundedPV
Manufacturing Initiative subcomractparmerships.The subcontracts
of Table 1 ail concernthe developmentof successfulprototyperhin
fti modules.Issuesof moduledesignandefficiency. and prototype
processes,are the focus. In contrast.PV Manufacturing Intitiative
parmershipsareintendedto takePV manufacturingtechnologiesrhat
are successfulaf the prototype level and optimize them for lowest
cost.
The annual U.S. DOE/NREL furding of the subcontncts in
Table 1 is about S42OOK.with another 01900K cost-share
contribution from the companies.Thus rhe totai invesunentover the
three-yearsubcontmctswill be about $18 million. In addition, we
have one other industrial subcontract.with Martin Marietta. Their
goals are to investigaterotating cylindrical magnetrondeposirionof
copperand indium (for CIS) andcadmiumandtellurium (for CdTe).
as well as some work in CdTe elecuodeposition.
-0.5 I
-5
Area = 3883 cm2
ISC= 2.44 Amp
Voc = 24 Volts
FF = 0.644
6
^,
’
I
I
I
I
5
10
15
20
:
Voltage (V)
Figure 2
Light I-V characteristics of a Siemens Solar
CIS power module
PHOTON ENERGY CdTe -MODULE
,Mostrecently. NREL has verified a Photon Energy CdTe module
with an apermreareaefficiency of 8.1%. The module size is 832
cm’ and was measuredoutdoors at NREL’s tesf facilities. The
insolation at the time of measurementwas 1018 W/m’ and
temperature33 ‘C. The I-V curve of this 8.1% efficient module is
Until this fiscal year, we were not able to addressthe infrastructural
weakness of polycrystalllle thin film research at universities.
However. we have under way a competition for university
subcontractssupporting CdTe and CIS research.We expectto fund
about six subcomractsat the $lOOKlevel each.In addition, we have
tie ongoing university subcontracts(totaling about %830K) at
Colorado State U&e&y.
Institute of Energy Conversion
(University of Delaware) and University of Toledo. Thus the
breakdownof our funds in FYI992 will be about 75% for industry
and 25% for universities. In addition. we fund an in-house ‘NREL
effort of about S33COKannually.
shown in Figure 3. The module parameterswere: maximumpower,
6.8 W: open-circuit voltage, 21 V; short-circuit cunmt, 0.59 amp,
and fill factor, 0.54.
IMPROVED CIS/Mo ADHESION (ISET)
Inremational Solar Electric Technology(ISlX) makesefficient CIS
solar cells (115%. total area verified by i4REU by selenizationof
Dunn precursorfilms. Analysis by Sites suggeststhat the junction
properriesof the ISET CIS cells (in termsof diode quality factors.
recombiition currents) are as good or bette.rthan any other ClS
devices and comparableto those needed for 15% cells (3). Like
others involved in CIS fabrication especially rhose who use
selenization.ISET hasencounteredadhesionproblemsat its CIS/Mo
interface.In December1990.ISET appliedfor a Europeanpatenton
an innovative solution to the adhesionprobiem:the depositionof a
very thin telhuium layer (10-500angstmms)betweenthe MOandthe
CIS. ISET claims in its patentsthat this layer reduces adhesion
problems significantly and also tmtlu in a device with improved
SIE,MENS SOLAR CIS POWER kt0DlJ-N
Of the numerousadvancesof the last eighteenmonths,perhapsthe
most significant was the achievementof a near-lo% encapsulated.
CIS power module by SiemensSolar Industries. Becausethin film
modules come in many sizes. it is somewhatdifficult to develop
perspectiveabout the relative imponanceof their efficiencies and
2-2
1.2
GlasJTOlCdSlCdTe P
Effect of CdSThickness $
a
Glass/TO/CdS/CdTe/C/Metal
-z
0.48
5
= 0.32
a
t
3
0.16 .
0
0.2 ’
v
FF = 0.54
Eff = 8.1%
,.A.-.
0.01.1.....
0.3
0.4
8
12
16
Voltage (Volts)
20
:
Light I-V characteristics of a Photon Energy
CdTe module
perfbmmce (4). By adding the Te layer, ISET was able to increase
the flexibility with which they addedsubsequentCu. In (and Ga. if
desixui) layers. For example, they prefer to deposit ln and then Cu.
the revelseof the previous conventional order (5). They believe that
the pmsenceof a Te layer allows the better depositionof iridium in
terms of avoiding “balling up” of the indium (which degradesthe
uniformity of the final film). The piogmss at ISET ln finding a
wotitable solution to the CIS/Mo adhesion problem. and the
subsequentimprovement in their cell efficiencies. is a significant
conuibutlon to the progressof ClS technology.
Figure 3
:*
-* .- I
0.5
1
0.6
I
0.7
I
0.8
0.9
I
1.0
Wavelength(pm)
Quamum efficiency plots of thin film CdTe
solar ecus
Figure 4
4
.L- 19.5mAlcm’
The Chu msult was achieved using an innovative solution-growth
CdS depositionprocess.Chu producedhigh-voltage cells using this
solution-growth method with his own Cd’Te(madeby close-spaced
sublimation) as well as with Photon Energy CdTe materials(up to
890 mV). The consistencyof thesehigh voltagessuggeststhat much
of the improvedefficiency can be attributedto Chu’s solution-grown
C&3 mther than the CdTe. Coniinnation of this requires further
verification. Chu’s solution-grown CdS is not yet optimixed for
omical thinness;this will be a focus in upcoming months.with the
aim of achieving total amasolar cell effi&enciesover 15%.
Gtass/TO/CdS/CdTe/ClMetaI
7+-y
ADVANCES I& CdTe CELL EFFICIENCIES: THE CHU
BRRAKTHROUGH
.
Several tecorclCdTe cell efficiencies were reportedduring the last
year and a half (3). In somecases,however. rhesereporrswere not
validated by independent measurements.In specific cases, for
instance. questions concerning excessive current densities suggest
that the reported efficiencies were lnaccumte. On the other hand,
NREL did measurecells from Photon Energy and then from T. L.
Chu at the University of South Rotida that surpassedpreviously
verified standards.In May 1991. Photon Energy produceda small
ama (0.3 cm’, cell with exceptionally high current density (26.2
mA/cn+). This cuirent density, which is about9lX of the theoretical
maximum for CdTe, was achieved by using an optically thin CdS
htyek The Eeli had a quamum &kiency 6180% at 400 mn
wavelength(Figum4), a~trbt@ndic&ion Witmuch ofthe improved
cment came‘fromreducing the absorptio?rof‘photonswith energies
above‘ihz Cds band gap.
.
Area
Jsc
Voc
.
FF
Eff
.
=
=
=
=
=
1.2 cm2
21.93 mAkm2
0.84 Volt
0.7264
13.4%
i
I
!
ThW weelm‘later. T. ‘L. Chu (a s&tier NT&L subcontractorof
&6fdn .Energy at University of South Florida) substantially
sulpasseaUrePhoton Energy result. ahd &I a way that suggestschat
Ii&her progms a&l be reladvebyeasy. Cttu achieved 13.4% cell
e@lckncy (total-atea= I .2cm,z~Flgute5) verified -byM&L. but did
so by +each@ very high voltages and fill f-m. ln his recordcell,
the voltage was 840 mV. and the till factor was 72.6%. h another
Chu cell (12.6% efficient), the fill factor was 74.6%. the highest
kmxvn CdTe fiu factor. Yet the curmnt densitiesin thesecells were
moderate (about 20-22 mA/cm*), leaving plenty of room for
improvement.
0.0
1
I
I
0.2 0.4 0.6
Voltage (Volt)
I
0.8
1.0
Light I-V characteristics of a high efficiency
thin film CdTe solar cell
OUTDOOR TESTS AND STABILITY ISSUES
Figure 5
Although intrinsic device stability appears good with aU
polycrystallii thin films. there are issuesat the module level rhat
require attention as these technologiesmove into me market. For
instance,CIS and CdTe are sensitive to chemicals used in sealing
modules,so various approachesneedto bedevelopedfor minimizing
chemicalinteractions.CdTe cells andmodulesam sensitiveto water
2-3
vapor, so careful sealing is a necessity.Issues with CIS modules
tend to be associatedwith specific layers: e.g., undesirableMoSe
layersthat impedecurmntflow betweeninterconnectedZnO andMO
contacts and defectsand adhesionissuesassociatedwith &MOand
US interfaces.Thus the achievementof a 30-year life for these
technoIogies appearsto require attention to specific processing
details as well as careful module design and sealing.
OTHER HIGHLIGHTS
Other key advancessponsoredby the U.S. DOENREL ptognun
during the last eighteenmondtsinclude progressat AstroPower (SiF&n), Georgia Institute of Technology (CdTe celI MOCVD),
University of South Florida (CdTe cell IMOCVD). and in-house
To this date, the results of outdoor tests on Siemens Solar CIS
modules are extremely promising. NRBL has testedonly two CIS
modules,but both have beenoutdoorsfor almostthree years.Based
on our most recent measurement.the two modules are producing
power at 99.8% and 97.5% of dteir original levels (Figure 6,
reference 6). These measurementswen taken outdoors at near
standardconditions, which accountsfor someof the scatterin the
data. The aperture-amaeffiiencles of the modules were near 8%.
Thus it appearsthat the intrinsic stability of thesemodulesls not an
issue.For CIS. any remainingstability issuesare likely to be process
and design specific.
10.0
CIS Modules
AstroPowet’sPmductI is currently in the manufacturingstagewith
a capacity of 0.5 MW for FYI991 with plans to expand it to 2-3
Mw in FY1992. In the U.S. DOEJNREL subcontract program.
-Power
ls developing elemenmof their product II and Pmduct
III technology. EssentialIy in product II,- a metallutglcal barrier
depositedon the low-cost conducting ceramic substratesb being
developed. This also semesas an optical mtlector to enhancethe
shott-circuit curtent and improve * cell efficiency. In addition+the
thicknessof the Si-film will be reduced to less than 50 microns.
product III will be a monolithlcalIy integratedmodule fabricatedon
an insulating ceramicsubstrate. The module size is 1200cm’ and
is shown in Figure 8. Qassical interconnectschemesusedin other
polycrystalllne thin film technology an being used in the
developmentof Product III. For more information on AstmPower,
seefor example,reference7.
26
‘I
7.01
0
loo
200
300
Outdoor
400
500
Exposure
600
700
800
9oa
- -Efficiency
1
-
2
Efficiency
- - Voltage
1
*Voltage
2
19
(Oays)
Stability performance of CIS modules tested
outdoors at SERI under load and open-circuit
conditions
Figure 6
‘Ilk? outdoor test results for Photon Energy CdTe mod&s are
promising but some issuesremain. As can be seen from Figure 7
(outdoor tests, near standardconditions), several Photon Energy
modules have shown good stability over reasonably long periods
outdoors. However, other modules (not shown) have degraded
somewhat Photon Energy believes that the instability of rhses
modulesis dependenton the encapsulationandedge-sealingscheme.
They have tried several of thesedesignsduring the coumeof their
testing. Thus it may be expectedthat Photon Energy moduleswiU
improve as theseissuesare dealt with successfully.
I:ir
CdTe
.
5.oc
Monolithically integrated Si-Film on insulating
ceramic substrate
Georgia Institute of Technology (GIT) and the Univenity of Soudt
Florida (USF) both have improvedthe performanceof their rhin film
CdTe solar cells deposited by the metal organic chemical vapor
deposition(MOCVD) technique. Total areaefficiencies in the range
of lo%- 11%have beenverified by NREL. Cells wem subjectedto
the CdC& chemicaltreatmentsand heat ueatmentsof 42OV for 20
minutes. Further improvements ate expected by reducing the
thicknessof the CdS film and improving the contacts to the high
resisdvity CdTe absorberlayers.
Figure 8
Modules
lg
n-layer
p-layer
1
I
--9466
.
--6711.m98981
- -6711-3la980
g 4.5
m
2
l
z 4.0 t
-
3-~------A
Figure 7
*178/9172
-2OW9173
3.5 1
a90 9
.
l
AA961191
74
Institute of Energy Conversion(IEC) at the Univemity of Delaware
has reported fabricating 10% small-atea CIS devices by the
selenization method using hydrogen selenide. Also, 7% efficient
smaIl-amadevices have been reported by depositing Cu. In. Se
layers and heat treating in excess Se aanosphen. From their
10 11 12’9Oli91 2
3 4
O~taoor Exposure (Monlns)
Stability performance of CdTe modules
tested outdoors at SERI
2-4
modeling studies,IEC has concludedthat the open-circuitvoltage of
CIS SO~U&IS C-t
be solely describedby a Shockiey-Read-Hall
(SRHI recombination mechanism.ln the caseof CdTe solar cells,
the device operatesas a p-n heterojunctlonwith curtent dominated
by SRH recombination ln the junction @on of the Cdl’e device.
IEC has also concluded that the 2nTe:Cu contacts ln an n-i-p
stmcture are the mon stable than either Au or C&Au contacts.
The emphasis of the in-house CIS program in the last year has
been on understanding the phase behavior and mictosuucture
of the Cu/In pmcursor used for selenixadon in order to learn
how to enhance the quality of CIS films for optimum device
efficiency.
In selenization, fabrication of a precursor structum containing
mainly Cu and In deposited onto a MO-coated substrate
precedes the actual selenixation step. A correlation of the
Cdb precursor microstructure with the post-selenized CIS
film and device characteristics is then very desirable. To this
end, a diffusion-controlled mechanistic model of alloy
formation for thermally evaporated CuAn precursors is
developed and will be presented separately in these
proceedings.
For co-evaporated films, the inter ,)mtxn&r microstructure is
dominated by the compositional and substrate temperature
dependence of Cu&e precipitation at grain boundaries and
free surfaces.
The con crystallite is exclusively
stoichiomeuic. or Cu-poor, with little deviation from optimal
valency and with the chemically soluble Cu.&e (&0.15)
minor phase accounting for the Cu-excess in Cu-rich film
compositions. The inua granular microsuucture of a nearstoichiometric grain is a phase-separatedmixture of ordered
chalcopyrite and disordered sphalerite. with CySe (x=0.5,-1.0.
1.5. 2.0) minority phase inclusions. Off-stoichiomeuic Cupoor film compositions additionally contain isolated grains of
the chalcopyrite-variant ordered-vacancy compound CuIn,Sq,.
The goal of this work is to ultimately optimize the fabrication
process by deliberate modification and better control to achieve
higher quality CIS films and devices.
TRANSlTION
for alI the stepsof module production, other challenges include
designing successful encapsulation schemes, confirming
reliability with outdoor and accelerated tests, and developing
market acceptance for these untried and relatively inefficient
(6%-g%) PV modules. In addition, environment, safety and
health issues such as plant safety, plant waste disposal, and
related matters must be fully addressedfor the first time. In
pamUe1, the technologies must stiB progress toward higher
efficiencies (10%15% modules) if they are to make the kind
of impact on global energy production that we in the U.S.
DOE/NREL program believe is possible. The tendency during
this period of transition will be to underestimate the difficulties
and also to underestimate the progress. The latter is a matter
of pemeption associated with the fact that the necessary
progress will occur in manufacture. with few outward rewards
except--once every few years--the introduction of a new or
better product.
ACKNOWLEDGMENT
This work was supported by the U.S. DOE under contract #
DE-ACO2-83CH10093.
REFERENCES
1.
J. R. Sites, “Separation of Voltage Loss Mechanisms in
Polycrystalline Solar Cells,” 20th IEEE Photovoltaic
Specialists Conference, Las Vegas, NV, September 2630, 1988
2.
K. W. Mitchell, C. Eberspacher, J. Ermer, D. Pier.
“Single and Tandem Junction CuInSq Cell and Module
Technology,” 20th lEEE Photovoltaic Specialists
Conference, Las Vegas. NV, September X-30. 1988
3.
I. R. Sites, “Role of Polycrystallinity in CdTe and
CulnSc, Photovoltaic Cells.” Annual Subcontract
Report April 1990 - March 1991 (XC-O-10046-1), Solar
Energy Research Institute, Golden, CO, 22 pp, 1991
wriift~
4.
B. M. Basol and V. K. Kapur, “Improved Group I-IIIVl, Semiconductor Films for Solar Cell Applications,”
European Patent Application, HOlL31/02,3l/l8.
Publication # WO 9005445, December 1990 (applied)
5.
B. M. Basol and V. K. Kapur, “Group I-III-VI,
Semiconductor Films For Solar Cell Application,” U.S.
Patent # 5.028.274, July 2, 1991
6.
L. Mrig, “Outdoor Stability Performance of Thin Film
Photovoltaic Modules,” 26th Intersociety Energy
Conversion Engineering Conference, Boston. MA.
August 4-9. 1991
7.
K. Zweibel and A. M. Barnett, “Polycrystalline Thin
from
Film Photovoltaics.” in Fuels and Electricity
Renewabie Sources of Energy, R. Williams (ed).
Island Press, Washington, DC (in press); prepared for
the U.N. Conference on Global Climate Change, Brazil,
1992
TO MANUFACTURING
Although three of the participants in the U.S. DOE/NREL
Polycrystalliie Thin Films Photovoltaics Project (Siemens
Solar Industries, Photon Energy, and AstroPower) were
winners in the most recent (1990) PVUSA Emerging .Moduie
Technology 2 (EMT-2) competitions, none have yet delivered
the required 20 kW PV systems. The developmental work that
exists between the achievement of an excellent prototype and
true manufacturing is quite significant. This is the source Of
the delayed delivety of 20 kW PV systems. The three
technologies in question (CIS, CdTe, Si-film) have reached an
excellent level of laboratory success;however, the transition to
true production has only recently begun. NREL is suppordng
these technologies during this transition, but we recognize that
delays and unexpected problems are natural. For example.
besides the obvious need to finalize a set of effective Processes
2214 IEEE
Photovoltaics
PHYSICAL,
Specialists
CHEMICAL,
Conference,
Las Vegas NV
act
7-11, 1991
AND STRUCTURAL
MODIFICATIONS
BASED PHOTOVOLTAIC
DEVICES
TO THIN-FILM
CuInSe2
John R. Tuttle, Miguel Contreras,David S. Albin, and Rommel Noufi
National RenewableEnergy Laboratory
(formerly the Solar Energy ResearchInstitute)
Golden, c01ofado
ABsTRAcr
Two approachesto the modification and improvement of
CuInSez-based photovoltaic devices are investigated. The
inanporatioll of wide gap CulnS~-based alloys at the interface is
discussed The results are inconclusive but suggestthat the choice
of alloy systemsis critical. The growth of enhancedgrain thinfilm CuInSq is accomplishedin two manners. The properties of
the films are consistentwith that requiredfor device fabrication.
INTRODUCTION
Present state-of-the-art CuInSe2 thin-film photovoltaic
devices exhibit efficiencies exceeding 14% (1). This record
efficiency has been attained through a proprietary process thaw in
part, relies on a solid-state chemical reaction between mixed Cu
and In metal precursors and F&Se gas to form the C&Se;!
absorber. The alternate fabrication technology of physical vapor
deposition (PVD) has,at besk produceddevices with efficiencies
exceeding 11% (2) with a CuInSg absorber and 12% with a
Cu(In,Ga)Sea absorber (3). A common result of these vastly
different approachesis the realization of high short-circuit current
densities (JEcL 40 mA/cmZ) due to the nearly complete optical
absorption and carrier generation within the field region of the
absorber. Therefore, the primary factor in attaining higher
efficiencies lies in the increasesof the open-circuit voltage (V,)
parametezfYom~44OmVtogreaterthan5OOmV. Theincreaseis
likely linked to the improved electronic nature of the absorber
mateaialfabricatedby the selenizationprocess.This issueis being
addressedby co-workers in thesepmc&ings.
The objective of this work is to use the flexibility of the
PVD technology to investigateprocessesaimedat improving upon
the present stateof the art. The common objective in the various
scenariosis to push the V, parameterabovepresentbarriersusing
both novel and conventional methods while still maintaining the
near-optimalJscvaluespresentlyrealized.
In this paper, we discuss two approachesto realizing this
goal:
1. Band gap enhancementand chemical modification of the
near-interfaceregion by introducing CuInSeZ-basedalloys.
2. Material quality improvement via grain-size enhancement
andchemical modification.
We will report on the incorporation of Zn- and Alcontaining~oysasanintermediatelayerbetweenthewindowand
absorber. Additionally, we will describeseveralapproachesto the
growth and characterizadonof polycrystalline thin-film CulnSe2
with observable crystallite. sizes of 2-10 times that of devicequality films grown by conventional techniques. The results
3-l
suggestthat the latter approach should be emphasizedover the
former until a better understandingof the multinary alloy material
systemsis reached.
EXI’ERIMENTAL APPROACH
The classical picture of the p-n heterojunction device
depicts the metallurgical junction as a position of maximum
recombination, i.e.. the location of type conversion from the ptype absorber@ase)to the n-type window (emitter). Experience
with actual CdS/C!uInSe2devices suggeststhat this is not so;
further investigation (4) indicates that the CulnSe2 base-near the
junction is inverted. Carrier collection, therefore,is accomplished
entirely within the base, thus elevating the importance of the
CuInSez quality in this region. Similarly, Turner et al. (5) have
reported that the V,, of CuInSeZ-based cells is limited by
recombination in the spacecharge. Our objective is to reducethe
recombination, either by inserting a wide-gap layer in the space
charge of the CdS/CuInSezhetero-interface,or by improving the
lifetime of the semiconductorin this region.
Experimentally, CuInSez-based thin-film absorber
structuresare fabricatedby PVD of the constituentelementsunder
a vacuum of 1O-6 torr. The wide-gap interface layers are
fabricated by alloying CulnSeZ (EB = 1.0 eV) with either the
binary ZnSe (2.65 eV) or the temaries CuGaSe2 (1.64 eV),
CuAlSe (2.65 eV), or CuInS2 (1.53 eV) during the final minutes
of the deposition (Table 1). Each alloy systemmay be examined
on the basis of required alloy content, phase stability, and
anion/cation exchangecharacter. Previousreportscan be found in
the literature on the Cu(In.Ga)Se;!system (6.7), so we will focus
here on the Al and Zn systems. The CuIn(Se.S)z system is
presently under investigation and will be reviewed in a future
publication.
Table 1. List of potential alloy partnersto CulnS~.
c
Compound
&SC
CuAlSe~
Band Gap Alloy
Content
(eV)
(%)
2.65
2s
2.65
2s
Concerns
Ailw< dwe
natureunknown
.
.2h dopingvs. allaying
l
l
l
a-2
1.67
60
Al reactivity with SC
Unsabk compound
.WCUChEXWld
27% Gauppu limit
l
cuIns2
1.s3
IS
l
Anion
exchange
supcdor
l GOOdStnad-aloncftbSUbU
;
A parameterof importancewhen evaluating the quality of a
thin-film semiconductor is the carrier lifetime. The relationship
between carrier lifetimes (711,rp) and the device performancefor
CuInSea-based solar cells has been derived via parametric
modeling studies. The resultsof device modeling studiesreported
herearebasedon the solar cell computermodel ADEPT developed
at Purdue University. The code uses a simple single-level
recombination model with the level located at mid-gap for the
CdSKuInSe2 system. The absorption coefficients and other
material parametersare generally derived from the literature. For
carrier lifetime and mobility parameters, where data for
polycrystalline thin films are scarce,adjustmentsare madein such
a mannerthat simulation resultsreflect actualdevice data
The results are plotted in Figure 1 for the CuInSe2/CdS
device structure, where a ~a,, = 3.4x10-9 s is representativeof
evaporatedmaterial. The data suggestthat enhancing the carrier
lifetime will have beneficial effects. Extensive work in the Si
material system (8) derives a direct relationship between carrier
lifetime and the grain size. An analysis of the C&-I&J system(9)
further deducesa direct relationship between V, and grain size.
Our approach, therefore, is to engineer thin-film CuInSe with
enhanced grain sizes and test their suitability in a device
application.
becauseof the presenceof excess Cu2Se. We have taken the
following approachesto producing devicequality films of similar
morphologies:
1. Using the morphology of thesefdms as a growth surface
for active layers
2. Converting the existing photo-inactive material to photoactive materid by a vapor-phaserecrystalhmtionprocess.
Unfortunately. the very nature of the substratesurfaceon
which thesefilms are fabricatedmakesdevice fabrication difticult
We have thereforechosento evaluate thesefilms in the following
1. Performing compositional analysis by electron probe for
microanalysis (EPMA) to relate the actual and intended bulk
compositions,and qualitatively examining the compositional
gradientnearthe film surfaceby variable beam-energyprobing
2. Performing compositional depth profiling by Auger
electron spectroscopy (AES) to examine the near-surface
region for conditions favorable with device fabrication (i.e., a
Cu-poor region with low net carrier concentrationsrequired
for chargedepletionand field generation)
3. Using scanning electron microscopy to examine
polycrystalline thin-film morphologies for enhanced grain
growth
4. Using x-ray diffraction of as-deposited and powdered
films to identify major and minor phases and to examine
preferredorientation.
RESULTS AND DISCUSSION
We have previously reported(6) enhancedV, in Al- and
Ga-containing alloys of CuInSe2. Each alloy system,however,
exhibits unique deficiencies. In the quatemary isoelectronic
alloys. CuInt-y(Ga,Al)ySe2. phase.instabilities for Ga contents
greaterthan 50% and Al-alloy sensitivity to 02 and moistureresult
in a generally poor spacecharge region and therefore lossesin
current and nonoptimal gainsin voltage.
lo-‘0
l#
16
16’
16
Carrier Lifetime (aec)
lo*
Figure 1. The relationship betweencarrier lifetime and the
V, and &vice performance in CuInSe2/CdSsolar cells as
modeledby the ADEPT code.
Experimentally, thin-film CuInSeZ is fabricated by a
variety of processes,including PVD. selenization. and reactive
sputtering, and can be found within a range of “crystallite” sizes
between 0.1 and 1.0 pm. The thin-film morphology and
microstructure has been studied extensively as a function of the
fabrication process for the PVD process (10). This study
concludes that the primary factor dictating the resulting film
morphology is the relative activity of Cu and In during film
growth, and that the secondary factors are the nature of the
substratesurfaceand its temperature. Cu-rich films depositedon
glass surfaces at 500 ‘C closely emulate the morphology and
microstructure of Cu$le. We have been successfulat producing
CuInSe2thin films with unique morphologiesand crystallite sizes
on the order of 2.0-10.0 Pm. The electronic properties of the
films, however, are unsuitable for photovoltaic applications
3-2
Introducing the binary ZnSe into the cuIn&~~ matrix, on
the other hand, may have the opposite effect; the single-cation
semicondnctmsgenerally contain fewer mid-gapdefectstatesthan
do theii ternary analogs. Additionally, the presenceof Zn in the
spacecharge region of a CuInSe2 device.will contribute to the
region’s type conversion from p-type to n-type. Zn. however, is a
quick diffuser and will be difficult to localize to the near-interface
region. Moderate Zn contents (< 50%) may also fall withii the
two-phaseregion of the ZnSe-CuInSq phasefield, which will not
be beneficial for our purposes.
Our latest attemptsat fabricating devicescontaining these
alloys haveproducedmixed results,but we have learnedaboutthe
dynamicsof the alloy formation. Figures2 through 4 presentdata
on modified devices in which -1000-2000 A of the alloys (20%
ZnSe, 12% CuAlSe2) were introduced at the CdS/CuInSe2
interface. The implication is that Zu and Al are incorporatedinto
the absorber in observably different manners. For example,
secondary ion mass spectroscopy (SIMS) (Figure 2) depth
profiling reveals Zn diffusion well into the bulk of the absorber
while Al maintainsthe profile of the depositionmcipe-.
behaves in a similar manner to a standard CuInSe2 device,
whereasthe device containing the (C!uIn)l&n&e2 alloy (Figure
4b) exhibits a deep burled junction with an extensive n-type
region.
Cu(In,Al)Se2
CuInSe2
MO
I
. . ... .. - -..__- -----.. 1 1
CdS
Depth (elm)
Figure 2. SIMS depth profile of two CuInSe2 device
stmctnres. One contains 1000 A of CuIr@.saAl&t2Se2,
while the other contains NOOA of ~q4(~u~.n)~.~2.
BecauseAl is sensitive to air and moisture,we fabricateda
series of samples with a pure CuInSe2 cap on top of an Alcontaining alloy. what is, we compareda device with 2000 A of a
12% alloy (CuIno,atQUo.t2Se2) at the interface with a device
containing 1000 A of a 24% alloy capped with ‘1000 A of
CuInSe2. In Figure 3, the relative spectral response these
modified devices is presentedin comparisonwith a standardthinfilm CdS/CuInSe2 device. The device incorporating the
homogeneousAl-alloy layer (sample D) exhibits a shift in the
onset of absorption and alloying well into the bulk, while the
device incorporating the burled layer of Cu(In,Al)Sg (sampleC)
doesnot display such a shift in the absorptionedge.
(4
MO
Zn(CuIn)Se2
1 CuInSe2l
1
CdS
(b)
Fig. 4 Electron Beam Induced Current linescans depicting
charge collection profile of two modified CuInSg-based
SOhU CdS COnadXting
the (a) cUh().8g&.12Se2 and (b)
Figure 3. Relative spectral response of a standard
CdSIcuTnSe2device.(A) and three alloy device structures:
(BY mm A ~a1.6zno4se2
attheinterface;
(c;wy ;
CuInu.7eAlo.24Se2
buried layer; (D)
~.88&12%
(c~h.f@O.4~2
attheimf&~-
Sample B, which contains -1000 A of the Zn-containing
~oy,issimilarinthatthereisno&~leshiftintheabsorption
edgeof the absorber. The enhanuadresponsein the near-infrared
region. however, suggests a wide space charge region. This is
likely caused by the type conversion of the Cu-poor CuInSe2,
which exists in these structures near the junction, by Zn
substituting on Cu vacancies. This phenomenon is observed in
the electron-beam-induced-current(EBIC) line scansof Figure 4,
where the signal represents the charge collection profile of the
device. In Figure 4a. the device comaining Ct,h,,,.a~lt,J$Q
3-3
d“p-
Thus, we see a general consistency between the SIMS,
sjxctral response,and EBIC analysis that suggeststhat both alloy
systemsrequire additional work to produce highquality devices.
Part of this work is to determine the single-phase region of the
CuInSepZnSe alloy system. preliminary optical characterlxation
identifies the Zn-minima boundary of the single-phase field at
-5.5%ZnwithanEa=1.09eV.
Detem&ationoftheZn-maxima
boundary is presently under way.
on C~Q& . In the bilayer approach to
thin-film cuInse2 device faknicaw film layas with siflcatltly
different Cu contentsare sequentially depot&d With the possible
22nd IEEE
Photovoltaics
Specialists
Conference,
Las Vegas NV
exception of the boundary layers, homogenization occurs via Cu
and/or In diftitsion and the bulk composition becomesa weighted
averageof the individual layers. There is a narrow range of bulk
compositionsthat producedevice-quality material, and it generally
must be Cu-poor. In our first approachto enhancedgrain growth,
this phenomenon of Cu interdiffusion is an important
consideration.
We have prepared a series of four samplesin which the
first thin-film layer is the binary CqSe deposited on glass at
500%. This layer is followed by a seriesof layers of decreasing
Cu content at substratetemperaturesbetween 350X! and 5OO’C.
The intended average composition may be calculated and
compared to that measured by EPMA. Table 2 and Figure 5
summarize the results of this series of films. For reference
purposes,a standardbilayer devicequality film will exhibit only
slight (112) preferred orientation with a total peak intensity of
4000-8000counts/s.
Table 2. Summary of enhancedgrain growth on CuZSe.
Ott
7-11, 1991
Several items of interest may be extracted from the data.
There is a general progressionin grain sixe from the Cudeficient
sample(841) to the Cu-rich sample(844), but the continuity of the
morphology is generally pear in this series.with the exception of
844. We can atmbute this to a threshold thickness above which
the morphology of the baselayer is lost. The possible exception
to this may be sample 840. Its recipe was nearly identical to that
of 847 except that it was cooled following the C!@e deposition.
The resulting orientation was nearly monocrystalline with a (112)
peak height of 140,000counts/s.
A final observation is the high orientation of the Cu-poor
sample 841. In previous studies (11). we have observed either
random or (220) preferred orientation in Cu-poor compositions.
Thii suggeststhat we have the ability to convert monocrystalline
Cu-rich thin films to an overall Cu-poor composition without
losing the potential benefits of e&anced m growth.
Our best effort to date in this series is shown in Figure 6
aud exhibits the type of morphology we have targeted Crystallite
sizes are on the order of 5.0 pm with a smooth surface texture.
The cracking, however, will present a problem for device
fabrication. The cracking may result from the thermal mismatch
betweenthe Cu$e baselayer and the 7059 glasssubstrate.
Tofurtherunderstandthemauuialpmperdesandthenature
of the growth process, we performed extensive Auger analysis.
The analysis consisted of surface and near-surface surveys on
crystallite surfaces, edges, and within a crack between the
structures (Figure 7). and a depth profile at a location within the
crystallite boundaries(Figure 8). The elementalsensitivity factors
used to calculate the atomic percent of the Cu. In, and Se
constituentsare determinedby forcing the composition within the
bulk to equal that measuredby the EPMA. This method lacks
sensitivity to changesin the material matrix at surfaces,but it is
adequatef&our analysis.
Fig. 6’ SEM micrograph of surface morphology of large
grain thin-tilm CuInSe2.
The resaltssuggestthat the crystallite surfacesare deficient
in In and/or rich in Cu, relative to the bulk. The crystallite
boundary is also Cn-rich relative to the bulk This is consistent
with previous studies (11). where we concluded that the
mechanismof thin-film growth is related to the precipitation of
CuzSe at grain boundaries and f&e surfaces. Within the bulk of
Fig.5 scanllin g electron micrographsof thin-tihn samples
descrii
844.
in Table 2. (a) 841, (b) 847, (c) 840, and (d)
3-4
22nd IEEE P&otwoltpics
Specialisb Conference,
Las Vegas NV
the m
we have beon successful at generating a compositional
gradieataptothesllrface(Figme8).~~ofwhichwas
disegsed earlier. Device fabrication, however, has been
unstlccessNduetoshnntingandhigh&reais~attheback
contact. We &e presently working on reproducing these
moxphol~onwrx3actillgs~..
act
7.11, 1991
bulk compositional changes resulting frrim the vapor exposure.
Table 3 describesthe processing wndltlons for several of these
films. The change in overall wmposilion ill&cam .suwess.in
.
nmpmtingInint.otbefilm.
Figme%.~
thesurfaceand
be. little difference in the crystallite
size among the selecti
samples.with the exception of run #77, where the very Cu-poor
natureofthesurfacepmducesa”mottled”appearance. Thenature
ofthecrackingisdifferentthandescribedearlierinthattbey
a@araswnnectedratherthandlsconnt?ctedgrowthplanes. This
may be moreconduciveto the fabricanon of devices.
Table 3. Summaryof vapor-phaserecrystallization se&s.
I
I
Run
No.
at.%
Cu
In, Se
CAM
Time,Temp
(min. ‘c)
FinalCompo&on
tCu/In/Se)
Figure 7 Augtr wmpositional analysis of sampledepicted
in Figure 6. The top number representsa surface survey
and the bottom number a survey following two minutes of
sputtaing. (A) Brain tops
srti edge, 03 grain top.
and(D)crwkbetweengrains.
74
50
1.0, 4.0
10, 300
42.2 J 14.5 J 43.3
84
50
2.0, 8.0
30, 500
26.1 J 24.7 J 49.2
85
40
2.0, 8.0
20, 500
26.1124.5 J 49.3
76
30
2.0, 8.0
20, 300
16.5 J 29.9 J 53.6
77
30
2.0. 8.0
30, 400
15.4 J 30.9 / 53.7
86
30
2.0, 8.0
14 500
26.3 / 24.4 I 49.3
I 7050
I alus
----
0
40
20
Sputter
Time
60
(min)
Figure 8 Auger wmpositional depth profile of bulk grain
region of thin-film sampledepictedin Figure 6.
Vanor phase recrvstallixation Our analysis of Cu-rich
thin-fihn CuInSe2 deposited on glass at 5oo’C shows a direct
relationship between excess Cu content above 25% and the
resulting crystallite sixe and degreeof preferred orientation. The
microstructoral model for this material system suggeststhat the
exoessCuexistsintheformofCu2Seatgrain~undariesandfree
surfaces. X-ray diffraction has identified the majority Cu2Se
phase as cubic with nearly identical lattice parametersto that of
CulnSq (11). We may therefore imagine a processby which the
Cu2Se is exposed to In and Se activity at elevated temperatures
and is feaystallhed into CuInSez. This hasbeenaccomplishedin
a process whereby CuzSe was exposed to 2.0 and 8.0 A/s of In
and Se, respectively, for 15 minutes at 5oo’C under I@6 torr of
vacuum. The chalwpyrite phasewas identified by powder X-ray
diffraction.
We report here on a series of films fabricated with Cu
wntents ranging from 30-50 at% Cu (i.e., 1867% Cu2Se) and
crystallite sixes ranging from 2 to 5 ltnr. We have systematically
varied the recrystallization parameters (i.e., In and Se
impingement flux, substratetemperature,and exposure time, and
characterized the resulting fihns on the basis of morphology and
I
1okx
Fig. 8 SEM micrographs of selected films from Table 3.
(4 74, @I 77. (-3 84, and (4 86.
I
Wearei&reskdinexamimngthemodeand&greeofIn
.
lnanporationduringthevapor~onprocess.
InFigure
9, Auger compccitionai depth profihs of selectedsamplesprovide
insight into the matter. The vastly different sputter times are
cansedbydifferentsputteringratesosedtoprobethefihn.
The
signal fluctuations at the rear of the fihn are artifacts causedby
penetration to the glass substrate. In Figure 9a. we examine a
pure Cu2Se film exposed to In and Se activity at 500% for 15
min. The indium has penetrateduniformly throughout the bull,
converting a portion of the Cu2Se to CttIrrS~, as detected by
powder XRD. The completepenetration of In is aiso observedin
samples84-86 (Table 3 and Figure 9d). In this series, the most
Cu-rich film was processedfor 30 min., while the least Cu-rich
fihnwasprocessedfor1Omin.
Thetlnalproductineachcase
was a fihn with a slightly Cu-rich uniform wmnosition.
o-9
(a)
DWh (run)
Dwth (rm)
0.0
1.0
2.0
a.0
CONCLUSIONS
We have wnsidered two approaches to improving
CuInSq-based solar cell technologies. Modifying the nearink&ceregionwithwi&-gapaIloysofCuIasqandeither%Se
orcuAI!+presenkaninter&ingchahengebecauseoftheitature~
of the wmpounds. Superior resplk with Cu(In~)S& are
attaimdwfienthealloyisceppedoffwitha~layer.?fCnInSe2.
ZnSepesentsaaniqueprobleminthatitmay~~alloy~~ar
dope the C!uhrS~ host, depending on the avaiIabiity of Cuvacancies for the latter phenomenon. Of the two, the
incorpm&onofAIiseasiertocontroL
The altemative is to improve the quaiity of the thin-ftim
material via grain size enhancementprocesses. We have been
successful at producing devicequality thin-film CuInSe2 with
observed crystallite sixes of 2-10 pm. We have also been
successfulatconvertingdaematerialfrofnitsas-depositedc11-rlch
form to Cu-poor by a vapor-phase recrystal&ation method.
Future work will emphasize the measurement of transport
propertiesandthefabri&onofdevices.
ACKNOWLEDGEMENTS
The authors wish to thank A. Tennant, A. Mason, A.
Franz. J. Dolan and A. Duda for technical assistance. This work
was supported by SERI under Contract No. DE-ACOZ83CH10093to the U.S. Departmentof Energy.
REFERENCES
1)
I6
4*
34
;2
2
0
~""'~"'~".....cu.............~...'
Q.................In...........
‘.B
0 6
2
'0
20
Sputter
40
60
lima
80
(min)
10
Sputter
16
2)
20
Tln~4 (min)
Fig. 9 Auger electron spectroscopy depth profiles of a
variety of enhanced-grain thin-films of CuInSeZ as
descrlbedinTable3.
In Figures 9b-d, we examine a seriesof fii
subjectedto
the recrystallization processat different temperaturesto elucidate
the relationship between In diffusion and substratetemperature.
Samples77 and 84 were eachprocessedfor 30 min. at 400-C and
5OO’C,respectively, while sample 76 was processedfor only 20
min. at 300-C. Samples 76 and 77 exhibit similar diffusion
profiles and compositional gradients near the film surface, while
sample84 exhibits a uniform composition throughouf suggesting
a temperature threshold above 400-C for rapid In diffusion.
Sample77 also implies a upper (lower) iimit on the inwrporalion
of In (Cu content). This stoichiometry is approximately that of the
v phase,CuIn;ZSe3.5.
of the In$?q-Cu$3e phasediagram. KRD.
however, has only weakly i&tit&d the presenceof this phasein
the recrystalhred films.
Our future efforts in attaining enhancedgrain growth in
thin-film CuInSe2 will focus on optimizing the processes
describedhere on conducting substrates. We-havealso begun to
characterizethe transportpropertiesof thesefilms, and the results
are very enwuraging. The goal is to correlate grain size with the
enhancementof canier lifetimes anddevice V,.
3)
4)
5)
6)
7)
8)
9)
10)
11)
K.W. Mitchell and H.I. Liu, Proceedings 20th IEEE
Photovoltaic Svecialists Conference. Las Vegas. NV 1988
(IEEE, New Y&k, 1989) pp.-1461-1468. ” ’
K.W. Mitchell, C. Eberspacher, J. Ermer, and D. Bier,
Proceedings20th IEEE Photovoltaic Specialists Conference,
Las Vegas, NV, 1988 (IEEE, New York, 1989) pp-- 13841389.
W.E. Devaney, W.S. Chen, J.M. Stewart- R.A. Mickelson,
IEEE Trans.Elec.Devices,37,1990, p.428.
R.J. Schwartz and J.L. Gray, Proceedings 2Zst IEEE
Photovoltaic Specialists Conference,Kissimme. FL, 1990
(IEEE, New York, 1990) p. 570.
G.B. Turner, R-I. Schwartz, and J.L. Gray, Proceedings
20th IEEE Photovoltaic Specialists Conference,Las Vegas,
NV 1988 (IEEE. New York, 1989) p. 1452.
J.R. Tuttle, M. Ruth, D.S. Albin, A. Mason, and R. Noufi,
Proceedings of the 20th IEEE Photovoltaics Specialists
Conference, Las Vegas, NV, 1988 (IEEE, New York,
1989) p. 1525.
R.W. Birkmire, W.N. Shaferman, R.D. Varrin, Jr.,
Proceedings of the 2Zst IEEE Photovoitaics Specialists
Conference, Kissimee, FL, 1990 (IEEE, New York, 1990)
p. 550.
H. C. Card and E.S. Yang, IEEE Trans. on Electron
Devices, 24(4). 1977,p. 397.
L.L. Kazmerski, Solid State Electronics, 21. 1978.p.1545.
J. R. Tuttle, Ph.D. Dissertation, May 1990.
J.R. Tuttle, D.S. Albin, and R. Noufi, Solar Cells, 34
1991, pp. 21-38.
A STUDY OF ITO/CdS/CuIn~2~1‘EIIN
FILM SOLAR CELLS
I<. Rnmnnothnn, R. G. Dhcre, and T. J. Coutls
Nalionnl Rrnewahle Energy I,ahoratory
(Formerly lhr Solar Energy Research Institute)
Golden, Colorado
ABSTRACT
Solar cells were fabricated on evaporated CuInGaSez
thin films by depositing thin CdS layers from aqueous
solutions and subsequently
depositing indium tin
oxide (ITO) by dc sputtering. The CdS thickness was
varied, and its effect on device quality was studied.
The blue response of the cells showed improvement
when a thin CdS layers was used, but the cell
performance was affected by shunt leakage. Using a
thick, solution-grown
CdS layer, devices with 8.3%
efficiency were fabricated.
INTRODUCTION
CuInSez and CuInGaSez based thin film solar cells
have demonstrated improved efficiencies in recent
years (1, 2). The incorporation of thin, solution-grown
CdS films and the use of ZnO as a transparent
conducting oxide (TCO) layer are considered to be the
major contributing
factors for these improvements.
In the CdS/CuIn(Ga)Se~
heterojunction,
it is
necessary to limit the CdS thickness to an absolute
minimum so that the short wavelength light in the
range of 350 to 500 nm can be transmitted into the
absorber and subsequently
collected. Appreciable
gain in the short-circuit
current density has been
achieved using this approach (1). Solution-grown CdS
films, typically less than 50 nm thick, possess the
unique properties
of being continuous
at such
thickness levels and also being able to coat the rough
surfaces of CuInSe;! films conformally. Other CdS
deposition
processes,
such as physical
vapor
deposition,
are not capable of meeting
this
requirement.
One of the problems associated with
the use of such a thin CdS layer is the possibi1it.y of a
direct
contact
between
the conducting
TCO
subsequently deposited and the CuInSe2 film at the
regions not covered by CdS. This results in tosses due
to shunt leakage. The preferred TCO in CuInSe~ cell
fabrication is ZnO. This is because the ZnOICuInSe2
junction
itself is a reasonably
good rectifying
junction.
A previous work (33 has shown that ZnO is indeed
more effective in minimizing shunt losses than other
transparent
conductors
such as ITG. In these
experiments, evaporated CdS layers several microns
thick were used in conjunction
with a highly
conducting IT0 layer. The effect of including a thin,
solution grown CdS layer remains to be established.
IT0 is still an attractive
transparent
conductor
because its technology is more advanced and its
physics is well understood.
For these reasons, a
systematic study of the ITO/thin CdS/CuIn(Ga)Se2
devices was undertaken.
EXPERIMENTAL
The CuInGaSe2 thin films used in this study were
supplied
by lloeing Aerospace and Electronics
Company. These films were grown by the elemental
coevaporation
method
on molybdenum-coated
alumina substrates (1). The Ga content in the films
was approximately 7 at. 9%.Three samples were used
to fabricated devices, and some variation in the cell
parameters can be attributed to the differences in the
CuInGaSex
film properties. CdS thin films were
grown from an aqueous solution containing CdClx,
NHdOH, NH,Cl and thiourea at 80°C. Deposition
parameters were optimized to yield CdS films of 60-70
nm in one run. Also, the films were grown slowly to
ensure maximum
coverage. Thicker films were
produced by multiple coating of the substrate. In all
cases, the iilms were also deposited on a clean glass
substrate
so that the optical properties,
film
thickness, and refractive index could be obtained.
The refractive index of the films deposited on quartz
substrates,
as measured by ellipsometry,
was
generally lower than the reported value for the bulk,
indicating that the films were somewhat porous. AS
hydroxide inclusions tend to lower the refractive
index, films with an index less than 2.0 were
The thickness
values obtained by
discarded.
ellipsometry
were found to be consistent with the
thickness of CdS in CuInSez as inferred from the
interference
color. After the growth of CdS, the
samples were annealed in air at 200-220% for 15 min.
Our approach to the deposition of IT0 differs from
that reported by Shafarman et al. (31, but it is similar
to the one developed by Boeing for ZnO (1). The IT0
layers were deposited by dc magnetron sputtering in
an argon and oxygen atmosphere. The initial 40 nm
of IT0 was sputtered under a high oxygen partial
pressure to yield a highly resistive (MWsq), highly
transparent layer. This layer is intended to act as a
blocking contact to CuInSex The second IT0 layer of
40 nm thickness was depos’ited at a predetermined
oxygen pressure to produce a conducting layer (200
Wsq). The IT0 thickness was not optimized to
losses.
Standard
minimize
reflection
phot.olithographic procedures were used to defil;e the
grid pattern. Ohmic contact to the IT0 was obtained
hy electroplating gold. The total area of the devices
was 0.25 cm2 and a total of nine devices were
fzthricntcd on a suhst.mt.r area of 1 in2.
RESULTS
0.02 ’
I
I
I
I
--I
1
First, the eFfect of varying the CdS thickness on the
reflectance of CuInSe2 film is shown in Figure 1. The
CdS thickness values are approximate estimates.
Also, some variation in the refractive index of the
films was noted. The bare CuInSez film has an
average reflectance of 15% in the spectral range of
interest. Applying a 70 nm thick CdS film imparts a
dark blue color to the surface and reduces the
reflectance to about 5% at 600 nm. This is the
characteristic of a single-layer, antireflection coating.
As the CdS film thickness increases, the reflectance
shifts
to longer wavelengths
with
minimum
concurrent changes in the visual appearance of the
sample.
The data show the sensitivity
of the
reflectance to the thickness of the CdS layer. This
becomes even more important after the TCO layer is
added.
Figure 2 shows the light 1-V characteristics of three
cells with varying CdS thicknesses. The external
quantum efficiency curves are presented in Figure 3.
Table 1 summaries the parameters of all the cells.
In device #769, the CdS thickness was 40 nm. Shunt
leakage and series resistance
losses caused a
Figure I. Rrllcc~anccl 0f CulnSrZ film with CdS
coatings of’vario~ls thirkncsscs. a. Inlc-oattd: b.
70 nmi c. 100 nni: d. 120 nni.
100
90
-2
-0.4
-0.2
0
0.2
Voltage (VI
0.4
0.6
200
400
600
800
1000
Wavelength (nm)
120
1400
Figure 3. External quantum efficiencies of
ITO/CdS/CulnGaSeP
cells. The CdS thicknesses
are: #759 (40nm). #1217 I150 nm) and #1220
1,750 nml.
FigIn-e 2. Light I-V characteristics of
ITO/CdS/CulnGaSe2
cells. ‘fhc CdS thicknesses
are: # 759 (40 nm). # 12 17 (150 nm) and # 1220
(350 nni).
4-2
Table
1. A comp3risnn
Cell Parameter
#759
CdS thickness
voc (VI
Jsc (mA cms2)
Fill factor
EfIiciency (70) *
40 nm
0.488
26.67
0.35
4.6
ol’ I.ho cell pnramctcrs
* SERI data, AM 1.5 Glol~al, 1000 W
#1217
150 nm
0.395
29.01
0.43
4.9
~11-2,
#1220
350 nm
0.516
‘26.42
0.61
8.3
total area 0.25 cm2
thickness is less than this
response is improved but it
diode quality. Thicker CdS
properties, but much of the
CdS.
reduction of the fill factor; these can he attributed to
the ITO/CuInGaSez contact areas and the ITO/CdS
interface,
respectively.
The external
quantum
efficiency (Figure 3) shows the effect of thin CdS.
Current collection is observed down to 300 nm. The
maximum quantum efficiency (QE) is only 70%,
indicating that the reflection losses and a wavelength
independent recombination
mechanism combine to
reduce the QE. The diffusion length in the nbsorher
appears to be poor, as inferred from the low QE at
I-ong wavelengths.
When the CdS thickness is
increased to 150 nm (device #1217), a reduction in the
short wavelength response is ohservcd (Figure 3).
The loss mec&misms in this cell are similar to the
first device. In device #1220, the CdS thickness is 350
there is no collection
for
nm. As expected,
wavelengths
below 500 nm, as this part of the
spectrum is absorbed in the CdS layer. However, the
open circuit voltage and the fill factor have improved
significantly.
In fact, these are close to the ones
reported by Devaney et al. (1) for their hest device.
Reflection
and recombination
losses are also
significant in -this cell and t.hey reduce the short
circuit current density and QE.
value, short wavelength
occurs at the expense of
films yield good junction
useful light is lost in the
ACKNOWLEDGMENTS
The authors would like to thank K. Emery for
providing the cell performance data, W. Devaney of
Boeing Aerospace and Electronics Company for the
samples, and X. Li for assistance in cell fabrication.
This- work was performed under Contract No. DEACO2-83CII10093
.to the U. S. Department of
Energy.
REFERENCES
1. W. E. Devaney, W. S. Chen, d. M. Stewart, and R. A.
Mickelsen, IEEE Trans. Electron Devices, 37. 428
(1990).
2. K. W. Mitchell, C. Eberspacher, J. Ermer, and D.
Pier, Proc. 20th IEEE Photovoltaics
Specialists
Conference, Las Vegas, NV; 1988, p. 1384.
CONCLUSIONS
From the above results, it appears that good quality
CuInSe.z- devices can be fabricated using solution
~-grown CdS and IT0 as the transparent conductor if
the CdS thickness-is at least-150 nm. When the CdS
3. W. N. Shafarman and R. W. Birkmire, Proc. 20th
IEEE Photovoltaics Specialists Conference, Las
Vegas, NV, 1988, p. 1515.
4-3
OPTICAL
CHARACTERIZATION
OF CuInSe9
SOLAR
OBTAINED -- BY
THE SELENIZATION
METHOD
--.--_._.
CELLS
R. G. Dhrre. K. Ramanathan and T. J. Coutts
National Rcnrwnhle Energy Laboratory
(Fornicrly Ihr Solar Energy Rrsrnrch lnslit~ilr)
Goidrn. Colorado 80401.
Tel :(FW3)23 I- 177 1. Fnx:(303)23 l- 1381
lnlrrnnl
1%.M. 13asol and V. K. Kapur
tonal Solar Electric Technology
lnglrwood, California 9030 1.
ABSTRACT
A systematic.
step-by-step
analysis of the
optical characteristics
of the ZnO/CdS/CuInSez
solar
cells
prepared
by the selenization
technique is reported.
The CuInSez layer was
found to be rough. with a surface feature size of
approximately
0.7-l pm. The final featrlre size of
the cell after ZnO deposition
is approximately
0.8 pm. The texturing provided by Cult&z
and
ZnO and the index matching combine effectively
to
reduce
reflectivity
of
the
the
ZnO/CdS/CulnSez
structure to about 5% over
the entire spectral range. An esl.itnatc of the
losses in the window layers and their cffcct on
cell performance is provided.
INTRODUCTION
Polycrystalline
CuInSe2 thin film solar cell
has become one of the leading candidates
for
large scale terrestrial
W applications.
CuInSez
thin
films have been prepared
by various
techniques such as sputtering (1). elemental coevaporation
(2). compound
electroplating
(3),
and the two-stage process (4). Even though all of
these techniques
produce
reasonable CulnSez
films. only the films prepared by co-evaporatlon
of elements
and the two-stage
process have
resulted
in devices
with
high
conversion
efficiency.
The structure
of a high efficiency
CuInSe2 solar cell consists of a glass or alumina
substrate,
the back contact. a CuInSe:! film, a
window layer. and a finger pattern. The window
layer consists
either of a thick,
evaporated
Cd(Zn)S film. or a thin Cd(Zn)S layer followed by
a transparent conducting oxide such as ZnO. The
use of a thin Cd(Zn)S layer (usually obtained by
the solution
growth technique,
also called dip
coating or chemical growth) and a transparent
conductive
ZnO film redudes the optical losses
associated
with the window layer of a thick
Cd(Zn)S/CuInSez
structure.
Junctions
formed
between solution grown CdS films and CulnSez
layers obtained by the selenization method have
also been found to be of better quallty (4). These
junctions
displayed low diode quality factors and
high open circuit voltage values (5.6). In this
paper we present an optical analysis
of the
ZnO/CdS/CulnSe2
devices. in which the CulnSe2
films
were prepared
by the selenization
technique and the CdS layers were grown by the
solution growth method.
EXPERIMENTAL
Glass substrates
were sputter coated with
approximately
2 pm thick MO layers. Cu and In
films were sequentially
deposited on the MO
coated substrates in an electron-beam evaporator
with a four-pocket hearth.The thicknesses of the
Cu and In layers were 0.2 pm and 0.47 pm.
respectively.
Selcnization
was carried out in a
IlzSe+Ar
mixture
at 400° C. CdS films were
deposited
by the chemical
solution
growth
method in an aqueous bath containing a Cd salt.
a complexing agent, and thiourea as the sulfur
source.
The devices
were completed
by
depositing
ZnO films
by a metal organic
chemtcal vapor deposition
(MOCVD) technique
using diethyl zinc as zinc source.
Two sets of samples (set 529 and 530) were
prepared
for analysis.
The main difference
between these two sets was the thickness of the
ZnO layer (Table-l). Starting with the CuInSez
films,
optical
data were obtained
as each
overlayer (CdS and ZnO) was deposited. CdS and
ZnO films were also deposited on witness glass
substrates
for the optical
analysis
of the
individual layers.
Table
1. Structural
cc11 ID
529
530
Parameters
CdS thickness
nm
70-80
80-90
for the ‘ko
Sets
ZnO thickness
w
2.5
1.3
Optical measurements
were performed on a
Cary 2300 spectrophotometer equipped with an
integrating
sphere. Reflectance
was corrected
using the National Institute
of Standards and
Technology
(NW? reflectance
standard as the
reference. Transmittance
was corrected for the
0% and 100% shifts. Integrated
and diffuse
reflectance
and integrated
transmittance
through
the film. were measured.
Sample
thicknesses were obtained using proillnmetry.
Wavelength,
nm
the ZnO films deposited on glass substrates is
presented in Figures 3a and 3b for the two sample
sets. It Is clear that
the surface feature for
sample # 529 is - 0.5 pm and that the feature size
of the sample # 530 is much smaller than the
feature size of Sample #529 which can be deduced
from the reflectance
data presented
here.
Srnnning Electron Microscopy (SEM) studies have
confirmed thls observation and demonstrated that
the surface roughness of the ZnO films increased
with the thickness. Because of the increase in the
surface roughness, these films became milky for
thicknesses
above
1.5 pm. When
similar
reflectance
measurements
were
made
on
ZnO/CdS/CuInSe2
structures, the overall surface
feature size was found to be about 0.8 pm. The
integrated
reflectance.
on the other hand. was
found to be reduced to -5% in the entire spectral
range of interest as can be seen in Figure 2.
Optical losses caused by device reflectance
and absorption in each of the window layers are
shown in Figure 4. Absorption losses in CdS and
ZnO layers were computed from the integrated
reflectance
and transmittance
measured on the
films deposited on glass substrates.
Figures 5a and 5b present the data on the
expected reduction
in the device performance
based on the optical loss analysis presented in
Figure 4 for two devices #529 and #530. The data
were obtained by multiplying
the AM1.5 global
Ffgure 1. Reflectance of CulnSez/Mo/glnss
film 530, (a) integrat.rcl reflectance (1,) diffuse
reflectance.
RESULTS
The diffuse component of the reflectance Is
expected to be dominant. for wavelengths smaller
than the feature size of the surface. Therefore, in a
reflectance
measurement,
the wavelength
at
which the total reflectance becomes equal to the
diffuse reflectance
is a good
indicator
of the
surface
feature
size. Integrated
and diffuse
reflectance
measurements
performed
on the
CuInSe2 film of sample set #530 are presented in
Figure 1. It is observed that the two reflectance
data diverge for the wavelengths greater than 0.7
pm. Indicating that the layer is textured, with an
average surface feature size of approximately
0.71.0 pm. We have observed similar behavior in the
CuInSe2 film of the sample set #529. The optlcal
measurements
performed
on the CdS films
deposited
on witness glass substrates
showed
indicating
that
negligible
diffuse
reflectance,
these layers were specular.
Reflectances of the
CdS/CuInSez/Mo/glass
structures
exhibited
no
change in the surface feature size, although the
overal integrated
reflectance was reduced from
about 17 % to the 4%- 14O/brange, as can be seen
in Figure
2. This reduction
is due to the
antireflection
effect provided
by the refractive
index match between CuInSez(N-3)
and CdS(N-2
for the solution grown films). The reflectance of
Wavelength,
nm
Figure 2. Integrated reflectance of
CuInSez film and the structures after
CdS and ZnO coating for set 530.
5-2
spectrum photon density with the optical loss in
the wavcicngth range of 320-1200 nm. A quantum
efficiency
of unity
was assumed
for these
calculations.
Calculated
short circuit
current
densities
(mA/cm2) for the two ceils indicatrd in
i’lgrlrr 5. tmmi on the opl.Iral losses arc prcscntrd
in Table 2. Current densities calculated here are
tllc nriivc arca vaiucs and lhev do not inciudc the
losses drle to grid shadowing. ”
q
ZnO Absorption
Waveirngl h. nm
Figure 3a. Integrated (a) anti difftlse (b)
reflectance of ZnO film 529 on a glass subsfmlr.
Wavelength.
nm
Figure 4. Optical losses in device 530 due to
various loss mechanisms.
d
z
3 0.06
DISCUSSION
Wavelength.
From the data presented in Table 2. reduction
in Jsc due to the ceil reflectance is in the range of
2-2.5
mA/cmz.
Thus, natural
texturing
of the
CuInSe2 layer. texturing of the ZnO window layer.
and the index
matching
combine
to provide
sufficient lowering of reflectance losses. Therefore,
the use of an additional antireflection
coating is not
necessary in these ceils. For the present device
structure. the major loss of performance arises due
to absorption in the ZnO window layer. In the case
of device #530 with a ZnO thickness of 1.3 pm, the
current density loss amounts to nearly 7 mAjcm2.
For a ZnO thickness of 2.5 pm (device #529). the
current
density
loss due to ZnO absorption
increases to 12 mA/cm2. These results demonstrate
that a significant
improvement
in the performance
of ZnO/CdS/CuInSe2
solar cells could be achieved by
nm
Figure 3b. Integrated (a) and diffuse (b)
reflectance of ZnO film 530 on a glass
substrate.
5-3
1
q
Jsc loss dur to reflccflon
q
‘Jqr loss due to ZnO absorption
,Jsc loss d11r fo CtlS nhsorptm
fl
<JRrloss d11c to CdS atxmrplion
Jsc aftrr nbovc losses
q
•J
Jsr loss dllc to rrtlcction
m
Jsr loss due to ZnO absorption
0
q
1 -.
Jsc aftrr ahvr
losses
0.8-
0.8
“E
T2 0.6
J
g
v 0.4
0.2
0
Wavelength, nm
Figure 5(b). Estimated Jsc for cell 530
consfdering various loss mechanisms.
Figure 5(a). Estimaled Jsc for cdl 529
considering various loss niechanisms.
Table
2. Calculated
Devices
Considering
Cell ID
529
530
Current
Various
Densities
for
Optical
Losses.
Jsc wif.h
no losses.
Jsc with
reflection
loss,
mA/cm2
mA/cm2
45.93
45.93
43.84
43.38
further
optimization
of the ZnO electro-optical
properties.
Loss due to the absorption in the CdS
layer is in the range of 3.5-4 mA/cma which can be
further
reduced
by using thinner
CdS films.
IIowever. the assumption that the photons absorbed
in the CdS layer are lost may not be valid because
the photo-carriers
generated
in CdS may be
collected due to their proximity
to the junction and
the
J .w with
refl +ZnO
absorption
toss,
mh/cd
I
1
1
31.84
36.43
Jsc with
refl + ZnO
and CdS
absorption
losses,
mA/cm2
28.49
32.57
as CdS layer is completely depleted due to Its high
resistivity.
We have not considered the shadowing
losses due to the front
grid coverage in our
calculations.
In the cells considered
here. these
amount to approximately
7O/6. Uslng an optimized
grtd design, It may be possible to reduce such losses
to under 5%. More recent cells fabricated by the
selenlzation
method have already yielded higher
5-4
eflIciencles by employing the findings of this st~ldy
(6). Further
improvements
arc possible
if the
lossrs, especially due to t.he ZnO ahsnrption.
are
mtnimixed.
ACKNOWLEDGEMENTS
The allthors would ltke to. thank A. Ilnlntli for
the help In device fabrication.
This wnrlc is
supported by the U.S. Department of’ Encrjiy imtirr
Contract No. DE-X02-83CJ
I 10093.
REFERENCES
I). J. A. Thornton
and T. C. Lommasson.
Cells, 16. 165 (1986).
Solar
2). W. E. Dcvaney. R. A. Mickelsen. and W. S. Chen,
Proceedines
of the
18th
IEEE
Photovo&&
Smcialists
Conference.
Las Vee;is
NV(Octuber
1985) IEEE. New York. (1985). 173%1734.
3). R. N. Bhattacharya and K. Rajeshwar, Solar Cells,
16. 155 (1986).
4). 13. M. I3nsol and V. K. Kaput-, Solar Cells, 30. 143
( 1991).
5). X. X. Mu. R. A. Sasala. and J. R. Sites, presented
at the 22nd
IEEE
Photovoltafc
Specialists
Confcrencc. Las Vegas. NV(October 1991).
6). B. M. Basal, V. K. Kapur and A. Malanl.
at the 22nd
IEEE
Photovoltaic
presented
Specialists
Conference,
Las Vegas, NV(October
1991).
5-5
-,TIIIN-FILM()YS
FOB
Rommel Nouft, John Tuttle, David hlhin. hli~url ~ontrcras, Jeff Carapella.Alice Mason, and Andrew Tennnnt
Natirlnnl Rrnewable Energy Laboratory
(Formerly the Solar Energy ResearchInstitute)
1617 Cnlc nlvd., Golden, CO 80401-3393 USA
Tel: (303) 231-7310, Fax: (303) 231-1381
ABSTRACT
We describethe fabrication and characterizationof thinfilm quaternaryalloys of CdxZnt-xSeyS~-yfrom combinationsof
the II-VI binaries. We measureoptical band gaps and lattice
constants of the quatemary alloys by spectrophotometry and
x-ray diffractometry. respectively, as a function of composition.
The optical band gaps range from 1.68 to 3.66 eV, and the
lattice constants range from 5.41 to 6.05 A. The work is
directed at presenting a matrix of optical band gaps and lattice
constantsfor the quatemary alloys, from which one can choose
a junction partner for a given absorber in a heterojunction
photovoltaic cell.
INTRODUCTION
The heterojunction partner materials in CulnSe:! (CIS)
and CdTe solar cells are as critical as the absorber itself in
determining the performance of the cell. Recognizing the
importance of this window layer and the fact that it can be
deposited on an already formed absorber,or vice versa, we set
out to fabricate and characterizethe quaternaryCd,Znt.,Se,S t-y
thin-film alloys. It is important to note that in this work, we deal
with solid solutions that include substitution between both
anions and cations. This allows flexibility to tailor the optical
band gap and the lattice constant for optimal lattice matching lo
the absorber. The valence and conduction bands of ZnS, CdS.
ZnSe, and CdSe are assumedto be composedfrom the localizcfl
electronic p-states around the anion and s-states around the
cation (I). Therefore.,it is expected that the mixed cation-anion
systemshave contributions to the band gap shift from both the
valence and conduction bands (2). This flexibility allows for
changesin the chemistry at the interface.
In achieving low resistivity by doping, most II-VI
semiconductor compounds usually exhibit electrical
compensationof the introduced donor or acceptorimpurities by
intrinsic defect centers of the opposite conductivity type. We
have attempted to dope the CdxZny.,SeySt.y system with In,
Ga. and F and did not succeedin achieving a reproducible wide
range of resistivity. However, in some applications, layers less
than 500 A thick are applied asjunction partnersto the CuInSeZ
films, and thus resistivity is not of great concern.
EXPERIMENTAL
Several alloy compositions of the Cd,Zn I-,Se,S tmy
quatemary were preparedfrom the binary systemsof CdS-ZnSe
6-l
and CdSe-ZnS at a substrate temperature of 2ooOC. Optical
band gaps were measured for films 5000 A thick grown on
7059 glass, using a Beckman spectmphotometerequipped with
an integrating sphere to measurediffuse and total reflectance,
and scattered and total transmission. Lattice constants were
extracted from x-ray diffraction (XRD) measurementsof films
2pm thick using a Rigaku Dmax vertical goniometer and
controller systemwith a rotating Cu anodex-ray generator.
RESULTS
AND
DISCUSSION
Figure 1 shows the relationship of the flux to the
resulting composition of the quaternary films producedfrom the
two systems,ZnSe-CdS and ZnS-CdSe. The deviation of the
actual composition from that expectedfrom the flux is due to the
different sticking coefficient and decomposition/reconstitution
rates of each binary at a substrate temperatureof 2OOY. The
composition of those films was obtained by electron probe for
microanalysis.
CdSe
ZnSe
1.
0.
0.
0. 4
m
1
0. 2 -
l
01
0.0
,
Zr IS
I
0.2
I
I
0.4
I
I
0.6
I
Ild
0.8
l **
1.0
CdS
Fig. 1 Flux of the binaries vs actual composition of the
quatemaries.
The absorption spectra for the samples represented in
Figure 1 are plotted in Figures 2 and 3 as a2 vs. energy so as to
extract the primary direct transition energy. The data show that
someof the films exhibit more than one optical transition. This
indicates that the fihns may not be single phase. Ilowcvcr, the
nature of the phase separation is inconclusive from the optical
data analysis.
Figures 4 and 5 show the XRD spectra for the hinarie%
and quaternary alloys. IJndcr the fahricatmn condition5
specified above, these films appearto hc multiphase. with the
cubic phasepredominant. As a consequence,extraction of the
cubic lattice constant, (a,), was somewhat difficult in certain
compositions. and thus deviated the most from the calculated
values (seeTable 1).
In order to examine the variations betweenexperimental
and calculatedvalues of both energy gap and lattice constant,we
have plotted the optical band gaps and lattice constants of the
above samplesas a function of composition (x and y) in Figure
6a and 6b, respectively. The calculated values are derived using
the equation proposed by Moon et al. (3) for quatemary alloys
described by the chemical name CdxZnt .,Se,S t -y. The data is
summarizedin Table 1 and suggestsseveralanomalies.
Fig.2 a2 vs. Energy for the CdSe-ZnSsystem.
Thin films fabricated in the ZnS-CdSe system, on the other
hand, exhibit observable band gap enhancementrelative to the
calculated values. This is somewhatcontra-indicative of mixed
phasebehavior since it is expected that one of the phaseswill
exhibit a lower optical band gap. It may be indicative of
localized strain effects resulting from incomplete alloying. The
presenceof mixed phasesis more likely in the CdSe-ZnSsystem
as this systemis expected to be partially miscible. This results
in a range of composition in which a single-phasesolid solution
is unstable. There exists an empirical relationship between the
excess free energy of mixing a number of II-VI pseudobinary
systemsand the relative difference in the cubic lattice parameters
of the two componentsof the alloy (4).
A quatemary alloy parameter such as band gaps and
lattice constants for CdxZnl.xSeySt-y can be described by a
surfaceQ(x,y) in which x and y define the composition plane (5)
as seenin Figure 7. The boundary conditions are defined by the
four ternary systems: Cd(SeS), (CdZn)Se, (CdZn)S, and
20
21
22
23
2.
23
28
27
26
29
30
31
32
33
34
35
2e
Fig.4 XRD spectra of selected compositions for the
CdSe-ZnS system showing a systematic shift in 28
correspondingto shift in the lattice constants.
Fig.3 a* vs. Energy for the CdS-ZnSesystem.
20 2, 22 A .;.
The difference between calculated and experimental
values of the optical band gap and lattice constant is generally
small in the ZnSe-CdS system, likely within the experimental
error. It is expectedthat thesefilms are nearly single phasedue
to the small structural variations betweenthe end-point binaries.
“sn
’
ie io il i
3.3 i4 is
Fig.5 XRD spectraof selectedcompositionsfor the CdSZnSe system showing a systematic shift in 28
correspondingto shift in the lattice constants.
6-2
X
Table 1. Summary of the data.
Measured Calcutated
I%
P
Y
E;
h’
--.
1.00
2.64
5.68
2.67
5.76
0.03 -0.09
%n(SeS). In Figure 7. the band gap and lattice constantcontours
arc ralrnlatcd using the equations proposedby Moon et al. (3).
The input to the equations for the lattice constants and bowing
parametersare taken from Ido (6).
We superimposeon the plane the experimental, flux, and
actual composition of the films to show their positions in relation
to the calculated values and in relation to the lattice parameterof
(XnSe2 (5.78 A). It is obvious, when we examine the data,
that obtaining an alloy composition with predetermined
zrlse
I
I
I
0.00
0.26
0.41
0.47
0.87
0.72
0.67
2zTkz-z
c&e
I
t
I
zns
1.00
0.46
0.21
0.08
0.00
1.00
0.57
0.38
0.33
o.cKl
2.39
2.28
2.27
2.30
5.74
5.78
5.77
5.78
2.33
2.24
2.21
2.29
5.74
5.76
5.77
5.78
0.06
0.04
0.05
0.01
0.00
0.00
0.02
0.00
24
1.66
2.51
2.92
3.26
3.64
5.82
6.09
5.69
5.60
5.50
5.43
2‘3
1.70
2.31
2.62
3.07
3.67
5w
6.05
5.74
5.59
5.53
5.41
0.0s
0.02
0.20
0.10
D.21
B.02
RQQ
0.04
-0.05
0.01
-0.02
0.02
composition to fit a specific lattice constant and band gap is
difficult becauseof the complexity of the alloy formation and the
presenceof multiphases.
For lattice matching between CuInSez and the
CdxZnt-xSe,S tsy alloy, we only need to look for those
compositionsalong the tie-line defined by the lattice constantfor
CuinSeZ (5.78 A>. A range of compositions with band gaps
between 2.46 and 2.16 eV satisfy the lattice-matching criteria.
The higher band gap composition will contribute to a higher
short-circuit current becauseof awider light energy transmission
window, while the lower band gap composition can contribute
to the current by allowing more photons to be absorbed and,
hence,carriers to be generatedclose to the junction. This carrier
generationin proximity to thejunction may benefit the device by
localized defect passivation. The latter requires a very thin layer
of this composition. The two compositions defined above are
shown in Figure 7 with the cross-hair symbols and consist of
the approximate
stoichiometry
Cdu.9Znu. t S and
Cdo.sZno.sSwSo.3.
,
ZnSe
CdSe
ZnS
CdS
Fig.7 Energy band gap and lattice constant contours of
the CdxZnl-xSeySl-y quaternary alloys calculated asper
reference3. Superimposedon the contours ate, the lattice
constant for CuInS~, and the flux and composition of the
thin film samples from Fig. 1. The cross-hair symbols
signify the optimum compositions as described in the
text.
X Meeeured
0 Calculated
Fig.6 Variation in (a) the measuredand calculated band
gap (Eg) vs composition, and (b) the measured and
calculated lattice constant (ao) vs composition, for the
CdSe-ZnS and Cd!&ZnSe systems.
6-3
ACKNOWLEDGEMENT
This work was performed under Contracr No.
DE-AC02-83CHlfJO93IOthe U. S. Departmentof Energy.
REFERENCES
1)
2)
3)
4)
5)
6)
J. L. Bitman, J. Phys. Chem. Solids, 8, 35 (1959).
A. Congiu. P. Manca, and A. Spiga, Nuovo Cimm, 5B.
204 (1971).
R. L. Moon, G. A. Antypas, L. W. James,JmElectron.
&&.L, 3,635 (1974).
L. M. Foster. J. Electrochem. Sot 121. 1662 (1974).
T. H. Glisson, J. R. Hauser. M. i Littlejohn, and C. K.
Williams, J, Electron. Mater,, 7, 1 (1978).
T. Jdo,,I. Electron, Mater., 9, 869 (1980).
J’REJ’ARATION.AND CHARACJBRJZATJON Or: J’OLYCRYSTALLJNE rf SPU-J-JERED
CdTe THJN FnMS POR PV APPLJCATJON
F. Abou-Elfotouh, M. Soliman, A. E. Riad*, M. Al-Jassim, and T.Coutts
National RenewableEnergy Laboratory (formerly the Solar Energy ResearchInstitute)
*Virginia Polytechnic Institute and StateUniversity
determine various materialsparametersthat influence the device
ABSTRACT
Jn this work, CdTe films were sputter deposited from a CdTe
performance. These include the type and concentration of the
target using a 2 in. rf planer magnetron S-gun system that
dominant defects, interface states, and deep trap levels. A
minimizes electron bombardment of the film surface.The as
successful method of preparation of p-CdTe thin films must be
grown films were polycrystalline and consistedof a closepacked
capableof yielding high quality material, and be capableof mass
array of preferentially oriented single-crystal grains of 0.5-2.0
producing devices with high conversion efficiency at low cost.
pm in size. Most grains were oriented with their [IOO], ]llO]
Preliminary results indicated that CdTe thin filtns with
and I1 111axes aligned perpendicularly to the substratesurface.
appropriate electro-optical and structural properties can be
After heat treatment,the cartier concentrationswere 1016 1018
achieved by rf magnetron sputtering. The purpose of this paper
cm-j, and Jifetimes were approximately lo-10 s. CdS/CdTe
is to report on the preparation of rf magnetron sputtered CdTe
heterojunction devices were prepared using rf sputtered CdTe
polycrystalline thin films and on their electrical, optical, and
pofycrystalline film. .A conversion efficiency of 6.7% was
structural properties. Despite the fact that sputtering from a
measured.The Voc and Jsc values ate 0.7 V and 2LlmAI cm2
compound semiconductortarget has commonly beenrejectedas
respectively. This preliminary result indicates that sputtering
a meansof preparing semiconductor films, the proJxrrtiesof the
from CdTe target, which has commonly been rejected as a
films fabricated so far appear to be satisfactory for making
device fabrication
devices after a post deposition heat treatment.
Jx&ystalline
technique, may have potential
for
thin-film CdTe cell production.
EXPERJJvlENTALMJI?J’JJODS
JJ’JTRODUCJ’JON
P-type polycrystalline CdTe thin films, doped with either Cu or
To date, CdTe/CdS poIycrystaiJinethin-film heterojunction solar
02. were deposited (on various substratesincluding glass and
cells and modules have demonstratedefficiencies in the 12-1396
alumina) using a 2in. rf magnetron sputtering S-gun system.
range (l-3). Among t-hemethods used to prepare the p-type
This electrode systemis free from electron bombardmentof the
CdTe thin films involved in these junctions are electro-
film surface. A typical deposition rate of I-20 pm/h was
deposition(3). physical vapor deposition (4), close spaced
obtained with an r-f power of 50-500 W. The substrate was
sublimation (S), and screenprinting and sintering (6). The CdTe
placed off-center from the target and positioned close to the
material properties and junction behavior have been found to
target (2-5 cm). The sputtering parameters were 2 mT for
depend strongly on the method of depositing the CdTe films as
pressure,an Ar gas flow of 3 SCcm and a substnte temperature
well as on postdeposition heat treatments(7). The latter
in the range 150”-4OO’C. The sputtering duration dependedon
the required thicknessof the film. The chemical compositions of
the polycrystalline CdTe films were measuredwith a CAMECA
electron microprobe(wavelength dispersive X-ray spectrometry)
and a Physical Electronics Model SPO scanning Auger
microprobe. JJigh resolution photoluminescence(PJ.,)emission
spec~rnwere obtained at different temperatures(10-300 K) using
the 6471 A Ar laser line at different excitation powers (10-50
mW unfocused).Transmission electron microscopy (TEM) and
X-ray measurementswere used to study the crystallinity, grain
size, and morphology of the film material. JJeterojunction
devices were prepared by sputtering CdTe on glass substrates
coated with indium tin oxide (JTO) and thermally evaporated
Fig. I As grown CdTe polyctystalline film structure.
CdS supplied by R. Berkmire at the Institute of Energy
Conversion(lEC). The devices were heatedat 4OO”c,and etched
selectedareadiffraction patternsof single grains indicate that the
twinning occurs perpendicularly to the <1 11>-zincblende
in Br/methanol before deposting the Cu/Au.back contact.
directions.
Current density-voltage (J-V) and capacitance-voltage (C-V)
measurementswere cartied out on thesedevices.
RESULTS
Carrier concentrations in the range 10*6- l@ cm-3 have been
Structural J+oDerties
measured(Table l), together with a lifetime of approximately
Many CdTe thin films have been examined using TEM. main-
10-10 s.
view investigation revealed the grain size and individual
crystallographic orientation of the grains. Furthermore, X-ray
The PL emission from the CdTe film,V- I surface(most resistive
analysis (EDS) was usedin the TJZMto study the distribution of
material) is shown in figure 2a, in which room temptratum band
the copper dopant. In general, these polycrystaJline(as grown)
to band transition at 1.63 eV was observed for the first time in
films are composed of a close-packed array of single-crystal
CdTe poJycrystalJinethin films.
grainr of dimension 0.25-2.0 pm (oxygen-doped),and0.3-4 pm
for Cu-dopd material (Figure 1 ).
TabJe1.Elect&al propertiesof thtee different as deposited
Jr has been observed that the individual CdTe crystallites grow
CdTe films
with a preferred crystallographic orientation. Most often the
[ 1001. [l 101and [ 1111CdTe axes are aligned perpendicularly to
Sample Grain Mobility Cartier Resistivity Carrier
No. Size [cm2/V set] Type [ohm-cm] Concentration
the substrate surface. Atomic resolution images have shown
V 1
v2
v3
that the predominant type of structural defects are coherent
microtwin boundaries. Both the real spaceelectron imagesand
7-2
2.8
4.9
3.0
11.5
46.3
55.6
P
P
P
81.34
0.014
14.98
6.6x1015
9.6~1017
4.5x 1016
650
800
650
800
Wavelength(nm)
Wavelength(nm)
Fig2n PI, emission from rf sputteredCdTe polycrystnlline film
heforeheattreatment.
Fig.2b PL emission from the rf sputteredCdTe polycrystalline
film of figure 2n after high temperatureannealing.
This figure alsn demonstratestransition at 1.58eV due to bound
atmosphere. The radiative recombination levels at I.52 eV
exciton emission associated with Vctt. Annealing the film at high
shown in Figure 4. on the other hand , were observed only in
tempratrtres (370-480%) enhancesthis peak (Figure 2h). The
oxygen-doped CdTe films.
PJ.specmlm of CdTe polycrystalline fiims consists of emission
from two different energy regions. The near-bandedge emission
These defect levels are consideredthe major defect responsible
is attributed to free and bonnd exciton transitions and was found
for the p-type doping. Jn addition,
indicative of the stnlcture perfection (8) as well as stoichiometry
responsible for band-gap levels that are also identified as
deviations. The lower energy emission, on the other hand, is
acceptors. In comparison. Cu atoms in the p-type CdTe can he
nttrihuted to donor-acceptor recombinations (9) involving Cu
involved in the formation of two defectsor complexes thnt have
Clr at Cc1 sites are
and oxygen or their complexes with Vc(t. The exact peak
location, however, depends on the film resistivity as detemlined
r
(eV)
/ 1.465
hy the electrical activity of the dominant defects. Determination
I
of the origin of the defect levels in CdTe films was hosedon the
r 1.42(eV)
change in the PI, peak position and intensity with measuring
temperatureand excitalion power.
A wide range of chemical compositions
and doping
concentrationin CdTe films was also investigated.
Figure 3 shows the PI, emission spectrum of Cd-rich film (
CdRe = I. I) doped with Cu. Several donor-acceptortransitions
nttrihuted to defect impurity complexes associatedwith Cu. Cd,
II
740
I II.1
700
I
1
a20
Wavelength
I
1
860
1
900
(nm)
and Te are demonstrated.Most of thesepeaksare enhancedand
hecame sharper after high temperature annealing in N2
Fig. 3 PL emission from Cu dopedCdTe film.
1
940
24
1.4
Energy
1.5
FF = 51.0%
1.6
WI
0.0
Fig.4 PI. emission from Oxygen dopedCdTe film
0.2
0.4
0.6
0.8
Voltage (v)
Fig.5 J-V relation from CdS/CdTecell preparedfrom rf
sputteredCdTe
heen identified as donors. The composition of thesecomplexes
is not determined yet. Therefore, further work is under way IO
correlate PI, data with the type of conduction in the CdTe and
improving this device performanceand thus for establishing rf
CdS/CdTejunction behavior. This includes the effect of various
sputtering as a viable techniquefor the production of CdTe solar
post-depositionheat heatmentsthat am. crucial to the CdS/CdTe
cells. In order to improve the performance of thctf sputtered
device performanceas well as the defect statesdominating both
CdTe devices, it is essentialto obtain CdTe films with optimum
the CdTe layer and its interfacewith CdS
concentrationof the pmper defectsthrough optimization of both
ered CdTe Devicg
the deposition condition and he post- deposition herI beatmuM
CdSICdTe heterojunction devices were prepared using rf
good quality back contact, and window layer must be
sputteredCdTe polycrystalline film. A conversion efficiency of
developed).
6.7 ‘7nwas measured.The V, and Jscvalues are 0.7 V and 21. I
mA/ cm* respectively. The J-V relation of this device is
The C-V
depicted in Fignre 5.
heterojunction device are pnsented in Fig. 6 in the form of C-*
characteristics obtained on the CdS/CdTe
vs. V. It is clear that the capacitanceis independentof voltage at
A numerical analysis computer programhas beendeveloped for
high frequencies (> SO0kHz) confirming a p-i-n junction typ.
modelling of theseheterojunctiondevices. This program is used
In comparison. a dispersion is found in the lower frequency
for numerical determination of the spectral response, photo-
nnge(<SOOkllz). Therefore, the presenceof an interfacial layer
current, and performance characteristics of the CdS/CdTe
with high density of statesis confirmed.
system (including J-V. V,, J, and output power P,,,,) in terms
of the present measured CdTe and CdS material parameters.
ACKNOWEDGMENT
Based on the materials properties measuredto date, a device
The authorswish to thank R. nirkmire and co-worker at IFC for
efficiency in excessof 8% is expected.The difference between
the fabrication of the CdS/CdTedevices and the supply of CdS
the measuredand computedefficiencies is attributed IO the poor
films. This work is performed under Contract No. DE-ACOZ-
characteristicsof the device backcontact.Ample room exists for
836~10093 to the U.S. Departmentof Energy.
74
2. I-J. Matsumoto. K. Kurubayashi, H. Uda, Y. Komatsu. A.
Nakano, and S. Ikegami. sol.,
1I, 367 (I 984).
+5OOkHr
010OkHz
M
b
x
N
0
3. G. C. Morris , P. G. Tanner. Proc. 21s 1EEE PV SC Conf.
IEEE, New York, 575 (1990) .
4. Paul Sharps, A.L.Fahrenbruch. A. Lopez-Otero. and R. 11.
Bube, Proc. 21a IEEE PV SC Con& IEEE, New York, 493
(19m.
0
4-
0
5. T. L. Chu, S. S. Chu. K. D. Han, and M. Mantravdi. b
@JEEE PV SC Qf.&. IEEE New York, 1422 (1988).
?-
I
0
-4
-3
-2
-1
0
1
vd%le M
Pig.6 Capacimnee-Loltagecharacruistics showing C2-vs-V
dependenceof CdS/CdTe heterojunction.
REFERENCES
I. S. P. Albright, B. Ackerman, and 1. F. Jordan, IEEE Trns.
n. Dev., 37.434 (1990).
6. N. Suyama,T. Arita. Y. Nishiyama, N. Ueno, S. Kitamura,
and M. Murozono, proC. @ IEEE PV SC Conf,, IEEE, New
York, 498 (1988).
7. B. N. Baron, R. W. Birkmire. J. E. Phillips. W. N.
Shafarman, S. S. Hegedus, and B. E. McCandless. NREL
Golden. Co., Rep. No. SERlfl-211-4133, (1991)
8. Z. C. Feng. M. G, Burke, and W. J. Choyke, ADDI. Phvs,
u
53, (2). I28 (1988).
9. N. C. Taylor. R. N. Bicknell, D. K. Blanks, T. II. Myers,
and J. F. Schetzina. J. Vat. Sci. and Technoi. A, 3. 1.76
(1985).
ADVANCED
HIGH-EFFICIENCY
CONCENTRATOR
TANDEM
SOLAR CELLS
M. W. Wanlass, T. J. Coutts, J. S. Ward, K. A. Emery, T. A. Gessert, and C. R. Osterwald
National Renewable Energy Laboratory (NREL)
(formerly the Solar Energy Research Institute)
Golden, Colorado, USA
ABSTRACT
Using the modeling
results as a basis, two novel
concentrator
tandem cell structures
that utilize
lowband-gap
bottom cells have been investigated.
The
structures
involve
a combination
of practical
and
idealized
design considerations.
The lattice-matched
Ga,InI~,AsyPI~y/lnP
system has been chosen for bottom
cell fabrication
in both tandems because it offers highquality bulk and heterointerface
properties as well as a
wide range of band gaps (0.75 - 1.35 eV) in the near IR.
Both mechanically
stacked and monolithic
structures
have been considered and the top and bottom cells have
been treated as independently
connected subcells in this
preliminary
work.
The first tandem
consists of a
mechanically
stacked combination
of a GaAs top cell
above a 0.95-eV-CaInAsP
bottom cell, which is the
optimum
bottom-cell
band
gap
for
terrestrial
concentrator
tandems, according to the modeling results.
The quaternary
bottom cell composition
is given by
x=0.25 and y=O.54. A three-terminal,
monolithic,
latticematched combination
of an InP top cell on a 0.75-eVGalnAs
bottom
cell constitutes
the second tandem
design under study.
For this tandem, the bottom cell
composition
Hereafter,
is Gao.471n0.s,As.
GaO.zs inO.,s As,,, P,,, and Gao.47 ln,s3 As are referred to
as GaInAsP, and GaInAs, respectively.
Computer
modeling
studies
of two-junction
concentrator
tandem solar cells show that infrared (IRIresponsive
bottom cells are essential
to achieve
the
highest performance
levels in both terrestrial and space
applications.
These studies also show that medium-bandgap/low-band-gap
tandem pairs hold a clear performance
advantage under concentration
when compared to highband-gap/medium-band-gap
pairs, even at high operating
temperatures
(up to 100°C).
Consequently,
two novel
concentrator
tandem designs that utilize low-band-gap
bottom cells have been investigated.
These include (1)
mechanically
stacked,
four-terminal
GaAs/0.95
eVGaInAsP tandem, and (2) monolithic,
lattice-matched,
three-terminal
lnP/0.75
eV-GalnAs
tandem.
In
preliminary
experiments,
terrestrial
concentrator
efficiencies
exceeding
30% have been achieved
with
Methods
for improving
the
each of these designs.
efficiency of each tandem are discussed.
INTRODUCTION
In the past, we have used computer
modeling
studies to provide guidelines
for designing
novel twojunction,
concentrator
tandem solar cells (1, 2). These
studies have shown that operation
under concentrated
solar illumination
produces a profound
effect on the
optimum
band gap values for the top and bottom
subcells.
Specifically,
the modeling calculations
suggest
that optimally
designed concentrator
tandems typically
have much lower subcell band gaps than their one-sun
The previous
modeling
work
was
counterparts.
concerned
with independently
connected,
two-junction
terrestrial
concentrator
tandems.
In the present paper,
we have extended
the modeling
to include
seriesconnected
devices, as well as operation under the AM0
spectrum.
As shown later, in the section describing the
modeling results, we conclude that IR-responsive bottom
cells are essential in all cases to achieve high-performance
two-junction
concentrator
tandem cells.
All of the device structures discussed in this paper
were grown by atmospheric-pressure
metalorganic
vaporphase epitaxy (APMOVPE).
The epitaxial growth, device
processing, and characterization
procedures used in this
work have all been described previously (2, 3, 4). In the
remainder of this paper, we discuss our latest computer
modeling results and conclusions along with descriptions
and performance data for the novel concentrator
tandem
cells mentioned above.
COMPUTER
MODELING
The basic framework
of our computer
model has
been outlined previously
(1, 2). However,
in the most
recent set of calculations,
we have effected two minor
modifications
to the model to make the results more
useful for practical purposes. The ultimate application
of
these results is to aid in designing
high-performance
8-I
concentrator
technologically
tandem
cells
attractive materials.
In the model, the equation
current density, given by
that
and in electrical series. The series-connected
ce!ls were
modeled with top cells that were both optically thick and
optically thin. With optically.thin
top cells, we assume
that the subcell currents can be forced to match at a
value that is the average of the subcell currents obtained
for the corresponding
case using an optically thick top
cell. This can be achieved by reducing the thickness of
the top cell to allow photons with energies greater than
the top cell band gap to pass into the bottom cell. The
assumption is justified because we have set AEQE=l for
wavelengths 2500 nm.
incorporate
for the reverse-saturation
J0=pT3 exp(-Eg/kT)
in which the pre-exponential
factor p depends on the
band gap and, possibly,
the temperature.
In this
equation, T is the absolute temperature,
E, is the band
gap at temperature
T, and k is the Boltzmann constant.
The band-gap
dependence
of B at 25’C has been
determined
by obtaining
a best fit to the reversesaturation
current
density
derived
from illuminated
current-voltage
data for state-of-the-art
solar cells
measured at our laboratory.
The latest fit covers a wide
range of band gaps (O-75-1.93 eV) and has the following
form
J3=3.165 expJ2.912
Iso-efficiency
contours at an operating temperature
of 50°C. are shown in Figures la-lf
for the abovementioned cases. After reviewing the contour plots, it is
clear that IR-responsive bottom cells are necessary in all
cases to attain high performance levels. This is particularly
the case for tandems
operated
under
the direct
spectrum.
Generally, bottom cells with band gaps in the
0.6-l .2 eV range are required.
For series-connected
tandems,
thinning
the top cell to achieve
matched
subcell currents has no effect on either the band gap
coordinates for maximum efficiency or the need for IRresponsive
bottom cells for high performance.
This
procedure
simply allows series-connected
devices to
attain higher efficiencies with lower top cell/bottom
cell
band-gap differentials.
The overriding
conclusion from
the modeling
results is that two-junction
concentrator
tandem cells require IR-responsive bottom cells for high
performance in both terrestrial and space environments.
E, (eV)J Am-*K-3
Data from twelve high-performance
cells with different
direct band gaps were evaluated
to determine
J3.
However, due to a lack of experimental
data at different
8 is assumed
to be temperature
temperatures,
independent
in the present calculations.
We plan to
include the temperature
dependence
of J3 in the model
when sufficient data become available, however, we do
not feel that the conclusions
drawn from the present
study will be affected.
We repeated the modeling
calculations
using an
operating
temperature
of 100°C to see if higher
temperatures
would effect the above conclusion.
The
resulting data (not given here) show that the conclusion
remains unchanged.
Hence, bottom cells with higher
band gaps are not required
for high-temperature
operation.
As we have mentioned in previous work (1, 21,
this result is due to a significant
improvement
in the
temperature
coefficient
of efficiency
for low-band-gap
cells when operated under concentration.
Realizing that it is physically
impossible to achieve
unity absolute external quantum efficiency (AEQE) at very
short wavelengths due to reflection and absorption losses
in practical cells, we have included a blue response in our
model that rolls off for photon wavelengths less than 500
nm. The form for the blue response was modeled after
data obtained
from a high-efficiency
(24.8%, one-sun
global spectrum,
25OC) GaAs cell fabricated
by Spire
Corporation
(5).
in the model, the AEQE decreases
monotonically
to 0.915, 0.781, and 0.394 at wavelengths
of 450 nm, 400 nm, and 350 nm, respectively.
The AEQE
is set to unity for wavelengths 1500 nm.
It is interesting to note from the contour plots that
medium-band-gap/low-band-gap
tandem
pairs are
heavily favored over high-band-gap/medium-band-gap
pairs for high performance,
in all cases. This observation
has important implications
for tandem cell research. For
example, tandems utilizing
GaAs fEp-1.42 eV @ 5OW
bottom cells (e.g., AIGaAs/GaAs and GalnP/GaAsL which
have been heavily
researched
in recent years for
are
at an extreme
concentrator
applications,
performance
disadvantage
when compared
to tandem
combinations
such as GaAslGalnAsP
and InPICalnAs.
The lower-band-gap
tandems
have the potential
to
outperform the higher-band-gap
combinations
by five to
eight percentage
points at 50°C, depending
on the
The modeled tandem cell efficiencies
have been
calculated as a function of the top and bottom cell band
gaps and for various appropriate operating conditions.
For
terrestrial tandems, we used the direct spectrum (ASTM E
891,
1000
Wm-2
total
irradiance),
500
suns
concentration,
and at a temperature
of 50°C.
Space
tandems were modeled at 50X under 100 AM0 (6) suns.
Additionally,
the calculations
were repeated using an
operating temperature of 100°C for the same spectra and
respective concentration
ratios.
For both spectra, we
considered subcells that were connected
independently
8-2
incident spectrum and subcell connectivity.
performance advantage exists at 1OO’C.
are shown in Figure 4. The cell has an efficiency of 9.4%
at 30.6 suns under the direct spectrum, 25OC. The high
values of the open-circuit
voltage (V,,) and FF (0.658 V
and 82%, respectively) are particularly noteworthy for this
low-band-gap
cell.
A similar
GaAs/GalnAsP MECHANICALLY
STACKED, FOURTERMINAL TANDEM SOLAR CELLS
A schematic
diagram
of the GaAs/GaInAsP
mechanically
stacked tandem concept is given in Figure
2. The GaInAsP bottom cell is grown lattice matched on
an InP substrate and uses InP as a window
layer to
An n/p doping
passivate
the emitter
surface.
has been
used
to minimize
the
configuration
emitter/window
sheet resistance and emitter grid contact
Positioned on the bottom cell surface is an
resistance.
Entech prismatic cover to eliminate optical losses due to
Reference 4 contains further details of
grid obscuration.
the GaInAsP cell construction.
As shown in the diagram,
the quaternary
bottom cells have been tested under IRtransparent
GaAs filters and also under actual GaAs
concentrator
cells
grown
on IR-transparent
GaAs
In
both
cases,
the
GaAs-based
top
structure
is
substrates.
mirror smooth on the front and back surfaces with
appropriate
antireflection
coatings (ARCS) on each of the
surfaces.
We have been successful in a preliminary
attempt to
fabricate
actual GaAs/GalnAsP
mechanically
stacked
tandem
cells.
The cells have been tested under
concentration
using an aperture to define the illuminated
cell area. Efficiency versus C data for our best stacked
tandem are shown in Figure 5. The performance
of the
GaInAsP
bottom
cell in the stack was hampered
somewhat by the use of the aperture because only about
one-third
of the total area of the bottom
cell was
illuminated
during the measurement process.
Likewise,
the quality of the GaAs concentrator
top cells that were
used in the stack are far from state of the art. Despite the
obvious deficiencies
in the stacked device, the tandem
efficiency
still exceeded 30% for C values ranging from
30-100.
A top cell/bottom
cell current-voltage
data
composite
for the GaAs/GalnAsP
tandem
at peak
efficiency is given in Figure 6. At 39.5 suns, the top cell is
23.1% efficient and the bottom cell has an efficiency of
7.1%, yielding
a tandem efficiency
of 30.2%.
With
improvements
in the top cell quality, stacking procedure
and optical coupling
into the bottom cell, we feel that
concentrator
tandem efficiencies approaching
40% may
be achieved in the future.
Efficiency versus concentration
ratio (Cl data for a
high-efficiency
GaInAsP concentrator
cell under a GaAs
filter are shown in Figure 3. The efficiency data show the
expected increase as C is increased initially (compared ‘to
the modeled performance data) and then exhibit a broad
plateau at about 9.4% for C in the 20-130 range. The fill
factor (FF) data show that the cell becomes seriesresistance
limited
at about 30 suns, thus prohibiting
further efficiency gains for higher values of C and resulting
in the broad efficiency
plateau.
It is clear that GaAsfiltered GaInAsP cell efficiencies
exceeding
10% at C
2100 could be achieved through a reduction in the cell
the modeled cell
series resistance
(R,). Furthermore,
performance
data suggest that the efficiency
could
improve by l-2 percentage points at low values of C even
with the present value of R,. An analysis of internal
quantum
efficiency
and absolute
external
quantum
efficiency data for these cells (not given here) shows that
the majority
of the discrepancy
between the modeled
and measured efficiency
data is due to external optical
losses. Therefore, improved optical coupling techniques
Nevertheless,
the
should lead to higher efficiencies.
present efficiency
boost offered by the GalnA.sP cells is
substantial and immediately
useful in tandem stacks.
The illuminated
current-voltage
concentrator
cell at peak efficiency
NREL-grown GaInAsP cells have also been applied as
bottom
cell components
in series-connected,
threejunction AIGaAs/GaAs/GalnAsP
tandem cells fabricated
by researchers at Varian Research Center, Palo Alto, Calif.
These tandem
cells are designed
for one-sun
AM0
conditions
and are described
in a related
paper
presented at this conference (71.
InP/GalnAs
MONOLITHIC,
THREE-TERMINAL
SOLAR CELLS
TANDEM
The InP/GalnAs
monolithic,
three-terminal
tandem
cell was originally
conceived
for space applications
because it has several advantages, including a radiationresistant InP top cell (3). However, it may also be very
useful in terrestrial concentrator
applications
because it
has a high modeled efficiency.
The tandem performance
AM0
under terrestrial
conditions
is presented ‘here.
performance
data and temperature
coefficients
for the
subcell performance
parameters have been presented in
a recent paper (8).
data for a GaInAsP
under a GaAs filter
8-3
An illustration
of the lnP/GalnAs
tandem
cell
construction is shown in Figure 7. The device consists of
twelve
epitaxial
layers, which
are deposited
in a
continuous
growth
sequence.
The lattice-matched,
monolithic structure consists of three major components,
including
the GalnAs bottom cell, a middle contact
region, and the InP top cell. The details and function of
each of these components have been outlined previously
The three-terminal
cell utilizes
a two ‘?vel,
(3).
interdigitated
top/middle
contact
grid system and a
contact on the back surface of the InP substrate.
The
Entech prismatic cover on the cell surface is an integral
part of the tandem design because it directs all of the
incoming photons away from the top cell gridlines and
middle contact
trenches and onto the InP top cell
surface.
In Figure 8, concentrator
efficiency data are shown
for a high-efficiency
InP/GalnAs tandem cell. The GalnAs
bottom cell has performance
characteristics
that are
extremely
close to the limits predicted
by computer
modeling,
which suggests that further improvements
in
the GalnAs junction quality appear unlikely.
The InP top
cell also performs quite well, reaching a broad efficiency
peak of 23% over the 20-40 suns concentration
range.
As C approaches
100 suns, the InP top cell becomes
series-resistance
limited which results in a broad tandem
efficiency
maximum
that approaches
32% from lo-50
suns. The concentrator
short-circuit
current density (J,,)
versus V,, data have been used to determine the ideality
factors (n) for the top and bottom cell junctions.
For the
InP cell, n = 1.02; for the GalnAs cell, n = 1.03. These
values reflect that both junctions are of excellent quality.
Further improvements
in the tandem cell efficiency are
still possible. A reduction of R, in the InP top cell would
allow the tandem to operate at a higher efficiency
at
higher concentration
ratios. Discussions of this problem
and its solutions are given in companion papers presented
at this conference (9, 10). Passivation of the InP emitter
surface would lead to higher top cell efficiencies
(3).
Solutions to these efficiency-limiting
problems are being
pursued and concentrator
terrestrial efficiencies
~35%
appear possible for this tandem design.
the series resistance
order of magnitude.
A composite
current-voltage
data plot for the
InP/GalnAs tandem at peak efiiciency is given in Figure 9.
At 50 suns concentration,
the top and bottom cell
efficiencies are 22.9% and 8.9%, respectively, which add
up to a tandem efficency of 31.8%. This result marks the
first time that a monolithic
tandem cell has exceeded
30% efficiency.
Realistic modeling calculations (10) show
that efficiencies close to 35% (250 suns, direct spectrum,
25/C) could be achieved
with this tandem by reducing
We wish to thank Mark O’Neill at Entech, Inc. for
helpful discussions regarding the installation
of Entech
prismatic
covers on the various
concentrator
cells.
Support
for this work was provided
by the U.S.
Department
of Energy under contract
No. DE-ACOZ83CH10093
and the Naval Research Laboratory under
Interagency Agreement
No. RU-1 1-W70-AD.
of the InP top cell by roughly
one
SUMMARY
Computer
modeling
studies have shown that IRresponsive
bottom cells are required
to achieve the
highest efficiency two-junction
concentrator tandem solar
cells in both terrestrial and space environments
over a
broad temperature range. Based on this conclusion, we
have investigated two promising tandem cell designs that
incorporate low-band-gap
bottom cells: (1) mechanically
stacked, four-terminal
GaAs/GalnAsP
tandem, and (2)
monolithic,
three-terminal
InP/GalnAs
tandem.
The
preliminary
performance
results for these designs
corroborate
the modeling
predictions
and are very
encouraging
because
both types of tandems
have
exceeded 30% efficiency
at mild concentration
ratios
under standard terrestrial measurement conditions.
Under a GaAs filter, GaInAsP concentrator cells have
efficiencies as high as 9.4% at concentrations
of 20-130
suns. With a reduction
in R, and improved
optical
coupling, efficiencies exceeding
10% are anticipated for
these cells in the future.
Preliminary
GaAs/GalnAsP
mechanically
stacked tandems have achieved efficiencies
as high as 30.2% at 39.5 suns. By improving the GaAs top
cell quality and the tandem stacking procedure, tandem
efficiencies
approaching
40% may be achievable
at
higher solar concentrations.
Monolithic
InP/GalnAs tandem cells have reached
efficiencies of 31.8% at 50 suns. This is the first report of a
monolithic
tand.em cell with an efficiency greater than
30%.
The GalnAs bottom cell has near-theoretical
performance at low concentration
ratios however the InP
top cell efficiency could be improved substantially with a
passivated emitter and an improved ARC. If the top cell
cell
series
resistance
were
reduced,
efficiencies
exceeding 35% could be realized by operating at high
concentration
ratios.
ACKNOWLEDGEMENTS
8-4
REFERENCES
1.
M.W. Wanlass,
K.A. Emery, T.A. Gessert, G.S.
Horner, C.R. Osterwald, and T.J. Coutts, Solar Cells,
27, 1989, 191-204
2.
M. W. Wanlass,
C.R. Osterwald,
1991, 363-371
3.
M.W. Wanlass, T.A. Gessert, G.S. Horner, K.A.
Emery, and T.J. Coutts, Proc. NASA Conf. Soace
>y,
NASA Lewis
Research Center, Cleveland,
OH, Nov. 7-9, 1989,
102-116
4.
M.W.
G.S.
21 st
172-l
5.
S.P. Tobin, S.M. Vernon, S.J. Wojtczuk, C. Bajgar,
M.M. Sanfacon, and T.M. Dixon, Conf. Rec. of the
21 st IEEE Photovoltaic
Soecialists
Conf., 1990,
158-162
6.
C. Wehrli, Extraterrestrial
Solar Spectrum, Physical
Meteorological
Observatory
and World Radiation
Center, tech rep. no. 615, Davos-Dorf, Switzerland,
July 1985.
7.
B-C. Chung, G.F. Virshup, M. Klausmeier-Brown,
M.L. Ristow, and, M.W.Wanlass,
Plenary Session 2,
these proceedings.
8.
M.W. Wanlass, J.S. Ward, T.J. Coutts, K.A. Emery,
T.A. Gessert, and C.R. Osterwald, Proc. NASA Conf.
Soace Photovoitaic
Research and Technologv,
NASA Lewis Research Center, Cleveland, OH, May
7-9, 1991, p. 16-l.
9.
J.S. Ward, M.W. Wanlass, T.J. Coutts, K.A. Emery,
and C.R. Osterwald,
Late News Session, these
proceedings.
10.
C.R. Osterwald,
M.W. Wanlass, J.S. Ward, B.M.
Keyes, K.A. Emery and, T.J. Coutts, Oral Session 9A,
these proceedings.
J.S. Ward, K.A. Emery, T.A. Gessert,
and T.J. Coutts, Solar Cells, 30,
Wanlass, J.S. Ward, T.A. Gessert, K.A. Emery,
Horner, and T.J. Coutts, Conf. Rec. of the
IEEE Photovoltaic
Soecialists
Conf., 1990,
78
AMO, 100 suns, 50°C
2.0 P”~‘~“‘;r”““““““‘\““1”‘1”4”~
2.0
1.9
1.9
5 1.8
2
4 1.7
u
s
2
0
u
z
2
1.6
7 1.6
m”
z
1.5
z
1.5
c8
1.4
I-8 1.4
1.3
1.8
1.7
1.3
1.2
2.0
1.9
1.9
5 1.8
*
4 1.7
u
5 1.8
i?!
4 1.7
u
7 1.6
m”
2
2
1.6
z
3
1.5
1.5
ld
I-tit 1.4
I-8 1.4
qmax = 36.5%
Series Connected
qmx = 42.0%
Series Connected
1.3
7
-b
1.9
1.9
5 1.8
2
i$ 1.7
u
p 1.8
2%
i3- 1.7
u
2 1.6
m”
z
2
1.6
z
3
1.5
1.5
I-E 1.4
c0” 1.4
1.3
1.3
1.2
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
1.4
1.5
0.6
0.8
0.9
1.0
1.1
1.2
1.3
1.3
1.5
Bottom Cell Band Gap (eV)
Bottom Cell Band Gap (eV)
Fi ures la-lf.
Corn uter modeled iso-eiiiciency
si %ered represent re Pevant operating conditions
0.7
contours for two-junction concentrator tandem cells. The six cases conand subcell connectivity -options ior terrestrial and space applications.
8-6
ARC
IR-transparent
GAS
-Filter or
Cocentrator
I
80 -
I
I
60-
I
20 i
; v,:
0.6577 V
I I,:
531.8 mAcm-’
1 FF: 82.0 %
1T
9.4 %
i-- -----------
I
ARC
7.5
F5
5
u
Entech cover
ARC and grid
0.2-l .5 pm n+-InP window
0.2 pm n+-GaInAsP emitter
I
II-
r
t
nn
7”
0
--
4 urn p-CaInAsP base
0.4 pm p&P
-2Ok’
buffer
-0.2
’ ’ ’ ’ ’ t ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ 1
0.0
0.2
0.4
0.6
0.8
Voltage (VI
Figure 2.
Schematic
diagram of the GaAsKalnAsP
mechanically
stacked tandem cell concept.
Details of the
GainAsP
bottom
cell construction
are shown.
The
GaInAsP cells have been tested under an IR-transparent
GaAs filter and also in actual tandem cell stacks.
Figure 4. Current-voltage
data for a GaInAsP cell at peak
efficiency
under
concentration
(30.6 suns, direct
spectrum, 25Q
35 c
30
a
25
;
20
.z
2
z
0.6
1
10
Concentration
100
f
15 -
-v-s
I Top Cell
A Modeled Ebt. Cell
10
Concentration
1000
Ratio
Ratio
Figure 5. Efficiency versus concentration
ratio data for a
mechanically
stacked
GaAs/Gal,nAsP
concentrator
tandem
cell.
The
modeled
efficiency
versus
concentration
ratio data for the GaInAsP bottom cell are
also included.
Figure 3. Efficiency versus concentration
ratio data for a
high-efficiency
GaInAsP concentrator
cell under an IRtransparent,
AR-coated GaAs filter. Also shown are the
fill factor data for the same cell along with the modeled
efficiency, assuming no resistance losses, as a function of
the concentration
ratio.
8-7
V,:
I :
f;:
V
60
GaInAsP
0.6264 V
556.7 mAcmm2
80.7 %
7.1 %
‘s
50 :
40 >
30
I
I
I
I Tandem q:
130.2%
I
GaAs
V,: 1.096 V
I,,: 990.3 mAcmS2
FF: 83.5 %
ll:
23.1 %
*Tandem
= Top Cell
A Modeled Bat. Cell
0 B0tth-nCell
GaAs
1
10
100
Concentration
1000
Ratio
Figure 8. Efficiency
versus concentration
ratio data for a
high-efficiency
InP/GalnAs monolithic tandem cell. The
modeled efficiency data, assuming no resistance losses, for
the GalnAs bottom ceil are also given.
Voltage (V)
InP
Composite
current-voltage
data for a
Figure 6.
GaAs/GalnAsP mechanically
stacked tandem cell at peak
efficiency
under
concentration
(39.5 suns, direct
spectrum, 25OC).
GatnAs
v,:
I,:
FF:
11:
0.9733 v
1416 mAcmS2
83.3 %
22.9 %
loo
c
v,:
I,:
FF:
II:
0.4448 v
1321 mAcm-’
75.7 %
a.9 y.
1
I
80
0
--t---------l Tandem n: 31.8%
-20 tIII’IIIIIII’lIIi)IL’I1lJ
-0.2
0.0 0.2 0.4 0.6
+-
0.8
1 .O
Bottom cell
Voltage (V)
Figure 9.
Composite
current-voltage
data for an InP/
GalnAs
tandem
cell
at peak
efficiency
under
concentration
(50.0 suns, direct spectrum, 25W.
Figure 7.
Cross-sectional
schematic of the InP/GalnAs
monolithic,
three-terminal
tandem solar cell. Important
features include( 1) two-level,
interdigitated
top/middle
grid contacts, (2) middle contact that is common to both
subcells,
and (3) an Entech prismatic
cover, which
eliminates optical losses due to grid obscuration and IOSS
of top cell area.
8-8
STATUS OF TI1E PI1OTOVOLTAIC MANUFACTURING
TECHNOLOGY (PVMaT) PROJECT
C. Edwin Witt, NREI.; I,loyd 0. llerwig, U.S. Department of Energy;
Richard I,. Mitchrll, NRIX; G. David Mooney, NREL
Nntional Rcncwablc Energy Laboratory
(formerly the Solar Energy ResearchInstitute)
Golden, Colorado
ABSTRACT
The Photovoltaic Manufacturing Technology (PVMaT)
project is a government/industry photovoltaic manufacturing
R&D project composed of partnerships between the federal
government (through the U.S. Department of Energy) an4
membersof the U.S. PV industry. It is designed to assist the
U.S. PV industry in improving manufacturing processes,
accelerating manufacturing cost reductions for PV modules,
increasing commercial product performance, and generally
laying the groundwork for a substantial scaleup of U.S.-based
PV manufacturing plant capabilities.
The project is being carried out in three separatephases,
each focused on a specific approach to solving the problems
identified by the industrial participants. These participants are
selected through competitive procurements. Furthermore, the
PVMaT project has been specifically structured to ensure that
these PV manufacturing R&D subcontract awards are selected
with no intention of either directing funding toward specific PV
technologies (e.g., amorphoussilicon, polycrystalline thin films,
etc.), or spreading the awards among a number of technologies
(eg. one subcontractin each area). Each associatedsubcontract
under any phase of this project is, and will continue to be,
selected for funding on its own technical and cost merits.
Phase1 of this project, the problem identification phase,
was completed in early 1991. Phase1 competitive bidding was
open to any U.S. firm with existing manufacturing capabilities,
regardless of material or module design. Twenty-two
subcontractsof up to $5OJXJOeach were awarded. Phase2 is
now under way. This is a sohttion phase of the project and
addressesthe problems of specific manufacturers. Subcontracts
for the first Phase 2 solicitation, called Phase 2A, will be
awarded in early FY 1992. Phase 2A was open only to
participants in Phase1. The envisioned subcontractsmay be up
to three years in duration and will be highly cost-shared
between the U.S. government and U.S. industrial participants.
A second,overlapping, and similar process-specificsolicitation
(Phase 2B) is planned to follow soon and will be open io all
U.S. PV manufacturing companies. A third portion of the
project, called Phase 3, is also under way though slightly
behind Phase2. In Phase 3, becauseof the general interest to
industry, some general issues related to PV module
development will be studied through various teaming
arrangements. The PVMaT project’s ultimate goal is to ensure
that U.S. industry retains and extends its world leadership role
in the manufacture and commercial development of PV
componentsand modules. The activities to date have received
outstanding support, and the level of interest in participation is
exceptional.
THE PVMaT PROJECT
A decadeago, U.S. companieshad captured60% of the
international market for PV modules. Today, that share has
dwindled to less than 35%. To help reversethis declining trend
in U.S. competitiveness, DOE initiated the PVMaT project, a
five-year, three-phase,$55million technology transfer program
that is expected to reduce PV manufacturing costs and expand
U.S. production capacity [ 1,2].
This paper will focus mainly on the description of the
PVMaT Project and a description of the work carried out under
Phase 1.
Phase 1
The Phase 1 portion of the PVMaT,project. the problem
identification phase,was completed in early 1991. This work
involved competitive bidding that was open to any U.S. fmn
with existing manufacturing capabilities, regardlessof material
or modnle design. Early in 1991 the competitive selection
process for this phase was completed with 22 subcontracts
being awarded Each of these subcontracts was funded at a
level of up to $50,000 and a duration of three months (see
Table 1). The problems identified by the researchin this phase
representedopportunities for individual industrial participants
to improve their manufacturing processes,reducemanufacturing
costs, increase product performance,and support a scaleup of
These
U.S.-based manufacturing plant capabilities.
opportunities have since been detailed in the approaches
suggestedby theseorganizations for Phase2 research. It is not
anticipated that another Phase l-type solicitation will occur.
The procurement under this phase was only meant to precede
and support the first Phase2 solicitation.
Phase 1 subcontracted research included five
subcontractors working on flat-plate crystalline silicon
technology, ten on flat-plate thin-film modules (one on thin-
Tnbie 1. PVMaT Phase 1 Suhcontrnrtars
Subcontractor
Location
Spire Corporation
Astropower. Inc.
Solarex Corporation
SiemensSolar Industries
WestinghouseElectric Corp.
Silicon Energy Corporation
GlasstechSolar, Inc.
Global Photovoltaic Spec.
Alpha Solarco, Inc.
Photon Energy, Inc.
Energy Conversion Devices
Mobil Solar Energy Corporation
Entech, Inc.
Boeing Aerospace
Solar Kinetics, Inc.
Chronar Corporation
Crystal Systems,Inc.
Iowa Thin Films Technology
Solar Cells, Inc.
Kopin Corporation
Solar Engineering Applications
Spectrolab.Inc.
Bedford, MA
Newark, DE
Rockville. MD
Camiuillo, CA
Pittsburgh, PA
Chatsworth, CA
Golden, CO
Canoga Park, CA
Cincinnati, OH
El Paso,TX
Troy, MI
Billerica, MA
Dallas, TX
Seattle, WA
Dallas, TX
I.awrenceville. NJ
Salem, MA
Ames, IA
Toledo, OH
Taunton, MA
San Jose, CA
Sylmar, CA
will allow for a reduction in slicing costs and an improvement
in their material utilization, thus reducing wafer cost.
Westinghouseis a manufacturerof dendritic web solar
cells. Researchersare working on reducing material and
process costs for the dendritic web process. Westinghouse
rexcarchunder Phase1 of the PVMaT project identified several
problem areas for improving their process. Among these are
(1) increasing the growth rates, average crystal area, and
productive furnace time; (2) using stacked diffusion,
(3) manufacturing larger cells and modules; (4) increasing
module efficiency by 1.4% absolute using surface passivation,
improved antireflective coatings, and cell textutixation; and
(5) optimizing module design using the bifacial conversion
characteristicsof dendritic web solar cells.
Solarex Corp. presently manufactures both
polycrystalline silicon cells and thin-film, single-junction,
amorphous silicon (a-Si) modules. Solarex research under
Phase1 of the PVMaT project identified several problem areas
for improving their polycrystalline silicon process. Among
these are (1) improvements in cell efficiency (from present
12516.5%) through improved control of crystal growth, (2) a
more efficient use of materials via increasedprocessyields, (3)
a reduction in labor through automation, and (4) a reduction in
module materials costs, particularly framing.
The fifth company in this group is Sietnens Solar
Industries. Siemensis presently the world’s largestproducer of
PV with products including single-ctystal silicon and thin-film
copperindium diselenide (CuInSG modules. Siemensresearch
under Phase1 of the PVMaT project identified several problem
areasfor improving their single-crystal silicon process. Among
these are improvements in (I) Czochralski c-Si growth, (2)
existing wafer sawing technology, and (3) cell processing and
module fabrication.
film crystalline silicon, five on amorphoussilicon, and four on
polycrystalline thin films), six on concentratorsystems,and two
working on general equipment/production options.
Flat-Plate Ctvstalline Silicon Modules. Crystalline
silicon (c-Si) is the most common semiconductor material for
PV devices. With Phase1 PVMaT support, five companiesare
detailing the problems of this technology. This group of
crystalline silicon researchorganizations includes Mobil Solar
Energy Corp. of Billerica, Massachusetts;Crystal Systems,Inc.
of Salem, Massachusetts; Westinghouse Electric Corp. of
Pittsburgh,Pennsylvania; Solarex Corp. of Rockville, Maryland;
and SiemensSolar Industries of Camarillo. California.
Flat-Plate. Thin-Film Modules. Modules madeof thinfilm materials have inherent cost advantages,including the use
of less semiconductormaterial and integratedmanufacturing for
cells and modules. However, at present, prices for a-Si
modules are comparableto those of crystalline silicon. Gther
promising thin-film technologies-such as CuInSe, cadmium
telluride (CdTe), thin-film silicon. and gallium arsenide-are
rapidly approaching commercialization.
Mobil Solar is a major manufacturer of c-Si modules
using the edgedefined film-fed growth (EFG) method. Mobil
researchers are presently supplying a 180 kW-ac array to
Photovoltaics for Utility-Scale Applications (PVUSA), a major
demonstrationproject in California. Mobil Solar researchunder
Phase1 of the PVMaT project identified several problem areas
that can be addressedto improve their processand reduce their
silicon material usageby 50%. Among these are (1) reducing
material usage by decreasingwafer thickness to an 8 mil (200
pm) thickness, (2) increasing laser cutting throughput, and
(3) improving the wafer mechanical quality.
Ten U.S. companiesreceived Phase1 support to identify
potential cost reductions for thin-film module manufacturing:
one working in thin-film crystalline silicon. tive that are
concentrating on a-Si, and four that have focused on
polycrystalline thin films.
AstroPower is a manufacturer of (and the only
organization that is focusing on) thin-film crystalline silicon
solar cells. AstmPower researchunder Phase1 of the PVMaT
project identified several problem areas for improving their
process. Among these are (1) improving the production rate
and quality of their proprietary silicon Blm, (2) improving cell
efficiency, (3) reducing material cost, (4) reducing labor cost
Crystal Systems, Inc. is a manufacturer of crystalline
silicon ingots, bars, and wafers. Crystal Systems research
under Phase1 of the PVMaT project identified several problem
areasfor improving their process. Among theseare optimizing
the fixed abrasive slicing technique (FAST) technology. This
9-2
through automation, and (5) increasing production capacity,
thereby reducing the cost per unit.
Additionally,
four organizations focused on
polycrystalline thin films. This group of researchorganizations
includes Siemens Solar Industries of Camarillo, California;
Boeing Aerospace & Electronics of Seattle, Washington;
Photon Energy, Inc. of El Paso,Texas; and Solar Cells, Inc. of
Toledo, Ohio.
The five a-Si researchorganizations include Glasstech
Solar. Inc. of Golden, Colorado; Iowa Thin Films Technologies,
Inc. (ITF) of Ames, Iowa; Energy Conversion Devices (ECD)
of Troy, Michigan; Silicon Energy Corp. of Chatsworth,
California; and Cltronar Corp. of Lawrenceville, New Jersey.
Siemens Solar Industries, as previously stated, is
presently one of the world’s largest producer of PV crystalline
silicon products. They also have a major research project in
thin-film CuInSe, modules for which they have a planned
production of 20-kW in deliverable modules under the PVUSA
project. Siemensresearchunder Phase1 of the PVMaT project
identified several problem areas for improving their CuInSe,
process. Among these are (1) improving materials use
efficiency, and (2) increasing yield and CuInSe, module
efficiency through automation.
GlasstechSolar, Inc. is a manufacturerof single-junction
a-Si photovoltaic modules. They are presently using a
proprietary glass-in/panel-outconcept for the in-line processing
of 40 cm x 120 cm a-Si modules. Glasstech research under
Phase 1 of the PVMaT project identified several problem areas
for improving their process. Among these are (1) the
incorporation of a vertical, double-sided reactor, (2) enhanced
electrode and gas flow designs. (3) improved back contacts and
tin-oxide layers, and (4) improved module designs.
ITF is a producer of monolithic a-Si modules on a
continuous polymer substrateusing automatedprocessing. ITF
research efforts under Phase 1 of the PVMaT project have
identified several problem areas for improving their process.
Among these are (1) developing a roll-to-roll deposition
capability for the a-Si and ZnO layers, (2) developing scteenprintable etching stepsfor the top conducting contact layer, and
(3) reducing the cost of the substratematerial.
ECD presently manufactutes continuous, roll-to-roll.
singk-junction, a-Si a!loy devices. The roll-to-roll process
produces a complete solar cell structure on flexible stainlesssteel webs, 1,000 feet long and 14 inches wide. ECD research
under Phase 1 of the PVMaT project identified several problem
areas for improving their process. Among these are
(1) incorporating narrow-band-gap material to improve
conversion efficiencies and stability, (2) incorporating
proprietary microwave plasmachemical vapor deposition CVD
manufactnting technology for high ptoduction throughput and
higher gas utilization, and (3) reducing the cost of materials and
assembly labor through new product designs and automation.
Silicon Energy Corporation is a manufacturer of
multiple-junction thin-film a-Si photovoltaics and is doing
business as the Utility Power Group (IJPG). The UPG has a
300-kW/year production facility dedicated to internal R&D
activities and various DOE PV projects. UPG researchunder
Phase 1 of the PVMaT project identified several problem areas
for improving their process. Among these are (1) optimizing
the automation of their manufacturing line, (2) improving and
reducing the material required for encapsulation, and (3)
introducing maI-time processing and quality control to their
production line.
Tlte ftfth company in this group is Clrronar Corporation,
a manufacturer of thin-film a-Si PV devices. Their research
under Phase1 of the PVMaT project identified severalproblem
areas for improving their process. Among these ate the
implementation of automation for cost reductions.
Boeing Aerospace is a supplier of thin-film solar cell
manufacturing systemsand a researchorganization working on
the development and scaleup of processing for CuInGaSe,
modules. Boeing researchunder Phase1 of the PVMaT project
identified several problem areas for improving their process.
Among these are (1) uniform large-scale evaporation soutces
for the constituent elementsCu. In, Ga, and Se, (2) a high-yield
processfor depositing ultra-thin cadmium-zinc sulfide (CdZnS)
conformal coatings onto CuInGaSe, films for highquality
heterojunctions,and (3) scaledup aqueousdeposition processes
and novel low-temperature organometallic chemical vapor
deposition techniques.
Two companiesare investigating improvementsto CdTe
technology. The lirst, Photon Energy, is a manufacturer of
CdTe/CdS cells and modules using low-cost, high-throughput,
production-line scalable spray processes. They are presently
scaling up for the production of 20 kW of 4-ftr CdTe modules,
deliverable under the PVUSA project. Photon Energy research
under Phase 1 of the PVMat project identified several problem
areas for improving their process. Among these are (1)
reducing labor cost through processautomation, (2) improving
module efficiency, and (3) reducing material usage through
improved equipment designs.
I
The fourth in the thin-film group, and the secondCdTe
company, is Solar Cells, Inc. (SCI), who is a
development/manufacturing company producing large-area,
thin-film PV modules for utility generating systems. They am
developing the technology for a high-throughput manufacturing
line to produce 60 x 120 cm thin-film CdTe PV modules
deposited by close-spacedsublimation (CSS). SC1 research
under Phase1 of the PVMaT project identifkd severa!problem
areas for improving their process. Among these are (I)
investigating problems associatedwith uniform CSS deposition
of large-areamodules, (2) developing equipment for very-highthroughput deposition, 3) developing patterning techniques for
large-area modules, and (4) developing cost-effective
encapsulation techniques.
(I) optimizing two specific multijunction concentratordesigns
based on discrete GaAs and Ge cells, (2) improving
manufacturing yields, and (3) developing larger, more emcient
MOCVD growth systems.-
Concentrators. Concentrator modules use lenses to
intensify the sunlight falling on banksof small, highly efficient
cells, which reduces semiconductor material costs per unit of
output. R&D issues include optimum cell packaging and
assembly,concentratoroptics, and low-cost tracking arrays and
support structures. Manufacturing cost reductions would occur
primarily through automatedassembly.
The sixth company in this research group, Solar
Kinetics, Inc., is a manufacturer of several crystalline silicon
concentr;ttor systemswith a 200-600 kW/year capacity. Solar
Kinetics research under Phase 1 of the PVMaT project
identified several problem areas for improving their process.
Among these are (1) designing and developing prototype
tooling to demonstrate low-cost injection molding of pointfocus Ftesnel lenses, (2) detailed design of a I-MW
manufacturing plant, and (3) preparing a detailed design for a
5-MW automatedmanufacturing plant for 100-Wconcentrating
PV modules.
The six concentrator companies working in this area
include Alpha Solarco, Inc. of Cincinnati, Ohio; Solar
Engineering Applications (SEAC) of San Jose, California:
Kopin Corporation of Taunton, Massachusetts;Entech, ~IIC.of
Dallas, Texas; Spectrolab,Inc. of Sylmar, California; and Solar
Kinetics, Inc. of Dallas, Texas.
Alpha Solarco, Inc. is a manufacturer of highconcentration PV modules. Researchersare presently installing
a second-generationsystem on the world’s largest automated
two-axis solar tracking structure to produce low-cost electric
power for utilities. Alpha Solarco research under Phase 1 of
the PVMaT project identilied several problem areas for
improving their process. Among these are (1) developing and
testing new manufacturing designs, methods and materials for
cell and module assembly;(2) designing a prototype automated
PV cell assemblyline; and (3) developing production controls
and training proceduresfor a prototype assembly line.
Other Phase 1 Activities. Two Phase 1 participants are
targeting improvements to their commercial lines of
manufacturing equipment. This diversified group includes
Global Photovoltaic Specialists, Inc. (GPS) of Canoga Park,
California; and Spire Corporation of Bedford. Massachusetts.
GPS is a PV company that provides the equipment and
knowledge for building integrated turnkey factories. GPS is
presently investigating the installation of a fully automated.
computer-integratedproduction line in the United States. GPS
researchunder Phase1 of the PVMaT project identified several
problem areas for improving their process. Among these ate
(1) improving the characteristics of the semicrystalline cast
wafers; (2) introducing processessuch as spray-on diffusion
sources,dual antireflection coatings. and ink-jet metallization
printing; and (3) developing and integrating all of the
requirements for full automation using computer integrated
manufacturing.
SEAC is a manufacturer of IOX concentrator systems
using one-sun cells with a 14-MW/yr extruded lens
manufacturing facility in place. SEAC researchunder Phase1
of the PVMaT project identified several problem areas for
improving their process. Among thesearc (1) investigating the
problems associatedwith, and the potential of, adding module
housing sides to lens extrusions, and (2) significantly reducing
labor cost by automating the process.
Kopin Corporation is a manufacturerof high-efficiency
thin-film GaAs concentrator cells with a 22-MW/yr prototype
production capability presently in place. Kopin researchunder
Phase1 of the PVMaT project identified several problem areas
for improving their process. Among these are (1) designing a
tandem cell structure for 1000X concentrators; (2) improving
thin-film cell processmanufacturing; and (3) integrating either
a chemical removal process,called chemical epitaxial liftoff, or
the cleavage of lateral epitaxy for transfer (CLEW process
into their cell fabrication process.
The second company in this group, Spire Corporation,
is a major manufacturer of PV test and module production
equipment. They also presently have small-scale production
activities for high-efficiency silicon modules, and material
researchin compound semiconductorsand a-Si. Spite research
under Phase1 of the PVMaT project identified severalproblem
areas for improving their process. Among these are
(1) developing module processing equipment that will handle
thin (c 200 pm) c-Si wafers, (2) increasing automation, (3)
increasing processing rates, and (4) increasing processing
yields.
Entech, Inc. is a manufacturerof concentratormodules.
Entech is presently producing 3 ft x 10 ft modules using largearea one-sun cells with line-focus Fresnel lenses. Entech’s
researchunder Phase1 of the PVMaT project identified several
problem areas for improving their process. Among these are
(1) working with key vendors to improve the products they
supply to Entech, and (2) dramatically reducing labor cost and
increasing yield by automating the assemblyprocess.
Phase2
Phase2 of the PVMaT project is now under way, with
an expectedduration of five years. It will consist of multiple
competitive procurements over this period, and subcontracts
awarded under any of thesesolicitations may be of up to three
years in duration. The first solicitation under this phase, 2A,
was open only to those organizations that tcccived awards in
the Phase 1 solicitation. The award selection process is just
being completed, and subcontractnegotiations ate expected to
Spcctrolab, Inc. is a leading supplier of GaAs and Si
solar cells and panels to the PV industry. Spectrolab research
under Phase1 of the PVMaT project identified severalproblem
areas for improving their process. Among these are
9-4
be under way soon. This phase of the PVMaT project is
considered the solution phase and will primarily addressthe
process-specific problems that the manufacturers identified
under Phase 1. The subcontractsenvisioned under this phase
are expected to be highly cost-shared between the U.S.
government and U.S. industrial participants. A second,
overlapping, and similar process-speciftcsolicitation is planned
to follow in about a year. Future solicitations under Phase2 of
PVMaT will be open to all organizations.
Therefore,
organixations that .were not ready for the first Phase 2
procurement cycle (that is now in its negotiation stage) will
have another chance to participate in the solution phaseof this
project.
identify the R&D appropriate under Phase 2 of this project.
Under Phase2, selectedcompanieswill develop and implement
solutions to their manufacturing problems. The final selection
of successfulbidders under this phaseis now being cat-tiedout,
and the award of research subcontractsis expected to begin
soon. It is anticipated that as many as six subcontractswill be
awarded under this phase, in which successfulbidders will be
supported for as long as three years as they realize
improvementsto their manufacturing processes.As with most
PVMaT projects, these companies will be expected to costshare their work.
Future activities in the PVMaT project are expected to
inclttde an additional solicitation focusing on company-specific
problems (open to all U.S. firms, including those who weren’t
yet ready for the Phase 1 call for proposals). Additionally, a
Phase3 solicitation is scheduled for releasein early 1992 for
subcontracted teamed research on module-related R&D
problems common to several PV manufacturers. The PVMaT
activities to date have received outstanding support, and the
level of interest in participation is exceptional for this project.
Phase3
There are generic R&D problems in the PV industry that
are relatively common problems to the industry as a whole, a
number of companies, or the design and deployment of PV
systems. The PVMaT project will address these generic
problem areas through a teamed research approach. A
solicitation on this type of generic manufacturing technology is
scheduled for release in early 1992. Participants for this
generic researchmay come from either a consortia of industrial
companies, individual companies, a university or group of
universities, combinations of companies and universities, or
other groups with special capabilities for solving a particular
problem. These researchorganizations will focus on modulerelated R&D problems found to be common to several PV
manufacturers. They will also work in tandem with material
and component manufacturers to help strengthen the PV
industry.
ACKNOWLEDGEMENTS
A number of persons are involved in developing and
implementing this project, including Robert H. (Bud) Annan,
David Hasti, Rick Sellers, Scott Sklar, Donald Schueler, and
Jack Stone.
This work was performedunderContractNo. DE-ACOZ
83CH10093 to the U.S. Departmentof Energy.
REFERENCES
CONCLUSIONS
The long-term goals of the PVMaT project are (1) to
assist U.S. industry in retaining and extending its world
leadership role in manufacturing and commercially developing
PV equipment, components,and systemsand (2) to encourage
the investment of U.S. capital in U.S. PV manufacturing R&D
and large-scaledomestic manufacturing facilities. Phase 1 of
this project has beencompleted,with each company identifying
and developing a specific set of R&D areasthat addresstheir
specific process problems. In I? 1991, a competitive
solicitation was directed toward these Phase 1 participants to
9-5
1.
Witt. C.E.; Herwig, L.O.; and Mitchell, R. “Progress of
the Pltotovoltaic Manufacturing Technology (PVMaT)
Project.” Proceedingsof the 26th Intersocietv Enerav
Conversion Entrineerinp: Conference, August 1991,
Vol. 5, p. 79.
2.
Herwig, L.O.; Witt, C.E. “Photovoltaic Manufacturing
Technology Project (PVMaT) Goals, Plans. and Status.”
Pmceedinas of the Biennial Cottatess of the
International Solar Enerw Societv, August 1991,Vo!. 1,
Part II, p. 709.
THE EFFECT OF MICROSTRUCTURE AND STRAIN IN ln/Cu/Mo/Glass
PRECURSQRSON CdS/Ct&&o PHOTOVOLTAIC DEVICE FABRICATJON BY
SE_LENIZATION
D. Albin, J. Carapella, A. Duda, J. Tuttle, A. Tennant, and R. Noufi
National RenewableEnergy Lahoratory (NREL). Golden, Colorado
(formerly the Solar Energy ResearchInstitute)
and
B. M. Basol
International Solar Electric Technology (ISET), Inglewood, California
ABSTJWCf
Fabrication of CuInSeZ polycrystalline thin films by
selenization can be considered a two-step process: the
fabrication of a “precursor” structure consisting of In and Cu
depositedonto a molybdenum-coatedglass substrate,followed
by a thermal anneal in JJzSe. Deftnite correlations were
observed between the initial state of the precursor, i.e.,
immediately following deposition by thermal evaporation, and
the resulting device Jzrformanceafter selenization. In addition,
the following were also a function of the precursor history-: 1)
melting kinetics of the Jn/Cu layers, 2) air-annealing sensitivity
of the final device. and 3) shifting of the spacecharge region.
Of interest to long-term device reliability was a definite
correlation between the precursor thermal history and
molybdenum film strain as quantitatively measuredby x-ray
diffraction. Final device VW was directly related to the grain
size of selenizedprecursors.
~ODUCTJON
Selenization is one of a variety of processesused to
form CuInSe2 (CJS) thin films for subsequent photovoltaic
CdSKuJnSe2 and/or ZnO/CdS/CuJnSe2device fabrication (I).
EFsentially, the processconsistsof two steps: 1) the fabrication
of a pm-selenization or what we term a “precursor” structure
containing copper, indium, and sometimesgallium (hereafter
referred to as a In/Cu/Mo/Glass structure) deposited onto a
metallic-coated (typically molybdenum) glass substrateand 2)
chemical transformation of this precursor to CuJnSeZ by
subjecting the metallic-alloy stack to a selenium-containing
environment. The first step involves the deposition of indium
onto copper with a slightly Cu-poor stoichiometry of Cu/ln I
0.95 using several techniques, including electrodeposition
(2.3). evaporation (4). and magnetron sputtering (5). Recent
modifications to the processinclude the use of gallium at the
CJS-MOinterface (6) (reportedly for adhesion control), direct
incorporation of elemental selenium into the precursorstructure
(7). and recently, completely inverting the structure, ie.,
Cu/Jn/Mo/Glass,where a thin tellurium layer is usedto promote
In wetting of the MO (8).
10-l
Ubiquitous to the current literature about selenization is
a lack of emphasis relating the initial step of this two-step
process (i.e., the Jn/Cu/Mo/Glass precursor) with final device
performance. In one of the earliest attempts, tokhande and
Hodes (9) showed that indium loss during selenization,
attributed to JnZSetransport, was determined by the type of
indium-containing phasepresent.with CuJn alloys being more
susceptible to this transport mechanism than pure indium.
Subsequently,Dittrich et al. (19) characterizedthe Cu-In phase
behavior for thermally evaporated material as a function of
substratecooling and selenization in elemental selenium vapor
in which selenization was observed to involve the relatively
quick transformation of Cu. In, and Cut tlng to the CuxSe and
JnxSebinaries followed by the diffusion-rate-limited growth of
CIS. In a recent paper by Basal et al. (1 I), a correlation was
suggested between elemental indium (present in
electrodeposited precursors but absent in evaporated
precursors)and large grain growth which subsequentlyresulted
in higher device efficiencies basedon electrodeposited(>iO%)
versus evaporated (-7%) films. Perhaps the most significant
finding reported recendy is that of Eberspacher,et al. (12) in
which a correlation is clearly shown between BIm adhesion
(largely determined at the precursor stage) and CJS module
performance under environmental cycling. This single
observation, more than any other, mandates a thorough
understandingof the tirst stepof this two-step process.
EXPERJMENTAL PROCEDURES
General De&n
Three distinctly different Jn/Cu/Mo/Glass precursors
(labled precursor A, B, and C), regarding the initial degree of
alloying and morphology were preparedat NOEL and subjected
to Jixed selenization and device fabrication conditions. To
insure relevant results, all precursor selenization mentioned in
this paper was performed at JSET under fixed process
conditions conducive to good efficiency devices. Roughly, this
consisted of subjecting the Jn/Cu/Mo/Glass layers to a flowing
Argon and H2Seatmosphereat 40 ‘C for about J h.
Molybdenum-coated (-8500 A thick) soda-lime-silica
(SLS) glass substrates used in this study were supplied by
JSET. In addition, to investigate any effects associated with
improper coefficient of thermal expansion, a, mismatch,
precursorswere also fabricated on Mo-coated (-2.5 urn thick)
Coming 7059 glass substrates.although theselatter precursors
went not selenized.
Device fabrication was conducted at NREL and
consisted of evaporating within the same run, 8OOOA of
intrinsic (highly resistive) CdS followed by 2.5 pm of Jn-doped
(highly conductive) CdS onto the as-received,selcnized films.
Aluminum pads and photolithography then defined six smallarea (0.042 cm2) devices on each selenized film. Device
current-voltage (J-V) characteristicsat AM I .S conditions and
spectralresponsemeasurementswere then performedon these
“as-selenized”precursors.
Parallel to device fabrication, additional Jn/Cu/Mo/Glass
precursorswere analyzedimmediately after the In and Cu layers
were evaporated (hereafter referred to as “as-fabricated), and
following thermal annealsin vacuum at 400°C for various times
(referred to as “as-annealed”precursors).
“Designs” of precursorsA, B, and C are as follows:
Precursor A - representsa precursor in which we attemptedto
minimize the degree of alloying by depositing In onto an
unheatedlarge-grain (I -2 pm) Cu base. Fabrication consisted
of depositing 2fMMlA of CII at 2 A/sonto a MO/Glasssubstrate
heated at 450 “C to promote Cu grain growth followed by
substratecooling to approximately 15°Cbeforedepositing 4600
A In at 4.7 A/s.
PrecursorB - representsa precursor in which we promote some
alloying by depositing Jn onto a smaller grain (looO-20Ot’1A>
Cu base. Fabrication consisted of depositing 2000 A Cu at 2
A/sonto an unheatedMO/Glasssubstratefollowed immediately
by 4600 A of In at 4.7 A/s.
Precursor C - represents a precursor in which we promote
complete alloying by depositing both In and Cu at a substrate
temperatureof 2OB’C; approximately 47” C above the in-rich
eutectic temperature.Film thicknessand rateswere identical to
those above.
X-rav Diffraction Techniaues
Verifying the extent of alloying in precursorsA, B, and
C was initially hindered by a Jack of pubiished phase and
experimental x-ray diffraction (XRD) data regarding this binary
system. The recent phase diagram by Subramanian and
Laughlin (I 3) suggeststhe following possible Jn/Cu precursor
intermetallic phasesat T I400 “C: I) a 6 phase,existing as a
single phase between -28.9 to 30.6 at.% In with nominal
composition of Cu7InJ; 2) a TVphase,existing as a single phase
between -33 to 38 at.% In with approximately five different
temperature-dependent
polymaphs and nominal compositionof
CupIn or possibly CuteJng; and 3) a monoclinic structure,
Cut tlng (45 at.% Jn), existing with limited or no tolerance for
compositional variance.
In contrast to these reported equilibrium phases, the
only intermetallic phasesreported in the Joint Committee on
Powder Diffraction Standards(JCPDS) sets J-38 were CuqJn
(#2-I 188; 20 at.% In), Cugln4 (#2-l 178;. 30.7 at.% In),
Cu7ln4 (#26-522-523; 36.4 at.% In), and a believed Culn
phase (#35-l 150; SOat.% In). Due to this lack of agreement
between equilibria diagrams and XRD data, combined with
literature reporting the presenceof metastablephasesin this and
the similar hg-In system (14, 15) and in thin film couples in
general(16). we compiled our own seriesof theoretical powder
patterns for a number of new phases in this system. These
patterns included the monoclinic Cut tlng (first identined by
Dittrich, et al. (10); tetragonal Culn2 (which was previously
and incorrectly reported by us as an unknown FCC structure
(17) and due to the similarity in processing is probably the
reported Culn phase of Simic, at al.(18)); hexagonal CuZIn;
and triclinic Cu7ln3. Single crystal parameters for these
calculations were obtained from Dr. H. Dittrich of the lnstitut
fur Sonnenenergie und Wasserstoff-Forschung in Stuttgart,
Germany.
XRD measurementswere performed with a Rigaku
Dmax system using step scan parameters of O.OY/step
incrementsand 4 s/stepcounting timesover a 29 rangeof IO0to
, 90’. Beam power was nominally 60 mA at 40 kV with
variations in beamintensity normalized betweensamplesusing
a barepolycrystalline alumina standard.
Initially, phase analysis was performed using the
molybdenum (1 IO) peakas an internal calibration for 28 values
of 545’; however, it becameapparentthat the d-spacing and
width (full width at half maximum net intensity JFWHMJ)
associatedwith this peakdependedon the thermal history of the
precursor. The observed MO peak distortion appears to be
directly related to strain in that layer. As shown at the top of
Figure 1,-shifts and broadeningof the peakcan be explained by
a combination of evenly and unevenly distributed strain, E, in
the film. Also shown in Figure 1 is an example of this effect
when comparing an as-fabricatedprecursordeposited on MocoatedSLS (bottom trace) and a similar precursordepositedon
MO-coated 7059 glass (top trace), both precursors having
identical thermal history. Shifts in the MO peaks were
subsequently measuredrelative to other non-MO paks which
did not exhibit any broadening, i.e.. Cut tlng, and CuJn2.Note
that in the event of film compression,a contraction of planes
oriented perpendicular to a substrate,will, due to the Poisson
effect, result in a dilation of the d-spacings parallel to the
substrate. Consequently, film compression can result in an
indication of plane difarionrather than shrinkage when film
strain is oriented parallel to the substrate.
RESULTS AND DJSCUSSION
XRD Philsr;Behavior and MnarhnlnerEffr&ts
The initial phase make-up (phase type and distribution) and
morphology of the precursor was determined by the process
conditions associatedwith the thermal evaporation process,in
particular the substratetemperatureand deposition rate. The
initial degree of alloying could easily and reproducibly be
controlled. PrecursorA was found to contain elemental Cu. In,
and the nonequilibrium Cult12 phase. Precursor B contained
only CuJn2and Cu. PrecursorC contained Cult Jngand free
10-Z
The lack of alloying betweenIn and Cu is short-lived in
the caseof precursor A. Within a matter of days, all free In in
precursor A will react with Cu to form the Culnz phase such
that the final pattern appearslike that of precursor B. Timedependent effects were not observed in the other precursors.
Phase transformation in precursor A was monitored
quantitatively by measuring the net peak intensity ratio of the
CuIn2 (21 I) and In (101) peaks. A relation between this ratio
and the appropriate mole fraction, F, of In transformed to
CuIn2 was then obtained by theoretically modelling the XRD
patterns of a seriesof combinations, each containing different
amounts of the two phases. In this fashion, differences in xray scattering efficiency between In and CuIn2 were
incorporated into the determination of F (an empirical
calibration curve is impossible given the metastability of the
CuIn2 phase). The resulting calibration was as follows:
Molybdenum
F=
= 7.1 x IO-3 + 0.4092 Y + 0.572 Y2 ,
X(CuIn?)
X(CuIn2) + X(In)
where
41.0
38.0
44.0
Figure I. The fleet of strain, &, on XRD line position (dllo),
and width (FWHM) both theoretically (top figure) and as
observed(bottom)for InKuJMoJGlass precursors.
In. XRD patterns for each of these precursors are shown in
Figure 2. Also shown in this figure is the XRD pattern for a
precursorin which In and Cu were coevaporatedin a 1:1 atomic
flux ratio onto an unheated substrate. Although not selenized,
this case is interesting in that two intermetallics, Culnz and
what we believe to be Cu2In. were formed.
32.0
36.0
I(CuIn~)
I(CuIn2) + I(Indium)
,
(1)
X is the mole fraction, and I is the net peak intensity for the In
(101) and CuIn2 (21 I) peaks. The value of F obtained as a
function of time was then modelled with various solid-state rate
expressions(19) including:
26
28.0
Y=
40.0
28
Figure 2. XRD patterns for as-fabricatedprecursors A, B, and
C. Top pattern is of an as-fahkated precursor not-studied in
this paper but shown as another variation available for possible
selenization.
10-3
l-dimensional diffusion-limited:
F2=(k/rqt
2-dimensional diffusion-limited:
(1-F) In (1-F) + F = (k/r2) t
3-dimensional diffusion-limited:
[l-(l-F)tn]2 = (k/r2) t
2dimensional (shrinking cylinder) reaction-limited:
[I-(I-F)‘J=(k/r)t
3-dimensional (shrinking sphere)reaction-limited:
[I-(I-F)‘B] = (k/r) t
The best fit model corresponded to that of a 3dimensional reaction-limited case. This model as well as that of
the 3-dimensional diffusion-limited case is shown in Figure 3
where it is obvious that the transformation of In+Cu --> CuIn2
is reaction-rate limited. Also shown in this figure for the
reaction-limited case are two sets of data; one set obtained by
continuously scanningover the rangeof 30” to 55” 29 using the
same sample, and one set of data obtained by each time
scanning a new, previously unmeasured sample. The
sensitivity of alloying to x-ray heating dramatically
demonstrates the ease at which this phase transformation
proceeds.
The phasemake-up of asiannealed (30 min. in vacuum
at 400 “C) precursors was similar, regardless of the initial
phase make-up and consisted of CuttIn and In as predicted
by the equilibrium phasediagramexcept that the Cul tIn9 phase
could be formed at lower than indicated temperatures(-1OOV).
Surfacemorphologies of as-fabricated,as-annealed,and
as-selenizedprecursorswere very much different. The
o.otl
1000
.
time
m
2000
’
3000
IO.0
4oou
morphology at eachstagefor each precursoris shown in Figure
4. The nonwetting behavior shown for the as-annealedsamples
appears10correlate linearly with the degree of alloying. The
observed effect was reproducible hut only after long exposure
of the as.fahricntecl precursors to air kforc annealing. which
suggests a connection between oxidation atid lhe beading
behavior. Final selenkd CIS grain size did not vary linearly
with alloying however. Grain growth was largest for :ISselenized precursor R, with slightly smaller grains obtained
with as-selenizedprecursor C. A significant decreasein grain
size was observed for as-selenized precursor A. Grain size
variations cannot be due lo Cu-content. as is the case for
evaporated fibs (20). hecnuseall as-selenizcd compositions
were slightly O-poor.
I-V Device Measurements_
(min)
The best device efficiency, 118, was iniGally mcasumd
on as-selenizedprecursor R (moderately alloyed) followed hy
precursorA (least alloyed) and then precursorC (most nlloycd).
Precursor A
Precursor B
Precursor C
Figure 4. Surface morphologies of as-fabricated (left colrimn; 8000 kX), as-annealed
(middle column; 2000 kX), and as-selenized(right column; 8000 kX),for precrrrsors A (top
row), B (middle row), and C (bottom row).
10-4
Average B% (basedon six small-areadevices per precursor)
was 9.32%. 7.5396, and 6.62% respectively. Corresponding
average values for V,. Jsc, and fill factor were as follows:
precursor B (0.425 V, 33.4 mh/cm*, 65.8%); Precursor A
(0.373 V, 33.1 mA/cm*, 60.9%); and precursor C (0.349,
32. I1 mA/cm2, 59.0%). AI this point, a linear correlation
between the degree of precursor alloying and device
performance is not evident, although the results indicate that
the initial state of the precursor affects device performance.
More noticeable was the stability behavior associatedwith each
device as a function of alloying. Figure 5 tracks V, and q% as
a function of time and annealing. After several weeks,q% had
decreasedfor as-selenizedprecursorsA and B while B% had
increased for devices based on precursor C. These changes
were mainly due to similar changes in V,. with the largest
sensitivity to air anneals exhibited by precursor C, the most
alloyed precursor. In general, devices fabricated by
selenization exhibit excellent stability (12) and the results
shown here are probably indicative of a lack of optimization
between our NREL fabricated precursors and the selenization
processin use at ISET. Nevertheless,after some stability was
attained, device efftciency was directly proportional to alloying:
the more alloyed the starting precursor, the better the device
efftciency. In addition, an increase in grain size (as seen in
Figure 4). resulted in higher Voc: precursor B - 0.424 V;
precursor C - 0.4 14 V; and precursorA - 0.367 V.
SDectralResDonse
A correlation between the degree of alloying and the
spectral response behavior of as-selenixed precursors was
observed in the blue and near-infrared (NIR) regions of the
quantum efficiency vs. energy curves. As shown in Figure 6,
as alloying increased (precursor A to B to C). quantum yield
decreasedin the high energy (blue) region while simultaneously
increasing in the low energy (NIR) region. This behavior
suggestsa shifting of the space-chargeregion away from the
junction towards the Mo/CIS interface. LJnlike the previous
correlations basedon I-V measurements,this behavior should
be less obscured by adhesion and strain effects in the film and
more readily indicates a true dependence of the device
performance on the initial state of the as-fabricated
In/Cu/Mo/Giass precursor.It is unclear why the CIS absorption
edge energy decreaseswith decreasedalloying of the precursor
(shown in the figure insert). One possibility for the lower
energy edgeof as-seleniredprecursorA may be the presenceof
band-tailing resulting from excessivenative defects.
Energy(eV)
Figure 6. Variation of quantum efficiency as a frtnction of
precursor alloying (precursor A - lecut alloyed; precursor R moderately alloyed; precrusor C - most alloyed). The jigtire
insert highlights tire responsenear the bandgapof the CIS.
Film Straia
J
.
*.
Aflsr 30
o-oA,
Orown
min Air
Anneal
.
.
.
.
Atlsr
10
Atlsr
5 days
min Air
Antl*Sl
.
Atier
2 wks
As mentioned previously, significant distortion of the
MO (I IO) diffraction peak (in both FWHM and dttu) was
observed as a function of the precursor thermal history. Peak
distortions were less in precursors fabricated on the ISETsupplied MO-coated SLS glass substrates compared to
precursors fabricated on MO-coated 7059 glass substrates,
partly because of differences in MO layer thickness. Peak
distortion was measuredfor as-fabricated,as-annealed,and asselenizedprecursors(only FWHM measurementsfor the latter),
the results of which are shown in Figure 7. The accuracy of
FWHM measurementswas ~.01” 26 while accuracy for the
(110) strain measurementswas based on an error in dt tu of
approximately fl.o()l A. The (110) peak strain, ~(1to), was
calculatedaccordingto the following eqtation:
.I
Allsr
10
mln Air
AlWlSSl
Figure 5. Variation of small-area &vice eflciency (bottom) and
V,, (top) with time and thermal anneals for precursors A, B,
and C.
10-5
when the SLS subctratea are used. The fatter anomaly is beqt
explained by considering
points 1 and 2 in Figure 7. The
thermal history of each is identical, the only difference king
the glass substrate on which the precursor is fabricated and the
thickness of the Mo layer. As shown simplilied in the model of
Figure 9. the large Q of the SLS glass and resulting pretension of the MO layer when heated lo 2OW’C results in a lower
value of compressive strain (due to the high Q of the alloy) in
the final as-fabricatedprecursor.
(2)
where du”straimd was set at the JCPDS value (#4-809) of 2.225
A and d strained was measured using either CuIn2 (21 1) or
Cul llng (020) peaks as an internal standard.
g
*
‘0
r
3
------.._____
Precursor
A
Precursor
B
x
1
tl
tar
Prec
c
Figure 8. Coefficients of thermal expansion for various
materials plotted as a fitnctinn of their melting temperatwe.
Vauesof Q for the e1ement.r
and CIS were obtainedfrom rejk.
(2 I) and (22) respectively. SLSand 7059glaw Q values were
obtainedfrom technical literature srqplied by the ven&w.
-2.0
I
Precursor
A
Precursor
B
Precursor
c
Figure 7. Molybdenum (1 IO) peak distortion (FIVI1M (top} and
e( 1IO) strain (bottom))for us-fabricated, ns-annealed,and asselenized (SIS substrates only) precursors A, t3, nnd C. The
effect of thermal history on observedstrain is explained in the
textfor points I-6.
The observed Mo peak distortion is undoubtedly due to
the large a mismatch that exists between Cu alloys and the
MO/Glass substrate.This disparity in 01 is shown in Figure 8,
which also includes o! values for a variety of other materials.
Note that as the melting temperature,
I’,, of the material
decreases,there is a gradual increase in 01 due to decreased
interatomic bonding. Consequently, the a of Cu-In alloys may
be even higher than that of Cu (16.6 x I O-6 (“Q-l), because the
alloys have lower T,.
Also note that the a of 7059 glass (4.6
x 10-6 (“C)-1) actually matches the MO layer (5 x 10-h (“Q-l)
better than does the SLS glass (9 x IO-6 (“C)-I),
even though
Figure 7 would seem to indicate that MO films are strained less
10-6
Additional
features of Figure 7 can he explained
in
much the same fashion.
For instance. why does the asannealed precursor B exhibit more strain (larger FWriM) than
the as-fabricated precursor (points 3 and 4) when the reverse is
observed for precursor A (points 5 and 6)? Again the answer
lies not so much in the degree of alloying but rather on the
thermal history of the precursor. The as-fabricated precursor R
is never heated and as such should exhibit the least strain while
the as-annealed precursor R is subjectedlo the strain associated
with cooling the Cu-In alloy, which consequently puts the MO
layer into a highly compressive state. In the cace of the asfabricated precursor A, rhe Cu layer is deposited at 450°C and
remains as a solid during the entire cooling process. The asannealed precursor A however, contains a liquid phase between
the 4WC soak and the In eutectic al 153OC and consequently
imparts less compression to the Mo layer during cooling.
The above discussions are all based on the premise that
struc~um somehow manifests itself
stress in these multi-layer
as strain
in the MO layer alone. At first, this appears
preposterous
since bulk Mo has a much higher Young’s
modulus than either Cu or In and therefore. is less likely to be
the layer to’deform. However, peak broadening has only ken
observed for the MO peaks and never for the alloy or CIS peaks
in our diffraction patterns. This localization of strain in the MO
layer has also been confirmed recently by Nomarski
microscopy observationsof strained MO aggregatestructuresin
CIS depositedon a variety of substrates(22).
Poinl 2
in Fig. 7
hint
1
in Fig. 7
Mo dqxdtd
mlo
Glnss Sllhslma
T=RT(ZT”C)
CU-In
Alloy dcpitcd
a12000C
<
.1.61 1 lo"-03(mmp
.__.. -.. -..-.
. _.
m
Cu.In
Allny
Ea
Moiyhdcnum
Ill-Cu/MdGlaSS
PlCClaSOl
Cmlcd to RT
Tc=153”C
(a H 17 x IO 6-6)
(a = 5 x 10 “-6)
Figure 9. A simple mechanistic model for expluining the
difference in ftlm srrain (quulitativeiy) measuredfor points I
and 2 in Figure 7.
Finally, with regards to the strain found in as-selenized
films, there appearsto be a good correlation betweenfilm strain
(FWHM) measurementsand final device efficiency, i.e., an
increase in the MO (1 Ift) FWHM correlated with higher device
efficiency. It is intuitive that strain will be present in any thin
film layer structure consisting of materials with different a
values, as long as there is good adhesion between the layers,
In the event of poor adhesion, strain relaxation is expected.
Consequently, an indication of strain in as-selenized films is
probably an indication of good adherence although the
magnitude of that strain should be minimized for improved
reliability.
CONCLUDING REMARKS
The initial stateor microstructureof the In/Cu/Mo/Glass
precursor can significantly affect final device performance.
Effects which can largely be attributed IO the initial
microstructure include the final as-selenizedmorphology (and
resulting V, dependence), systematic changes in spectral
response,melting behavior, and air-annealing sensitivity. The
presenceof film strain (and consequently adhesion) must also
take into account the more significant effects associatedwith
processing,in particular, the thermal history of the precursor.
ACKNOWLEDGEMENTS
This work was performed under Contract DE-ACGZ83CfllOO93 to the U.S. Department of Energy. The authors
thank individuals at the Institute of Energy Conversion, in
particular SandeepVerma, Rob Birkmire. Rohert Varrin, Jr.,
and Frasier Russell for their help in corroborating someof the
study results and for general discussions.
REFERENCES
1. K. Zweibel, H.S. Ullal, and R.L. Mitchell, Proc. 21st IEEE
Photovoltaic Specialists’ Cont. 1990, IEEE, New York,
4% (1990).
2. V.K. Kapur, B.M. Basal and ES. Tseng, Solar Cells. 21,
65 (1.987).
3. V.K. Kapur, IJ.V. Choudary, and A.K.P. Chu. 1J.S.
Patent 4.S81,108, (1986).
4. B.M. Basal and V.K. Kapur. ,&I. Phvs. L&. 54 (19),
1918 (1989).
5. J.H. Ermer and R.B. Love, U.S. Patent 4.798.660, 1989.
6. G. Pollock and J. Ermer, European Patent 8930818.3,
(1989).
7. C. Eberspacher,J. Ermer and K. Mitchell, European Patent
89311197.3, (1989).
8. B.M. Basal and V.K. Kapur, U.S. Patent 5.028.274,
(1991).
9. C.D. Lokhande and G. Hades.Solar Cells. 21,215 (1987).
IO. H. Dittrich. U. Prinz, 1. Szot and H.W. Schock, in W.
Palz, G.T. Wrixon and P. Helm (eds.), Proc. Ninth E.C.
Photovoltaic Solar Enerev Conf,, Kluwer. Dordrecht, 163
(1989).
1I .B.M. Basal and V.K. Kapur, m&
Specialists’ Conf,, 1990, IEEE. New York, 546 (1990).
12. C. Eberspacher,J. Ermer, C. Fredric, C. Jensen,R. Gay,
D. Pier, and D. Willett, Proceedinesof the 10th Euronean
Lisbon, Portugal,
P-e,
1991.
13. P.R. Subramanian and D.E. Laughlin, Bulletin of Alloy
PhaseDiagrams. 10 (S), 554 ( 1989).
14. W. Keppner, R. Wesche,T. Klas, J. Voigt and G. Schatz,
Thin Solid Films, 143,201 (1986).
15. Z. Marinkovic and V. Simic, Thin Solid Films, 195. 127
(1991).
16. U. Gosele and K.N. Tu. 1. ADD]. Phvs. 53 (4). 3252
(I 982).
17. D.Albin. G.Mooney. J. Carapella, A. Duda, 1. Tuttle, R.
Matson, and R. Nouti, Solar Cells, 30.41 (1991).
18. V. Simic and Z. Marinkovic, ,!, Les-~,
72,
133 (1980).
19. A. Blazek. Thermal Analvsis, Van Nostrand Reinhold
Company, ~64-65 (1972):
20. D.S. Albin, Ph.D. Dissertation, University Microfilms
international, (1989).
21. R. Weast and M. hstle, eds., mdbook of Chem&rv and
Phvsics. 60th ed. CRC Press,Inc., Boca Raton, (1979).
22. L. Margulis. G. Ifodes, A. Jakubowicz. and D. Cahen. L
&I. Phvs. 66 (8). 3554 (I 989).
FUNDAMENTAL
RESEARCH IN CRYSTALLINE SILICON PHOTOVOLTAIC
PROGRAM PERSPECTIVE
MATERIALS:
Bhushan L. Sopori and John P. Benner
National Renewable Energy Laboratory
(formerly the Solar Energy Research Institute)
Golden, Colorado 80401
ABSTRACT
This paper identifies the directions and goals for future
research in crystalline silicon materials for high-efficiency
solar cells. Currently. the substrates produced by technologies
of commercial interest have lower chemical purity and crystal
perfection than those of microelectronics-grade wafers. The
material quality of these substrates can be improved by
post-growth treatments that can be incorporated in the cell
fabrication process schedule. Such treatments, applicable to
low-cost single- and polycrystalline substrates, include
impurity gettering, defect passivation, and defect annihilation.
Development of these processesrequires a better understanding
of point-defect phenomena involving interactions of defects
and impurities. The future program must also include research
toward the growth of substrates in thin film configurations
which are mom tolerant of less-than-ideal crystalline quality
and can exploit light trapping and passivation of surface and
bulk defects; The research objectives for achieving DOE goals
for the cost effectiveness of silicon solar cells are defined.
BACKGROUND
Low-cost crystalline silicon substrates, obtained from
ingots or grown in sheet (ribbon) form, ate now extensively
used for the commercial production of solar cells. The
efficiencies of these cells typically range between 12% and
14% under one-sun illumination. Higher efficiencies obtained
in the laboratory have been a result of applying a host of
processes that are not well understood. A considerable effort
in fundamental research is necessary to understand these
complex processes to a degree where they can be included in
a commercial cell processing schedule. Results to date have
shown that a variety of phenomena can be exploited to
improve the quality of low-cost substrates and, hence, yield
cell efficiencies of 16% to 18%. These processes include
gettering, defect passivation, and defect annihilation. Impurity
gettering by phosphorous diffusions or aluminum alloying has
been proposed to improve the performance of polycrystalline
silicon solar cells (1,2). Hydrogenation by techniques such as
radio
frequency (RF) plasma or low-energy ion implantation
have been applied to produce passivation of crystal defects and
of some impurities. The passivation effect is more pronounced
for lower-performance cells, typically when the cell efficiency
is less than 12%13% (3.4). A significant reduction in the
defect density can be brought about by suitable hightemperature annealing processes ($6). Although, the basic
approaches of post-growth treatments are known, the details of
such complex mechanisms are not well understood. It is
recognized that these processes can be strongly influenced by
point defect phenomena. It is the purpose of the NREL silicon
research program to develop such an understanding to a degree
that can allow these processes to be incorporated into celi
fabrication.
Other approaches that can lead to higher-performance cells
involve improvements in the crystal growth techniques. Recent
studies suggest two cost-effective approaches in crystat growth
that offer promise for producing high-efficiency substrates.
First, improvements in magnetic Czochralski (MCZ) and
Czochralski (CZ) growth may reduce the need for post-growth
treatments for material enhancement. Initial evaluations of
MCZ wafers indicate that these substrates can be processed to
fabricate cells with efficiencies comparable to those on float
zone material (7.8). It has also been proposed that MC2
material may be produced at costs comparable to that of CZ
material. Likewise, there are some renewed efforts to optimize
CZ growth for growing solarquality silicon. Further studies
are needed to characterize these substrates and evaluate their
potential for high-efficiency celis. Second, new growth
technologies may be developed to produce crystaIIine silicon
in configurations compatible with advanced thin-cell designs.
It has been recognized that by providing effective light
trapping and surface passivation these devices can yield cell
efficiencies in the 16%-l 8% range with the material quality of
the current polycrystalline substrates. However, the growth of
silicon films with thicknesses of about 50 pm may require a
supporting substrate to facilitate handling during film growth
and in subsequent cell processing steps. Clearly, the choice of
such a substrate depends on the role of the substrate in the
overall cell design. In general, such a composite film/substrate
structure must satisfy many compatibility requirements in
addition to being cost effective.
There are other possibilities, including the growth of the
film on a temporary substrate followed by transfering the film
to a different substrate to act as a support for the solar cell (9).
In either case, the presence of the substrate can influence the
quality of the film material and the performance of these
devices.
11-l
.
RESEARCH DIRECTION
.
.
The NREL crystalline silicon materials research program
will foster a strong interaction with the photovoltaic industry
and assist the industry by carrying out fundamental research on
the following crucial topics:
Novel Concerts for the Growth of Thin Silicon Films for High
Efficiency Solar Cells
Detailed Characterization of Photovoltaic Silicon Substrates
It is expected that the research will address thin-film
growth for high-efficiency cell designs using novel approaches.
Of particular interest are thin silicon films in configurations
that can be compatible with optical confinement, surface
passivation, and hydrogenation for defect passivation. Topics
under this research task include
It is clear that future research will require an in-depth
analysis of the characteristics of commercial substrates, as well
as of material grown by improved CZ and MCZ methods.
Such analyses will include the characterization of defects,
impurities, residual stresses, and other pertinent information
that can be deemed essential in controlling solar cell
performance. Various aspects of substrate characterization are
.
l
.
.
.
.
.
The spatial variation of the defects
The nature of the defects in the bulk of the material
A quantitative determination of the influence of defects
and impurities on solar cell performance
A determination of the concentration of various impurities
in the substrate.
.
Basic ProDetties Related to Imuuritv Diffusion and
Imouritv-Defect Interactions in Silicon
.
.
.
.
.
These aspects of research are expected to benefit future
development work by strengthening the understanding of the
basic mechanisms that affect processesthat can (i) enhance the
quality of the low-cost substrates and (ii) provide alternate
novel methods for growing thin-film silicon for high-efficiency
solar cells. This research is expected to address the basic
issues that can be applied for post-growth quality enhancement
of low-cost silicon, and it addresses basic research into the
issues that influence the quality of silicon growth in the novel
thin-film regime. It is expected that high-efficiency solar cells
can be fabricated by incorporating such processes in the
commercial scale, thus assisting in meeting the goals of the
Department of Energy’s Photovoltaics Program.
The diffusivity of silicon self interstitials and vacancies
The diffusivity of hydrogen in substrates containing
crystal defects
Gettering and the release of impurities by defects
An enhanced diffusion of hydrogen by point defect
injection
Defect annihilation by point defect injection
Gettering aided by point defect injection
An improvement in the minority carrier lifetime by
thermal processes involving point defect injection.
NREL is now evaluating of proposals submitted under a
solicitation for research in these areas. The resulting research
subcontracts are anticipated to be awarded by February 1992.
This work was performed under Contract No. DE-ACOZ83CHlOO!?3 to the U.S. Department of Energy.
Develoument of Post-Growth Oualitv Enhancement Techniaues
Improving of the material quality of photovoltaic silicon is
necessary to fabricate cells of efftciencies in the 164-l 8%
range in the near term. Various mechanisms that can be
exploited are
.
Impurity gettering using extrinsic as well as the intrinsic
approaches
Novel thin-film configurations for silicon solar cells
Novel growth concepts
Film substrate interface effects on impurity diffusion into
the film
The characterization of the photovoltaic properties of thin
silicon films.
SUMMARY
Some concepts of solar cell processing for improved cell
performance can benefit from a knowledge of the basic
Of particular
properties of point defects and impurities.
interest are processes involving point defect injection or
extraction, which can significantly alter the diffusion of a
variety of impurities in the silicon lattice. Consequently, it is
important to investigate processes in which point defect
phenomena can be applied to enhance the materials
characteristics during the device fabrication processes. Such
processes, which must be compatible with standard solar cell
fabrication schedules, include
.
.
Defect passivation by methods such as hydrogen
implantation
Defect annihilation using thermal annealing
Techniques for improving the minority carrier lifetime.
11-2
REFERENCES
I.
2.
3.
4.
5.
6.
7.
8.
9.
S. Narayan, S.R. Wenham and M.A. Green, A~pl. Phvs.
J&t&. 48, 1986, p. 873.
S. Martinuzzi, H. El Ghitani, D. Sarti, and P. Torchio,
Conference Record of the 20th IEEE Photovoltaics
bcialists Conference, 1988, p. 1575.
J.W. Corbett. J.L. Lindstrom, S.J. Pearton, and A.J.
Tavendale, Solar Cells. 24, 1988, p. 127.
H. Yagi, Conference Record of the 21st IEEE
Photovoltaics Suecialists Conference, 1988, p 1600.
B.L. Sopori, J. Benner, and J.D. McBrayer, 21st IEEE
PVSC, 1990, ~653.
L.A. Verhoef, S. Roorda, W.C. Sinke, and R.J.C. Van
Zoligen, Conference Record of the 20th IEEE
Photovoltaics Soecialists Conference, 1988, p. 1551.
T. Higuchi, ADD]. Phvs. Lett., 53, 1988, p. 1850.
S. K. Pang, J. Electrochem. Sot., 137, 1990, p. 1977.
R.P. Gale,
Conference Record of the 20th IEEE
Photovoltaics SDecialists Conference, 1988, p. 446.
11-3
HYDROGEN
IN SILICON:
DIFFUSION
AND DEFECT PASSIVATION
BhushauL. Sopori, Kim M. Jones,XiaoJun Deng, R. Matson, M. AI-Jassim and S. Tsuo
National Renewable Energy Laboratory
(formerly the Solar Energy Research Institute)
GoIden, Colorado
Alan Doolittle and A. Rohatgi
Georgia Institute of Technology
Atlanta, Georgia
in silicon, passivation mechanisms and undesirable effects of
hydrogen in silicon. In order to take full advantage of
hydrogenation, it is important to know several aspects of
hydrogen in silicon including
ABSTRACT
This paper discusses the nsults of our studies on hydrogen
diffusion and the passivation of crystal defects and impurities
in single and polycrystalline silicon obtained from several
different vendors. We show that enhanced diffusion of
hydrogen can occur in some of these materials, both in the
bulk and along grain boundaries, with an effective diffusivity
of about an order of magnitude higher than previously reported
;values. Hydrugen incorporated for defect passivation can
induce defects in silicon. We discuss these defects and their
recombination characieristics, and propose that these defects
pose the ultimate limit on the degne of improvement
manifested by a cell. The observed behavior of hydrogen
plays an important role for defect passivation in solar cells and
can be explained on the basis of point defect interactions with
hydrogen. We describe a back-side hydrogenation technique
for solar cell passivation that takes advantage of the enhanced
diffusion mechanism and circumvents many drawbacks of the
front-side hydrogenation.
.
.
.
how hydrogen can be most effectively and efficiently
introduced in solar cells and in a state that is most
beneficial
the effects of impurities in siticon on hydrogen
diffusion and defect passivation
the influence of hydrogen-induced defects on solar cell
performance
In this paper, we report results of our studies on hydrogen
diffusion and defect passivation in single and polycrystalline
silicon, aimed at understanding the above-mentioned issues.
We show, for the first time, that the bulk diffusivity of
hydrogen in some polycrystalline silicon can be higher than
that in float zone (FZ) wafers. This result is contrary to the
general belief that the diffusivity of hydrogen is highest in FZ
material. We will discuss probable explanations of this effect,
and its application for solar cell passivation. We will also
describe hydrogen-related defects. In particular we will identify
the difference between defects caused by the surface damage
due to ion momentum, and those induced by high
concentrations of hydrogen in the silicon lattice. Again, we
will show for the first time that hydrogen-induced defects can
produce deleterious effects on the cell performance, thus,
limiting the effectiveness of hydrogen passivation.
._~
EXPERIMEN’iAL
PROCEDURES
INTRODUCTION
Hydrogen in silicon is gaining increasing technological
importance because of its ability to passivate defects both at
the interfaces ( e.g Si-SiO,> and in the bulk. However, the
behavior of hydrogen in silicon is quite complex because
hydrogen can interact with the silicon lattice, with the crystal
defects as well as with impurities in silicon. The syn&rgistic
effects arising from these activities are only beginning to be
understood. However, it is now known that hydrogen can
interact with the shallow impurities such as B, Ga, and Al to
deactivate their acceptor behavior, passivate dangling bondi
arising from lattice discontinuities such as surfaces or at the
crystal defects, and passivate deep levels due to metallic
impurities (l-9). The photovoltaic community has primarily
focussed on the bulk defect passivation aspect of hydrogen in
silicon, which, in itself, has been largely on a qualitative basis.
It is well established that hydrogen can passivate some solar
cells as manifested by improving their performance, but it is
not known why the improvement occurs only for low
performing devices. Furthermore, there is insufficient and
conflicting information on the basic parameters such as the
difisivity of hydmgen and its dependence on the impurities
Silicon substrates were obtained from different vendors and
included single crystal Czochralski (CZ) and FZ,
polycrystalline ribbons. and cast wafers. In some special
cases, samples grown by the same technique, but with different
impurities, were obtained. Samples were characterized prior
to hydrogenation in terms of carbon and oxygen
concentrations, xesistivity, and nature of defects and grain
boundaries. All samples were slurry polished in a way that
leaves no surface damage, and then hydrogenated in a
Kaufman ion system. The samples were implanted typically
at 250 “C with ion energies between 0.5 and 2 keV. Cun-ent
densities ranged between 0.2 and 0.6 mAlcm2 with the
resultant flux densities in the range of 5 x 10’7/cm2. The
implanted samples were analyzed as follows: planar and crosssectional transmission electron microscopy (TEM and XTEM)
12-1
characterization of these detects has been done (lo,! I). we
find that these defects are in a disc-like configuration and lie
in ( 111) planes and are elongated along [I IO] directions.
Figure 2a shows a (001) planar view TEM image of such a
platelet. It is seen that such a platelet exhibits a contrast
identifying a core-like structure associated with the defect.
Figure 2b is a lattice image of such a shuctun?, indicating a
loss of contrast associated with the defect core. We believe
that the co? represents the entrapment of hydrogen or an
aggregate of vacancies or both. As discussed in a later
section, the region of surface defects has ve
high
concentrations of hydrogen, typically from 1019to 1dr ’ cm3,
which is considerably higher than the solubility limit of
hydrogen in silicon. It is thus conceivable that the hydrogen
trapped at the defect core could be molecular in nature. The
platelets are seen to extend deep into the bulk of the material
(deeper than the surface damage described above). The
tendency of the platelet formation appears to be related to the
impurities in silicon. For example, low-oxygen/low-carbon
materials have a higher tendency to generate platelets. At this
time, we do not have sufficient data to determine if the
propensity for platelet formation depends on the crystal
orientation. It should be pointed out that under the same
conditions of implantation, the samples having lower
concentration of oxygen exhibit deeper penetration of defects.
This can be seen from a comparison of Figures la and lb; the
oxygen concentrations of these samples am 5 ppma and 20
ppma, with the penetration depth of platelets - 2 and - 0.6
pm, respectively.
to anaiyze the defects induced by hydrogenation and to
determine the depth of hydrogen diffusion as described later in
the paper; Fourier transform infrared spectroscopic (FTIR)
analysis was done to determine interactions of hydrogen with
silicon and boron; secondary ion mass spectroscopy (SIMS) for
hydrogen and boron profiling; scanning ellipsometery to
characterize hydrogenated surfaces: resistivity measurements;
defect analysis by chemical etching; and electron beam
induced current (EBIC) analysis. Similar experiments were
repeated with deuterium, particularly for SIMS profiling; this
technique has higher sensitivity for deuterium, thus, it has the
capability to measure deeper profiles.
In addition, we have used a new approach to determine,
semiquantitatively, deep hydrogen profiles. In this method
samples are selected to have bulk dislocations so that the
implanted hydrogen can segregate at the defect sites. The
probability of segregation is related to the hydrogen
concentration. The segregated hydrogen can then be observed
under a TEM. Therefore, this method allows us to determine
semiquantitative depth profiles of hydrogen deep inside the
bulk of the substrate.
RESULTS AND DISCUSSION
The results of our experiments will be described in the
following sections.
1.
Hydrogen defects
In addition to producing its own defects, hydrogen
interacts with extended defects, such as dislocations, in the
material. TEM analysis shows that hydrogen can segregate at
dislocation sites, as seem in an XTEM photograph of Figure
3. Qualitatively, we have observed that hydrogen segregation
occurs mainly at dislocation nodes and is more pronounced at
“clean” dislocations. However, no segre ation is observed if
the hydrogen concentration is below 101%
/cm’.
We have analyzed hydrogen defects at the implanted
surface as well as in the bulk. Defects due to hydrogen
implanted at low energies in silicon may be divided into three
categories: surface damage, defects extending into the bulk,
and defects due to hydrogen interaction with the extended
defects such as dislocations. The near-surface damage is
believed to he caused by the combined effects of energetic
ions and high surface concentrations of hydrogen. Figure la
is a cross-sectional transmission electron microscope (XTEM)
micrograph of a sample showing typical structure of the
defects due to surface damage caused by implantation at 1.5
keV for 30 min. These defects appear as dislocation loops,
stacking faults, and hydrogen entrapment (shown by the mow
in the figure). FTIR analyses of h drogenated samples show
absorption peaks around 2100 cm- 7 due to Si-H coordination.
However, after polishing off 0.5 pm from the surface to
remove the damaged layer, the absorption peaks are strongly
diminished. This analysis also shows that optically active
hydrogen is predominantly confined to the near-surface
damaged region. The depth of the surface damage clearly
depends on the ion energy; the higher ion energy results in
deeper damage.
An important issue related to hydrogen defects is to
identify whether the hydrogen defects have a significant effect
on the performance of the cell. Our EBIC studies have shown
that the hydrogen-induced surface damage results in a high
surface recombination velocity, indicating that hydrogen does
not passivate self-induced defects. These results can be
derived from Figure 4, which shows EBIC signals on a
Schottky diode. The Schottky diode was fabricated on a
region in which a part of the surface was implanted and the
rest was masked during the implant. The signals a,b, and c,
shown in Figure 4. are taken at electron beam energies of 10.
20, and 30 keV, respectively (at different amplifications). It
is seen that at 10 keV the relative EBIC signal is lower on the
hydrogenated side than on the masked side. As the energy is
increased the signal from the hydrogenated side increases
relative to the masked side. These results indicate a high
surface recombination at the hydrogenated surface but a
passivation below the surface. Likewise, platelets can be
The near-surface region also shows a preponderance of
“platelets”. Figure lb is an XTEM micrograph of another
sample implanted under the same conditions as those in Figure
la, showing “platelets” extending into the bulk of the material.
These defects have been postulated earlier and a limited
12-2
‘imaged by EBIC contrast, thus identifying them to be high
carrier recombination regions. We have not been able to
effect of hydrogen segregation at the
determine the
dislocations.
These results have two important implications: (i) In a
conventional front-side hydrogenation, the formation of similar
defects is expected. Being close to the junction, these defects
are expected to have a pronounced compensatory effect on the
solar cell performance. Clearly, this effect can be minim&d
if hydrogen for passivation is intmduced from the back side of
the ceil. (ii) Defects induced in the bulk of the cell can limit
the improvement in the minority carrier diffusion length. We
believe that these effects, mlated to the defects induced by
hydrogen, are one of the limiting mechanisms of the
effectiveness of the hydrogenation process.
2.
We have studied hydrogen diffusion in the bulk as well as
along grain boundaries. A comparison of hydrogen diffusion
in different materials of about the same resistivity ihows a
unique featurt: the hydrogen diffusion can be very deep in
some polycrystalline wafers as compared to that of highquality FZ wafers of the same resistivity. Figure 5a shows a
SIMS profile of hydrogen implanted at 1.5 Kev in a single
crystal FZ wafer; the profile is similar to that published in the
literature (1). Figure 5b is a corresponding SIMS profile of a
silicon ribbon of the same tesistivity and implanted in the
same run. It is seen that the diffusion profile in the ribbon is
extended deeper into bulk (data taken within one grain). Due
to limitations in the SIMS measurement of hydrogen, we have
carried out an extensive analysis using deuterium implantation.
These measurements also showed that deeper diffusion of
deuterium occurs in the ribbon samples as compared to the FZ
and CZ wafers. Depth of diffusion within large grains was
found to be the same or even man than along the grain
boundaries (determined as described below).
Hydrogen Diffusion
The mechanisms of hydrogen diffusion play an important
role in understanding passivation of solar cells as well as
developing production-compatible processes that can employ
hydrogenation for imprbving solar cell efficiency.
It is
expected that, solar cells being bulk devices, would require
hydrogen to enter the entire bulk for an effective passivation.
One can determine the typical duration of a hydrogenation
process from the data on hydrogen diffusivity available in
literature. Unfortunately, the= is an enormous scatter in the
.experhnental values of the diffusivity of hydrogen. in silicon.
There are a number of reasons for the large variations in these
values, which are related to the fact that hydrogen diffusivity
depends on a number of parameters:
.
Substrate parameters such as Esistivity, type of dopant,
and presence of other impurities;
.
Surface concentration of hydrogen, the temperature
during hydrogen diffusion, and the method of
hydrogenation.
Previous results have shown that diffusivity of hydrogen
is higher in FZ silicon than in CZ when doped with the same
dopant of the same concentration. It is also found that
hydrogen diffuses faster in some grain boundaries; however, to
date there is no suitable explanation for this observation (12).
It is generally accepted that hydrogen diffusivity
at
temperatures below 500°C follows:
Dt = 4.2 x 10” exp ( -
The most convincing evidence of deep diffusions was
obtained by exploiting hydrogen segregation at the dislocation
sites as a semiquantitative detector for hydrogen depth profiles.
Selected samples were examined in cross-section with ‘I’EM
to determine the degree of hydrogen segregation. Figure 6 is
an XTEM micrograph of a ribbon sample showing hydrogen
segregation manifested as “bubble-like” structures associated
with dislocations. The sample was implanted at 1.5 keV at
250 ‘C! for 30 min. A decreasing concentration of hydrogen
is manifested as a reduction in the number and size of
hydrogen “bubbles” below the surface. The segregation
characteristics depicted in Figure 6 represent an approximate
hydrogen concentration of 10”/cm3 at a depth of 20 pm. In
comparison, a similar concentration of hydrogen is reached at
a depth of about 4 pm below the surface in FZ wafers of the
same resistivity. This implies that the diffusivity of hydrogen
in the ribbon sample is enhanced by a factor of 25. We
believe that this is the first observation indicating a higher
diffusivity of hydrogen in the bulk of a polycrystalline silicon
than in the FZ wafers. Although the exact mechanism of this
enhanced diffusion is not well understood at this time, we
believe that it is similar to that which causes enhanced
diffusion along some grain boundaries or dislocations. It is,
however, important to recognize that the diffusion shown in
Figure 6 is a bulk diffusion.
In order to understand some similarities and differences
between the enhanced bulk diffusion and the grain boundary
diffusion, we have also investigated hydrogen along grain
boundaries. In general, there are many common features in the
enhanced bulk diffusion and the grain boundary diffusion.
These include segregation effects, impurity dependence, and
oxygen related effects. The effect of oxygen on the diffusion
of hydrogen along a grain boundary may be seen in Figures
7a and 7b. These figures an XTEM micrographs showing
hydrogen segregation along the grain boundaries of two
different substrates implanted under the same conditions. The
oxygen concentrations of the materials in Figures 7a and 7b
are: <l ppma and 20 ppma, respectively. A deeper diffusion
0.56 eV
-__-_-__) cm2 .S-’
kT
This typically yields a vaIue of e 10“’ cm2.S’
for
hydrogen diffusivity in silicon (13) . Using this value for
diffusivity one can determine the typical duration of a
hydrogenation pmcess to be several hours, which is clearly not
acceptable for a commercial process.
12-3
is seen in Figure ?a as compared to that in Figure 7b. This
result is similar to the deeper bulk diffusion occurring in FZ
than in CZ wafers. From this type of behavior one may
conclude that hydrogen diffusion is not purely interstitial and
that a vacancy mechanism is involved in the enhanced
diffusion.
The mechanism of enhanced diffusion is clearly important
for solar cell applications since solar cells are bulk devices and
we are striving to diffuse hydrogen through the entire thickness
of the wafer. Some of the unique features of hydrogen
diffusion observed in this study are:
.
.
.
Hydrogen diffusion in the bulk appears to be strongly
enhanced (compared to a single crystal silicon) in some
substrates, with a diffusivity approximately an order of
magnitude higher than published value. In some cases
hydrogen decoration in XTRM was observed more than
50 pm from the hydrogenated surface. This behavior
of hydrogen can be explained on the basis of point
defect interaction with hydrogen and is clearly
important in making hydrogenation a commercial
process.
Oxygen in silicon appears to retard diffusion of
hydrogen. Low temperature FJlR
spectra of
hydrogenated samples of the same resistivity
(implanted under the same conditions) show inverse
correlation between hydrogen and oxygen peak
amplitudes.
Correlation between hydrogen diffusion and carbon
concentration was not observed.
Our data show that in some polycrystalline substrates the
enhanced diffusivity will favor the hydrogenation process by
reducing the implantation times to be only about 15 min. for
hydrogen to penetrate through the entire wafer.
3.
State of hvdroeen
We have determined that a significantly large fraction of
implanted hydrogen stays in an electrically and optically
inactive state. As described in the Hydrogen Defects section,
the IR absorption peaks related to Si-H coordination were not
observed below a certain depth, even though hydrogen was
present in fairly high concentrations (as observed by
decoration). Likewise, we do not observe any measurable
change in the resistivity after hydrogenation. This observation
implies that boron deactivation is not significant. during our
process. We have, however, observed an new phenomenon of
out-diffusion of boron during hydrogenation. Such an outdiffusion occurs from a thin region near the surface. Figure 8
is a boron concentration profile, measured by SIMS, of the
sample whose hydrogen profile is shown in the Figure 5b. It
indicates a boron out-diffusion depth of about 0.5 pm. Deeper
out-diffusions have been observed in some other substrates. In
general, there is a good correlation between the hydrogen
diffusion depth and the out-diffusion depth of boron. Since
boron diffusion is assisted by interstitials and vacancies, boron
out-diffusion may be explained by invoking point defect
supersaturation near the surface. This result is clearly
important for front-side hydrogenation becausea change in the
boron profile near the junction can have a significant effect on
the dark characteristics of a solar cell. It should he pointed
out that formation of a surface inversion due to hydrogenation
has been reported previously; however, it has been thought to
be a manifestation of the donor type of behavior due to surface
damage (14). Hem, we propose boron out-diffusion to be a
factor contributing to the surface inversion.
4.
Defect Passivation
Although the mechanisms of passivation are not well
understood, it is known that hydrogen can be present in silicon
in a variety of states, and thus capable of many different
interactions (15.16). It is clear from our results that hydrogen
also interacts with point defects, presumably forming a highly
mobile complex. We believe that hydrogen associated with
defects and impurities as a passivating species is in dynamic
equilibrium with the total hydrogen content. If so, then in a
hydrogen passivation process, it is sufficient to ensure that the
minimum concentration of hydrogen in the bulk of the cell is
above that of the density of states due to defects and
impurities.
BACK-SIDE
HYDROGENATION
TECHNIQUE
It is clear that the influence of hydrogen-induced damage
and the dopant out-diffusion can be minimized if hydrogen is
implanted from the back side of the cell. However, back-side
hydrogenation requires that hydrogen diffuse rapidly through
the thickness of the cell, typically 300 pm, in order to be
effective in improving the junction properties. Such a
technique has many other advantages that can make it well
suited as a production-compatible process (17).
In a
commercial cell fabrication schedule back-side hydrogenation
may be implemented prior to metallization, making the entire
back side available for hydrogenation. Various steps needed
for such a process are illustrated in Figure 9. The solar cell is
implanted from the back side and then coated with a thin layer
of aluminum, typically about 2ooO thick. The aluminum is
then alloyed in an optically heated furnace. similar to a Rapid
Thermal Anneal (RTA) process. This step drives hydrogen
deeper into the cell and also dissolves the damaged region to
produce a Si-AI alloy to form an ohmic contact. In addition,
it compensates for the out-diffusion of boron discussed in the
Figure 10 shows the tffect of such a
previous section.
hydrogenation process on a row of solar cells. The figure
shows the open circuit photovoltage of the devices before
(solid line) and after (dotted line) the hydrogenation process.
A
The back-side hydrogenation can also be readily applied
to finished solar cells, provided the back-side metallization is
in a gridded configuration to allow access for hydrogen to
enter silicon through open areas. Figure 11 shows the effect
of hydrogenating such a finished cell on the spectral response
of the cell; for comparison, the other parameters of the cell,
12-4
before and after the hydrogenation, are also indicated. From
this figure, it is seen that the impro+ement in the cell response
is primarily due to an increase in the long-wavelength
response, indicating an increase in the minority carrier
diffusion length due to the passivating effect of hydrogen.
Experiments on defect passivation in solar cells (using our
back-side hydrogenation process on finished cells) indicate that
the increase in celf perfbrmance due to hydrogenation also
depends on oxygen concentration. Cells containing high
concentrations of oxygen (typically above 20 ppma) do not
show significant effect of passivation. Results to date also
indicate that hydrogen-induced defects could possibly set a
limit on the degree of passivation in a solar cell. Clearly, in
high efficiency cells one may expect that hydrogen-induced
defects may even lower the cell performance.
Our experiments have shown that impurity/defect
passivation in silicon solar cells is possible even for cells with
initial efficiencies greater than 12% if the substrate has low
oxygen content, typically less than 20 ppma. We also suggest
that hydrogenation could be applicable to single crystal solar
cells if deep diffusions can be performed, provided that the
iriitial cell performance is limited by the impurities/defects.
during a process that is required for solar cell fabrication. in
particular, we believe that processes that produce a
supersaturation of inter&i&
at the surface and a vacancy
supersaturation in the bulk is likely to produce an enhanced
diffusion of hydrogen and an out-diffusion of boron.
However, such a process should also occur at relatively low
temperatures where the passivation mechanisms are stable.
These arguments clearly explain passivation observed in nitride
deposition using Plasma Enhanced Chemical Vapor Deposition.
Acknowledgement:
The authors are very grateful to John
Benner of NREL for his support and interest in this project.
This work was supported by the U. S. Department of Energy
under Contract No. DE-AC02-83CHlOO93
5.
References
1.
S. J. Pearton,I. W. Corbet$and T. S. Shi. Appl. Phys. A43.
153 (1987)
See pertinent articles on hydrogen in silicon in Oxygen,
Carbon, Hydrogen and Nitrogen in Cfystailine Silicon, eds.
J.C. Mikkelsen, Jr., S. J. Pearton. J. W. Corbett and S. J.
Pennycook,MRS Plttsburg, 1986.
J. I. Pankov, D. E. Carlson, J. E. Berkeyheiser and R. 0.
Wance,Phys. Rev. L&t., a, 2224 (1983).
C. T. Sah, J. Y. Sun and J. J. Tzou, Appl. Phys. Lett., G
204(1983).
M. Stavola. S. J. Pearton.J. Lopata and W. C. DautremontSmith, Appl. Phys. Lett., 3.1086 (1987).
S. Martinuzzi,M. A. Sebbar and J. Gervais, Appl. Phys.
Lett., 47, 376(1985).
T. Zundel. A. Mesli. J. C. Muller and P. Siffert, Appl.
Phys. A 4& 31(1989).
N. M. Johnson, C. Herring and D. J. Chadi, F’hys. Rev.
L&t., & 769 (1986).
A. J. Tavendale and S. J. Pearton, J. Phys. C16, 1665
(1983).
S. J. Jeng, G. S. GeMein and G. J. Scilla. Appl. Phys. la,
$$1735(1988).
N. M. Johnson, F. A. Ponce, R. A. Street and R. J.
Nemanich. Phys. Rev. B.S.4 166(1987).
L. Kazmexsld,Proc. 18th IBBB PVSC, 993(1985).
T. Ichimiya and A. Fund&i, Int. J. Appl. Rad. Isot.,
l9373(9168)
T. Thou. Z. Radzhnskl, B. Pamailc,G. Rozgonyl and B. L.
Sopori, Appl. Phys. Let& s, 1985 (1991).
J. C. Muller, Y. Ababou, A. Bad&Ii. B. Courcelle, S.
Unamuno. D. Salles and P. Siffett, Solar Cells, II. 201
(1986)
C. H. Seager and D. S. Ginley, J. Appl. Phys..$&
1050(1981).
B. L. Sopori, J. Appl. Phys., $4,5264 (1988).
2.
3.
CONCLUSION
We have described some characteristics of defects
produced by low-energy hydrogen implantation. The surface
damage, consisting of dislocation loops, stacking faults, and
trapped hydrogen, are related to the ion energy as well as
hydrogen concentration at the surface. It is shown that the
near-surface region also develops a preponderance of
“platelets,” which are disc-shaped structures that reside in
(111) planes and are elongated along [l 101 directions. We
have also shown that hydrogen can segregate at dislocation
nodes, and have exploited this feature to determine,
semiquantitatively, the hydrogen profiles deep in the bulk. It
is determined that hydrogen-induced defects exhibit high
carrier recombination. It is proposed that this mechanism
could limit the effectiveness of hydrogen for improving solar
cell performance.
We have shown that enhanced bulk diffusion of hydrogen
can occur in some polycrystalline substrates, with a diffusivity
of at least an order of magnitude higher than that in FZ
wafers. We have shown that such an enhanced diffusion is
similar to that taking place along some grain boundaries. It is
also seen that hydrogen diffusion can be retarded by certain
impurities. A back-side hydrogenation techniques is described
which requi~s only 15-30 min. to produce effective
passivation and minimizes the effect of hydrogen-related
defects.
Our results of enhanced hydrogen diffusion point out that
deep diffusions of hydrogen can occur if the process used for
hydrogenation is accompanied by injection of suitable point
defects. This opens up the possibility of introducing hydrogen
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
Figure
1.
Figure 2.
TEM micrographs showing : (a) [Oil] plan view of
a platelet having a core surrounded by a disc-like
st~cture,
and (b) a lattice image of the platelet.
Figwe
TEM micrograph
a dislocation.
XTEM micrographs of the near-surface
damage due to a hydrogen implant at
1.5 keV for 30 minutes, showing:
(a) dislocations and hydrogen entrapment
(indicated by arrow), and (b) stacking faults
and platelets (indicated by the arrow).
12-6
3.
showing
h*rogen
segregation
at
to” t .._ I-l~I~~I-I~l_I~l~~~-L~.l-J-I-LLL
1.0
1.5
2.0
2.5
3.0
0.0
0.5
Depth (microns)
3.5
4.0
Depth {microns)
Figure 5.
Ffgure 4.
EBIC scans of a schottky diode showing relative
responses under different electron beam voltages
fmm implanted and unimplanted regions.
12-7
SIMS profiles of hydrogen implanted at 1.5 keV on hvo
different substrates, (a) FZ wafer, and (b) ribbon.
Figure 6.
An XTEM micrograph showing observation of deep
diffusion of hydrogen by hydrogen segregation on
dislocations
Figure 7.
12-8;
Grain boundary diffusion observed by XTEM in samples
of (a) low. and (b) high oxygen concentrations.
610
_....._..-..___-._--.---
IOM
_ _. _.-. _
(4
r
a.
before implant
4.
alter implnnl
600
10’9
oc
f
10’8
.E
.g
10’7
it
t2
10’6
3
10’5
560
10’4
0
I
7
8
9
10
Cell Number
Figure
8.
SIMS profile of bomn showing
due to hydrogen implantation.
out-diffusion
of boron
Figure
10, Effect of hydmgen implantation on the V,, of solar
cells. Solid and dotted circles correspond to before
and after hydrogenation.
Figure
1 I. Spectral responses of a finished
after hydrogenation.
I c i 1”’
P-type substrate
Hydrogenated
surface
Alloy Al by optical processing
Figure 9.
A schematic
of the back-side
hydrogenation
process
12-9
solar cell before and
1fOT SPOT SUSCEf’TIfflLfTY
AND TESTfNG
OF PV MODULES
E. Molenbmek. D.W. Waddington, K.A. Emcry
National Renewable Energy Laboratory
(formerly the Solar Energy Research Irwti~r~tc)
Gnklen. Colorado
I.aboratory
(JPL) (3) and Underwriters
LaboTatnry
(4)
developed an’ intmsive test, in. which individual
cells are
reverse biased in the dark or under reduced ilhimination.
The
Joint Research Center in Jspra. Jtaly under the auspices of
the Commission of the European Communities develapd
a
nonintrusive
test under a simulatot,
in which cells in the
modules are shaded (5). Various national and international
standards organizations
are developing module qualification
procedures
involving
hot spot. susceptibility.
Because
cnrrent hot spot tests were developed
specirtcrlty
for
crystalline
silicon
modules.
these tests should
be
mevaluated for use in evaluating amorphous silicon and other
thin-film material systems. The first goal was to investigate
and compare the behavior of amorphous silicon (a-Si) and
crystalline
silicon (c-Si) modules under reverse biasing.
This study addressed the basic question of what is hot spnt
heating, what causes it and how can it be simulated?
The
experiments
done on the a-Si and c-Si mndules show that
the procedures for the different hot spot tests have problems
as far as cell selection
arid determination
of the test
conditions are concerned.
ARSTRACT
Localized
heating nr hot spots in a photnvaltaic
module can occnr by any combination
of cell failure,
inrerconnection
failure, partial shading. and variation in the
photncurrent
from, cell to cell (mismatch).
To probe the
sensitivity
for hot spot heating of commercial
amorphous
silicon
and crystalline
modules,
several
intrusive
and
nonintrusIve
experiments
have been performed.
In the
intrusive
experiments
each cell in several commercial
amorphous silicon modules was evaluated separately acd in
groups for localized heating effects. Damage in amorphous
silicon mod$es occurred under reverse-bias conditions in the
dark above a 5-20 mAcm-2 cell current density at the
interconnection
between cells. Shading can cause a larger
temperature rise than current mismatch.
For the monolithic
amorphot!s
silicon
modules
investigated
the current
mismatch
between
each cell was substantial,
but the
temperatnre rise was negligible
because of the rather low
shunt resistance.
What is Hot Snot lleatine?
INTROIXICTION
llot spot heating occurs when a cell in a string of
series connected cells is negarively
biased and dissipates
power in the form of heat instead of producing electrical
power. This happens when the current produced by a given
cell is lower than the string current. This can occur when the
cell is shaded. damaged, or simply generates less current
One example of mismatch is given in
than the module.
Figure I, which shows a thermal image of a short-circuited
crystalline module. The temperature of the mismatched cell
is about IttY~ versus SfPC of the rest of the module. When a
single cell in a series string generates less current than the
module, localized
heating will occur because the current
flowing through each cell in the string must be equal.
Figure 2 shows a module operating under standard
test conditions with a voltage at maximum powet of Vmp and
Localized heating (hot spot) of photovoltaic
modules
has been documented since the early spacecraft days (1). It
was.argued that cell failure, interconnection
failure, or partial
shading by parts of the spacecraft could cause cells to heat
np and possibly damage.
In 1979, damage due to hot spot
heating was observed at test sites at Mead, Nebraska and
Arlington,
Texas (2). At Mead, a hail storm caused cells IO
crack. which in turn. caused some cells to become reverse
biased and heat up. At Arlington. hot spot heating occurred
after the modules, each with a bypass diode, were shortcircuited for washing and inspection.
In the past years, several tests have been developed
to test the ability of a photovoltaic
(PV) module 10 withstand
hot spot heating. When bypass diodes are used to limit the
reverse bias voltage across a module to less than I V hot
spot heating is minimized.
Standards have been developed
to determine the degne of hot spot heating when bypass
diodes are properly installed in a module. The Jet Propulsion
current at maximum
power of Imp with a single shaded cell in
the series string. Because of Kirchoff’s
voltage and current
laws the shaded cell will operate at a voltage of -V’ and
current of Imp, causing power to be dissipated in the cell
13-1
60°C
80°C
70°C
.rmnlphnus
l
l
l
man-cryslat
.
0
Si
l
l
.$ 0.5
Figure I.
Thermogram
of a mismatched
short-circuited
mono4
8
B
2
cell in a
module.
-10
Figure 2.
Reverse
0
5
10
-5
Modulc or ccl1 vohagc (V)
I-V
mismatched
module.
equal to V’+ Imp.
cell voltage
module
curve
type-B
of
a partially
cell
and I-V
15
l
fl
shaded
voltage
at the maximum
to be type-A
in the dark is at -V,p
power point (V,p)
I
0
Reverse
bias I-V curves
of a “typical”
commercial crystalline and an amorphous silicon
module.
or
AMORPHOUS
A cell is considered
e..
-I
-2
Rcvcrsc hias vohagc (V)
of the
if the
with a
A cell is also type-A if the cell voltage
and the current is less than Imp.
SILICON
UNDER
REVERSE BIASING
The behavior of crystalline
silicon modules under reverse
biasing (7,8) has been well studied, but little information has
been published on amorphous silicon (9). The intrusive hot
spot test in wide use was developed by JPL (3). In this. test
wires are attached to individual cells. The advantage of this
method is that the I-V characteristics
of each cell can he
known for positive
as well as negative voltages under
varying degrees of illumination.
As can be seen from
Figure 3, amorphous
silicon
behaves differently
from
For the six
crystalline
silicon
under reverse biasing.
in he dark at -V’ is greater than the illuminated
current equal 10 Imp.
-3
m
Based upon the resuhs af this study the worst
possible condition of hot spot heating is to completety shade
a single cell. To minimixe these effects the array should be
mounted to avoid dust or snow collection on the module, and
shadows from objects in the foreground should not obscure a
single cell.
The term current mismatch refers tn any
mechanism that can cause a reduction in the short-circuit
current of a cell compared to other cells in the series string.
Manufacturers
of crystalline
silicon cells and modules sort
ttieir cells by current at a fixed voltage in order to minimire
hot spot heating and optimize module power production.
If a
cell degrades
because of a crack or other intrinsic
mechanisms, the current mismatch may be severe enough to
damage the encapsulation.
20
curve
l
a
Figure 3.
-IS
D
The
reverse characteristics
of a type-A cell are voltage limited.
A cell is considered to be type-B if -V’ is less than V,p.
The reverse characteristics
of a type-B cell are current
limited.
A type-B cell often has a low shunt resistance or
has been damaged.
Two typical
examples
of reverse
current-voltage
(I-V) curves for commercial crystalline
and
amorphous silicon modules are given in Figure 3.
amorphous silicon modules from three manufacturers when
the number of cells in series exceed 15, the cells can all be
considered to be type-B (9). This is nor surprising because
of the relatively
low shunt resistances of the a-Si modules.
Most amorphous silicon modules are fabricated by depositing
amorphous silicon on tin oxide coated glass. The individual
cells are defined by a scribing process. In the intrusive hot
spot test, electrical contacts are made to each test cell, and a
negative voltage is applied in the dark or under reduced
The power dissipation for any given faulty or shaded
cell depends on series-parallel
configuration
of cells in a
module (h). In general, increasing the number of cells in
series increases the power dissipation
and increasing the
number of cells in parallel decreases the power dissipation of
the faulty cell.
13-2
iiluminaGon. depending on whether it is a type-B or a type-A
cell.
Jlowevet;
the individual
cells in many thin-film
technologies
like*amorphous
silicon are very difficult
to
contact without damaging.
For the commercial
amorphous
silicon
modules
evaluared. electrical contacC1 to the individual cells was made
to the back of the cells. Because the aluminum back contact
thickness is approximately
1 pm and a single scratch. would
destroy the.cell, great caution was required. Two wires, one
at each end, were attached tq the back of every cell with
silver epoxy that cured at room temperature.
Localized
heat@
because of current crowding
was measurable but
small because the &II length w.9 JO cm and the use of. two
wires, one at each end, minim&d
these effects.
By reverb biasing individual cells in the dark, it was
found that the voltage never exceeded the value of -7 to -8
V. At lhi4 voltage. only the.current still rises. The current
level at which damage occurred varied widely, from 0.5 Jsc to
2 I,,
or SO-Z& mA (for 1 by 10 cm culls).
Damaged
In 24 of the 37
damaged &Is
from two different
manu’facturers,
damage
started at the cell interconnection.
Jnspection
of the
damaged region under a microscope showed that the damage
developed right next to the third scribe line. Figure 4 shows
an example. One possible explanation of this phenomenon is
the following:
Wheti the third scribe line is made, the laser
beam causes the a-Si near the scribe line to crystalliirz
to
some extent (JO).
Because c-Si has a much higher dark
condtictivity
than a-S& these parts may act as shunt paths
and break down earlier because of the high current density.
Another
possible
reason ‘that damage
occurs
at the
interconnection
is that ihe particular
shape of the
interconnection
causes the current to concentrate
at the
edge. This would mean that other thin-film
solar cells that
are made in the same way might have the same problem.
In case the damage did not start at the cell
interconnection,
usually pinholes develop (small light spots
with a diameter of less than a millimeter).
During reverse
biasing,
sometimes
small
light
emitting
spots were
observed, which could lead to the formation of pinholes.
Jn
the case of crystalline
silicon, this has been associated with
avalanche breakdown (7).
As is evident
from the temperature
profile
of a
reverse biased cell, the power dissipation is very nonuniform.
Reverse biasing of a module with 30 cells in series clearly
shows
this nonuniformity
(see Figure 5).
Because
temperature is the important factor that determines whether
the module will be damaged or not, it is interesting to know
how much the local temperature
rise will be at a given
current level.
Therefore,
several cells and modules were
reverse biased and their maximum temperature as a function
of the current was measured.
The advantage of reverse
biasing whole modules is that 30 cells are probed at a time,
so that there is a tiasonable
chance that all possible cell
qualities
are represented.
The disadvantage
is that the
Glass
Figure 4.
Scribe lines viewed from the aluminum back, 25X.
whoie module is heated instead of one celi, but because the
heating is locali;rxd, the hot spot temperature will probably
not be influenced very much by the other cells.
El 22°C
fl24”C
i 26°C
Figure 5. Thermal image of a reirerse biased amorphous
silicon module in the dark showing hot spots.
The maximum temperature Bp a function of the curtent
is depicted in Figure 6 for commercial I-!? amorphous silicon
on glass modules from three different manufactures.
figure 6
shows that for modules of manufacturers
A and B, the
temperature
rise is moderate,
even at the short-circuit
current level, which is about 350 mA. In the module of
manufacturer
A, several pinholes developed.
The range of
temperatures
fell within the range given in the literature
earlier (9). A much greater temperature rise was observed
for the module from manufacturer
C. Visible damage was
done to several
cells,
but not to the encapsulation.
Measurement of the 1-V curve after the damage had occurred
showed no anomalies.
The fraction of the cell that was
damaged was so small that the power reduction was less
than the resolution of the measurement system (-1%).
was
done
by
temperature
reverse
manulacturcr A
l
manufacturer B
eizo
n
A
-
n
Bm-
n
j
u
Z
80:
g
60
l
A
l
m
A
l ee
40
l
e
l
l
l
l
20
/
0
so
loo
150 200 250 300 350
Revcrsc bias current (mA)
Figure 6. Reverse
biasing
amorphous silicon
4w
cell that gave the highest
450
of three l-ft2
commercial
modules, at room temperature.
temperature
Is,.
However,
two small pinholes.
showed
hot spot test was also performed
was selected.
The 4 -ftz module
some more damage,
After
middle.
Measurement
probably
anomalies.
two
more
of -the I-V
One
of manufacturer
at one end as well
weeks
curve
nothing
showed
C
as in the
had changed.
some degradation;
all due to the Staebler Wronski
effect, but no
The encapsulation,
glass on glass as well as
glass on polymer,
weeks of exposure
was not damaged.
the result ias
SKJCON
After
three
more
still the same.
UNDER
REVERSE
BJASJJ’JG
totai irradiance of la40
Wm-2, global light. As part of a
nonintrusive
hot spot test every cell in the module was
individually
shaded with black tape and the maximum
temperature was measured. The maximum temperature that
each of the 30 cells achieved when shaded as a function of
fhe reduced current is shown in Figure 7. The reduced
current is defined as the short-circuit
current of the module
after the illumination
of one of the cells in the module has
been reduced by shading. The module short-circuit current is
tbe module current under uniform illumination.
The current at
maximum power for the module was 3 A,.giving
10 type-A
and 20 type-B
cells in the module shown in Figure 7.
Figuti 8 shows the temperature
rise for as a function
of
percentage of the area shaded for one of the type-A cells in
the module evaluated in Figure 7. With the uniform shading
I’ n
-
type-A o-
8 190 T!
1 170 -
on a few
modules.
For amorphous silicon, three modules of three
different manufacturers
were placed outside. Cell selection
’ -&e-B
if
n
mm
j:;;
m
short-
circuiting a a-Si module does not show cells that are hotter
than the rest, as does short-circuiting
crystalline
silicon
modules. This is because the shunt resistance is so low that
hardly any power dissipation can take place. It also implies
that, in the case of mismatch, when the cell is illuminated,
the current distribution
is more uniform than under complete
shading.
An outdoor
until
After one week of outdoor
!a
All these experiments were performed in the dark or
under shading.
Could mismatch of amorphous silicon also
lead to a significant
temperature rise? Measurement of the
short-circuit
currents of 20 cells in a submodule (a module
without encapsulation)
showed that the lowest Jsc can be
than the module
module
as in Figure 7. The
exposure, nothing had happened to the 1-ft2 moduk of
manufacturer B. The l-ft2 module of manufacturer A showed
E2lO
5% to 8% lower
whole
Crystalline
silicon modules from two manufacturers
with 30 cells in series were evaluated.
A module was
illuminated
with a filtered Ar-arc (Vortek simulator)
to a
A manufacturer C
0
the
became visible,
cell in every module was shaded.
CRYSTALLINE
l
biasing
differences
n
n
l
irn
mm
m
:
I
p110 )
, Jsc=3.28A
i3 90
4
E
b
Imp’
70 -
I
I
. ’
I.
’ .
. . ’ . . .
2.5
2
Short-circuit current with one cell shaded
Figure 7. Maximum
temperature for each of
connected
c-Si cells shaded individually
in a
module as a function
of reduced short-circuit
current.
WI
--1.5
’
a transhlcent cellophane filter was placed over the entire cell
while in the case of nonuniform
shading the light was
completely blocked from part of the cell. Figure 9 compares
the results of this experiment
for a type--A ceil from the
n10th1lc in Figure 8.
In the JPL, block V procedure (3). type-h cells are
tested by applying a voltage equal to the negative of the
maximum power voltage (-Vmp) and adjusting the inadinnce
level until the current is equal to the current at the maximum
power point (Imp). This situation for an intrusive test can hr
translatrd into a nonintrusive
test by applying shading to a
cell in a short-ciicuited.
AlIly, illuminated
module. until the
current
is equal to Imp.
Ilnwever.
in standnrdixcd
l
nonintrusive
tests such as the one given in reference 4,
type-A cells are partially
shaded with opaque shading.
These two situations
are not equivalent,
as shown in
0
l
a
’
.
’
.
m
l
0
m
mm
.
n
-m
w
tl
10
20
so 6n 70
30
4n
Fcrccnt ofccll shadowed
no
90
100
Figure 8. Reduced short-circuit
current (W) and maximum
localized temperature (0) as a percentage of the
cell area that is shaded.
SUMMARY
Translucent shading
.-
OF HOT SPOT TESTS
The a-Si modules from manufacturers
A and B
exhibited neither degradation in the I-V characteristics
nor
physical damage. In the modules from manufacturer C, some
visihle pinhole damage occurred with the nonintrusive
test
but the decrease in performance was insignificant.
Several of the type-B cells in the crystalline-Si
module from two different
manufacturers
showed small
cracks and bubbles in the rear encapsulation after continuous
one sun illumination;
the module shorted and one cell
completely
shaded. This implies that these modules would
fail the nonintrusive hot spot test in reference 4. With proper
siting of a PV array, the shading of only one cell in a module
should not occur.
The hot spot tests assume that the worst case of
heating of type-A cells occurs when the module current is
increased to the maximum power current (I,,,).
It is true
that the maximum power dissipation
for the entire cell is
greatest at the maximum power point. However, in the case
of partial shading, the power dissipated per unit area is not
uniform.
In fact, the smaller the illuminated area, the higher
power dissipated per unit area in the illuminated region of the
cell. The maximum temperature rise in a shaded module is
determined by region with the maximum power dissipated
per unit area and not the total power dissipated.
Figure 9
shows that the tempemture rise as a function of the fraction
of the total cell area shaded varies from 50 - 140°C.
n
m
4
/
2.65
2.1
Short-circtlit
2.75
2.8
2.84
2.9
current (A) with one cell partially shaded
Figure 9. Comparison
between partial shading
uniformly mduced ilhlmination (W).
(0)
and
For the c-Si modules investigated,
the temperature
rise in type-A
cells is below the damage threshold,
regardless
of whether translucent
or opaque shading is
applied.
For
the
crystalline
modules
investigated.
unlikely
to occur
lfowever,
provided
these conclusions
that the reverse
and
failures
amorphous
silicon,
PV
due to hot spot heating
bypass
diodes
are
are employed.
are based on the assumption
bias across a module
will
not exceed -I V
(a bypass dinde is installed).
The array design influences
the pnssihility and severity of hnt spot heating. Therefore, a
hot spot test should take into account
evaluating
what
simulating
these
mismatch
cnuld
go wrong
circumstances.
can occur in short-circuit
could lead to long term degradation.
hot
spot
immediately.
test
is to detect
this array design by
in a certain
Some
conditions.
However,
problems
design
heating
that
and
due to
Maybe
this
the aim of the
arise
almost
ACKNOWLEDGEMENTS
Thiv work was supported hy the Department
contract No. DE-hCO2-13CIIIOO93.
of Energy under
REFERENCES
I.
2.
3.
PA. Make and K.I.. ilnnson. !%~r~~$b Intersoc. EnK
@IV. Ene. Conf., Washington. DC, 1969. Am. Inst.
Chrm. Eng. New York, 575, (1969).
S.t?. Forman.
M.P. Themelis.
Proc. 14th IEEE
motvoltric
Suecirlists’
Conf.. San Diego, CA, IEEE,
New York. 1214. (19gO).
f&ck
V Solar Cell Module
Desien and Test
Specification
for Intermediate Load Applications, JPL
internal
document
5101-161.
Jet Propulsion
1.abnratory, Pnsadcna, CA, (February 19111).
4.
&alification
Test Procedure
for Photovoltaic
Jvlodule~, Commission
of the Europe.an Commrmities
standard specification number 502, issue 1, (1984).
_5 .
&u&trd
for Flat-Plate
Photovoltaic
Modules and
P,lnels. Underwriters
Lnbomtories
number UL 1703,
Augnst
1. 19116, Underwriters
Laboratories
Inc..
Research Triangle Park, NC (1986).
M.M. Alkaisi and N.A. Aldawody, Solar Cells, 28, 11,
(1990).
J.W. Bishop, Solar Cells, 26, 335, (1989).
C. Gonznlex, R.W. Weaver, R.G. Ross, R. Spencer,
J.C. Amett, Proc. 17th IEEE Photovoltaic
Soecialists’
Conf., Kissimmee, FL, IEEE, NY, 1984, 668, (1984).
C. Gonzalez
and E. Jetter, Proc.
10th IEEE
f-i.
7.
8.
9.
10.
Photovoltaic
Soecialists
Conf.,
Las Vegas, NV,
IEEE, New York, 1041, (1985).
S. Yamaxaki, K. ltoh, S. Watabe, A. Mose, K. Urata,
K. Shihata, and Il. Shinohara.
Proc. 17th IEEE
Photovoltaic
Snecialists’
Conf.. Kissimmee,
FL,
IEEE, New York, 206. (1984).
13-6
WEATHERING
DEt;RAI)ATION
OF EVA ENCAI’SULANT
AND TItE EFFECT
OF ITS Yt$L!.~)\VINt:
ON
SO1
AH
<‘FI
I
F,FFICIKN(:Y
--..---L”F12
F..I. tkrn.
A.W. t’zuxlrrua.
K.A. Emery. and R.G. Dhcre
N:rticmnl Rcnrwnhlr
Energy I.ahoratory
(ftrrmrrly
tbr Solar Energy Rpsrarch Institute)
Golden, Colorado
ABSTRACT
The encapsulant
materials
provide optical coupling,
mechanical
support, electrical
isolation,
physical
isolation/
protection. and thermal conduction for the solar cell assembly
(4.5). EVA copolymer (33% vinyl acetate) is extensively used
for crystalline
Si-based PV module encapsulation.
EVA is
formulated
and processed to provide the desired mechanical
strength and stability (4,5). The processed EVA contains a
65% to 70% gel content (degree of cross-linking),
0.30 wt %
Cyasorb (IV 531 (a UV absorber identified as “Cyasorb” here),
and two antioxidants (4-6). Upon weathering degradation. the
EVA between the cover glass plate and solar ceils in PV
modules tnay hecome discolored.
In previous papers, we
studied tbe structural effects and relationship among the extent
of degradation, gel content, and Cyasorb concentration in EVA
(6.7). In this paper, we etiphasize
more the effect of EVA
yellowing
an the solar cell efficiency.
After five or more years of weathering,
the dcgradntion
of
ethylene-vinyl
acetate (EVA) encapsulant in photnvoltaic
(PV)
modules resulted in a yellow to dark brown color. Degraded
EVA shows a substantial increase in the gel content and a
large to complete
loss of the ultraviofet
((iv)
ahsorbcr,
Cyasorb UV 531. The EVA discoloration
is causrd by the
formation of polyconjtigated
(C=C), double hontls of various
lengths. Acetic acid and other volatile organic cnmpnnents arc
also produced from the photothermal
decomposition
of the
EVA. The solar cell efficiency was reduced by --9% by a light
yellow brown EVA and -50% by a dark brown EVA. WMthered PV modules with dark brown EVA nlsn show a -.SO%
decrease in efficiency.
EXPERIMENTAL
EVA Materials
In the past two decades, significant imprnvemcnts
in the
efficiency
and reliability
of new semiconductor
tnnterials for
PV devices and modules have resulted in PV becoming an
increasingly
viable energy alternative.
While efficiency
improvements in large-area CulnSq, CdTe. and a-Si-based thinfilm PV modules are increasingly
promising.
single-crystal
(and polycrystalline)
Si-based PV modules have been the most
dominant products widely used. Typical efficiencies
of these
PV modules range from about 11% to 14% (1). A long-term
stability of over 20-30 years for all types of PV modules is
one of the requirements to be cost-cnmpctitive.
and Analvticat
Procedure
The EVA materials we studied, our analytical
procedures. and the analytical instruments we used are essentially
the same as those reported previously
(6.7). The laminated
and cured EVA in PV modules that are stored in the dark fnr
six years remains clear. Clear and discolored (degraded) EVA
samples were cut from PV modules weathered outdoors for
more than five years. The Cyasorb concentration
(wt %) and
gel content (8) are the two crucial measurements of EVA
degradation (6).
Measurement
However, some Si-based modules deployed nutdoors for
five years or more develnp a light yellow, to yellow-brown,
to
dark brown color. depending on their locations, use tetnperature. and configuration.
A common factor in these discolored
modules is their use of EVA films as the encapsulant.
The
consequence
of the EVA weathering
degradation
can be
significant.
For example, Gay and Berman reported that EVA
browning resulted in a -30% loss in the annual energy output
from the six-megawatt
PV systems at the Carrisa, California,
power plant (2). A detailed study on the module pcrfnrmance
of the Carrisa PV systems revealed that the EVA degradation
is highly nonunifnnn
from module to module, and that the
average module power otttpttt is 35.9% below that of the initial
The mismatch between neighboring .modules,
rating (3).
caused by nonuniform
EVA degradation,
cnntributed
an
additional power loss of 11 .l%.
of the Effect of &A
Yellowine
on Sutar
Cells Efficiency
The effect of EVA discoloration
on solar cell efficiency
was evaluated as before (6). using a calibrated Spectrolab X-25
solar simulator.
Quantum efficiency
(spectral response) was
measured on a computerized
system with periodic (440 Hz)
monochromatic
light directed through one of an array of lontn band-pass interference filters. The lower wavelength limit
was 290 nm. In both cases, the virgin or degraded EVA films
were pressed above the reference solar cells with or without
adding a cover glass above the film.
Similar measurements
were also perfomd
for a solar cell specimen (26 cm’) taken
from a broken PV module.
The I-V performance
of three
large PV modules was measured on a Spire Model 240A solar
sitnulatnr.
14-1
RESIJI.TS
EVA Degradation
AND DISt:IISSION
and Discoloration
Figure I shows the results nbtaincd from nnnly-rinp: a
large number of non-degraded
and field-drgratlrcl
EVA
samples. The gel content increased substantially
from (is%70% in virgin EVA to gS%-88% in clear degradrd EVA, to
90%-92% in light yellow EVA. to 96%-97% in yellow-brown
and dark brown EVA, while the Cyasorh concrnlration
ticcreased concomitantly
during the degradation
prncesc.
No
Cyasorh was observed in the extensively
drgratlcd
EVA.
When the Cyasorb concentratinn dropped bcinw.abnut 0.21 WI
%, the EVA films became discolored (6). As the cxtcnt of
degradation increased, the color darkened from light yellow. to
yellow.
to yellow-brown.
and to dark brown.
The EVA
degradation across each solar ceil unit on weathered PV modules was not uniform.
The Cyasorb concentration
was much
lower and the gel content was higher in EVA from over the
central region than from over the cell edges (6.7). No nnliceable change in either the gel content or the Cyasorb rnncentration was observed in EVA samples taken from unweathered
mndules from similar incations.
‘lhe EVA mostly rrmnined
clear around the cell edges and in Ihe areas between nrighhoring solar ceils. in addition tn the discnlnration,
acetic acid and
volatile organics were present in the &graded EVA.
1
_ 100%
Arbltrnry
Exient
01 EVA degradation
(?G)
Figrrre 1. The Cynsorb IIV S3i concentration.
pi content.
and EVA cnlor at different extents of degradation, summarixd from the results of analyzing a large number of undegraded and field-degraded
EVA samples. The actual degradation is nnt known except for the virgin EVA.
Figure 2 compares the transmission spectra for virgin and
degraded (to various degrees) EVA samples. As the cnlor on
the EVA darkens. more UV and visible light is absorbed. The
yellow-browning
of degraded EVA is caused by the formation
of poiyconjugatcd
carbon-carbon
double bnnds (pdyenes)
(6.7). The discolored EVA iuminesces strongly upon illumination by UV and visible light. In fact, the extent of EVA degradation can he detected directly using flunrescence analysis
from the shift in the emission peak position (from shnrter to
longer wavelengths in the 500-650 nm region) and the increase
in peak intensity (6.7). intense acetic acid odors and other
volatile organic components were detected when the wealhered
PV modules were cut open. The presence of acetic acid is significant.
In various simulated degradation experiments. acetic
acid was produced from EVA exposed to UV light at 45°C or
healed in the dark at 130°C. Acetic acid has been found tn
catalyze EVA yellow-browning
from 8S’ lo 130°C. A detailed
discussion will be given elsewhere (7).
200
300
400
500
600
700
8
Wavelength (nm)
Figure 2. Transmission
spectra measured for (1) a clear
EVA film. (2) a light yellow-brown
EVA liim. (3) a darker
yellow-brown
EVA film, (4) a brown EVA film, and (4) a
darker brown EVA film. Samples #I. #4. and #5 were iaminated in two glass plates and exposed to an RS4 sun lamp at
90°C for ifi00 h. Samples #2 and #3 were taken from weathered PV modules.
In summary, weathering degradation of the EVA encapsulant in PV modules results in an increased gel content, a loss
of the UV absorber, a yellow-browning
of the EVA, and the
production of acetic acid and volatile organic components (6).
Effects of EVA Discoloration
1
0.01 0.0
’
on Solar Cell Efticiency
The effect of yellow-browning
on the light transmitsion
is shown in Figure 2 for samples that are clear to dark brown
EVA.
The effect of EVA yellow-browning
on the electrical
performance
and quantum efficiency
(spectral response) was
obtained
using single-crystal
Si reference ceils with and
without a degraded EVA sample laid over them. The results
are given in Tables 1 and 2(h). respectively.
Table 1 gives
the measured ripen--circuit voltage (V,). short-circuit
current
densily (J,), maximum power (P,,,,,). and calculated loss (45)
for two virgin clear<and two degraded yellow-brown
EVA
films. When covered with a virgin clear EVA film. the Si cell
lost about 6.7% in J, and 0.8% in P,,,,,. However, the loss
increased lo l2%-14% in J, and 14%16% in P,., with the
14-2
yellow-brown
EVA films.
The loss in V, was reInlively
small.
In effect, the Si cell’s 11.6% efficiency
changed to
10.7% with the virgin EVA and 9.7% with the yellow-hrown
EVA; this corresponds
to a net loss of 9.3% of its original
efficiency
when the EVA color changed from clear to yrllnwbrown. These effects also appear clearly in the rclativr quantum efficiency
ratios to the bare Si refcrcncc cell, as S~OWII in
Fi8ure 3. The spectral responses for the two virgin cured
EVA films are virtually identical, with a cutoff hrlow 360 nm
from the absorption of Cyasorb. For the two degraded yrllowbrown EVA films that differ slightly in color, the spectral
responses begin at about 290 nm because of the absence of
Cyasorb. Because of light absorption by polyenrs, the spectral
response is lower between 360 and about 900 ntn than thar of
the virgin EVA (Figure 2, curves 3 and 4).
As the EVA color became dark brown, the change in
measured cell efficiency
increased. Table 2(A) shows the results obtained for two dark brown EVA samples with or without a 7059 glass plate pressed tightly onto the EVA over the
reference Si cell. For comparison, a solar cell specimen taken
from a broken PV module with dark brown EVA was also
similarly analyzed.
With the dark brown EVA, a nearly SO%
loss in P,,, was observed; this is -35% greater than that for
the yellow-brown
EVA seen in Table 1. The large difference
is simply because the dark brown EVA absorbs more UV and
visible li8ht (Figure 2) than does the yellow-brown
EVA (6.7).
The presence of a cover glass plate reduced the P,,. an
additional
2%.
For the solar ceil specimen, the measured
efficiency
for the three tape-defined
areas ranged from 3.0%
to 4.4% (see Table 2B). further supporting our earlier conclusion that the degradation of EVA is non-uniform
across the
cell surface (6). More importantly,
the measured efficiencies
650
650
are 67%-77% lower than the typical 13%-14% efficiency
of
new 4”x4” mini-modules
made by the same manufacturer.
The
decrease is 17%-27% greater than the changes measured with
dark brown EVA films alone (see Table 2A). suggesting that
other factors such as increased series resistance may have also
contrihutrd
to the efriciency drop. This argument seems to be
supported hy a fill factor of ahout 0.50 as compared to a fill
factnr of about 0.70 for new mini-modules.
The relative spectral responses measured for the variously degraded EVA samples are compared in Figure 4.
Tahle 1. Eflect of EVA Yellowing on the I-V Performance
of a Single-Crysfal Si Reference Solar Cell’
Sample
EVA
V,
09
No EVA
0.474
39.87
A9918
0.470
1529X
P
C:%
V,
37.21
3.74
0.84
6.67
0.79
0.471
37.23
3.75
0.63
6.50
0.76
Yellow- I
0.466
35.0
3.48
1.69 12.21 1428
Yellow-2
0.465
34.5 I
3.41
1.90
l
250
Figure 3. Relative quantum efficiency measured for a
single-crystal
Si reference solar cell covered with virgin
clear (curves #I and #2) and degraded yellow-brown
(curves
I
450
I
I
f
650
Wavelength (nm)
650
Fipre 4. Quantum efficiency
crystal Si reference solar cell,
with virgin cured EVA A9918,
with yellow-brown
EVA. and
dark brown EVA. Curves #I
100%. Curves #2 and #3 are
#3 and #4) EVA films. The spectral responses an ratioed to
that for the bare Si reference cell. The virgin EVA films are
EVA A9918 (curve #I) and EVA 15295P (curve #2). The
two degraded EVA films differ slightly in color.
14-3
Loss (a) in
J,
P,,,,
13.44
16.01
An aperture of 0.35 cm* was used over a single-crystal Si reference
solar cell for the measurements. The EVA film was placed beneafh the aperture. No cover glzs was used on top of the cell or
EVA. The loss (Q) would lx slightly larger if a cover glass were
used. The P,, loss (%) would be about 7% less than the values
shown if they were corrcctcd for Ihe light scattering (see text).
Data for Yellow-l and Yellow-2 are identified as samples #2 and
#3 in Pigurc 3 and Yellow-2 as sample #3 in Figure 4.
al/II:
Wavelength (nm)
J,’
(mA/cm*)
I
I
1050
I.250
measured for (I) a single(2) the cell in (1) covered
(3) the cell in (1) covered
(4) a solar cell specimen with
and #4 are normalized to
normalized to curve #l.
Table 2. Effect of Dark Rrawn EVA on the I-V I’erhrtnante of (A) a Single-Crystal Si Rcferentc War (‘cl’ and
(R) a Sotar Cell Specimen
EVA changed from clear to yellow-brown
and -42% when it
changed to dark brown.
While the integrated reflectance
measured for the solar cell specimen is about 35% lower than
that for a dark brown EVA with a smoother Si-side surface
(Figure 5h). this small difference is not sufficient to adversely
affect the interpretation
of results in Table 2.
(A) Two dark brown EVA films peclcd fmm tlcgrntlrtl PV mtwlulcs
-.-..-----.
- ._.. . _ .-.. _-- _
I.ozc t%) in
P
Sample
VCC
J,
(mhlcm’)
m”’
EVA
09
(mW)
V,
J.,
P,,,
----.
. ~.-_
..-. __
3.54
----.-___0.398
22.54
Si Cell
Brown-1
+ slideb
0.347
0.341
13.15
12.64
1.X0
1.70
12.81
14.32
Bmwn-2
+ slide’
0.345
0.343
13.21
12.76
1.79
1.72
13.32 41.39
13.82 43.39
.----~
Effect of EVA Discoloration
41.66 4915
4.3.92 5l.S
4944
51.41
(B) A solar ccl1 specimen with dark hmwn EVA’
SW
No.
Area
(cm*)
V,
(V)
I
0.170
2
3
Jnc
Pmm
Fill -- Efficiency
Factor
c47,)
(mhlcm’)
(mW)
0.330
26.62
0.746
0.50
4.4
0.190
0.325
21.A2
0.665
0.50
3,s
O.lh3
0.308
20.16
0.489
0.48
__-.__
3.0
* The results shown in the tahlc arc not atljustrd for the loss of light
transmission thmuglr the EVA film due to light scnltrring fmm rhc
coarse Si-side surface. SW text for clctails. Data fnr Pmwn-I are
identified as sample #3 in Figure 5(a).
’ A glass plate was pressed tightly on top of the EVA film. An aclive aperture area of 0.967 cm* was used for the mcasuremmts.
’ The solar cell .specimen (26 cm*) was taken fmm a hmkcn PV
module with dark hmwn EVA and the supcrstratcs and substrates
intact. The three areas were detincd by black tape. Contacts were
made by soldering wires to the fmnt and back huslinrs.
on PV Module Performance
The effect of EVA discoloration
on module I-V performance was atso evaluated for three PV modules of the same
type (33 4”x4” cells).
The EVA color ranged from light
yellow to dark brown. The shunt and series resistances were
determined from dark I-V measurements. Table 3 summa&es
the results.
The original data on the I-V performance
and
resistances are unavailable, which prevents an accurate interpretation of the data. However. the results seem to suggest
that an increase in the EVA yellow-browning
is related to a
decrease in the shunt resistance and an increase in the series
resistance, which in turn results in a corresponding decrease in
the fill factor and hence the efficiency.
The losses in fill factor and efficiency
in the yellow-brown
and dark brown EVA
modules are about 50% when compared to a typical fill-factor
value of about 0.70 and an efficiency
of 210% for a new
module of the same configuration.
Our results on three modules are in good agreement with the field results obtained at
the Carrisa Plains power plant, where EVA yellow-browning
caused an averape 36% loss in power output (2) and where
some modules with mirror-enhanced
configuration
lost nearly
70% more than average peak power (3).
0.20
- (4
*'
#---.
\
B/c--
\
1
II
,
/'
/
,
,
__---_
O.lO-
One concern in our measurements of efficiency and spectral response is the accuracy offset by light scattering from the
textured surface of EVA films, and the extent of noncoupling
between the EVA and the reference Si cell. This concern is
particularly
important for the dark brown EVA films (used for
measurements in Table 2A) that have a coarse surface side
after being forcefully
removed from the micro-pyramidically
etched surface of the Si solar cells. The light-scattering
effect
was assessed from the integrated reflectance measurements
from 250 to 1250 nm for a virgin clear EVA sample, a yellowbrown EVA sample, two dark brown EVA samples (one of
them with a coarse R-side surface as described above and the
other with a smoother Si-side surface). and the solar cell
specimen.
The nsults are shown in Figure 5. Figure S(a)
shows that the integrated reflectances are nearly identical for
the virgin clear and yellow-brown
EVA films (about 7%). and
double that for the dark brown EVA with a coarse Si-side
Therefore,
the measuted solar cell performance
surface.
parameters in Tables I and 2(A) should be corrected by 7%
and 14%- 15% of their listed values, respectively.
Accordingly,
the corrected decrease in solar cell efficiency
is -8.6% when
HC
2
------A
1
L
2
0.20-
(b)
b
2
E
o.ool
250
,
,
450
1
I
650
I
I
850
I
1
1050
I
1:
Wavelength (nm)
Figure 5. Integrated
reflectance measured for (a) a virgin
EVA A9918 film (curve #I). a yellow-brown
EVA film
(curve #2), a dark brown EVA film (curve lt3) with a coarse
Si-si& surface, and (b) a dark brown EVA with a smoother
Si-side surface (curve #I) and a solar cell specimen
(curve #2).
14-4
Table 3. Electrical Performance Measured using a Spire
Simulator and Resisfances Determined from Dark I-V Data
for Three Weathered PV Modules*
Mod. EVA
V,
cold w
I,
Pm,,
IN
(W)
All
Fartnr
.~
- -___..Hf.
R,,
R,.
(%) (kohm) (Ohm)
-__--_
-___
A
II ycl
6.7.5s
1.217
34.0
0.10
9.34
I.35
0.109
B
yel-hm
6.733
6.428
16.35
0.38
4.49
0.577
o.sox
C
drk bm
7.047
6.137
IS.99
0.37
4.41
0.123
0.499
to about SO%, depending
from the degradation.
on the extent of EVA discoloration
ACKNOWLEDGMENT
The authors thank R. DeBlasio for his interest in this
work and S. Rummel and E. Beck for their technical contributions. This work is supported by the Department of Energy
under Contract No. DE-ACU2-83CHlOO93.
REFERENCFS
The three mndulcs were of the same type. and e,wh conlnins 33 4”x4”
squnm SolN cclla.
b II yel = light yellow. ycl-hm = yellow-brown. dfk hm = dark hmwn.
l
1.
2.
CONCLUSIONS
We have reported that weathering
degradation
of the
EVA encapsulant in PV modules results in an increase of
copolymer cross-linking,
a large decrease in CJV absorber concentration, the production of acetic acid, and a yellow to dark
brown color. The discoloration
results from the formation of
polyenes of various lengths. The polyenes absorb UV and visible light and luminesce in the visible region; thus, they may
partially compensate for the loss of solar cell efficiency due to
reduced light transmission.
The light-absorbing
nature of
polyenes reduces the net solar cell efficiency
from about 9%
3.
4.
5.
6.
7.
J. P. Thornton. R. DeBlasio, and K. Zweibel, Energy Engineerinp, 87, 1990, pp. 63-79.
C. F. Gay and E. Berman, Chemtech. March f990, pp.
182- 186.
A. L. Rosenthal and C. G. Lane, Proc. PV Module Reliabilitv Workshop, Oct. 25-26, 1990. Lakewood, Colorado,
SERl/CP-4079,
pp. 217-229.
C.G. Gebelein, D.J. Williams, and R.D. Deanin, ed.. polvmers in Solar EnerEv Utilization,
ACS, Washington D.C.,
1983, Ch. 22 and 23, pp. 353-385.
E. Cuddihy, C. Coulbert. A. Gupta and R. Liang. &
Plate Solar Arrav Proiect Final Rewrt. Vol. VII--Module
Encapsulation. JPL publication 86-31, DOWJPL-1012-125.
1986.
F.J. Pem and A.W. Czandema. Solar Cells, in press.
F.J. Pem and A.W. Czandema, to be published.
JvllNORITY-CARRIER LIFETIME OF POLYCRYSTAM
CdTe IN CdS/CdTe SQI AR CFI &
R.K. Ahrenkk 8. M. Keyes, and L. Wang
National Renewable Energy Laboratory (formefiy the Solar Energy Research fnstitute)
Golden, Colorado
and
S. P Albright
Photon Energy, Inc.
El Paso, Texas
of these devices could be improved if the minority
ABSTRACT
carrier lifetime in the p-CdTe is increased.
Our
technique provides a way to easily monitor minorityPhoton Energy, Inc. has produced CdSKdTe
carrier lifetime while developing the processing
solar cells by a low-cost, spray technology.
The
technology.
NREL-measured efficiency of the best device is 12.7%
at air mass 1.5. For the first time, the minority-carrier
EXPERIMENTAL TECHNIQUES
lifetime of the polycrystalline
CdTe grains was
measured by time-resolved photoluminescence.
The
A schematic of the Photon Energy backwall
longest measured lifetimes were over 5 ns and were
device is shown in Figure 1. The devices have the
material
found in large grain (-4pm). high-density
backwall configuration
of glass/tin oxide/n-CdSI
near the CdSlCdTe interface.
p-CdTe/electroda with deposition on a glass substrate.
INTRODUCTION
The improvement of low cost solar cells for flat
plate applications is one of the goals of the national
photovoltaics (PV) program.
The improvement of
device efficiency
using low-cost
processing
technology is a key component of the program. Thinfilm CdSXdTe devices are one of the more promising
candidates
for low cost. large-area,
terrestrial
applications.
The ultlmate device efficiency is
intimately related to the minority-carrfer lifetime of the
polycrystalline material. We wish to maximize the
electron lifetime in the polycrystalline p-CdTe base of
the device while not substantially
increasing the
processing costs.
The measurement of electrical parameters
of
polycrystalline
materials is more complex than of
Minority-carrier lifetime is
single crystal materials.
especially difficult to determine in pofycrystalline
semiconductors. Here we will describe the first direct
in
measurement
of minority-carrier
lifetime
polycrystalline CdTe using a laser technique.
The
technique has been widely used for the lifetime
analysis of crystalline semiconductors but this is the
first
report on its application to polycrystalline
materials. This technique’ is a form of time-resolved
photoluminescence
(TRPL) that is called timecorrelated
single
photon
counting2J.
It is a
contactless, optical technique that is capable of
measuring the minority-carrier
lifetime in small
(-several pm2) areas and of making plan view lifetime
maps of the material. We analyze the data in terms of
models that provide bulk minority-carrier lifetime and
surface recombination data. Here we measured the
minority-carrier lifetime of the p-CdTe of a number of
devices. This study enabled us to understand many of
the important electronic processes which control
device performance and efficiency. The performance
ltacts
Ii-0 i;dS
Figure 1.
15-l
Schematic of the solar cell and the measurement conflguratlon
Early work1 on these devkes repotted an AM1 efficiency
of 6.7%. Recent devices2 have demonstrated AM1 efficiencies of 12.7% as measured at SERI. These CdTe
films were ail prepared by the Photon Energy proprietary. low cost spray technology and were fabricated
with a variety of additives. The device is fabricated by
first depositing indfum tin oxide (ITO) on a glass substrate followed by n-type CdS (-1~10~~ cm-s). This was
followed by the deposition of a Mlevel p-type CdTe layer
that was about 6ym thick The initial layer was composed of a high density CdTe layer with an average
grain size of about 3pm. The ffnai layer consisted of a
variable grain size (l-4pm) composite of low density
material (about 2/3 theoretical density). Capacitancevoftage measurements on the heterostructure indicated
an effective carder concentration of about 2x1015 cm-s.
The additives included compounds containing arsenic,
selenium, zinc, and sulfur, which were analyzed for
possible passivation of the grain surface recombination.
Some PV devices, that were measured by the timeresolved PL technique, were also measured for efficiency and correlations noted.
The PL was excited by a pulsed dye laser wavelength that was tuned to 600 nm and was therefore
transmitted
by the gfass/CdS window layer and
completely absorbed by the CdTe layer.
Photoluminescence spectroscopy measurements were run on
all devices prior to the ltfetime measurements. The band
edge PL peaked at about 1.47 eV and was reasonably
strong in most devices that we measured.
The PL
lifetime
measurements
were made when the
spectrometer was tuned to the peak wavelength of the
PL. Some devices were made without the back graphite
electrode so that the back CdTe grains could be optically excited. Thus we measured the lifetime in the
large-grain, dense CdTe and the smaller grain, porous
CdTe of the upper (back) layer. The TRPL response of
the CdTe grains in these two regions was quite different
as will be shown.
RECOMBINATION
THEORY
The recombination processes in semiconductor
devices have been extensively reviewed in the iiterature.
These included bulk radiative recombination
producing light emission and Shockley-Read-Hall
(SRH) recombination at deep defect levels. in addition,
SRH recombination
at surfaces and interfaces is
described by a surface recombination velocity.
In a
polycrystalline
semiconductor,
the grain boundary
recombination velocity is the relevant parameter. The
band-to-band radiative recombination
processes in
semiconductors in terms of a bulk recombination rate R
are expressed as follows:
R=Bpn
s-l. Van Roosbroeck and Shockieys derived a reiattonship between the absorption spectrum a(E) and the Bcoefficient. Because CdTe and GaAs both have a direct
band structure, comparable band gaps, and similar a
(E), we have assumed that the B-coefficients are comparable. We will define the excess minority-carder density
as An = n-no. After pulse excitation, one may show4 that
the PL intensity varies with time as:
IPLW
=
An(r,t)dv
where V is the volume of the semiconductor which
contains excess carriers. in this case, V is the volume of
a particular grain in the polycrystalline film. The PL
intensity then tracks the excess minority-carder dens&y
in time allowing a measurement of lifetime. The excess
carrier density is An and defining the majoritycarder
density as N, one can write the radiative recombfnatfon
rate as:
- B (np-nf) = -B (f&An +An2)
3)
In low injection (An << NA), one can easily show
that the radiative lifetime is ~/BNA. in high injection (An
>> NA), the solution of Equation 2 is:
An(t) =
Ano
1 +BAnot
4)
Here Ano is the the excess minorftycarrfer
den&y
generated by the laser pulse. The PL decay in this case
is nonexponential and is called bimolecular decay.
The semiconducting material may also contain
SRH defects in the volume and at the surface. The
surface SRH defects will be described in terms of a
surface recombination velocity (S). The total bulk iifetime ~6 is given by:
1
1
-=-+‘50 OR
1
%RH
GRAIN BOUNDARY RECOMBINATION
Recombination
mechanisms in poiycrystaiiine
semiconductors have been analyzed by a number of
researchers. Very early worW3,7 developed models for
Other
grain boundary recombination
in silicon.
researcherss.9
have developed models of poiycrystalline solar cells incorporating grain boundary recombination.
1)
To incorporate the grain boundary recombination,
one must assume a geometrical model and solve the
time-dependent
diffusion equation in that particular
geometry. Here we represented the grain boundary as
a sphere of radius with a surface recombination velocity
where B is a constant that is dependent upon the band
structure of the semiconductor.
To our knowledge, the
The B
B coefficient has not been published for CdTe.
value is well known for GaAs and is about 2 x lo-10 cm-s
15-2
S. Here we used a spherical coordinate system (r, 0. $J)
and equated the diffusion current and recombination
currents at the surface as a boundary condition. A Beers
law absorption in the spherical grain was incorporated
into the model to- give An&.@, the initial excess
minority-carrier
density, after the laser pulse is
absorbed. The solutions to the diffusion equation are :
-&=&+DG
12)
For small values of S (Sa/D < 1). one can solve equation
8 to find the effective PL lifetime:
13)
At large values (S a/D >> l), the solution to equation 8 is
ko - n/2a and therefore:
c c A,exp(-Dk&J)
n-0 m-0
x jnl(~~Prn(~S
0)
An(O)
= exp(-t/te)
f-5)
where.jm and Pm are the mth spherical Bessel function
and Legendre polynomial, respectively. Also D is the
minority-oarrfer diffusion coefficient and Amn is determined by the initial value of An&,@. By integrating over
the polar angle’ 9. we eliminate all terms except the m =
0 term and get:
00
An(r,t)
=exp(-Uru)
c
n-0
A, jo(k,r)exp(-Dk$)
7)
The quantity kn is a solution to the transcendental
eigenvalue equation that comes from the boundary
condition on S.
.,_
$=9-g
In the situation described by equation 14, the lifetime is
limited by the diffusion transit time to the gram surfaces
and only depends on the minority-carrfer dfffusivity.
Figure 2 is a plot of equation 12 versus the grain
boundary S with grain radius a as a parameter and D =
20. A range of four grain sizes is used corresponding to
the sizes in this polycrystalline CdTe. The p-type doping
of the grains is about 2x1015 cm-s.
Using the
B-coefficient of GaAs, the radiative lifetime in this doping
range is about 1 ps which is the value used for sg here.
We see from the figure that the surface recombination
will dominate the lifetime unless the S value is less than
104 cm/s. This calculation will be discussed further in
relationship to the data.
tan(k,a) = &-
loo0
The n = 0 solution lies in the first or second quadrant
with fJ < koa < R. The higher order solutions to the
eigenvalue equation 8 lie in quadrants 3 and 4, etc. It is
easy to show then that the solutions to the argument kn
increase rapidly with n.
By integrating An(r,t) over the volume of the grain
according to equation 2, one calculates the total PL
intensity as afunction of time:
An(t) = exp(-tits) c Cnexp(-Dkzt)
tl=O
9)
Here Cn is An times the integral of j,(knr) over r. The
n&l term dominates at long times.
t>l_
100
z
c
'O
c-,
1
grain diameter
.l
.Ol
AC&
Dk: Dx*
S (cm/s)
10)
Flgure 2. Mlnortty-carrler Iltetlme of a gratn In ns versus the
sur’tsco recomblnetlon velocity wlth gmln m&s a
18 a pammeter
And therefore:
An(t) = Coexp(-the) exp(-Dk$t)
14)
11)
From equ&fon 11, one can see that the lifetime at times
is greater than that of equation 10:
Photolum&escence
Soectrurct
The room temperature PL spectrum of the largegrain and small-grain sides of the CdTe film were
obtained prior to measuring the TRPL. The TRPL data
are only useful if one is measuring the band-to-band
recombination luminescence.
Typical PL spectra of a
higher quality solar ceil is shown in Figure 3. The
primary light emission emanates from the strong band
gap transition of CdTe that peaks at 1.47 eV. Curve A is
from a front surface (adjacent to the CdS) that is the
large-grain region. Curve B is from a back surface
(small grain area adjacent to the contact) and is about
one order or magnitude weaker. The minority-carrier
lifetime is larger in the front surface than in the back
surface. This result will again be verified by the TRPL
data.
l
Small grab
by the laser. At longer times (t :, 12.5 ns). the extended
PL “tail” is related to delayed or secondary !IJminescence. This signal is produced by electrons that are first
trapped, reemitted to the conduction band, and then
recombined with holes to produce photons. These
trapping effects will be discussed in a later section.
(x10)
0.0
12.5
25.0
t (ns)
Flgure 4.
1.25
1 .so
1.75
EW
Figure 3. The photolumlnesconce spectrum of the polycrystaltlns Cdl0 using SW nm laser l xcttatlon.
The Iargo gnln snd 8malt gmln response In
photon counts Is shown.
Our measurements on a number of solar cells
always indicated a larger lifetime for the front-surface
CdTe than for back-surface grains. Many solar cells had
front-surface lifetimes in the 1 to 1.5 ns range and subnanosecond back surface lifetimes and will not be
reported here. As noted, some of these devices were
made without the back contact so that the small grain
regions could be analyzed by TRPL.
Figure 4 shows low injection data from the best
device that was measured. The decay is generally very
nonexponential
indicating a variety of recombination
rates within the CdTe film. Nonexponential behavior is
expected because of the variation in grain size when
grain boundary recombination is dominant. This follows
from equations 12, 13, and 14. The data from the larger,
dense-grain CdTe shows a lifetime of 5.67 ns after the
excitation pulse is followed by slower decay at longer
times. The initial decay represents the signal from the
smallest grains in the grain recombination model. The
decay times found here are surprisingly large for a polycrystalline film. The initial back surface lifetime of this
device is 2.0 ns. indicating a smaller average grain size.
We believe that the initial part of the decay represents
the primary radiative decay in the grains being excited
TRPL response (photon counts vsmus ttmo) ot
the large-graln end smrti-gnln mgtons at lowlnjectkn
As noted, the low injection radiative* lifetimes
(l/BN) are relatively large compared to the measured
values. If one assumes that grafn boundary recombfn
nation is the dominant mechanism, the Me!fn?a fs
by equation 12. We will assume that the pnm&enI fzt surface grain radius Is 2.0 pm and determine S from
Figure 3. From the measured valueof %pc; thereoomt&
nation velocity is calculated to be 5x104 uWs with a - 2
)lm. These S values are surprfsingfy smafl compared to
For
what one might estimate for a grain surface.
example, the S of bare GaAs has been measured at
about 1xl 07 cm/s. Our preliminary measurements of
lifetime in single crystal CdTe also indicates that S is
relatively large.
The low-injection lifetime has a nonexponential
decay component.
This is easily accounted for by
realizing that a distribution of grain radii are being
analyzed by the measurements. A more exact analysis
would use a model of a weighted distribution of grain
radii. The net lifetime is then a sum of the lifetimes of
equation 12 multiplied by the population factor for that
grain radii. A complementary scanning electron micrograph of the specific area being analyzed will aid that
analysis.
Lifetime measurements were made as a function
of laser power. As the power density was increased
beyond a certain limit, the initial lifetime begins to
decrease, and a bimolecular decay described by equation 4 becomes evident. These data on the same device
are shown in Figure 5. Calculations of the initial An
using the laser energy per pulse and the absorption
coefficient of CdTe indicate that the Ano may easily
15-4
exceed 10” cm-s at our higher laser powers. The
majority-carrier density was depleted in grains near the
CdS interface and is less than the background doping.
Therefore, the onset of high injection conditions will
occur at fairly low light levels in this material.
The
decrease of lifetime, as seen in Figure 5. is easy to
explain by high-injection effects. The high-injection,
bimolecular response was seen whenever the radiative
recombination rate exceeded the grain boundary rate.
K. A fit of the two data points to exp(-AEIKT) form indicates that AE is on the order of 20 meV and therefore a
very shallow trap. A more rigorous data fit would take
into account the conduction band density of states and
the variation of Fermi level with temperature. One might
suspect that a continuum of surface states exist that
interact with the electron as the Fermi level changes with
temperature.
The initial decay, on the other hand, becomes
faster as the temperature is decreased. The data of
Figure 7 show the inifial high injection decay at 300 K
and 77 K. The effective lifetime decreases from 2.0 ns at
300 K to 0.5 ns at 77 K. This is the expected behavior
for radiative recombination in direct band-gap semiconductors. As the B-coefficient increases inversely with
temperature11 (B m T-1.6) the radiative recombination
rate increases accordingly and dominates the low-temperature lifetime. As the high-injection lifetime is controlled by B, this temperature behavior substantiates the
high injection explanation of the data.
t W
m
Flgure 5.
hieh 1. 308 K 1
TAPL response (photon counts versus time) of
the large-grain and smell-grab regions at highlnjectlon
The delayed PL response was clarified by the
temperature effects shown in Figure 6. For thermally
activated processes, the “lifetime” increased as the temperature was lowered. Delayed emission or phospho-
\ 7 = 0.51 ns
0
2
4t(ns)6
7 = 402 ns
1 .
0
Figure 7.
100
at grain boundary passivation were
certain compounds to the p-CdTe. The
compounds containing arsenic, selesulfur. However no significant changes
found as a result of any of the above
DISCUSSION
Flgure 6.
The delayed phololumlnescence
or phosphorescence regimes (photon counts versus time) of
the CdTe at 300 K and 77 K
lo
The lnltlat photolumlnescence
response (photon
counts versus time) of the same CdTe device at
300 K and 77 K
Attempts
made by adding
additives were
nium, zinc, and
in lifetime were
additives.
t ;tG,
6
OF RESUiTS
We have focused here on the best material found
after measuring a number of devices. Devices made
from this deposition had AM1 efficiencies of 8.7% to
9.0%. The open-circuit voltage (V,,) of these devices is
a better indicator of basic performance because the
back contact appears to vary in quality on the same set
of devices. The V,, of the l-2 ns devices ranged from
0.75 to 0.79 volts. This compares well with the best’*
reported device of 0.83 V. However, the short-circuit
rescence1° was produced when electrons are emitted
from traps and radiatively recombine. The “lifetime” in
this case reflects the emission rate from traps. The
behavior is indicative of a trap emission rate that varies
as exp(-AEIKT) where AE is the trap depth. Here, the
lifetime is 31.5 ns at 300 K and increases to 402 ns at 77
15-5
SUMMARY
The TRPL technique has proven to be a very
powerful diagnostic tool for CdS/CdTe solar cell
fabrication.
The technique has provided, for the first
time. information about the recombination processes
and lifetime measurements in polycrystalline grains of
CdTe. Our best film showed a low-injection lifetime of
over 5 ns. The minority-carrier lifetimes in many films
were much larger than might be expected in a polycrystalline semiconductor. The technique provides a quick,
contactless estimate of the CdTe film quality. Future
work will correlate the minority-carder lifetime with the
morphology and cell performance. Work will be undertaken to analyze surface state passivation. Our work
indicates that either larger grain sizes or surface state
passivation will be needed to achieve higher open circuit voltage.
11. H. Karamon, T. Masumot, and Y. Makino. J.
ehus, 57, (1985); p. 3527.
12. c
f”l 1Ssa (Solar Energy Research Institute, Golden,
Colorado, 1991). p. 110.
ACKNOWLEDGMENT
This work was performed under Contract No. DEAC02-83Ch10093 to the U. S. Department of Energy.
REFERENCES
1.
V.P. Singh, R.H. Kenney, J.C. McClure, S.P.
Albright.
B. Ackerman
and J. F. Jordan,
of
the
19th
IFFF PB
. .
-IEEE.
New York, 1987); p.
216.
2.
mc
Pros
(Solar Energy Research
Institute,
Colorado, 1QQO),p. 115.
3.
Golden:
W. ,van Roosbroeck and W. Shockley, Phvs.
94, (1954); p.1558.
15-6
HIGH-EFFICIENCY
HETEROEPITAXIAL
InP SOLAR CELLS
M. W. Wanlass, T. 1. Coutts, 1. S. Ward, and K. A. Emery
National Renewable Energy Laboratory
(formerly the Solar Energy Research Institute)
Golden, Colorado, USA
ABSTRACT
High-efficiency,
thin-film one-sun and concentrator
InP solar cells grown on GaAs substrates are reported. A
novel, compositionally
graded heterostructure
is used to
grow high-quality
InP layers. One-sun cells have AM0
efficiencies as high as 13.7% at 25°C (equivalent to 15.7%
under the global spectrum).
For the concentrator cells, at
2S°C, a peak conversion efficiency of 18.9% under 71.8
AM0 suns has been achieved.
Under the direct
spectrum, the equivalent efficiency is 21.0% at 88.1 suns.
At 80X, the peak AM0 efficiency is 15.7% at 75.6 suns.
These are the highest efficiencies
yet reported for InP
heteroepitaxial
cells.
Temperature
coefficient
data for
the concentrator cells are also presented. Approaches for
further improving the cell performance are discussed.
INTRODUCTION
InP solar cells are particularly
attractive for space
applications
due to their resistance to radiation damage
and demonstrated
high energy conversion
efficiency
under the AM0 spectrum (1, 2). Single-crystal InP wafers,
however, have characteristics
that make them generally
undesirable for solar cell fabrication and operation.
These
include high cost, high fragility, high mass density, and low
thermal conductivity.
Thus, in order to promote the
widespread
use of InP cells in space it is critical that
techniques are developed for fabricating high-efficiency,
thin-film InP cells. Three approaches are currently under
investigation
for solving this problem and they include
cleavage of lateral epitaxial film for transfer (CLEFT) (31,
using a bulk InP wafer, chemical separation (4) from an
tnP wafer, and heteroepitaxy
onto single-crystal
materials
with more desirable characteristics.
Of the three options,
heteroepitaxy
may prove to be the preferred
choice
because, ultimately,
large-area thin films of InP may be
too difficult
to handle and process on a large scale.
Furthermore,
it is uncertain
whether
the InP bulk
substrates used in the CLEFT and chemical separation
processes will actually be reusable. Heteroepitaxial
cells
have the advantage of being fully compatible with existing
cell processing technologies
as well as being based on
mature, single-crystal wafer technologies in materials such
as Gags, Ge and Si.
Due to the large differences in lattice constant and
thermal expansion
coefficient
between
InP and the
above-mentioned
materials, problems generally arise that
inhibit the growth of high-quality
InP heteroepilayers.
For
example, the lattice constant mismatch is 3.7% between
InP and GaAs and 7.5% between InP and Si. Such large
mismatches
result in high mechanical
stresses in the
resulting epilayers, which in turn, lead to the generation
of a high density
of defects.
The defects include
dislocations,
stacking
faults, and even microcracks.
Several techniques
have been investigated for reducing
the density of defects in the InP layers, thereby reducing
their deleterious effects. These have included thermally
cycled growth, post-growth annealing, and inclusion of an
intermediate GaAs layer for the case of InP grown on a Si
substrate.
Limited success has been realized with these
procedures and InP epilayers with dislocation densities of
-3 x 108 cm-2 and minority carrier lifetimes of -1 ns or less
in undoped material are reported for the best cases when
grown on GaAs substrates (5). Unfortunately,
InP layers
with these properties are of insufficient
quality for the
fabrication of high-efficiency
solar cells. Using post-growth
annealing,
the highest efficiency
for InP cells grown
directly on GaAs substrates is 10.8% (one-sun, AMO,
25°C) (6). Even lower efficiencies have been reported for
InP cells grown on Si substrates (7).
In previous work (81, we reported on the use of a
novel structure for the growth of high-quality InP epilayers
on substrates such as GaAs, Ge, and Si. A full description
of the device structure concept is given in reference 9.
The structure utilizes a compositionally
graded Ga,ln,,As
layer disposed between the bulk substrate and the InP
device layers.
This serves to reduce substantially
the
dislocation
density
in the InP device
layers when
compared
to the conventional
techniques
discussed
above. In this work, substrates of GaAs and GaAs/Si were
placed side by side in the growth reactor and identical
structures were deposited on each.
The resulting InP
epilayers were then characterized
using transmission
electron
microscopy
(TEM), electron-beam-induced
current (EBIC), and photoluminescence-decay
(PL-decay)
lifetime techniques
to assess the defect density and
n+/p shallow homojunctions
minority carrier lifetime.
were grown into the InP layers and solar cells with grids
designed for one-sun operation were processed from the
structures
grown
on the GaAs
substrates
only.
Additionally,
structures with three different Ga,In,-,As
graded layer thicknesses (8, 12, and 20 Pm) were grown
and characterized;
however, the InP material and solar
cell quality were essentially independent of the thickness
With this structure, dislocation
chosen in this range.
densities of 3 x 107 cm-z and minority carrier lifetimes of
over 3 ns were achieved in the InP layers using either
Furthermore,
the InP
GaAs or CaAs/Si substrates.
epilayers were completely
free of microcracks
in both
cases, which is an extremely
important result for highquality solar cell fabrication.
InP solar cells with one-sun
efficiencies of 13.7% (AMO, 25°C) and 15.7% (global,
25°C) were fabricated on GaAs substrates using an 8 pmthick Ga,In,.,As graded layer. Unfortunately,
pinholes in
the InP layers grown on the GaAs/Si substrates, resultine
from surface contamination
prior to growth, precluded
the fabrication of cells in this case. However, it seems
reasonable to assume that InP cell efficiencies similar to
those achieved using GaAs substrates should be possible
on 5i substrates due to the similar dislocation
densities
and minority carrier lifetimes observed in the InP layers
grown on either substrate type.
in the remainder
of this paper, we describe the
epitaxial
growth, fabrication,
and characterization
of
concentrator
heteroepitaxial
InP solar cells grown on
GaAs substrates,
using a compositionally
graded
intermediate
structure similar to that described above.
The cell performance has been determined as a function
of the concentration
ratio
and the operating
temperature.
We have also investigated the behavior of
the cell performance
parameter temperature coefficients
as a function of the concentration ratio. The details of this
work are described in the sections that follow.
DEVICE STRUCTURE
A schematic diagram of the heteroepitaxial
(HE) InP
solar cell structure grown on a GaAs substrate is given in
Figure 1. The structure is initiated with a thin buffer layer
of p-GaAs, which is then followed by the p-Ga,ln,-,As
linearly graded layer (LGL), which has a thickness of 8 pm
for the results reported here. The LGL is followed by a
buffer layer of Gaoe4, In,,, As, which is lattice matched to
InP. The InP solar cell layers are finally deposited at the
top of the structure and these comprise a high-efficiency
n+/p shallow homojunction
(SHJ) cell structure. (In Figure
1, BSFL is an acronym for back-surface field layer.) A back
contact of pure Au is applied to the exposed bottom
surface of the GaAs substrate.
The top grid contact on
the surface of the InP cell emitter is also composed of
pure Au. A two-layer antireflection
coating is deposited
on the front surface of the cell structure and an Entech
prismatic cover is also incorporated
into the structure to
allow for a high top-contact-metallization
coverage
Further details of the device structure are
(-20%).
discussed below.
EXPERIMENTAL
The heteroepitaxial
solar cell structures were grown
by atmospheric-pressure
metalorganic
vapor-phase
epitaxy (APMOVPE), using a specially designed, radiofrequency (RF)-heated vertical reactor vessel (101, which
yields highly uniform epilayers.
The growth system is a
home-built,
run-vent type and uses palladium-purified
hydrogen as the carrier gas through the main mixing
manifold and through each of the metalorganic
source
cylinders.
The primary reactants used in the growth
..A
process included trimethylindium,
trimethylgallium,
pure
phosphine, and pure arsine. The sources for p- and ntype doping were diethylzinc
and SOO-ppm hydrogen
sulfide in hydrogen,
respectively.
Zn-doped p+-GaAs
wafers oriented
2O oif the (100) were supplied
by
Sumitomo Electric, Inc. and used as substrates.
These
were loaded directly into the growth reactor as received
from the vendor (i.e., without any pre-growth cleaning or
etching steps). Prior to growth, the GaAs substrates were
heated to 7OO“C for 10 min with arsine flowing into the
reactor vessel. Growth was then carried out at a constant
The structures were grown at a
temperature of 650°C.
rate of 75-175 nm min-l in a continuous sequence of
without
stop-growth
periods
at the
steps (i.e.,
heterointerfacesj.
A typical growth run takes about 2.5 h,
including the time required for warm-up and cool-down
of the reactor vessel. The entire process is controlled and
monitored using’s home-built, PC-based control system.
16-2
The epitaxial
structures were then processed into
completed
concentrator
solar cells, using conventional
techniques.
Ohmic, low-resistance
contacts were made
to both the back surface of the p+-GaAs substrate and the
n+-InP
emitter
surface,
using electroplated
AU as
deposited.
The back surface of the GaAs substrate was
etched in 1% by volume bromine in methanol for 5 min
at room temperature
prior to applying the metallization.
The top contact
and device mesa geometries
were
defined by photolithographic
techniques, using positive
photoresist.
The top contact
grids were specially
designed to accommodate
an overlying Entech prismatic
cover, which was originally
designed for concentrator
GaAs solar cells (11). A center-to-center grid line spacing
of 127 pm was used, and the individual grid lines have a
cross-sectional area of -125 pm2 (-25 pm wide by -5 pm
high), A busbar is included at both ends of the grid lines
in this design to allow for the simultaneous placement of
test probes at both ends. This aspect of the grid design
results in better performance
under concentration.
Through the use of the Entech cover, it is possible to
cover -20% of the cell surface with the grid metallization
without
incurring any photocurrent
losses due to grid
obscuration.
This allows for ample grid metallization
on
the cell, which results in low electrical power losses within
the top contact. As such, the Entech cover has proven to
be a very important component in the fabrication of highefficiency concentrator
cells.
Electrical isolation of the
individual
cells was accomplished
by etching moats
through the n+/p InP junction with concentrated HCI. A
two-layer antireflection
coating of ZnS (-55 nm) followed
by MgF, (-95 nm) was then deposited on the front surface
of the device wafer.
The concentrator
cells were
completed by installing the Entech cover. A typical array
of completed
heteroepitaxial
InP concentrator
cells is
shown in Figure 2. The effect of the Entech cover is also
illustrated in this figure. Each individual cell has an area of
0.0746 cm* which is computed by subtracting the areas
of the two busbars from the total device mesa area (this is
a standard area definition for concentrator solar cells) (12).
The performance
of the concentrator
cells was
characterized
by measuring
the absolute
external
quantum efficiency (AEQE) as a function of temperature
as well as the illuminated
current-voltage
characteristics
as a function of the temperature
and the concentration
ratio.
The latter data sets were used to calculate the
dependence
of the cell
performance
parameter
temperature coefficients on the concentration
ratio. ‘The
measurement techniques have been described previously
(13). All of the results reported here are referenced to
the AM0 spectrum
(14).
A discussion
of the cell
performance
is given in the following
section.
RESULTS AND DISCUSSION
Initially,
the current-voltage
characteristics
for the
cells were measured as a function of temperature under
one-sun AM0 conditions in order to obtain the necessary
information
for evaluating
the efficiency
under
concentration
(i.e., the one-sun short-circuit current (I,,) is
needed
to calculate
the concentration
ratio
for
concentrator
measurements).
To within experimental
of temperature.
error, we found I,, to be independent
The AEQE data shown in Figure 3 illustrates why I,, is
temperature
independent.
As expected, the InP band
edge shifts to longer wavelengths
as the temperature
increases, and one would normally expect an increase in
I,, due to this effect. However, a concomitant decrease
in the short- and mid-wavelength
response is also
observed for these devices as the temperature increases,
which offsets any increase in I,, due to the band gap shift.
Thus, I, remains essentially constant as the temperature
is increased. Note that the blue response for these cells is
relatively low. This characteristic
is typical of SHJ solar
cells that have a high surface recombination velocity. We
have shown in previous work that graded emitter doping
profiles can be used to improve the blue response in
these cells (15). However, a technique for effectively
passivating the emitter surface needs to be developed in
order
to realize
InP cells
with
near-theoretical
performance characteristics.
The HE InP cell performance
was then tested as a
function of the temperature and the AM0 concentration
ratio, and the results from these measurements are shown
(Figure 4)
in Figures 4 and 5. The AM0 efficiency
increases rapidly at low concentration
ratios and then
reaches a broad plateau for concentration ratios of -40 or
more. At 25“C, the cells have efficiencies of close to 19%
over a broad range of concentration
ratios. This value
decreases to -16% as the temperature
is increased to
80°C. The broad plateau in efficiency can be understood
by examining the open-circuit
voltage (V,,) and fill factor
(FF) versus concentration
ratio data given in Figure 5. The
behavior of V,, is as expected.
In fact, when the V,, data
are plotted against In (concentration
ratio), a straight line
is obtained.
However, the FF data indicate that the cells
quickly
become
series-resistance
limited
as the
concentration
ratio is increased
beyond
-20 suns.
Additionally,
this effect appears to be enhanced as the
operating temperature
is increased.
An analysis of the
resistance components
contributing
to the overall series
resistance for these cells shows that the emitter sheet
resistance
is primarily
responsible
for limiting
the
concentrator
cell performance.
A lower emitter sheet
resistance or a smaller grid line spacing will be necessary
to improve this aspect of the ceil performance.
The
broad plateau in efficiency versus concentration
ratio is
seen to be due to offsetting effects of the V,, and FF as
the concentration
ratio increases.
Current-voltage
data for an HE InP concentrator cell
at peak efficiency are shown in Figure 6. At 250~ the
efficiency
reaches 18.9% under the AM0 spectrum at
71.8 suns. As shown in Figure 4, the peak efficiency at
8O“C is 15.7% at 75.6 suns. Under the direct spectrum at
25’C, the peak efficiency is 21 .O% at 88.1 suns. These
values are very encouraging and demonstrate that HE inP
cells have the potential to reach high efficiencies at high
concentration
ratios and high temperatures.
Additionally,
these results show that the HE cell efficiencies improve
dramatically
when operated under concentration.
Using the data shown in Figures 4 and 5, we have
calculated
the temperature
coefficients
for the HE InP
cell performance
parameters
as a function
of the
concentration
ratio. As a basis for comparison, we have
also fabricated homoepitaxial
(HO) InP concentrator solar
cells on single-crystal
InP substrates
with junction
structures that are similar to those used in the HE InP
Similar
concentrator
measurements
and
cells.
temperature
coefficient
calculations
have
been
performed for the HO InP cells. In Figure 7, we compare
the V,, temperature coefficients for the two types of cells
At low
as a function
of the concentration
ratio.
concentration
ratios, the HO cells clearly outperform the
HE cells. However, at high concentrations,
the HE cell
temperature
performance
improves
substantially
and
approaches that oi the HO cells. This result highlights an
additional
advantage
of operating
the HE cells under
concentration.
realize higher efficiencies
at high concentration
ratios.
SUMMARY
High-efficiency
heteroepitaxial
InP solar cells have
been fabricated
on GaAs substrates using a novel,
compositionally
graded, intermediate
layered structure.
One-sun cells have AM0 efficiencies as high as 13.7% at
25’C.
The concentrator
cell performance
has been
characterized
as a function of the temperature and the
AM0 concentration
ratio.
Peak concentrator
AM0
efficiencies of 18.9% at 71.8 suns, 25X, and 15.7% at
75.6 suns, 80°C, have been obtained with these cells,
which are the highest efficiencies
yet reported for InP
heteroepitaxial
solar cells. It has also been shown that the
conversion-efficiency
temperature
coefficient
for these
cells improves substantially
as the concentration
ratio is
increased.
The advantages of operating the HE InP cells
under concentration
include reduced cell area, higher
conversion
efficiencies,
and improved
temperature
performance.
The cell performance
is presently limited by three
main loss factors: (1) recombination
at the surface of the
emitter layer, (2) high emitter-layer
sheet resistance
leading to reduced FF values at high concentration,
and
(3) high density of threading dislocations in the active cell
layers. Improvements
in any of these areas will lead to
increased cell efficiencies.
Technologically,
it would
be important
and
immediately useiul if the results obtained in this work for
InP cells grown on GaAs substrates could be duplicated
using Si substrates. Such a result would make HE InP cells
a viable contender
for space power applications,
and
efforts toward this goal are currently under way.
ACKNOWLEDGEMENTS
Efficiency and FF temperature
coefficient
data ior
the HE tnP cells as a iunction oi the concentration
ratio
are plotted in Figure 8. The data indicate that the
temperature
periormance
of the FF actually degrades
with increasing concentration.
This behavior is linked to
the series-resistance
problems
discussed previously.
Nevertheless,
the temperature
periormance
of the
conversion
efficiency
actually
improves
as the
concentration
ratio is increased due to the behavior of
the V,, temperature coefficient (shown in Figure 7). The
temperature
coefficient
of efficiency
would improve
much more rapidly with concentration
if the cell series
resistance were reduced.
This problem remains as an
important
one to solve for these devices in order to
Support for this work was provided by the U.S.
Department
oi Energy under contract No. DE-AC0283CH10093 through an award from the NREL Director’s
Development
Fund.
REFERENCES
1.
M. Yamaguchi,
Shibukawa,
2.
V.E. Haven, and S.M. Vernon,
21 st
Conf., 141 (1990).
16-4
A. Yamamoto,
and A.
Jap. I. Apol. Phvs., 23, 302 (1984).
C.J. Keavney,
Record
C. Uemure,
IEEE
Photovoltaic
Conf.
Soecialists
3.
R.W. McClelland,
C.O. Bozler, and J.C.C. Fan, Appl.
Phvs. Lett., 37, 560 (1980).
4.
M.B. Spitzer,
Conf.
B. Dingle,
Record
J. Dingle, and R. Morrison,
21st IEEE Photovoltaic
Specialists
Conf.. ‘I96 (1990).
5.
S.M. Vernon,
‘Karam,
M.M.
C.J. Keavney,
Al-Jassim,
E.D. Gagnon,
N.M:Haegel,
N.H.
V.P. Mazzi,
and C.R. Wie, Proc. Mat. Res. Sot. Svmo., 198, 163
(1990).
6.
C.J. Keavney,
Spire Corp., private
communication,
(Mar. 1991 I.
7.
C.j. Keavney,
Record
S.M. Vernon,
20th
IEEE
and V.E. Haven,
Photovoltaic
Conf.
Soecialists
_
Conf.. 654 (19881.
8.
M.M. Al-Jassim,
R.K Ahrenkiel,
.Olson, and S.M. Vernon,
Back
Contact
M.W. Wanlass, J.M.
Schematic diagram
Figure 1.
structure on a GaAs substrate.
Proc. Mat. Res. Sot. Svmp.,
of the HE InP solar ceil
198, 235 (1990).
9.
M.W.
Wanlass
4,963,949
10.
M.W.
and P. Sheldon,
U.S Patent
No.
(Oct. 16, 1990).
Wanlass,
U.S. Patent No. 4,649,859
(Mar.
17, 1987).
11.
M.J. O’Neill,
12.
1987).
Terrestrial
Report
13.
M.W.
U.S. Patent
Photovoltaic
ERDA/NASA/l
Wanlass,
T.A. Gessert,
No. 4,711,972
Measurement
02277/l
J.S. Ward,
(Dec.,
Procedures,
6, (June, 1977).
K.A. Emery, T.J. Cows,
and C.R. Ostetwald,
Solar Cells, 30,
363 (1991).
14.
C. Wehrli,
Extraterrestrial
Meteorological
Solar Soectrum,
Observatory
and World
Center, tech. rep. no. 615, Davos-Dorf,
Physical
Radiation
Switzerland,
(July 19851.
15.
M.W. Wanlass,
Coutts,
Norman,
G.S Horner,
Plan-view photomicrograph
of a typical array
Figure 2.
of HE InP concentrator cells. The cell in the center of the
micrograph has an Entech cover properly installed.
T.A. Gessert, and T.J.
Proc. 1st Int. Conf. on InP and Rel. Mat.,
OK, U.S.A.; March 1989, 1144, 445-458,
SPIE (1989).
16-5
950 ,
, 81
z 850
sl
rs 800
P
Wavelength
Voltage
**.* Fill Factor
I
0
(nm)
AEQE data for an HE InP concentrator
25°C and 8OT.
Figure 3.
-
79
-
78s
77 50
Iif
1 76z
600
F?s,9s,3U.H8
80
-
650
0
-
20
I
I
I
-
75
-
74
I
73
40
60
80
100
AM0 Concentration Ratio
120
Figure 5.
Open-circuit
voltage and fill factor data for an
HE InP concentrator
cell as a function of the operating
temperature and AM0 concentration
ratio.
cell at
19
18
I
I
AMO, 71.8 suns, 25T
I
I
I
V 06
0.902 v
2588 mAcm-2
ll:
18.9%
I
I
I
14
13
12
20
Figure 4.
concentrator
temperature
40
60
80
100
AM0 Concentration Ratio
.2
120
AM0 conversion efficiency data for an HE InP
cell as a function
of the operating
and AM0 concentration
ratio.
0
0.2
0.4
0.6
Voltage (VI
Figure 61 Current-voltage
data
concentrator
cell at peak efficiency
AM0 illumination.
16-6
0.8
for an HE InP
under concentrated
HE InP cell
/
I
0
20
I
I
40
60
I
AM0 Concentration
I
80
100
Ratio
120
Figure 7. Open-circuit
voltage temperature coefficient
data as a function of the AM0 concentration
ratio for HO
and HE InP concentrator cells.
-2800
-500
-.
-600
y -2900
:
E
3% -3000
6
&
?T
32
= -3100
;z
b
0
<
*.
-.
-.
*.
-700
-800
-900
..
-.
*.
-.
$ -3200
&
r;
%
6
&
‘0
$
=.LL
-1000
-3300
-1100
0
Figure 8.
temperature
concentration
20
40
60
80
100
AM0 Concentration Ratio
AM0 conversion
120
efficiency
and fill factor,
coefficient
data as a function
of the
ratio for an HE InP concentrator cell.
ASTM PHOTOVOLTAIC STANDARDS DEVELOPMENT STATUS
CR. Ostcrwald
National Renewable Energy Laboratory
(Formerly the Solar Energy Research Institute)
Golden, Colorado USA
(titled Photovoltaic Electric Power Systems) of ASTM technical main committee E44 (titled Solar, Geothermal, and other
Alternative Energy Sources). Other subcommittees of E44
deal with subjects such as wind energy conversion, solar
thermal systems, materials performance,
and geothermal
E44.09 currently has responsibility for
energy conversion.
nineteen standards and draft documents that fall roughly into
three categories: photovoltaic (PW reference cell calibration,
device and module characterization,
and module environmental testing. At the present time, there are four reference
cell calibration standards: one for secondary cells and three
for primary cells. A reference cell packaging standard that
specifics physical characteristics is currently being revised.
Eight standards and draft documents cover cell and module
characterization,
includingsolarsimulation,
electrical performance, spectral response, linearity, spectral mismatch, and
module insulation resistance and ground path continuity.
Finally, module environmental tests cover: temperature and
humiditycycling,solar-UVweathering,
marineenvironments,
and hail resistance. Tables 1 and 2 list the exact titles and
current status of all standards and draft documents that fall
under the responsibility of E44.09. All adopted E44 standards
can be located in reference (2).
ABSTRACT
ASTM technical subcommittee E44.09, titled Photovoltaic Electric Power Systems, has been developing consensus
standards for photovoltaic
measurements and characterization for both cells and modules since 1978. This paper
presents a brief summary and lists the current status of each of
the E44.09 standards and draft documents.
fNTRODUCTION
ASTM is a well-known nonprofit organization devoted
to the development of voluntary full consensus standards for
materials, products, systems, and services. Although ASTM
standards are primarily used in the United States, they are also
widely used and referenced throughout the world. ASTM
standards are developed by technical committees whose
members represent producers, users, consumers, and general
interests. All participation
by members is voluntary and
therefore standards developed by ASTM committees reflect
the desires of organizations that feel the content of standards
is important for their business. ASTM technical committees
are generally organized into subcommittees charged with
developing standards for a particular problem or technical
area.
One special category of ASTM standards is test methods
that produce numerical results. ASTM requires such standards to have a precision and bias statement that presents the
results of an interlaboratory
intercomparison
(known as an
interlaboratory study, ILS, in ASTM standard language) of the
test method.
New test methods
may state that an
intercomparison
is pending, but cannot be reapproved after
the first five years until the intercomparison
is completed.
Several E44.09 standards are in this category, and the subcommittee is attempting to organize several intercomparisons
that will need industry participation for completion.
These
intercomparisons will cover reference cell calibrations, spectral response measurements, and module and cell performance measurements.
ASTM standards must pass ballots at three separate
levels before they are approved: subcommittee, main committee, and finally at the ASTM society level. Any negative
votes must be resolved according to ASTM rules before a
document can proceed to the next level. A negative vote that
is ruled persuasive requires thatthe ballot process begin again
at the subcommittee level. Negative votes can be ruled not
persuasive
by the subcommittee or main committee if a twothirds majority agree. Every ASTM standard must be reapproved every five years. If a standard is not reapproved, it must
either be revised or balloted for removal as a standard (1).
ASTM subcommittee E44.09 has been involved in developing and revising industry-consensus standards for photovoltaics since 1978. E44.09 indicates subcommittee 09
17-1
Table 1
Status of E44.09 Standards
status
Title
Standard No.
for Solar Simulation
Standard revised and adopted in
1991
E 927-91
Standard Specification
Photovoltarc Testing
for Terrestrial
E 948-83
Standard Test Methods for Electric& Performance of
Non-Concentrator
Terrestrial Photovoltaic Cells Using
Reference Cells
Standard revised, title changed,
needs ILS prior to ballot
E 973-91
Standard Test Method for Determination of the Spectral
Mismatch Parameter Between a Photovoltaic Device and
a Photovoltaic Reference Cell
Standard revised and adopted in
1991
E 1021-84
Standard Test Method for Measuring the Spectral
Response of Photovoltaic Ceils
Standard revised, needs ILS prior
to ballot
E 1036-85
Standard Methods of Testin Electrical Performance of
Nonconcentrator Terrestria $ Photovoltaic Modules and
Arrays Using Reference Cells
Standard revised, title changed,
needs ILS prior to ballot
E 1‘038-85
Resistance of
Standard Practice for Determinin
Photovoltaic Modules to Hail by 7mpact with Propelled
Ice Balls
E 1039-85
Standard Method for Calibration and Characterization
Non-Concentrator
Terrestrial Photovoltaic Reference
Cells Under Global Irradiation
E 1040-84
Standard Specification for Ph sical Characteristics of
Non-Concentrator
Terrestrial b hotovoltaic Reference
Cells
Currently
E 1125-86
Standard Test Method for Calibration of Primary
Non-Concentrator
Terrestrial Photovoltaic Reference
Cells Using a Tabular Spectrum
Standard to be revised, needs ILS
prior to ballot
E 1143-87
Standard Test Method for Determining the Linearity of a
Photovoltaic Device with Respect to a Test Parameter
Revision has passed subcommittee
ballot
E 1144-87
Standard Test Method for the Calibration of
Non-Concentrator
Terrestrial Photovoltaic Reference
Cells Under Direct lrradiance
Standard to be ballotted for
withdrawal Fall 1991 (replaced by
E 1125)
E 1171-87
Standard Test Method for Photovoltaic Modules in Cyclic
Temperature and Humidity Environments
Standard revised, needs adoption
of Draft 196 prior to ballot
E 1328-90
Standard Terminology
Energy Conversion
New standard adopted in 1990
E 1362-90
Standard Test Method for the Calibration
Nonconcentrator
Terrestrial Photovoltaic
Reference Cells
Relating to Photovoltaic
Solar
of.
Secondary
17-2
‘-
of
Standard revised, changed~to test
method, needs adoption of Draft
196 prior to ballot
Standard revised, title changed,
needs KS prior to ballot
being revised
New standard adopted in 1990
Table 2
Status of E44.09 Draft Documents
status
Title
Draft No.
191 R8
Standard Test Method for Saltwater Immersion and
Corrosion Testing of Photovoltaic Modules for Marine
Environments
Passed subcommittee ballot, needs
adoption of Draft 196 before
proceeding
192 R4
Standard Test Method for Natural and Artificial Solar
Radiation Weathering Tests of Nonconcentrating
Photovoltaic Modules
Currently
196 R3
Standard Test Methods for Insulation Integrity and
Ground Path Continuity of Photovoltaic Modules
Simultaneous sub-main committee
ballot Fall 1991
198 RO
Standard Test Method for Saltwater Pressure, Immersion,
and Tern erature Testing of Photovoltaic Modules for
Marine Pnvironments
New title, no draft yet developed
199 RO
Standard Test Method for Wet Insulation Resistance of
Photovoltaic Modules
Currently
CELL AND MODULE
CHARACTERIZATION
under revision
under revision
tors. Performance measurements are reported at Standard
Reference Conditions (SRC), that specify the measurement
spectral irradiance, total irradiance, and temperature. SRCs
can be 25”C/lOOO Wm2, Nominal Operating Cell Temperature (NOCT)/800Wm2, or specified by the user. Measurements that deviate more that specified amounts from SRC are
corrected back to SRC. Procedures for measuring NOCT and
temperature and irradiance correction factors are included in
the test methods.
STANDARDS
This specification classifies solar simulators according
to their spectral match to the global reference spectrum, their
temporal stability, and their spatial uniformity.
This test method is used to determine electrical performance parameters of solar cells such as efficiency, maximum
power, open-circuit voltage, and short-circuit current in a
solar simulator.
Spectral errors due to the simulator are
corrected using spectral mismatch (see E 973).
El143
This test method describes a way of quantifying nonlinearities in PV device parameters by performing a linear
least-squares fit through the origin.
Draft 196
This draft document provides methods of testing the
ground-path continuity, insulation resistance, and insulation
leakage current of a module (insulation leakage current is
commonly called hi-pot or high-potential
testing of insulators). All of the module environmental
tests reference this
document as part of the before-and-after
characterization
tests (see module environmental
testing section). Because
ASTM regulations require a standard to be adopted before it
can be referenced in another standard, balloting of the module environmental tests at the main committee level must wait
until draft 196 is approved.
This test method standardizes the calculation of the
spectral mismatch parameter between a test device and a
reference device. It sets requirements on spectral irradiance
measurements and shows how to correct spectral measurement errors.
E 1021
Spectral response/quantum
efficiency measurements
on solarcellscan beperformedwith
this tcstmethod. It places
restrictions on allowable measurement procedures in order to
eliminate the most common errors in these measurements.
E 1036
These test methods are analogous to E 948, the cell
performance measurement, but are applied to modules and
arrays instead. Light sources included in the test methods are
natural sunlight, steady-state simulators, and pulsed simula-
Draft 199
This draft standard is to be used for measuring the
insulation resistance of a module while the module is wet. It
is considered to be a much more sensitive test of the module
17-3
.
insulation than the dry insulation measurements. The test is
performed by either immersion of the entire module or by
immersion of a module edge in a surfactant solution and
measuring the resistance between the liquid and the shorted
module leads.
REFERENCE CELL CALIBRATION
MODULE
ENVIRONMENTAL
TESTING STANDARDS
All of the module environmental tests are structured as
a series characterization tests prior to the particular environmental exposures, with a repetition of the characterizations
following the exposures. These tests have their roots in thesocalled jet Propulsion Laboratory’s BlockVteststhatwere
used
to qualify lots of modules for block purchases.
STANDARDS
Reference cells are classified to be either primary or
secondary. Primary reference cells are calibrated in sunlight
against a wide-spectral-range
detector, while secondary cells
are calibrated against a primary reference cell in either
sunlight or a solar simulator.
E 1038
This standard tests the ability of modules to withstand
hailstorms by propelling ice balls onto the module surfaces.
For a given ice ball size, a free-fall terminal velocity is
combined with a selected windvelocitytoobtain
therequired
ice ball speed needed.
E 1039
This primary procedure is designed for silicon reference
cells calibrated against a global pyranometcr.
Atmospheric
conditions measured with a sunphotometer must be within
specified ranges. Spectral corrections are not used. The
procedure applies only to silicon devices because the restricted ranges were selected to minimize the spectral errors
over the silicon spectral response range.
El171
Two separate tests are contained in this standard, a
thermal cycling test and a humidity-freeze
cycling test. The
humidity-freeze
cycle consists of 20 hr at 85”C, 85% RH,
followed by a 0.5 hr -40°C freeze period, with a total of 10
cycles. Thethermalcyclingtesthasatotalof
2OQcycl.esofthe
module temperature ramped between -40°C and 90°C.
El125
Draft 191
This draft specifies a marine environment test intended
for modules designed to be deployed on or near shorelines. A
48 hr seawater immersion test followed by a 21 day 9O”C,
85% RH corrosion test.
This primary calibration is called the tabular calibration
method because spectral errors are corrected using a tabular
spectral irradiance of the solar spectrum at the time of the
calibration.
Reference cells are calibrated against a pyrheliometer, or, for reduced error, an absolute cavity radiometer.
Although the measurements are performed under direct normal solar irradiance, the results can be numerically corrected
to any desired reference spectrum.
Draft 192
This document contains three separate solar-UV weathering tests. All three simulate three years of exposure at
southern U.S. latitudes.
The first is a 36 month outdoor
exposure, the second an accelerated outdoor exposure using
concentrators, and the third an indoor accelerated exposure
that uses Xe arc lamps.
E 1144
This primary calibration is very similar to E 1039, the
silicon global calibration procedure, but is calibrated against
a pyrheliometer rather than a global pyranometer. It therefore
gives reference cells calibrated to the direct reference spectrum. Because of lack of use and because it has been largely
replaced by E 1125, the subcommittee has decided to withdraw this standard. To withdraw a standard a ballot must pass
each of the three balloting levels.
Draft 198
This draft is the pressure-immersion-temperature
(PIT)
test that was developed by the U.S. Coast Guard for modules
intended for offshore floating aid-to-navigation
power applications. The tests use a cyclic hot-and-cold pressurized
immersion in seawater that simulate modules being underwater for periods of time because of wave action.
E 1362
This test method specifies the secondary calibration
procedure where the test cell is calibrated against a primary
cell. Spectral errors are corrected using the spectral mismatch
parameter.
17-4
E 1040
This specification
for reference cells.
E 1328
Termscommon
document.
recommends
physical characteristics
Preparation of this paper was supported by the U.S.
Department
of Energy under Contract No. DE-ACOZ83HC10093.
KEFERENCES
to E44.09’s standards aredefined
in this
E 891 and E 892
Two other standards that are the responsibility of subcommittee E44.02 on environmental
parameters are referenced many times by E44.09 documents are E 891 and E 892.
These specify the direct normal and global reference solar
spectral irradiances.
1.
ASTM Technical Committee Officer Handbook,
ASTM, Philadelphia, Pennsylvania, 1985.
2.
1991 Annual Book of Standards, Vol. 12.02, ASTM,
Philadelphia, Pennsylvania, 1991 (revision published annually).
17-5
DESIGN
OF A FIBER OI’I-IC
l3&ED
SOLAR
SIMULATOR
Bhushan I,. Sopori nncl Craig Marshall
National Renewable Energy Laboratory
(formerly the Solar Energy Rrsrarch Institute)
Goklen. CO gO4O I
iwmwx
A new solar simulator is described which camhines three
light sources to produce an output beam that can closely match
a desired solar spectrum. The three selected light sonrces are
srritably filtered, nud their output beams RIP mixed hy ~WYWS
of a trifurcated, mndomi7ed liher cable to produce a highly
uniform intensity d&ibution
of the output benm. 7lte criteria
for selecting sauks,
filters, and other optical elements,
required to prodnce an AMI.
spectrum are discussed. 711~
output spectrum is cornpilnd
with the reference AM 1.5
spectral irradiance, and the results of single and multijnnction
solar cells tested under this simulator are compatPd with those
from other simdntors.
IHl-RODUCTlON
To date. the commercial solar simulators use R single light
sounze, typically
a xenon a: lamp, to deliver a beam of
crdlimated light that rpproximately
matches a desired solar
spectrum. A host i$ such simulators, with a variety cif O~II~III
are available
for both commercial
and
power
ratings,
labomtoty testing of solar cells (1‘2). Although the spectral
output of a xenon arc lamp“bas perhaps the most desirable
envelope of any single light source for simulating
a solar
specmlm, the output from a xenon arc lamp cannot he wsecl
directly because (i) the output of the lamp contains intense
emission lines that cover n wide range of the s~ctrun~.
particularly
the wavelength range between 0.8 and 1.2 1lm. and
(ii) the UV region of the output spectrrnn is too rich RS
Manufacturers
have designed
compared
to even AMO.
proprietary notch filters to rednce UV and IR contents of the
OUtput. Consequently,
a variety of optical filters are often
Further
necessary
to produce
a useable
spectrnm.
improvements
in the spectral matching can be produced hy
using a water caviry. However, even rfter extensive filtering,
the xenon arc lamp has several disadvantages:
(i) the output
ntains stmng emission liner, (ii) the peak of the envelope of
the xenon ~pechum is not located at the same wavelength as
that of the solar spectrum, and (iii) the absorption spectrum of
the liquid water cavity, typicnlly used with such a simulator,
is not :he same AS that of the water vnpor.
Although
a well-filtered
simulator is adequate f6r the
majority of generi+Infrpose
testing of .solar cells, accurate
testing
requires
that appropriate
corrections
he made
corresponding
to spectral mismatch between the simulator
outpnt and the standati solar spectrum, (e.g., AMI, AMI 3.
etc).
This procedure
involves
making
an additional
mrasuiement
of the spectral response of the cell and
identifying a suitable reference cell that can he used to set the
light intensity in the test plane. The methodology
for this
process has been well established. but it is recognized that
such corrections are also approximate
(3). A more serious
need for an accurate
simulator
has arisen in testing
mttltijnnction
cells where a current mismatch hetween vnrhtts
cells, due to mismatch between the refmnce md the simnlator
spectra, can pro&ice large errors, particularly jn the Ii11 factor
of the cell. Some of these difftcultier can be circumvented by
employing
a procedure which uses IWO different beams to
independently excite each cell. Ilowever, it has kcome clear
that the desired degree of spectral accuracy
cannot be
nchieved hy .a single source.
To reduce the complexity of such rneasnrennnts, attempts
are made tr, develop simulators cspahle of mulching
the
desired reference spectrnm with much better precision.
This
endeavor has led to the fabricrtiun
of two-sntmz
sirnuIatnr.c
consisting of a filtered xenon arc and a tungsten lamp which
have clearly demonstrated the capability
for significantly
Modifications
to the
improved
onlput spectra (4,5,6).
Spectrolab x-25 hy an integrating optics package have also
Recently, we have
altowed a more precise match (7.g).
proposed that an accurate match to AMO or AM13 can k
produced by combining output of three pmperly selected light
sources (9). One of the methods proposed to *‘mix” light from
three sources involves using a randomized, trifurcated optical
f&r cable. In n previous paper, we hrve shown that three
&II
in UV,
visible,
and IR contents,
soirrce-beams,
respectively, an sufficient IO produce a spectrum well matched
ro AM13 [ASTM E892 t3lobat]. in this pafur, we de.scribe
the essential design featuns of such a simulator and compare
test data of solar cells with the results obtained mler other
simulators.
We also descrlk Improvements on the previous
18-l
system that can yield a superior spectrum than previously
described. These improvements involve (a) A new combination
of filters, aud (b) the use of two sources to deliver three
beams.
whereas the UV leg consists of fused silica fikrs dispersed in
a glass mahix. 71te glass and the silica fikrs are 25.1 pm and
St?.8 ftm iii diameter, respectively.
The output from tk
com111on leg of the fiber is passed through the autpul aperture.
A,,, and collimaled hy the lens. 1,. The aperture A, serves
to control the intensity in the test plane. Notice that the oulput
l~aiir eiiinnntea from nearly f!!Nl fikrs. ench of which acts as
one element of an opIicni
integrator,
resulting
in an
exceedingly high nniformity of the ontpttt beam.
PRINCIPLE OF 7-l lE NEW SlMUl,ATOR
The major objective of our simuiatnr is to produce
closely matched spectrum for solar cell testing that:
I.
2.
3
-.
4.
5.
a
Obviates tk need for spectral mismatch correcGons
Simplifies cell measurements to Ihe extent that only a
few minutes are required to carry out accura1c cell
measurements
Does not require determination of the spectral response
of the cell
Does not require a reference cell which closely matches
the test cell
Makes testing a multijunction
cell AS simple as a
single-junction
ceil
The basic irincipie of the simulator is as follows: the light
kains, ultraviolet
(UV), visible (Via). nnd infrared
WV,
from suirahiy selected source-filter combinations, are focussect
onto the input ends of a trifurcated, randomized optical fiber
cable. The optical fikr combines these beams 10 produce an
output kam of linearly superposed inputs. 771e randomized
nature of tire fiber ensures that the intensity distribution
of
each kam in rhe out.put of rhe fiber is the same. Furthermorr,
since each branch of the fiber cable consists of ahout 2@0tI
individual fibers, the system is cnpable of prcwh1cirrg beam
uniformity
equivalent
to that of a 2000element
optical
integrator.
SYS’IEM
CONFIGURAI-ION
f;igun I is a schematic of our &er optic simulator.
it
con&r
of two light sources - a xenon arc lrmp and a lungslen
filament lamp. The xenon arc lamp is configured to produce
two beams which .serve as UV and Vis sources.
Ibis
arrangement, shown in Figure I, for exfrarting
two beams
from a single souse results in a much more efficient operalion
of the lamp as compared to the conventional
use in solar
simulators.
Such a configurntion
captures nearly all the light
emanating from the xenon arc lnmp. The output kams from
the xenon sotnce are modified by suitable lens systems and
passed through fillers F1 and I$. F, and Fz nre borh low-pass
filters which block the region of sharp emission lines. The
tungsten lamp has no filter.
The output from the filters and rhe direct ourput from the
JR source are passed through individual apertures viz A,. AZ,
and A,.
These apertures are used to control the power
contained in each kam and, thus, allow us to adjust the
spectral distribution in the output kam. Ihe output from each
nperture is f6cus.sed hy individual lenses. I ,,, I,. i3, onto the
correspanding ends of a trifurcated optical fiber cable. The JR
and Vis legs of the fiber cable are made from glass fibers
PRIMARY
CONSlDE~ATfONS
The primary syslem design. following
the selection of
sources, involves selection of filters
nnd the design of the
fiber optic cable that can produce the desired output spcctntm.
One of the criterion for filter selection is to suppress lhe
emission lines from tke xenon arc in tile UV and Visible
beams; this establishes a cutoff around 0.7-0.8 Itrn fm these
The achtal selection of each filter is based on an
filters.
integral system design that takes into account the changes in
through
the spectnun of each kam due to transmission
corresponding opkal elemenls, and clelermines the total output
spectrutn. Such a design can k performed with a computer
program Ihat cakes Ihe spectrum of each source and determines
the rntxiilied
spcclrum
due to the transmittance of fillers.
lenses. and Ihe optical fikr
and arrives at the combined
spectrum at the output.
This progmm can determine the
optimum filler combination
for the hest .fit to the spectnnn
It should be pointed ant that
requirccl at the lest plane.
ahhough use of all-quartz optics can significnntiy
reduce
design effort, such a system would be quite expensive.
Concomitantly,
we have employed quartz optics only in the
jhis feature also ~iiaws a greater control of
UV path.
individual
UV snd Vis b&nis.
We have measured optical
transmittance
of each element and developed a computer
program that selects suitable fihrr combinations to tninimize
Ihe specrrnl mismntch between the outpul spectrum and the
desired solar spcctnun. it should be pointed out that since our
system can independently
control power and the spectral
distribution
in each beam, then
are many
possible
combinations
of filters that cnn produce the same ontpttt
spectrum.
The
feattires:
I.
2.
.1 .
18-2
DESKIN
fiber
cable
is designed
to achieve
the following
High coupling erriciency nt ench input end to ensure a
minimum loss of optical power, Clearly, this is related
to the optical design of the source, Ihe focassing optics,
and the nunlcrical aperhire of the optical fikrs.
in our
current fiber, a coupling efficiency of tW’%-65% can be
achieved.
The loss wiGn the fikr should be minimized to avoid
undesired dissipation.
The effective oufput (optical) diameter of the cable and
the size of each fikr should k compatible with the
reqrtirements
of the si7. and the uniformity
of the
output beam in the test plane.
4.
The distribution of Fibers in the orttpttt cnti of the cable
should be random in order to product: a sparially
uniform superposition of each input hrm.
It is instructive to track the spectral changes that eaclt
knm develops ns tky pmpapnte, nrrd pnxlrtce the cotnlGnctl
output. Figure 2a ltows the spectrni contents of the UV bran,
a( the inpu( and the mtptt
of’ filter F,. 7’he cort~spontlinp
spectra for the Vis beam are shown in Figute 2h. Figute 3
shows the spectra of the individual UV. Vis anti iR hcams in
the lest plane; rhe total output spectrum corresponding In these
individual
ksms
is also sl~ow::.
Figure 4
sllows
R
comparison of die spectrum from the fiber optic simuialor and
the standard Global AMi.S spectrum.
it is clear Ilint a very
close match of AM1.S
simuiator.
can be produced
by the fiber opric
it should be emphasized fhat our present objective is only
to replicate the envelope of the AM I.5 spec~nrm. At a later
stage, we plan to incorpomte a suirahle filter tlrnt can introduce
absorption equivalaiuent
to that of the atmosphere.
ADVANTAGES
OF TJJE FIBER
OT’TJC SIMULA’I’OR
In addition to its excellent speclrai matching capahilily,
this system has other advantages.
I.
The uniformity of the ontput kam in the test plane is
extremciy high. This is because of the fact that ench
fiber acts like an element of an vptical integrntca.
Typicnlly. we ohtain helter than f I % uniformity in the
kam.
2.
The intensity of each kam can be controlled without
changing the uniformity of the mrtpttt beam.
3.
Because of a high coupling efficiency,
the nvernII
optical throughpul of the system is high. This fealute,
along with the design attribute
discussed above,
significantly
relieves the power rrqirirefnrncs
of tfie
input sources.
4.
The system is compact and low-cost.
5
-.
Use of independent
IR beam offers
a unique
advantage of providing
a spectral malching
in an
extended wavelength range as compared 10 that of a
single source.
One of the major advantages of our design is that the
sources are run at a constant output while the input to each
fiber is independently
controlled
to obtain the desired
spectrum.
This feature reduces changes in rhe spectrum of
by maintaining a constant
each light source.
Furthermore,
color temperature
of each source, tfle variations
in the
specrrum are minimized.
Another advantage of our system is
the ease with which the spect~~tm cnn k corrected
to
accommodate changes in the optical elements such as dttc to
aging of the lamps nnd filters.
18-3
,
TEST RESULTS
We have used the oull~ut spettrum
sl:0w:: in Figtin: 1 10
pcrforni I-V measuTcmen1s of single- and mullijunclion
solar
cells mnde from different material systems. These inch& Si,
Althattgh
the
n-Si, GnAs. CulnSe2. GJJe, and GaInP.
spcctnrnt of Figure 4 does not replicate AMI .S withonf a
“wntrr filter”. tk p&llary purpose of these measurements is to
develop a data base 10 assess the sensitivity of diffcfenl cells
to the sprclml contents of the light. This data base will k
nl~plirtl to tlelennine
lhe accuracy of spectral mntching
required 10 mnke solnr cell measrtrements within IQ accuracy.
‘i’ahle I shows cell parameters (isc, VW. and fill fnctor) of
some singlcjunctim
cells measured with the fiber optic
simulator. Far comparisoh, the values of these parameters,
obtained by tire standard NREI, measurements, are also given.
The two-junctinn cell data obtained under fiber optic simuia&tr
and using standard NREi,(SERi)
measurements is given in
Table 2. It is seen thnt the data obtained with the fiber optic
simulator shows somewhat lower values of I, than those
We klieve
the error
mcnsured hy standard test procedures.
is simply hccause of tlie’way we define I -sun power condition
and is not related to the spectrnl content of rhe nimnlator. We
are in the process of developing a better method to measure I sun intensity condition. 1
it is itnportan~ to point out that the actnal solar cell
measurements done with the fiber optic simulator cake less
than a few minutes. Tkre nre no corrections made in the data
S~I~~II in Tables I and 2. Clearly, the cell measurements
made with the fiber optic simrrlator are in good agreement with
those made hy standard rglethods. We believe thal measurment
accuracy will improve further when a “water vapor” filler is
introducetl in the syslen!.
IMPROVEMENTS
We arc in Ilie process of incorparnting
severnl
mcvtificntions
it: the system IO improve the spectral match as
well as simplify the use of the system for day-to-day operation
as a self-contained
piece of test equipment
snitabie for
commercial applicationsi
These modifications
rue expected to
provide (i) better couplitip to the optical fikrs,
(ii) a feeclhack
cnnlrol to maintain a Ipre-set input power ccrndition, (iii)
cnpnhitity
lo make trteasu~emen~s ott individual
cells of a
tnultijunction
solar cell, ,and (iv) replication of absorption due
to atmospheric gases and water vapor.
CONCLUSION
We have developed a fiber-optic-based
simulator that
niixrs light from rltrce; sources to produce an output that
closcJy matches the envelope of an AMl.5
spectrum.
The
initial tesfs on single- ahd multi-junclion
cells sltow a good
agreement between the cell parameters mensared with the fikr
optic
simulator
and hy standard
pfocednres.
Ihe
measurements made will: the fiber optic simulator did not
incorporate nny spectral ,mismatch correclions and did not ttse
II reference
cell. Furlher
thal will allow individual
cliaraclcrizrt1.
in~povenwils
are heine
crlls of a nitillijtinclion
iticwpornted
&vice
lo be
I.
XI’- 10 atld
(bpwation.
2.
Oriel
X-25
Corporalion,
are
man~r~actwed
by
2.50 Imng beach lUvd.,
Spcclrolah
Slratford,
c’l
06497.
The
.1,
‘1’. Glalfclter
4.
M. Krrsuhara. US l’atenr Number 4.641.227, Feb. 1987
.5 .
hf. Rennet er.al, Proc. 21~1 lEEE WSC,
6.
WRCcm
co.
IAd.,
3, Nihonhashi,
Chome, Chnnko, Tokyo 103, Jqwm
7.
Q. Virshnp,
8.
K. Ileidler et.nl. Measurement of Mtdtijunctian
Cells. 10th PVSEC, Lisbon. 1991
Solar
R. I,. Snpori
I I16
are very gralefrrl IO Keith Emery of NREI, for
suggestions
and Ihe measw-ements he has made
on the vnriorts devices we have wed for cornpariscw
studies.
~t~tfmrs
lmny
valuable
Ibcy
uwld
Hat-in
Illlal
like IO thank
for providing
Sarah Kurtt,
the cells
eI.al,
Proc.
Stafford
ml
in this slcttiy.
7%
I~evel~plnrnl
I:und
itsed
ACO2-83Cll lM93.
9.
l’roc. tlst
I 187 (1987)
1438 (1990)
et.&
Fyp/lV
_____-.
Gahsf
---- WG I .4
Si/Sl
------
-- I&b4) .__
A
n
--.--0V
A
MS)
-_
Ieill Paclnr(%)
-.-
_..--n
A
__--_-.
Proc.
---_l.M
03
I.
_-116.I
.__-
19.2
Am+(C,“‘J
-___n
WI9
2.24 --O.RP .-I_
--.79.2 --.79.1
-_2.2s -.-.-_
74.8 ----74.8
.-I-35.9 I:-.-.xi.1 ---0.65 --O.GS ..-?.!I4 3.02 O.Ql
00 I
Gil.2
N1.O
0.25
--- 1.f.w
0.75
--
R7.0
49.G
--- 0.25
I .w
Meromnchi,
IEEE PVSC.1249
( t 990).
‘taldc
PVSC,
Hymn
work was sq~porled by the SERl Director’s
and by the U.S.Deparlment
of Energy under contract No. DE-
Cell
---
19th IEEE
21st IEEE
4-
(1990).
PVSC,
System
Parabolic
Cotifiguralion
rekctors
Jern~ralure-corltrolled
ate
Figure
J.
lamp
A schematic of the Fiber Optic solar simnlator
of the system
showing
o.oLL.-J300
major elements
1
100
500
600
111
700
800
Wavelenglli
----
Figure
2a.
slap
Vis bean1 belore
Figure
Comparison of the spectral conterils
of lhc
UV beam before and after the filter P,
18-5
liller (F2)
900
I
I
1000
1100
(nm)
-
Vii beam after liller (F2)
21~. Comparison of the spectral contents of the
Vis beam before and after filter Fz
300
400
500
GO0
700
800
Waveleriglll
-----
Figure
3.
2001,
300 100
0
Figure
4.
1000
I t 00
(tinI)
.-. -.-. Visible
--In
Total oulput bean
uv
Spectra of individual
rhc lest plane
900
UV, Vis and IR lxn~ns anti the Kohl output beam in
\*
L-.-L~_
I
pII.-.---m-
500
600
700
Waveler,gfli
1300
(twl)
I-900
I-I
1000-
1100
A corrqmison of the spec~nm of the output bean1 from the fiber optic
simulator with AM I .S Global spectnm
18-6
INTERIM
QUALJFICATIQN
PHOTOVOLTAIC
TFSTS
AND
PROCEDURES
THIN-FILM
FLAT-PLATE
FOR
TERRFSTRIAL
MODULFS
R. DcRlasio.
L. Mrig, and D. Waddingtnn
Fnerpy Lahnmtny
National
Rcnewablc
(formerly
the Solar Ettcrgy
Rcscarch
Institute)
Goldctl.
Colorado
ADSTRACT
This papct pmvidcs
recommcndcd
pmccdurcs
and spceificationo
fat qualification
Icsta IO cvahtatc
tcrre&al
thin-film
llaI-pla~e
phoIovoltaie
(PV) nonconcentrating
modukr
&sign&J
for power generation
[ I]. The qualifieaIion
ksls
arc dmigncd
IO evaluate
PV module
dcsigtt pcrfotmance
and susccptihilily
to
known
failure
mechanisms.
Emphasis
is plaecd on testing
and evaluating
module pcrfmmattee
characteristics
and design fcalurcs that will affcel the degradation
of mndulc
performance
and physical
pmpctties
resulting
from solar
cxpcurc.
environmental
wcathcting.
mechanical
loading,
corrosion.
and module
shadowing.
Our primary
intmt
is to provide
the minimum
tests and itt~edotts
rcquircd
to cvaluaIe
PV modules
and IO provide
a common
approach
(e.g.,
bclwccn
pmduccr
and purchasct)
in conducting
qualification
lests.
With the
tcs~s and pmccdutcs
povidcd
hctc. additimrl
IcsIs and pruccdurca,
beyond the
ted scqucnccr
rpceified.
may ha e+tahlishcd
hetwcen
the prodveer
and purchaser
I practices.
that rcprcscnt
cutmnt
pmcutemctt
1. INTRODUCTION
Exirling
thin-film
flat-plate
moduledcrigns
arcundcrdcvclopmcnI.
and new
cnnceptr
continue
to appear.
Consequently.
qualification
Icsts must he flexible
enmtgh IO allow a reasonable
stseswncnt
of new designs. yet cumplete
enough
to identify
weaknesses
that would
lead IO pmbktnr
in Ihc lield.
Ideally,
modules
IhaI cxpcrime
early failures
in field operation
should fail qualificsIimt
tests.
Thcrcfute.
it is imp&anI
IO note thaI dtc tush apecilicd
here wm
devcbpcd
on an “interim”
basis and ate not intended
lo sctvc ac a uniform
btandard fnr the thin-Iilm
flat-plate
module
pholovoltaie
indushy.
Rccaurc
of limited
thin-film
module
ficield operation
expcticnce
and the
evohttionary
naIurc of new thin-film
module malctial
le&nologics
and designs.
these tesIs shmtld nd hc emwkkmd
definitive
or cample1c.
ttur do they pmvide
a basis to prcdicI
30.year
field life.
Our eurrenl
understanding
of faihttc
and
dcgradatim
mcchattirtm
and Ihe relationship
between
aceelcrated
IesIs and fiild
reliability
arc nn~ cnnugh
for us to esGmaIe life expcctaney.
nor ate the cycling
tests given here cunsidcrcd
IO be cquivaknt
to a full 30-year
field cxposurc.
$1owcvcr,
the test and evaluation
pmccdurcr
givctt
hem provide
* tommm
approach
for conducting
qualification
tcsIs. Aeeeptahle
rcsuultr from these teds
should provide renwnahle assurance that Ihc modula
that pass these ksts will
pcrfomt
rcliahly
in the liild
but fur an unspceificd
period of lime. In addition,
chamctctizatinn
tr?;Ia lo dctctminc
mudule
energy rating,
and Ic.sIc to evaluate
the cffceb
on module
performance
due to comhincd
UV/%cnnal
cnvimnmcntr
dcgradatinn, as well as the design of a non-inhusivc
hot-spol
and light-induced
lest method,
arc ertrtcntly
under develupmcnl
and are not included
hcte.
2. RACKGRfXJND
AND OVERVIEW
QUALfFICATION
TRST
2.1
The results of this work and expcticttee
gained wcrc rttililcd
IO fotmulaIe
new tests and modify
earlier tests for inclusion
in this docutncn~.
For example.
one modiftcation
includes
an itttxeasc
in the humidity-frccm
test cycle from IO
lo 20 cycles lo accelerate
Ihc potential
for moisture-ingtcss+tduccd
cffcels on
pcrfotmance
and safeIy.
Also inch&d
are new t&s for wct itt.eulaIion
nrktattec and rurfaee-cut
wrccpIibiliIy,
which
further rcprcscnt
an ineteascd
focus
on identifying
watn-ingress-induecd
faihtrcs
and elechical-shock
hamrd.
2.2
Overview
of Qualificalimt
Tes(
The overall
qualification
IcsI is dcsigncd
to minim&
the time to conduct
~hc tqrtircd
Icslr snd fhc numhcr
of test module
specimens.
A minimum
of six
test modules is rcquircd
to compkIe
all the spccificd
tcsla. The qualificatim-test
acqucncc
is illurhatcd
in Figure
1. and lhc required
leaf and inspccIintt
pmccdnns
arc provided
in Se&on
4. Module
qualification
ted mqui-tr.
which provide
detailed
guidance
and minimum
rcquircmcnts
for conducting
the
ovaall
qualification
test. arc given
in Scetintt
3.
Test and evaluation
requirements
pmvidcd
in SceIion 3 in&de
(I) initial
~cstr and inspcctiotts.
(2) scqucnce
“A” and “R” tcs~s and inspections,
(3) final Ierts and inspetions,
and (4) evaluation
of qualification
test results.
Initial test+ and inspections
(SccIion
3.2) ectahlish
selection
and accepIancc
of as-tee&cd
module
as I& module
spccimcns.
They aku establish
hasclinc
meacunment data lor
visual-inspeedon
information
and elcehieal-pcrformanee
comparison
and determination
of the effcectr on module physical
characteristics
nnd eketrical
pcrfotmanee
following
exposure
and endurance
tests.
Squcncc
“A’
&IS
and impcctinnr
(Section
3.4) imludc
a vaticly
of
envinmmental
and s~rcss tests designed
to induce
and identify
failure
mechanisms
that may lead IO module
failon,
degraded
ekchical
pctfotmancc.
physical
defotmitics.
a sheck hazard.
Sequenec “R” I&s (SC&m
35) it&t&
II hypaasdicde
dtctmal
IC~I Id mess
the ~ttcrmal design adequacy
of hyparc
dicdcs.
if used in Ihc module
conshuclion.
and a hot-spot
cndumttce
test to
detetmim
Ihe long-term
effcets of module
hut-spoI
heating
as+ociaIed
with a
condition
such as shadowing.
The littal tests and inrpcetions
(Section
3.6)
esIablish
a final database for cumparison
with inilial
telpl and inspcelion
results.
The criteria
for passing
the qualification
t&s
arc provided
in Sation
3.7
(Evaluation of Qualification
Test Results).
3.
MODULE
TEST AND EVALUATION
RRQUlREMRNTS
This scetinn .specifics Ihc minimum
tcstr and inspections
to he pcrfmnled
and the required
ted scqucnec
to ha followed
IO evaluate
photovoltaic
Lin-film
flat-plate
moduk-?.
The rquired
test scqttettec is ilhtsbalcd
in Ihc flow diagram
pmvidcd
in Pigurc
I along with designated
in+peIion
and Iesl identification
numhrts
corrc+ing
to ingrctiona
and 1-t pmccdurcs
rpceilicd
in Scctbn
4.
OF
Racknmmd
3.1
In dcvclnping
this
qualifiiacim-Ieti
doeumcnI,
-al
sources
of
infmmation
wcrc rcfcrenecd
and utilized
in designing
and formulating
the
majority
of the tests given.
kimaty
sources
included
the Jet Prnpulxintt
Lahomtnty
(JFL)
Rloek V rpceif~ation
fat flabplate
modules
121 and the
UndctwriIcn
Lahnratnties
standard
fur flat-plate
modules
1.31. Since
the
puhlicalinn
of thcsc documents.
phntovoltaic
thin-lilm
matcriab
technology
and
module
dcve@tncnt have advanced rapidly.
Major effottr
hy the photovoltaic
industry.
JPL. and SERI in module
rcliahility
research
have pruduccd
a better
undcrztanding
of known
and potential
faikrc
mechanisms
awxistcd with thinlilm moduks.
cspceially
with regard to the effects of moisture
ingress.
Test
Specimens
A minimum
of six module
lesl spccimcns
of ,cach module type rcprescntativc of the modules
to hc deployed
in the field is required
IO conduct
all Utc ledz
spccificd
in this docummt.
One test module trquind
for the hot-sp,at cmlttrattcc
ted shall be specially
fahricatcd
with Icads brought
out for aeecssing individual
cells IIL rpecifkd
in rcfcrcnce
141. The basclinc
power outpul
determined in
accordance
with Scetion
3.2.2 Tot each of the six modulw
shall hc within
fl@%
of Ihe average power output of tnttduln
pmposcd
or planttcd
for field deplnyment.
Each module
test vimctt
shall have a serial numhct
m label ur some
other means of identification
capahk
of surviving
both test scqttcttecs.
19-1
Selection
and acccptancc
of as-received
modules
as mndule test specimens
shall k hased on passing the initinl inspectinns
and teats sftecilial
in Section
3.2
and in the sequence shnwn in fignre
I and mecting
tk power output uniformity
requirement
specified
akve.
Any mndule
design or fnhicatinn
changes (e.g.,
mnkrials.
manufacturing
process.
or assemhfy)
of a module
type previcntsly
&ted
and cvrhtatcrl
may require reevaluation
and testing.
Such dctcrtnination
may k made between
Ihc mnduk
producer
and purchaser
of the modules.
3.2
Initial
Tests
and Inspections
ml
Initial tcstc and inspections
II illustrated
in Agure
I.
shall
k
performed
in the nrder
described
hrlow
3.2.1 Baseline
Visual ftusnection.
Each module shall k visually
inspccfed
in accordance
with the speeiEcatinns
given in Sectinn 4.1.1 IO obtain a baseline
determination
nf tk presence m absence of defects in tk module
for purposes
nf detecting
any changes
aitet
each required
3.22
Baseline
Electrical-Perfurms
tneacured
in accordance
with the speciftcrlionc
a has&e
ckctrical-output
power that will
detetmittatinn
of the effects nf qualification
test.
Teat.
Each module
shall
k
given in Section 4.2 In atahlish
serve as the comparisnn
value for
testing on electrical
perfnrmsttrc.
3.2.3
Grouttd-Continuity
TCSI.
Each module
having
conductive
cttrfac~
(i.e., frame. structural
members.
or edge
tested in accordance
with the specitications
given in Section
continuity
exists between
all such surfaces
and the module
Modules
that use direct attachment
to an array structure
to
shall k tcstcd after asremhly.
3.2.4
Electrical-Isolation
Test (Drv
Ni-put).
suhjectrd
to a direct
current
(DC)
hi-pot
test
speciftcatiuns
given in Section 4.4 IO assure electrical
from cxtrrtwlly
enftosed conductive
parts.
3.tS
inspected
3.26
accordance
measured
(mcacvrcd)
3.3
Intern&iaIc
in accordance
Visual
Impaction.
with the sfwcitications
Ekctrical-Ferforman
Test.
with the snecificatinns
niven
for each mndute
tested shall
baseline
power as determined
Completintt
..
of fnitial
Tests
enpnsed
external
&surer)
shall k
4.3 la verify that
grounding
point.
ohfain grnundittg
Each module
shall
k
in annrdartcc
with
the
ixolalion
nf the FV circuit
Each module
shall k
given in Section 4.1.2.
Sequence
“A”
Tests
3.4.3
ffumiditv-Reezc
Cycle Test.
The twn mndulcr
allocated
for the
humidity.freere
cycle test, following
the Stlcycle
thermal test. shall k tested in
accordance
with the specilkatitms
given in Section 4.7. At tk completinn
nf
this test. visual inspections
and elrcbicntperfrummncc
tests shnll k conducted
on kth modules.
After these tests are cnmpktcd
tuccessfully.
tk Iwo mndules
shall k used in cnttducting
the mechanical-lnading
test (Sectinn
34.4).
3.4.4
Mechanical-Loading
Test.
Tk
Iwo mndutes
allocated
frtr tk
mechnnicnl
loadina
test shall he tested mcordina
to the snecif~ations
eivm in
Sectinn 4.11. At IL cnmpktion
of this lesl. a &al
irt&tintt
and el&icalperfurtnnnce
test shall k cnndttcted
en kth.
Wkn
successfullycnmpIeted.
tk
two mndules
shall k used in conducting
tk hail-impact
test (Section
3.4.5).
3.4.5 Bail-Impact
Test. The two modules allocated
fnr the hail-impact
test
shall k tested in accordance
with tk .meeiBcatiotta
aivm in Seetion 4.9. At the
completion
of thB test. visual inspections
and elech%al-ycrfnrmance
tests shall
k canducted
an kth
mnduks.
Next.
tk
two mnduks
shall k used in
conducting
the surface-cut
rusceptihility
test (Sectinn
34.6).
if required
(we
Sectinn 4.10 fnr applicability
nf test).
3.4.6
Surface-Cut
Susceptibility
Test.
lk
four modules
(Iwo modules
fnllowing
the 2Of-cycle
thermal
test and two mnduln
following
the hail-irnfract
test) allocated
for he surface-cut
susc+ihi)ity
test shall k tested, if rquired.
in accordance
with the speciBcatinns
given in Section 4.10.
Wkn
tk test is
succesdut)y
completed
or it is determined
tok
not spFlicah)e.
the four modUks
shall he used in conducting
the wet inntlatinr-resistance
test (Section
3.4.7).
3.4.7 Wet Insulation-Resistance
TCSL The four mnduln
allocated
for the
wet insulatinn~rerictanee
test shall k tested accordina
to tk sneciticatinttc
aivm
in Sectinn 4.5. When Ihe lest is successfuRy
completed.
the four modules shall
k used in conducting
the final tests and inrpectinns
specified
in Seelion
3.6.1
fnr Sequmce
“A.”
.
visually
Each module
shall k measured
in
in Section 4.2. The maximum
nnwer
not k less than 96% of the ‘initial
in Scetion 3.2.2
and Inspectiotts
Each of the six mndule test sftecimens
must meet therequirements
specikd
in Sectinns
3. l and 3.2 kfnre
testing can pmceed.
When dtesc requirements
are
met. one of the tnedufes shall k stnred in tk dark at mom temprature
In setve
as a entttrnl
module
for future refermee
and until all planned
tests have ken
cnmpleted.
Tlte remaining
fwc m&tks
shall k allncated
for Sequence “A” and
Sequence
“B” tests~and
inrpeelinns
*C follnws:
four mnduln
fee conducting
Sequence “A” tests and inspctiuns.
as speciliedin
Section 3.4. and one module
for cnnducting
Sequcwze
“B” tests. as specikd
in Section. 3.5.
3.4
enpnsure.
At the completion
of an additinnal
150 cycles. these two mndules
shall undergo
visual
inspection
and an dectrical-perfnrmrncc
ted.
When
successfully
cnmpleted,
these two modules
shall k used in conductinp
the
surfacecut
rusccptihility
test (Section
3.46).
and Inspections
llte fnur mnduks
selected for Sqnence
“A” tests shall k subjected
to the
fnllnwing
tests. which
are deserihed
in &tail
in Section 4. and shall
k
cnnducted
in the nrrkr
indicated
klnw
and illustrated
in Figure
1. Module
intermediate
visual im~ctionr.
in aeeordance
with Section 3.25, atut eketricalperfnrmattce
tests, in accordance
with Section
3.26. shall he cnnducted
after
each rFplicahle
test as indicated
in Figure I.
3.5
Scquet~~~
Tk
following
conducted
“B”
.
Tests
module
selected
fnr Sequence
“B” tests shall k subjected
to tk
tests which
are descrikd
in detail in Seetinn 4. The tests shalt k
in the nrder indicated
ktow
and illustrated
in Rgun
I.
3.5.1
Bypass-Dioda
Tktmal
Teat. Tk tttoduk
akcated
fur tk bypassdiode lhermal
test shall k tested in acemdance
with tk speeifieatitms
given in
Section 4.f2.
When the test is mceessfuBy
fapkted
or wkn
it is determined
to k not applicahfe.
the module
shall k used in cnnducting
tk
hut-spnt
mdurmcc
test (Section
3.5.2).
35.2
Hot-St&
Endmmce
Teat.
The module
allucated
fur tk hot-spot
en&trance
test shall k tested aceutdlttg
to tk specificatinnr
givm
in Section
4.13. After tk test is successfully
completed.
tk module shall k used to eonduet the final ksts and insfectiorts
SpeciEed in Se&m
3.62 fnr Sequence
“A.”
3.6
Fmaf
Tests
and fttsoectiutta
At the completion
of all the Sequence
“A” and Squence
‘R’ tests and
inspectinns.
the following
fmal tests and impcctionr
shall k cnttducted
in tk
n&r
indicated
klow.
3.4.1
Wet fttsulation-Resistance
Test.
Seqttence
“A” modules
shall k
tested to evaluate
each mndttk’r
electrical-insulation
system hy measuring
the
electrical
resistance
ktween
the module
circuitry
and the frame. or mnunting
surface in accordance
with Seetion 4.5.
3.6.1 Semtence “A‘.
Following
tk wet insulation-resistance
test. tk fnur
mndulea
allocated
for tk finnl tests and inspeetiotts
shall k suhjated
to tk
electrical
isolatinn
test (wet hi-pot)
in aaotdattce
with tk qrecifiiationr
givm
in Section 4.11.
followed
try a gmtmdeorttinuity
test in aecnrdanee
with
Section 3.2.3 and a final visual
inspeetim
and electrical
perfnrmattee
test in
accordance
with Seaionr
4.1.3 and 3.2.6.
3.4.2
Thermal-Cycle
Test.
The four modules
selected for Sequence
“A”
shall k suhjcct~
to the thermal-cycle
test in acenrdance
with tk spe%catiortr
given in Section 4.6. At the end of 50 thermal cycles, tk test shall be stnFped.
and visusl insFectintts
and elechieal-perfmmance
tests shall k prfortned
m al\
four modules.
After these procedures
are completed
successfully.
two modules
shall k used for cnnducting
the humidity-freeze
cycle lest (Section
3.4.3). and
the other twn mntfttka
shall underpo
an additional
IS0 thermal
cycles
of
3.6.2 Sequence
“B”.
The module akcated
for final tests and inspectionc.
following
the hut-spnt
endurance
test. shall k suhjeeted to an electrical-iu-tlatio
test (dry hi-pot)
in accnrdance
with the specifications
givm
in Satinn
4.4.
followed
hy II final
visual
inspection
anf electrical-perfnrmame
test in
accntdance
with Sections
4.1.3 and 3.2.6.
19-2
3.7
Evrluntion
of Qualification-Test
Results
The crikrir
for parring
the quurlif~ntim
tests are thrill each modale,
as
alioertd.
mu4
pnzr
ail of the folkwing:
Initial
Icsts and inqwctionr
Ckctims
3.1, 3.2, nnd 3.3).
sqacnce
“A”
and scqucnce
“it’
tests and
inspections
(Sections
3.4 and 3.2). and final ksts and inspections
(Satinn
3.6).
I pupwe
for the test nr inspeelinn.
detailed
steps
cnnducting
the 1-f or inspcctirm.
and criteria
for pnrsing
when critctia
me not +fkd
in Section 3.
4.1
Visual-lnstxction
The putpose
4.
MOIWLP,
TRSI’
AND
1N.SPR~ION
intamedink,
determine
beginning
PROCEDURES
The f&wing
mnduk
test and inspection
pracdwer
provide
1he detailed
stw
and spccifknlinm
reqnircd
to conduct
and mat
the overnli
q~~nlificntim
test reqniremenIr
given in Section
3. Each of the following
pnxedums
provides
4.1.1
viwrlly
1. QuallfkMkn
Test
fr
Roadun
of this procedtm
is to provide
gidcli~
in obtaining
baseline.
nnd finill viwal-inspCClinn
infannatim
mpircd
to idenlify
nnd
any physical
changes
of defects
in module
cnntmctim
at the
and completion
of each required
test. n spxified
in Section 3.
RIscline
inspected
c
Ftpurc
and specifications
each test 0T imp&m
Sequence
Visual
bp&kn.
Modules
shall
for good
workmnnrhii,
shipping
be photognphed
and
damage,
mechanicnl
mounting
dimcncinnc.
nnd other
modok
producer
and the purchaw
incpcctim
criteria
of the moduk..
Any ohsewed
defects or rdmonnalities,
or inclusions.
scratches.
or color that might
or rcliahility
shall
show the location
k d~umcntcd
of the defects.
with
estahlishcd
such ax minor
adversely
affect
spproptialc
ktwccn
the
flaws, dclaminations
module performance
sketches
or photographs
to
Information
obtained
above shall be well documcntcd
and formulated
for
use in cnmpnring
hrselim
obscrvntionr
with the results of all intermcdiste
and
Anal impcctimx.
and for use in ectnhlixhing
a sound basis in dcIcrmining
lk
effects on module
physical
charsctcrixticr
following
each tact. Serious damage
or drfcctr
arc a ha-ix for s mcdulc
to k rcjcctcd
without
further testing.
4.1.2
lntcrmcdiate
Visual
lnspcctions.
After
each required
ted. the
modules shall k visually
inspected.
okctvatians
regarding
damage or dcgradatinn observed during inspcctiom
shall k documcntcd
for comparison
with bascline inspection
results.
The inspccdcm
focus shdd
be n fmms of damage or
physical
apdegradation
expected
after each test. Any change in tk module’s
pcsrance chnll k documcn1cd
in the same detail required
for basclinc in.spccIion.
4.1.3
Final Visual kvpcctions.
Modules
shall k thoroughly
inspected
and
photographed
after testing.
Any damage.
degradation,
or almormalitics
that
might affect long-term
reliability
shall k documented.
Gbscwationx
regarding
damage
or degradation
observed
during hasclinc
and inkrmcdiatc
incpcctions
shall k included
with the dncumcntafion
for the tinal incpcction.
llccauw
of the wide variety
of possible
module
designs, detailed
visual
in~pcctirm
accrptance
criteria
arc not provided
in this document.
Detailed
incpcction
acceptance
criteria
shall k developed
be1wcc.n the module
menufacturcr
and the module purchaser.
Application
of the agreed-upon
criteria
will
rely heavily
on the expertise
and engineering
judgment
of the inspcctora
and on
their evnluatin
and comparison
of inIcrmcdiale
and Iinal visual-inxpcction
rcnrlts
with kscline
inspection
results.
4.2
Ekctrical-Performance
Test
The purpose of this test is IO characterin
the electrical
performance
of lest
moduk~
and to detmminc
each module’s
peak oulpu1 power.
The test is the
mcarurcmcnt
of data IO plot an I-V curve from short-circuit
current
to opcncircuit
voltage
with a minimum
of 317 data p&us.
All dcctrlcal-performance
mearurcmcnta
shall k comiuc~ed
in sccordancc
with the following
standard lest
and rcpcrling
conditiom:
AM I .5 globs1 solar rcfcrcnce
spectrum
PC provided
in ASTM
ES92 lSj. hurl illumination
lntcnsity
of l000 Whn’(lO0
mW/cm’)
and
module
tcmpcra1urc
of 23°C.
using
rccommcndcd
ASTM
mcawwrrncitt
proccdrwca
as nutlincd
in ASTM
El036
161.
4.3
Ground-Conlinuitv
The module shall k observed during the test and there shall k rm sign- of
arcing or flash over. DC-leakage
current shall k mnnitcrcd
during tk lest and
shall not exceed SO pA. (Note:
Raced on texts conducted
on thin-fdm
modules.
tk lenkagc current
for this test should k much lower than the 50 PA crikrintx
experience
has shown thal tk requisite
wet innrlatinn-resistance
and wet hi-pot
kxts arc more scvcm than the dry hi-pot
test.)
AC-leakage
current
resulting
from power supply ripple and DC current
resulting
from capacitive
chsrging
shall not k camidcrcd.
DC-leakage
cutrcnt during lk test shall k recorded for
each polarity.
4.5
circuitry.
where it may cause
safely haxard to pcrxormcl.
Electrical-Isolation
The
purpoxc
Test (Dw
of this
Hi-pot)
text is to ensure
elcctrlcal
isolation
of tk
PV circuit
from any extcmally
exposed
conductive
parts
Each lest module
shall k
suhjcctcd
to both positive
and negative
polar@
DC hi-pot testing conducted
at
room tcmpcrsturc
with Ik output terminations
&ort-circuited.
Test leads from
a suitable
DC-voltage
power supply shall k connected
to the shorted terminals
and the module grounding
point.
in the case of modules
not provided
with an
equipment
grotmding
stud, a conductive
foil ln contact with the mtirc insulated
Voltage
shall k
surface of the module shall k used as the rltcrnate
test point.
applied
at a rate not to exceed 500 V/s up to tk test voltage of two times the
system voltage
plus ItIM) V DC and held rl this voltage
for I minute.
(%Ic:
maximum
system voltage
anticipated
for system applications
for tk modules
king
tested should k provided
hy the module produqr
or USn.)
Tal
corrosion,
ground
faults,
or pass
an ekctrical
.
7bc module
is tested by immersing
esch edge in turn in a watcrAveIting
The rcsistivity
of the sololion
shall
k
agent
(surfactant)
solution*.
35Mt ohmcm
or less and the sorface ten&m
of tbc solution
due to the wetting
agent shall k 30 dynes/cm
or less. The tcmpcraturc
of Ik module
and tk
solution
shall k 22 f 3°C.
The
inxulation
rcsistsnce
is meawed
ktwcn
the shorted
output
tenninationc
and the solution
by applying
a voltage
of 500 V DC (such ax a
Mepgcr)
in each polsrity.
The test prowdwc.
is as follows:
Apply the test voltage and record the wet insulation
rcsistarwe
aher 2 min
for each edge of rk mcdule.
bnmcrsc
lk side with tk lowest wet inwlatinn
resistance.
apply the lest voltage,
and thoroughly
wet all exposed surfaces of Ik
module for spproximately
10 s. psrticularly
the elcchicsl-termination
mea. using
a handkld
spray containing
Ibe same solution.
Record tk insulation
ndstsncc
reading afkr 2 min.
(Note: All wiring
connections
should k rcpmscntative
of
ck raommcndcd
field wiring
installation:
mwrc
that leakage cuncntv
arc not
originating
from
any
instrumentation
wiring
attached
to the module.
Terminstiom
and terminal
boxes shall be maintained
at least 127 mm (0.5 in.)
above the solution
level but shall be thoroughly
wetted with Ik spray.) All
resistance
mcasurcmcnts
shall exceed IO0 megohms
as measured with a suitable
high-impcdsnce
ohmmeter
(McgohmmcIcr
or Meggcr).
4.6
Thermal-Cvcle
Test
The pqmsc
of the thermal
cycle test is to dctcnninc
whcthcr
test modules
have adqunte
resistance
to failure resulting
horn diffcrcntisl
thermal expertvim
of component
parts mrd bonding
materials.
the
4.4
Imulation-Rcvistance
The purpo.sc nf this &I is to cvaluatc
tk module’s
&ctrkal
insulation
system rmdcr wet operating
conditions
and to verify
that moburc
cawed
by
rain, fog. dew, or melted sno’u will not enter tk active portions
of the module
Tcsl
Each module having cxpoxcd cxkrnal
conductive
surfacc.o (i.e., frame, btrttcturd
mcmkm
or edge cnclnwrcs)
shall k tested to verify
that ckctrical
continuity
exists between
all such surfaces and the me&k
frame or grnunding
point.
lvlodulex
that use direct
attachment
to the array stmcturc
lo obtain
grounding
shall k tested after assembly
to a suitable
simulated
panel and army
moonting
hwnc.
A current
of twice tk
sbort&cuit
current
rating
of the
module
shall k applied
hctwccn
tk
grounding
pnint
and each accessible
cnnductivc
part unda test. The resistance
shall k computed
by measuring
the
voltage drop klwccn
the grounding
point and a point on the cmulucIive
surface
within
12.7 mm (0.5 in.) of the point of current
injection.
The resistance
shall
k 0.1 ohm or lcsv.
Wet
Test modula
shall k subjected
to tha thermal
cycles in rcordance
with
prolllc
illustrated
ln Plgum
2 showing
the module
tcmprxaturc
varying
ktwccn
40°C
and 90°C.
lltc tcmpcrstw
shall vary approximately
linearly
with time at a rate not exceeding
lZOYZ/b
and whh a pcdod not greater thsn
6 h&le.
The modules
shall k instmmcntcd
and monitored
thrcughoot
the test
lo detect *try open circuits
or ground faults during the lest. Tha modules
shall
not have exhibltcd
cithcr of these conditions
during tbs text. Prcoau6ons
should
k taken to avoid condcnration
on the apccimat
during the test. (l%le:
An air
vent throogh
s contsincr
of silica gel will aid ln reducing
moixmrc
concentration
in the cnvirmuncntal
test chamber.)
At the end of .SO cyclec. the test shall k rtoppcd and visual htspcclion
and
electrical-perfotmance
tests shall k performed
on all four modules in accordance
with Sections
3.2.5 end 3.26. rcspcctively.
If all modules
pass. two moduks
shall k transferred
to the humidity-fen
cycle test as shown
ln the lest
squcnce
(Pigum
I). and tk remaining
two modules shall undergo an sddilirmal
I SO 1krmnl
cycles of expowrc.
*Surfactantr
such as Triton
X-RIO (Robm and Ham Company)
or its quivrlcnt
A 0.1% solution
of Trilon
X-1013 in trp waler is
may k used for this &I.
currently
king
used. (Tests conducted
at IPL have indicated
that a surface tmsion of 30 dynes/cm
provides
adquatc
wetting
condi1ions.)
Tbc mcdule should
k thoroughly
rimed after the test.
19-4
At least 10 of tk test moduler’
most scmitive
points shall k sclccted for
impact.
The candidate
points
selected should
include
(when
spplicahle)
the
following:
Center p&Ix
of cells: comers and edges of Ihe module:
pointfs)
dircclly
near bypnqs diode(s).
if installed;
and soldcrcd
or bonded
metsllic
intcrconncc@
ktwccn
cells. if present.
Errnr~ of up IO 13 mm (0-S in.) in tk locstion
of a hit are accep(nhle.
Either pneumatic
CR spring~acIuatcd
guns for projecting
the ice halls againd
the
modules are acceptable.
Ice-hall
velocity
at impact mud k conbolted
to within
f ~5% of the spccificd
velmify.
and the rcgiwd
haiknne
deviation
in dinmelcr
shall k less than f 3 mm (0.125 in.).
The ice halls shall k cooled to -IO
f 2°C as measured
in the compartment
where they are stora-l.
The module rhall
k supported
in a manner representative
of that used for actual installation
of the
module
in tk
any.
Note Ihat ice kllx
arc tk
only acccptahle
hailntonc
simularion.
Steel balls.
for example.
shall
not k
uxcd.
Resistance
of
Photovoltaic
Solar
Panels
to Simulated
Hail,
JPL S 101-62
18). descriks
techniqun
and equipment
suitable
for pcrlormancc
of this lest.
Figure
2. Mndulc
Tcmperafurc
Vsrlatlnn
During
the Thermal
Cycle Tesl
(A shorter cycle time is acccpIxble
if lM”C/h
maximum
time ix not exceedcd.
The Chamkr
may k opcmd at 2%cycle
intervals
for visual inspection.)
4.7
Humi&-Frecm
Cycle
4.10
modules
shall
k
SuhjccIed
IO tk
humidity-lrccr..
cycle
Nonplasr
ex,crinr
surfaces of the module
shall k suhjccted
lo the cut tcc1
in paragraphs
23.1 through
23.3 of UL Standard
1703 131. Drawing
of the tool illustrated
on page 21 of tk UL Stnndrrd 1703 shall k as dc.scrikd
in paragraph 23.2 except that three passes shall k made across each surface and
nt leaa two of these paws
shall k owr the tops of grid lines, cell edges. cell
interconncctr.
bus krs,
and similar
nonhomogcncour
areas.
IesI in-actor-
saturate and condense at lower temperatures.
The modules
shall k inrtrumcnted
and monitored
throughout
the Tess 10 detect any open eircuitx
or ground
faulta
during tk Iec(. The modules shall not have exhibkd
e&r
of thexe conditions
during
Ihe test. Twenty
cycles shall k performed.
A risk of electric
shock exists if the hlade of tk tool comactx
a pati
involving
a risk of electric
shock or if such a pars is rendered
accexcihic
(hancitory
or pcrmsnen~)
ax a result of the placement
of Ik blade on or Ihe
drawing
of the blade across tk surface.
4.11
Electrical-Isolation
The
3. hlndale
Text
Tcmperrture
Vnrlslion
During
pupose
Test (Wet
of this
Mechanical-Loading
the Ilomldlty-Freeze
4.12 Bwssn-Diode
The modules
shall
de~st any open circuits
exhibited
4.9 Hail-lmpacl
Tlw
loading
k insIrumenlcd
or ground faults
of these conditions
and monitored
during
the test.
during
isolation
of
tk PV circuit
k
observed
during
the text, and then shall k no signs
The leakage cuncnl
shall k less than .SO pA.
Ik
to tk
test module
subjected to normal
traveling
at a terminal
Test
The purpose
of this test is to awes
1he thermal
design adequacy
snd
relstive
long-term
reliability
of byFarr diodes used to limit
Ihe detrimental
effects of module
hot-spot
suaeplihilily.
Thix test is required
only if bypass
diode(r)
sre part of the module conxtroction.
of (1) establishing
(he diode’s characteristic
curve (voltngc
drop vemus junction
temperature);
(2) meamring
the diode junction
voltage
drop under simulntcd
125 mW/cm’,
40°C ambient
condition%
and (3) converting
the voltage drop to
the junction
tcmpcrahne
wing
the established characteristic
curve.
The test
procedure
consiix~s of the following
three xIep in the order given:
lent.
(if any)
k
Thermal
The procedure
iq based on mcasurlng
the diode’s junction
tsmpcranwe.
in
situ, in its undisturbed proposed field mounting
configuration,
using Ihe diode’s
forwant-voltage--v~s-te~aturr
as the measuring parameter.
It conricts
Throughout
the lest lo
The modules
shall not
Test
Purpose of this lest it to asxess damage
hail impact.
l’bc modules shall
with 254.mm
(l-in.)
diamctn
ice balls
m/s (52 mph).
a simslatcd
of 23.2
either
elcchical
Test
The purpose of this IC~I ix to cnwre
that the test modules
can withstand
a
wind gurt losding
of 30 lh/ft’.
Modules
shall k subjected
to a dynamic
wind
loading
tcs( acing IO,flfXJ cycles of front and hack surface loading
with a load
of 30 Ih/ft’ according
to UL Standard
1703, Section 3g 131 and Cyclic PressureLoad Developmcnlal
Testing for Solar Panels. JPL 5101-19 (7). The cycle rate
shall not exceed 20 cycleslmin.
have
Hi-Do@
text is to ensure
horn my externally
exposed conductive
parts when the module is wet. The test
shall k performed
hy dipping
each edge of tk test module inlo a welting
agent
and wster solution
in a trough.
The solution
used for thh test shall k tk same
solution
used in the wet insulatim-resistance
test. The connation
or junction
box, if any, zhali not k submerged.
The front and back of the module shall k
sprayed with the wetting
solution
(Note:
The module shall k Ihormrghly
wcI).
A vollage
equal to twice Ihe system voltage plus lo0 V DC shall k applied
II a rate not to exceed SO0 V/s up to the tcsI voltage
and then kld
at the
required
text voltage
for 1 minute
in each polarity
ktwan
tk shorIed leads ol
the module
and the metal frame of the moduk.
In the case of a plastic frame
or no frame. the applied voltnge
shsll k between
the shorted lerminalx
and an
aluminum
foil wrapped around the submcrgcd
edge of the test mndulep
a metal
plate submerged
in the liquid.
Each module
edge shall k submerged in the
liquid in mm snd the lent repeated.
The module
shall
of arcing OT flssh-over.
4.8
Test
descrikd
dance with tk profile il!ustraIed
in Figure
3 showing
that tk humidity is
controlled to RS f 2.5% relative when the temperature
is 85°C and allowed
to
FIgwe
Cydc
SuxceptibiliIy
The purpose of thix @I is 10 ensure that the test modules
shall k capable
of withntxnding
the applitatinn
of a sharp objet:
drawn
acron
its nonglnrr
sorhccx
(front and hack) without
creating
a risk of electrical
shock.
Tcsl
The purpoxc
of rk humidily-frcen
cycle test is IO determine
whctha
lest
modules have ndqaate
rcsis~ance IO the dcnimental
effcc~s of humidity.
condcnsalion, and freezing
and the rcwltant
humidity-induced
expansion
of materials.
Tea
Surface-Cut
after
impact
velocity
I.
19-5
Determine
acteristic
the diode
cutve for
rotward-voltage
drop versus junction-tnnperature
charconss~an~ meaxuremcnt
current
by placing
Ihe
a chosen
mnduk/diodc
acccmhly
in an oven.
Afln
achieving
temperslure
stability
of
the diode juncIion.
mcawrc
the diode forward-voltsge
drop
st three oven
tcmyrahrm
(approximately
47.70. lrnd IIXPC). The meaauremcnt current
I, (typically
Sll to 100 mA). rhould
k selected to provide
a good linear
rrsponce
with tcmpcrature
(i.e.. not to heal Ik junction
significantly)
and k
aI lesct Iwo orders of magnitude
ahove module
kaksgC
current
(with
PV
ccllc in darknew).
Once sckctcd.
I* mutt k kept identical
(f I’%) for all
junction-vollage
mranrrcmmtr.
Thiq is most easily implenrcnlcd
wing
an
acmrsIc
con*tanI-currcnt
power cupply.
Accurately
measure
the tcmppraIure
of tk module/diode
nsccmhly
(at lk
actual time tk vollagc
mmsurcmmt
is made) using a thermocouple
actached
in
the
proximity
nf the
diode.
However,
do not
distnrh
tk
Ihennnl/mcchanicsl
properties
of the diode or its heat-transfer
paths.
Plot the measured
IcmpcraIure
of Ik module/dink
diode jnncGon
vnl~rgc:
a lincsr response should
thcrmncouple
he achieved
against
The lest mnhrle
sf nllocnted
in Section
35.2 chsll undnpo
tk hot tp1
endurance
test m dctcrikd
in rcfcrcncc
141. This approach
is “intm&e”
in
nawre
and thus rquirer
Ik L&ication
of s xpccirlly
built Ied module
wui!h
leads brought
ouI for scccwing
individual
celk.
The moduie
completion
ol
of dcgradatinn
cell cracking,
idcntificd.
If
xtring may hc
A nnninImsivc
hoI-cpoI endurance
Icd it mrrcntly
under dcvelopmmI
hy the
International
Elecchotechnical
Commission
(lEC), PV Energy Systems Technical
Committee
TC-82.
When adnptcd and published
hy tk IEC. this approach
will
k evaluated
and concidcred
for inclusinn
in the rcvkcd
vcrxion of this document
as an alIcmaIe
IO tk inImrive
sppmach
descrikd
in reference
(4).
the
ACKNOWLElXMENT
2. Dckrmine
the diode junction
voltage drop hy covering
tk module IO prcveuI
ilhrmination
of Ihe solar
cells.
m otherwise
Prevent
pholovoltaically
gcrmrrtcd
voltagw
and currents
from king
impressed
on tk bypass diode.
Using an infrmcd
(IR) rndiant heater, heat the normally
unlit
aurracc of the
morhde or dindc mnunt to 3S”C shove rnnm IempcraIurc;
Ihis simulntes
the
typical
Iempcrnmre
rice shove ambient
atsociatcd
with an inadiame
level of
100 mW/cm’.
Alter the module and diode reach thermal equilibrium.
apply
a Iec( currcm to the dinde equal to 1.2s rimes the shortcircuit
currcn~ of the
module
nt IO0 mW/cm’
it-radiance.
Maintain
lhc IR henting &wing the current application.
The authors
wish to express appreciation
for the manuxcriPI
review
and
helphd
supgcdionx
provided
by tk
pholovoltaic
community
at large, with
special
thanks
IO William
Bottenkrg
(ARC0
Solar
Im.),
Joxqh
Burdick
(ECD/Sovonicr),
Rokrt
D’Aicllo
(Solarex
Cotp.). Madeleine
lohnxm
(Chranar
Corp.).
Ronald Rw
(Jet Propulsion
Laboratory).
Michael
SIcm (7JIiliIy Power
C,roup).
Walt
StolIe (Bechtcl
National
Inc.). and Micharl
Thomas
(Sandia
National
Labnratnies)
for as+Iing
in the development
and detailed reviews
of
initinl
draft manuxcriptr.
This work wax pcrformcd
under Contract
No. DEACOZ-83CHlIXI93
In tk U.S. Dcpartmcnt
of Energy.
After the diode n-aches thermal
cquilihriom
(spproximately
05 h for diodes
inkgral
IO tk module),
instantaneously
(cil.5 x) replace
tk
tect currmt
(1.25 x Iw) with lm and immediately
meawrc
~he dindc forward
voltage
drop. Wpurc 4 show% the ted circuil
for measuring
tk hyparrdiodc
voltage
drop. (Note:
volIngc and cuncnt
indrumcnIation
should provi&
accuracy
lo
three or four placn:
we a faa-acling
switch to remove tk high teat cumnt
and pats I, Ihrmrgh
the diode.)
3.
REFERENCFZS
I.
IEEE Recommended
ANSl/lEEE
Standard
Electronics
Engines
2.
Block V Solar Cell
Load Applications.
Lakratory.
3.
UndcrwriIcr’a
Labnratorics
Inc.. August
I. 1986, Standard
la Safety
Flat-Plate
Photovoltaic
Modules
and Panelr
UL Standard
1703.
4.
Interim f.&slification
Tests and Procedures
for Termstrial
Plmtovoltaic
Film Flat-Piale
Modules,
January
1990. SERl/TR-213-3624.
Golden.
Solar Energy
Research
InstiIute.
Dctcrminc
Ihe dinde junction
temperature
T, under 1.25 I current conditions
using Ihc curve cctahlishcd
in slcp I, and adjust T, from%
previous
step to
correct fnr sn ambient tcmpersture
of 40°C hy using tk folloting
equation:
T, ( 125 mW/cm’.
40°C)
= T, + 40°C
Thit
is the cxpcctrd
diode
ambient
field conditiom.
junction
T,,o,
Iempcraturc
under
125 mW/cm’.
40°C
The pans/fail criterion
Tar this test ix that the diode junction
temperature
compuIcd wing
EqunGon
I should nnI exceed 125°C for p-n silicon diodes or 75°C
for Schottky
diodcc.
Flgtrrc
4.13
4. Teat
Hot-Spol
Clrcult
Endurance
for
Menwring
Bypaw-Diode
Junction
is cnnsidcrcd
to have pawed the hn(-vpol endurance
text if aI the
this IctI. there ix no vicihk
damn8e In Ik Ik module.
Evidence
(including
delamination.
outgasring
or hlislcring
of cncspwlsntc).
mldcr melting.
or olhcr defects rcwlting
from this test should k
module damage shnuld nccur. I diode or mnrc dindcq per rcricr
rquirrd
lo mitigate
the hot-spot
healing effccI.
TemperrItrre
Test
The purport
of this text is lo evahrale
the ahility
of a module to endure the
long-Inm
effeclr
of periodic
ho(-cpoI
hesting
associated
with common
fault
conditions
.xuch as xcvcrely
cracked
or miamntchcd
cells. single-point
op”
circuit
fnihwcr.
or nnnonifnnn
shadowing
(prrlinl
shadowing).
19-6
Criteria
for Terrestrial
928-1986.
New York:
(IEEE).
Module
1981.
StrecIrai
Standard
Photovoltaic
The Institute
Pora
Systems.
of Electrical
and
Desian and Test Spccilication
JPL SIOI-161.
Paradcna.
CA:
for
ThinCO:
5.
Terre&al
Solar
m.
ASTM
and Mmrisls.
Ii.
Electrical
Performance
of Non-GmccntraIor
Terre&al
Photovoltaic
Modules
and Arrays
Using
Reference
Cells, ASTM
SIandanl
E1036.
New York:
American
Society
ol Testing
and Materials.
7.
Cyclic
1979,
8.
Resistance
JPUSIOI-62,
Pressure-Load
JPUSIOI-19.
irradiance
Table
al Air Mass
E892. New York:
American
fm Intermediate
JeI Propulsion
Developmental
Pasadena.
CA:
Tcstina
for
1c1 Propulsion
L5 for a 37” Tilled
Society of Tnling
Solar Panels.
Laboratory.
of Photovoltaic
Solar Panels to Shmtlatcd
l’asadena,
CA:
Jet Propulsion
L&oratory.
Hail.
Fcbrumy
Api1
l97b
TJJE U.S. DOE/NREL
R.L. Stafford,
SILICON
AMORPJIOIJS
PJJOTOVOLTAICS
PROGRAM
W. I.uft, R. von Roedem, R. Crandall, and W.L. Wallace
Nntiounl Rpncwable Energy Laboratory
(Tom&y
the Solar Energy Research Institute)
1617 Cole Blvd., tiolden. CO 80401, USA
phone: 305231-IO(H);
fax: 303-231-I 199
ABSTRACT
This paper reviews the recent advances of the 1J.S.
Department
of Energy (DOE)/National
Renewable
Energy
Laboratory’s
(NREL) amorphous silicon photovoltaics program.
Reseah
conducted at universities,
industry, and government
laboratories
is addressing the critical technological
issues of
increasing photovoltaic
module reliability
and improving
the
stabilized performance.
Multijunction
device structues have
demonstrated higher stabilized efficiencies than those of singlejunction devices. In addition. novel deposition techniques and
modifications
to conventional
deposition
techniques
have
produced intrinsic amorphous films with improved stability that
have not been fully realized in devices. Results are given for
multijunction
module performance after continuous 1000 hours
of illumination
at one-sun
intensity
and 50°C module
temperature.
These conditions
are used as a predictor for
stabilized performance.
A new “Amorphous
Silicon Utility/
Industry Power Project” has begun at NREL with a goal of
deploying amorphous silicon photovoltaic
power systems with
improved stabilized performance over present systems.
INTRODUCTJON
Since the fast commercial
amorphous silicon module
was produced 7 years ago, the amorphous silicon technology
has made pat
progress -- especially
in reducing the lightinduced degradation
and improving
the stabilized efficiency.
The amorphous
silicon
photovoltaic
technology
has a
significant
market potential that will develop further as the
knowledge base matures and the industrial base solidifies.
In
1990. amorphous
silicon photovoltaic
shipments were 14.7
MW.
repenting
one-third
of the world
photovoltaic
shipments totalling
46.5 MW (1).
These shipments were
primarily small cells for powering consumer electronics such as
calculators and watches. Amorphous silicon modules currently
on rhe market are priced comparable
to crystalline
silicon
modules on a per Watt basis and have comparable warranty
the development of this broad applications base for amorphous
silicon materials is a result of the federal government-supported
research for amorphous
silicon materials and photovoltaic
devices in the latter half of the 1970’s and the early 1980’s.
Worldwide,
there are approximately
20 companies
pursuing
amorphous
silicon
photovoltaic
research
and
development
and manufacturing
activities.
U.S. companies
involved
in amorphous
silicon
for PV applications
are
Advanced
Photovoltaic
Systems (APS); Energy Conversion
Devices @CD); International
Display Materials (IDM); Iowa
Thin Film Technologies
(ITFT); Solarex; Utility Power Group
(UPG); and United Solar Systems Corporation
(USSC), a joint
venture of ECD and Canon.
Since 1984, amorphous
silicon modules have been
installed in a number of demonstration
systems throughout the
world.
Lessons learned from system characterization
are that
reliable module and system performance
requires necessary
manufacturing
quality control, and proper module design and
System performance
has been observed to
encapsulation.
stabilize within the first six months of operation.
Thereafter,
reliable and predictable system performance has been obtained
for several years in many systems.
Utility-interconnected
system tests are being conducted
in collaboration
with a
number of electric utility companies, including Pacific Gas &
Electric under the Photovoltaics
for Utility Scale Applications
(PVUSA)
contract,
Philadelphia
Electric.
Alabama Power,
Detroit Edison, and Maui Electric (PVUSA).
The biggest
concern of system engineers is not the initial light-induced
power loss but the low stabilized efficiency
of these systems
which results in higher balance-of-system
costs compared to
systems using more efficient modules.
For market growth of
amorphous silicon modules into the larger utility power sector,
higher stabilized power outputs and lower costs are needed.
PROGRAM
ORGANIZATION
The Amorphous
Silicon Research Project (ASRP) at
NREL has oversight
for all research and development
of
amorphous silicon photovoltaics
funded by the DOE.
The
primary
objective
is to develop
the amorphous
silicon
photovoltaic
technology to be a major contributor for the future
needs of the utility power sector. Major components of the
ASRP include internal research at NOEL and subcontracted
research at universities,
industry, and government laboratories.
A new “Amorphous Silicon Utility/ Industry Power Project” has
pXiOdS.
Amorphous silicon photovoltaic
products represent less
than 10% of a much larger market which includes amorphous
silicon materials used for thin film transistors in active matrix
flat-plate displays, for electrophotographic
drums in photocopier
machines, and for optical sensors in facsimile machines. Total
worldwide
amorphous
silicon
product
sales in 1990 are
estimated at close to one billion dollars (2). The impetus for
20-l
begun at NREL with a goal of deploying amorphous silicon
photovoltaic
power
systems
with
improved
stabilized
performance over present systems. Awards for this project are
expected soon. New three-year subcontracts wem s~nrtrd in
1991 for fundamental research on specific tnntcri:ll ml tievicr
issues. Table 1 lists the major areas of research nltpportml by
the ASRP, along with NREI, subcontractors.
MODUJ,E
DEVEIAWMENIS
Because improvements
in the stabilized efficiencies
of
amorphous silicon modules will result in more cost-effective
systems, there has been a concerted effort by industry and
NREL to improve the stabilized efficiencies.
In 19841985,
when commercial
amorphous
silicon
modules
were first
introduced the stabilized module efficiencies
were in the 2%4% range. By 1987-1988, the stabilized module efficiencies
had improved to the 3%-S% range. In 1991, NREL measured
stabilized efftciencies
of modules (late 1990 vintage), both
commercial and prototype, in the range of 64-74
(3). Fuji
recently
reported
a prototype
module
with a stabililad
efficiency
of 8.5% (4).
Demonstrating
stabilized efficiencies
PJREL INTWAL
Silicon Projects at NREL
AMORPHOUS
SILICON RESEARCkJ
Metastability
Hot Wire Deposition
Process
Device Fabrication
and Engineering
Low Bandgap a-SiGe:H Alloys
Solicitation
released in March
expected in Fall of 1991.
GOVE-NDUSTRY
Solarex (Catalano)
USSC (Guha)
One other company
1991.
Initial module efficiency
is not a reliable indicator of
expected long-term performance. consequently
it is no longer
used to measure research and development
progress.
The
ASRP now reports stabilized module and cell efficiencies.
Module and cell performance
after continuous 600 to 1000
hours of illumination
at one-sun intensity and 50°C module
temperature is used as a predictor for stabilized performance.
Table 2 summarizes
stabilized
performance
of diffennt
modules measured at NREL.
NREL
published
“Interim
Qualification
Tests and
Procedures for Terrestrial
Photovoitaic
Thin-Film
Flat-Plate
Modules”
to evaluate thin-film
modules intended for power
generation applications (5). The term “interim” is used because
thin-film
module design and cell material technologies
are
undergoing evolutionary
changes in their development.
The
tests and procedures will be revised in the future to incorporate
the latest information
on failure mechanisms and the rtlationships between accelerated tests and field reliability.
Current
understanding of failure and degradation mechanisms and the
relationship between accelerated tests and field reliability is not
sufficient
to allow accurate estimation
of life-expectancy.
Acceptable results from these tests should provide reasonable
assurance that the modules will perform reliably in the field but
for an unspecified period of time. Also, the test and evaluation
procedures
provide
a common
approach
for conducting
qualification
tests.
commercial and prototype modules with
of 8% and 10%. respectively,
in two
TABLE 1. Amorphous
years seems achievable.
These higher stabilized efficiencies
will result from improving
light management, using thinner
intrinsic layers, and using different
band gap multijunction
device structures. Improved light management will result from
less absorptive
transparent conducting
oxides, thinner and
wider-band-gap
p-layers,
and
highly
reflective
back
metalizations.
Multijunction
device structures will be used with
narrower-band-gap
ceils using a-Si:H or a-SiGe:H
alloys.
Stahiiized module efficiencies of 8%-IO%, coupled with lower
manufacturing
costs, would lead to cost-effective
systems and
larger markets.
Awards
PROGRAM
APS is developing their Eureka module (nominally
1.2
m2 in area) for the utility power market.
The large size
manufactured
in high volumes
and coupled
with APSs
innovative balance-of-system
approach is expected to be costeffective for utility-scale
systems. The module and system will
be demonstrated in a 480-kW system at the Photovoltaics
for
Utility-Scale
Applications
(PVIJSA) site. The module design
uses single-junction
cells monolithically
interconnected
on a
glass superstrate and is encapsulated with ethylene vinyl acetate
(EVA) and glass on the back side. NREL has measured the
initial power output of one module at 77.8 W outdoors under
prevailing conditions.
Earlier, NREL had measured a different
module with a power output of 50 W, APS reported that this
module was stabilized by 130 days of outdoor exposure.
under negotiations
Colorado School of Mines (Williamson)
Harvard (Gordon)
Harvard (Paul)
Institute of Energy Conversion
(Baron)
Iowa State University (Dalal)
Jet Propulsion
Laboratory
(Shing)
National Inst. Standards & Technology
(Gallagher)
North Carolina State University (Lucovsky)
Pennsylvania
State University (Wronski, Collins):
Syracuse University (Schiff)
University of Illinois (Abelson)
University of North Carolina (Silver)
University of Oregon (Cohen)
Xerox (Street)
USSC is shipping
same-bandgap
(nominally
0.4 m2 in area) at a rated
(stabilized)
with a ten-year performance
purchased 11 modules (Model # UPM-880)
testing and evaluation.
The average initial
20-2
tandem modules
power of 22 W
warranty.
NOEL
from USSC for
aperture efficiency
TABLE 2. Stabilized
Efficiencies
of a-SI Modules
Measured
at NREL
l
efficiency
reported
by Fuji
has an aperture area of approximately
900 cm2. The first year
of the program is concentrating
on scaling the small-area cell
technology to larger areas of around 900 cm2.
of these modules is 7.0% measured outdoors under prwniling
conditions.
This average efficiency corresponds to an average
initial power output of 26 W. Five modules have been placed
outdoors at NREL under load to determine their stabilized
performance.
Four of the eleven are being subjected to thermal
cycling
and humidity-freeze
cycling
tests to assess the
durability of commercially
available modules.
INTERNAL
AND
FUNDAMENTAL
RESEARCH
Critical
research
issues
for
amorphous
silicon
photovoltaic
technology
include reducing or eliminating
the
metastability;
improving
low and high band gap materials for
use in multijunction
devices; understanding and controlling the
deposition
processes used; and understanding
device issues
related to interfaces, modeling, fabrication
and performance.
Progress in some of these areas is presented.
Sohuex sells single-junction
amorphous silicon modules
(nominally
0.1 m2 in area) at a nominal power of 5 W with a
five-year performance warranty.
NREL purchased 15 Solarex
modules (Model # SA-5) from a commercial vendor for testing
and evaluation.
The average initial aperture efficiency of these
modules is 5.9% measured under a Spire simulator and is 6.2%
when measured outdoors under prevailing conditions.
These
average efficiencies correspond to average initial power outputs
of 5.6 and 5.8 W, respectively.
Nine modules have been placed
outdoors at NREL under load to determine their stabilized
performance.
Six of the fifteen am being subjected to thermal
cycling
and humidity-freeze
cycling
tests to assess the
durability of commercially
available modules.
GOVERNMENT/INDU.STRY
total area
Modeline
and Measurements
of Metastability
Stretched exponential
curves are used frequently
to
describe the degradation
of films and devices because the
degradation,
in one way or another, is a hierarchically
constrained phenomenon.
Mechanisms
proposed that could
lead to this behavior are silicon bond breaking balanced by
thermal annealing (8), and charge trapping and recontiguration
balanced by thermal annealing
(9).
McMahon
observes
quenched-in ESR center formation and annealing in ultra-highpurity a-Si:H (10). Although McMahon can analyze the data
in terms of a simple two-level configuration
coordinate systems,
having a formation energy of 0.35 eV and an anneal barrier of
2.1 eV, additional measurements by McMahon show this simple
picture to be incomplete.
PROGRAM
Solarex is developing triple-junction
modules using an
a-SiGe:H alloy for the bottom cell and an indium tin oxide
(ITG)/silver
back reflector.
The goal is to demonstrate
a
module with 12% stabilized efficiency.
The baseline module
design uses a superstrate glass encapsulated
by a spray-on
polymer encapsulant, and has an aperture area of appmximately
900 cm2. NREL has measured the initial aperture efficiency of
one of these modules at 9.3%.
After 1000 hours of light
soaking (1000 W/m2) at a module temperature of SOaC, NREL
measured an average efficiency
of 6.9% for three different
The 22% power loss after 1000 hours of lightmodules.
soaking results from the characteristic
light-induced
losses in
the amorphous silicon materials and also from light induced
shunts related to module processing conditions of the ITO/silver
back contact (6). Modules using lower-reflectivity
aluminum
hack metalization have only a 15% power loss and do not show
any losses from light-induced
shunts.
von Roedem reports photocurrents and minority carrier
diffusion
lengths measured on one sample that degraded at
This
different
rates and on different
time scales (11).
observation is inconsistent with models of a single defect level.
While significant
pmgress has been made in reducing the
metastability
using simplified models as guides, it is clear that
the next level of progress will result from more comprehensive
The
models consistent with ail experimental
observations.
presence of charge-trapping
defects and changes in the amount
of trapped charge is experimentally
established in amorphous
silicon films. The discussion within the research community
needs to focus on the density of charge-trapping
defects and
their effect on stabilized device performance.
USSC is developing triple-junction
modules using an aSiGe:H alloy for the bottom cell based on ECD’s earlier cell
developments
of >13% initial efficiency
(7). The goal is to
demonstrate a module with 12% stabilized efficiency.
The
baseline module design uses a flexible stainless-steel substrate
with a Ag/ZnO back-reflector
encapsulated by EVcVTefzel. and
Accelerated light-soaking
is a quick method to reach a
stabilized state in a-Si:H films and devices.
The challenge is
interpreting
the results and keeping the sample temperature
constant.
Clearly, degradation of a device efficiency
by 80%
m-3
under 100 suns intensity does not represent a clrgratintion 104s
that will be observed under normal operating conditions at onesun intensity.
Raising the temperature of the device while
using high-intensity
illumination
can empirically
estnhlish only
a singular point of equivalence (12). Solarex prcsscntcd n srlfconsistent model for high intensity light-soaking
of single
junction devices assuming only neutral defects in the intrinsic
layer and stretched-exponential
kinetics (13). 1lowever, much
more must be done to reliably use accelerated liphr-soaking
as
a predictor of solar cell performance in the field.
ACKNOWLEDGMENTS
Many researchers have established relationships, but not
cause and effect, between metastnbility
and hydrogen content
or motion.
Xerox has published experimental
evidence for
light-enhanced hydrogen diffusion in a-Si:H films at 175OC and
above (14). The enhancement is attributed to an increased
release rate of hydrogen from silicon-hydrogen
bonds in the
presence of photo-generated
carriers.
Deoosition
amorphous silicon technology as a viable photovctlnic
option
(18). USSC has upgraded its production line and is se!!ing a
significantly
improved module.
Solarex is developing a iOMW/yr multijunction
pilot-production
line for larger modules.
17T;T is producing flexible and lightweight
amorphous silicon
modules on 13-inch-wide
rolls of polymer.
Research and
development
supported by DGE/NREL
will be continually
inctwporatrd
into newer modules
resulting
in improved
efficiencies,
increased reliability,
and lower costs.
This work is performed under Contract No. DE-ACOZ83CH 10093 to the U.S. Department of Energy.
REFERENCES
P. Maycock,
Studies
PV News. Vol. 10. No. 2. February 1991.
A. Madan, private
Modifications
of the conventional deposition method of
plasma enhanced chemical vapor deposition (PECVD) are being
investigated to reduce hydrogen content in a-Si:H for improved
stability.
Increasing the deposition temperature above 250°C
and increasing the dilution of silane with hydrogen results in
limited success in reducing the stabilized defect densities of
films.
communication.
W. Luft, B. von Roedem, B. Stafford, D. Waddington,
L. Mrig, “Controlled
Light-Soaking
Experiment
for
Amorphous Silicon Modules,” this proceeding.
Also, alternative deposition methods such as electron
cyclotron resonance PECVD (ECR-PECVD)
(15) and hot-wire
assisted CVD (HW) (16,17) have produced films with apparent
improved stability.
The HW films are remarkable in that the
hydrogen content ranges from 1% to 4%, and they have more
and larger voids, as determined by small-angle x-ray scattering
(SAXS), than those of conventional
PECVD films. Generally
voids are considered detrimental and researchers try to reduce
the density and size of voids. The best low hydrogen content
HW
films
with
good
pmperties
required
deposition
temperatures above 4OO’C. Plans are under way to incorporate
these films in devices.
All of these films, deposited
by conventional
or
alternative
deposition
methods, have been reported
with
improved stability; but the films still must be incorporated into
device structures for full verification
of improved stability. The
amorphous
silicon research community
is hindered by an
incomplete
understanding
of what material properties
are
necessary and sufficient
for improving
the stabilized device
performance.
The most progress in improved stabilized device
performance has been made primarily by reducing the intrinsic
layer thickness and improving the optical light trapping of the
device.
SUMMARY
Significant
progress has been made in improving
the
stabilized efficiency
of amorphous silicon modules.
Private
sector announcements and commitments reflect confidence in
20-4
4.
Hiroshi Sakai and Yukimi Ichikawa, presented at 14th
ICAS, Garmisch-Partenkirchen,
FRG, 19-23 August
1991, J. Non. Crvst. Solids. in press.
.5.
R. DeBlasio,
L. Mrig.
D. Waddington,
“interim
Qualification
Tests and Procedures
for Terrestrial
Photovoltaic
Thin-Film Flat-Plate Modules,” SERIIIR213-3624,
Golden.
Colorado:
National
Renewable
Energy Laboratory
(formerly
Solar Energy Research
Institute), January 1990.
6.
M. Bennett, J. Newton. C. Poplawski, K. Rajan, “Jmpact
of Defects on the Performance of High Efficiency
12”
x 13” a-Si Based Three-Junction
Modules,”
this
proceeding.
7.
S. Guha, “Research on High-Efficiency,
Multiple-Gap,
Multi-Junction
Amorphous Silicon-Based
Alloy ThinSERlIIP-21 l-3918,
Golden,
Film
Solar
Cells,”
Colorado:
National
Renewable
Energy L&oratory
(formerly
Solar Energy Research Institute),
August
1990.
8.
R. A. Street. AIP Conf. Proc., 234, 1991, p. 21.
9.
Howard M. Branz, Richard S. Crandall,
AJP Conf. Pmt., 234, 1991. p. 29.
IO.
T. J. McMahon,
11.
B. von Roedem, unpublished. Similar results are
reported by E. Sauvain. J. Hubin, A. Shah, P. Pipoz,
Phil. Map. Letts. 63. 1991, p. 327.
Marvin
Silver,
AJP Conf. Pmt., 234. 1991. p. 83.
2.
T. Tonon. X. Li. A. E. Delnhoy, AIP
d-.-L,C’onf I’roc
1991, p. 259.
3.
Liat;g-fan Chen and Liyou Yang. prrsrnlrti at 1~1th
ICAS, Garmisch-Partenkirchen, FR(;. 1Q.L.I Attpw~
1991, J. Non. Cwt. Sulids, in press.
14.
15.
221,
P. V. Santos, C. Dobnd. N. M. Jtd~nsc~, R. A. Strcrt.
ptesented at 14th ICAS, Garmisch-Wrteukircbm, FRG.
19-23 August 1991, J. Non. Crwt. Solids, in press.
Vikram L. Dalal, Ralph Knox, Greg RaIdwin, N.
Kandalaft. “Growth and Pmpenies of Amorphous
Silicon Films Grown Using Pulsed-Flow Reactive Ream
Epitaxy,” this proceeding.
20-s
16.
A. l-1. Mahan and M. Vanecek, AIP Conf. hoc., 234.
1991. p. 195.
17.
A. H. Mahan. Y. Chen. D. L. Williamson. G. D.
presented at 14th ICAS, GarmischMmney.
Partenkirchcn, FRG, 19-23 August 1991.1. Non. Cwst.
Solids in press.
d*
18.
D. Carlson, “Markets, Manufacturing, and Technical
Progress in Amorphous Silicon
in the U.S.,” this
..
prnceeding,
EFFECTS OF HELIUM DILUTION ON GLOW DISCHARGE DEPOSITIONS OF
a-Si,$e,:H
ALLOYS
Y.S. Tsuo, Y. Xu. 1. Balberg*, and Richard S. Crandall
National Renewable Energy Laboratory
(Formerly the Solar Energy Research Institute)
Golden, Colorado, U.S.A.
l Racah Institute of Physics, The Hebrew University
Jerusalem, Israel
ABSTRACT
We have studied the effects of helium feed gas
dilution on the properties of a-Sit-xGe,:H alloys deposited
using radio-frequency glow discharge decomposition of silane
and germane gas mixtures. Comparing a-Sit-xGex:H films
deposited using 65% helium dilution with films deposited
using 65% hydrogen dilution, we find that films deposited
with helium dilution have a longer charge carrier diffusion
length and higher quantum efficiency-mobility-lifetime
product values. We also find that the incorporation of Ge
atoms in the a-Sit-xGe,:H film is more efficient with helium
dilution than with hydrogen dilution.
INTRODUCTION
The development of multiple-junction solar cells is
important for improving the conversion efficiency and
stability of amorphous-silicon-based photovoltaic modules.
Although the “same-band-gap” a-Si:H/a-Si:H tandem cell
approach has achieved high efficiency and stability (I), such
cells have limited potential for further improvements in cell
efficiency because of insufficient light absorption. To further
improve the efficiency of amorphous silicon multijunction
cells, we need to develop highquality, low-band-gap alloy
materials. So far, only a-Sit-xGex:H (a-SiGe:H) alloys have
achieved high enough quality to be used in multijunctiotr cell
However, problems. such as preferential
~struiiures.
attachment of H to Si rather than to Ge and poor
microstructure, limit the electronic property of a-SiGe:H
alloys. Electronic properties of a-SiGe:H deteriorate rapidly
when the Ge content is increased beyond 40 at.% and when
the optical band gap, Eg, is less than 1.5 eV (2,3).
Methods of improving the quality of glow-dischargedeposited a-SiGezH that have been studied recently include
increasing the substrate temperature (4), hydrogen dilution of
the feed gas mixture (4-S), argon dilution of the feed gas.
I
21-1
Increasing the ’
mixture (2). and pulsed discharge (9).
substrate temperature by about 50°C above the normal
deposition temperature of a-Si:H usually can improve the
photoelectronic properties of a-SiGe:H (4). Tsuo et al. (4)
observed no significant improvements of the transport
properties of a-SiGe:H with up to 83% hydrogen dilution of
feed gas mixtures of SiH4 and GeH4. They also noted that
the film deposition rate decreases rapidly when the hydrogen
dilution ratio is above about 50% because of feed gas
depletion. Crandall et at. (10) observed increased disorder in
the microstructure of a-SiGe:H films deposited with hydrogen
dilution. Godet et al. (8) observed reduced disorder but no
improvements of the transport properties of a-Ge:H with 99%
hydrogen dilution. However, Bennett et al. (7) observed not
only improved diffusion length but also better photostability
in a-SiGe:H alloys deposited with hydrogen dilution. Argon
dilution was observed to be detrimental to electrical
propenies of a-SiGe:H (2.4).
A major advantage of plasma deposition of thin
films is that it is a product of simultaneous etching and
deposition. It is believed that the etching process during
deposition preferentially removes weak bonds during film
growth. However, radio frequency (RF)-generated hydrogen
plasma does not etch a-&H whereas it etches a-Si:H at
about 2 nm/min. The lack of hydrogen plasma etching of
deposited germanium atoms may be a major reason for the
poor microstructure of glow-discharge-deposited a-Ge:H and
a-SiGe:H. It is well known that hydrogen dilution of the
feed gas does not have significant effects on the film
properties of a-Si:H (11). This is probably because in the
glow-discharge of SiH4, there is already an abundance of
hydrogen in the plasma even without hydrogen dilution.
‘Physical sputtering by ions in the&ma
may also affect
film properties. Argon dilution causes too much sputter
damage and is detrimental to a-Si:H and a-SiGe:H film
properties (4,ll).
Helium dilution may provide the right
amount of sputtering to be beneficial to a-SiGezH film
properties. In this paper, we report our study of the effects
of helium dilution on the properties of a-SiGe:H deposited
using glow discharge decomposition of SiH4 and GeH4 feed
gas mixtures.
EXPERIMENTAL PROCEDURE
All a-SiGe:H samples used in this study were
deposited in a RF glow-discharge deposition system with a
single-chamber reactor, which has a base vacuum better than
1 x IO-* ton; and a load-lock vacuum chamber. The
electrode spacing between the 2.5-in-diameter round
electrodes is 1.2 cm. This electrode spacing should be small
enough to avoid the problems caused by Ge clustering in the
dark space of the plasma (9). The sample substrate is placed
on the unpowered electrode during deposition. The substrate
temperature during deposition is 270°C. which is calibrated
using a 0.005-in.-diameter thermocouple attached to a film
surface with the process gas flowing so that the temperature
is close to the actual film temperature during deposition.
The combined flow rate of SiH4 and GelI is about 32 seem
and the flow rate of the diluent gas, either He or H,, is 60
seem. We use this dilution ratio of 65% for both He and H,
for a fair comparison of the effects of these two diluents.
The deposition rate for He-diluted gas mixtures varied from
0.254 rim/s for a film with E = 1.59 eV to 0.444 IX/S for
a film with E = 1.26 eV. he total gas pressure during
deposition is Of8 torr. The RF power density used is slightly
above that needed to maintain the plasma, which is about 20
mW/cm? All the films studied have a thickness between
500 and 700 nm.
1.3
1.4
1.5
Oplicd
Figure 1
1.6
Ihd
Gap
1.7
(eV)
Quantum efficiency-mobility-lifetime
measured using 600-nm wavelength
function of optical band gap for
films RF-glow-discharge-deposited
dilution and with Hz dilution of
GeH4.
product,
light, as a
a-SiGe:H
with He
SiH4 and
n
a
RESULTS AND DISCUSSION
H
l
m
:;I_-‘-
The values of the quantum efficiency-mobility-lifetime
(rjp~) product, determined from photoconductivity
measurements using 600-nm wavelength light, for a-SiGe:H
films deposited with 65% He dilution and with 65% Hz
dilution are shown in Figure 1 as a function of the optical
band gap.
The ambipolar diffusion length (Ld), as
determined by the steady-state photocarrier grating technique
(12). for a-SiGe:H films deposited with 65% He dilution and
with 65% Hz dilution are shown in Figure 2, also as a
function of band gap. Films deposited with He dilution show
notably higher qjts and Ld values than films deposited with
H, dilution. However, the photoconductivity, measured with
a 100 mW/cm’ unfiltered ELH lamp illumination, and dark
conductivity values of these two types of films are similar,
as shown in Figure 3.
We also note that, for films deposited with the same
GeH4 gas concentration, films deposited with He dilution
have a lower optical band gap than films deposited with Hz
dilution.
Figure 4 shows Eg as a function of GeH4
concentration in the feed gas for films deposited with 65%
He dilution and films deposited with 65% H, dilution. This
difference in Es is caused mainly by the difference in the Ge
content of the films ([Gel). The Ge concentrations in the.
21-2
sool-
-1
I
1
I
1.4
1.5
1.6
Optiral
Figure 2
Itand Cap
Ambipolar diffusion
optical band gap for
discharge-deposited
Hz dilution of SiH4
I
1.7
(eV)
length as a function of
a-SiGe:H films RF-glowwith He dilution and with
and GeH4.
film, determined by electron microprobe analysis, for films
deposited with He dilution and films deposited with Hz
dilution are shown in Figure 5 as a function of GeH4 gas
concentration in the feed gas mixture. This means that less
GeH4 gas is needed to deposit a-SiGe:H films with a given
band gap for He dilution than for Hz dilution. Since GiH,
costs much more than SiH4, He dilution may be an important
cost reduction method.
In addition to a-SiGe:H, we also studied $e
differences between a-&H films deposited with He dilution
and with H, dilution. For an a-Ge:H film with E = 0.98
eV, deposited at 270°C with a 3-seem GeH, flow, a 81O-seem
1.2 -
nD>L-“w
I I
h----v
1.35
1 .&I
1.45
<)pticrl
Figure 3
1.50
nand
1.55
Gap
1.60
1.0 -
I
1.65
0.8b
I
I
0
20
40
I
60
I
80
I
100
1
120
[GeJJ4/(GeJI4+SiJJ4)] (%)
(CV)
Photo- and dark conductivity values as a
function of optical band gap for a-SiGe:H
films RF-glow-discharge-deposited with He
dilution and with Hz dilution of SiH4 and
GeH,.
Figure 4
Optical band gap as a function
concentration in the feed gas for
films RF-glow-discharge-deposited
dilution and with Hz dilution of
GeH4.
of GeH4
a-SiGe:H
with He
SiH4 and
He flow, a 0.68-torr total gas pressure, and a deposition rate
of 0.142 rim/s,, the photoconductivity at 100 mW/cm2 is 8.79
x lV5 S/cm, and the dark conductivity is 6.86 x 10m5S/cm.
This photo-to-dark conductivity ratio of 1.28 is the highest
we have observed for a-Ge:H films. The highest photo- to
dark conductivity ratio for a-&H films deposited by us
using Hz dilution is 0.4.
We believe the beneficial effects of helium dilution in
q&t% L4. and [Gel come from the enhanced bombardments of
the growing a-SiGezH film surface by helium ions. Film
growth steps on the substrate consist of surface diffusion of
deposition radicals and hydrogen elimination (11). The
removal of weak, distorted bonds can be achieved by
chemical etching or physicai sputtering. Because a hydrogen
plasma etches Ge-Ge bonds less efficiently than Si-Si bonds,
the added He ion bombardment in the case of helium dilution
may assist the rearranging of Ge-Ge bonds at the growing
film surface to lower energy contigurations.
It is also
possible that He ion bombardment enhances the surface
mobility of Ge-containing radicals on the growing film
Surface.
The fact that He dilution increaseses the
incorporation of Ge atoms in the deposited aSiGe:H film for
a given germane gas concentration indicates that the ion
bombardment-assisted rearrangements of Cie-Ge bonds also
reducethe amomt of Ge atoms escaping ftom the growing
film surface. These arguments of the benefits of enhanced
ion bombardment are consistent with the results obtained by
Paul et al. (13) &owing that a-GezH films with better
microstructure were obtained when the substrate was on the
powered elect&e, which has more ion bombardment, rather
than the unpowered ekclrode in an RF glow discharge
deposition. The poor quality of a&H
and a-SiGezH films
deposited with argon dilution of the feed gases may be due
20
40
60
80
100
120
(GeJJ4/(CeJJ4+SiJf4)I (%)
Figure 5
Ge concentration in the film as a function of
GeH4 concentration in the feed gas for aSiGezH films RF-glow-discharge-deposited
with He dilution and with Hz dilution of SiH4
and GeH,.
Lo excessive surface sputtering damages by the very heavy 5
argon ions.
CONCLUSXONS
Comparing a-SiGezH films deposited with 65% He
dilution and with 65% Q dilution, we find that films
deposited with He dilution of SiH4 and GeH4 have higher
Rpr (measured using 6O&tm wavelength light) and L4
21-3
‘values. In addition, He-dilution of the feed gas results in’
higher Ge concentration in the film than H2 dilution of the
same feed gas mixture. We believe these beneficial effects
of He dilution come from He ion bombardment of the
growing film surface.
ACKNOWLEDGMENT
This work was performed under Contract No. DEACO2-83CHlOO93 to the U.S. Department of Energy and by
the U.S.-Israel Binational Science Foundation. We thank
Alice Mason of SERI for the electron microprobe analysis
measurements.
REFERENCES
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
Y. Ichikawa, S. Fujikake, T. Yoshida. T. Hama, and
H. Sakai, Proc. 21st IEEE Photovoltaic Specialists
Conf., pp. 14751480 (1990).
M. Stutzmann, R.A. Street, C.C. Tsai, J.B. Joyce, and
S.E. Ready, J. Appl. Phys. &, 569 (1989).
W. Luft. Appl. Phys. Comm. 2 43 (1989).
Y.S. Tsuo, Y. Xu, E.A. Ramsay, R.S. Crandall, SJ.
Salamon, I. Balberg, B.P. Nelson, Y. Xiao, and Y.
Chen, Materials Research Society Symposium
Proceedings, 219.769 (1991).
A. Matsuda, M. Koyama. N. Ikuchi, Y. Jmanishi. and
K. Tanaka, Japn. J. Appl. Phys. 25, L54 (1986).
J. Wind, 9. Ktitz, V. Petrova-Koch, G. Miiller. and
P.P. Deimel. J. Non-Crystal. Solids 114,531 (1989).
A. Catalano, Proc: 21st IEEE Photovoltaic Specialists
Conf., pp. 36-40 (1990); and MS. Bennett, A.’
Catalano, K. Rajan, and R.R. Arya, hoc. 21st IEEE
Photovoltaic Specialists Conf., pp. 1653-1655 (1990).
C. Godet, V. Chu, B. Equer, Y. Bouizem, L. Chahed,
I. El Zawawi, M.L. Theye. S. Basrour. J.C. Bruyere,
and J.P. Stoquert, Materials Research Society
Symposium Proceedings 192. pp. 163-168 (1990).
T. Yoshida, H. Fujisawa. T. Hokaya, I. Ichikawa, and
H. Sakai, Materials Research Society Symposium
Proceedings, 219. 655 (1991).
R.S. Crandall, Y.S. Tsuo, Y. Xu, A.H. Mahan, D.L.
Williamson, Solar Cells, a 15 (1991).
W. Lufi and Y.S. Tsuo, Appl. Phys. Comm. & 1
(1988).
I. BaIberg, A.E. Delahoy, and H.A. Weakliem, Appl.
Phys. Lea. a 992 (1988).
W. Paul, SJ. Jones, F.C. Marques, D. Pang, W.A.
Turner, A.E. Wetsel, P. Wickboldt, and J.H. Chen,
Materials Research Society Symposium Proceedings,
219. 211 (1991).
21-4
mm
T.A . Gessert, X. Li, and T.J. Coutts
National Renewable Energy Laboratory
Golden, Colorado 80401
(Formerly the Solar Energy ResearchInstitute)
N. Tzafaras
AT&T Microelectronics
Reading, Pennsylvania, 19612
ABSTRACT
This paper presents and discusses the pmcedures used
in this pilot production of 4cm2 ITO/InP cells. The discussion includes analyses of the performance range of all production cells and device performance data of the best cells
produced so far. Additionally, processing experience gained
from the production of these cells is discussed, indicating
other issues that may be encountered when larger-scale productions are initiated. The paper also presents the presently
available information concerning another 4-cm2 ITO/InP
solar cell that was launched on the UoSAT-5 satellite. This
information describes the mounting of the cell into the flight
“coupon,” and indicates performance changes after each
mounting step.
This paper presents the experimental results of a pilot
production of 32 4-cm2 indium tin oxide (ITO)InP space
solar cells. The discussion includes analysis of the device
performance of the best cells produced as well as the performance range of all production cells. The experience
gained from the production is discussed, indicating other
issues that may be encountered when larger-scale productions are initiated. Finally, available data on a 4-cm2
ITOiInP cell that was flown on the recently-launched
UoSAT-5 satellite is reported.
INTRODUCTION
InP homojunction solar cells have becomeimportant for
spaceapplications in recent years becauseof the& combiiation of radiation resistance and high efficiency. Thus far, the
highest reported efficiency has been achieved with epitaxially-grown cells (19.1% Air Mass Zero (AMO), NASA measurement, 4-cm2 cell, Spire Corporation) (1). However, the
majority of the InP cells that have been used on actual spacecraft are produced by a simpler process of closed ampule
diffusion, yielding cells with lower efficiencies (highest
reported: 16.6% AMO, NASA measurement, 4-cm2 cell,
Nippon Mining Company, Japan) (2). In comparison to
these results of diffused-junction cells, remarkable performance and radiation hardness has been achieved with
ITO/InP cells (17.0% AMO, SERI measurement, 0.1-cm2
cell) (3.4). In this latter cell design, the IT0 is sputter
deposited onto an InP p- base, at room temperature, using
either radio frequency (if) or direct current (dc) magnetron
sputtering of an In203/Sn@ composite target. The sputter
deposition process causes type conversion of the InP surface
and forms a shallow-homojunction solar cell beneath a lowsheet-resistanceIT0 window. In addition to this technique’s
relative simplicity, it has been shown that it can be easily
scaled to produce large-area cells with respectable efficiencies (16.2% AMO, NASA measurement,4-cm2 cell).
Since the results mentioned above indicate that the
ITOAnP technology is nearing practicality, a pilot-production
project has been completed in which 32 4-cm2 ITOAnP cells
have been produced. Through this project, not only has a
mote representative assessmentof the performance of largearea ITO/InP cells been established, but the previously
assumedadvantages of production scale-up has been tested.
This larger volume of cells has also created the opportunity
to gain a better understanding of the effect of fabrication procedureson cell performance and has allowed several recently
developed process improvements to be further optimized.
These improvements include two-gun sputtering, premetallization plasma cleaning, and grid/mctallimtion optimization.
EXPERIMENTAL
. .
abncatto Procedures Used for Ptlot-Productton cells
The &erials and processes used for the pilot production of lTO/InP cells have been developed over many years
of research and are discussed in several publications (5,6,7).
However, to provide a benchmark of these evolving processes, they are outlined here. The InP substrates were
supplied by AT&T
Microelectronics
(Reading,
Pennsylvania), had carrier concentrations of 0.5-2.0 x 1016
cm-3 [Zn-doped, (100) orientation], and were supplied
polished on the front side and chemically etched on the back
side. Photoluminescence measurements of the as-received
substrate material indicated bulk lifetimes very similar to
those of other -1 ~10~6 (Zn-doped) materials used in
previous research, demonstrating up to -10 nsec on unpassivated surfaces. The substrateswere cleaved by the vendor
to 1 in. x 1 in. squares, on which was fabricated a single 4cm2 cell. Prior to junction formation, back surface metallization was performed. This multistep process involves the
vacuum dewsition of 120 nm of AuBe (1 wt 8 Be) usine an
ULTEK vacuum system, annealing at 380°C chemTca1
etching in concentrated NaOH, and electrochemical plating a
thick layer of Au (1.5 urn, Aurovel UP24 Plating Processof
Lea Ronal Inc.). Since the thickness uniformity of the AuBe
vacuum deposition is not critical, up to eight substrateswere
metallized at one time. Except for cleaning in organics after
back contacting, no additional front surface preparation was
performed before cell fabrication.
The IT0 deposition was performed in the sameULTEK
vacuum system using 2-in. planar US Guns in a sputter-up
orientation with a source-to-substrate distance of -12 cm.
The IT0 targets were purchased from KEMA (Gustine,
Texas) and were 91 molar 46 In203 and 9 molar % St@.
Earlier studies indicated that adding a small amount of H2 to
the Ar sputtering gas substantially increased the open circuitvoltage (V,,) of the resultant solar cell (4). However,
22-1
continued sputtering of IT0 in this Hz-rich atmosphere
progressively altered the target material, resulting in poor
control of the optical and electrical properties. Thus, to
provide greater compositional control of the IT0 Elm(s), two
US Guns have recently been incorporated into the vacuum
system. The fit gun deposits IT0 in an Ar/H2 atmosphere
at a total pressure of -1 x lo-* torr, a partial pressureof HZ
of -3.6 x 10-3torr, and a very slow deposition rate of -0.01
nm set-I. For the Ar/H2 deposition, the pressure is measured using a Granville-Phillips Convectron gauge. Since
the optical transmission of this Hz-rich IT0 is poor, tne
thickness of this layer is limited to 5 nm, as indicated by a
crystal monitor. Once this first 5 nm of Hz-rich IT0 has
been deposited, the shallow-homojunction formation is
complete. However, to reduce the emitter sheet resistance
and to form the first layer of a two-layer ITO/MgFz antireflection coating (ARC), an additional 55 nm of IT0 is
deposited using the second US Gun source. For this deposition, an Ar/OZ/HZ ambient is used (“02~rich” ITO); the a
and HZ partial pressures are adjusted to yield optimum
electrical and optical properties. Both the Hz-rich and the
@rich depositions are performed without breaking vacuum;
however, due to thickness uniformity considerations for the
IT0 ARC, only a single cell is produced per deposition.
Following deposition, ellipsometry and four-point probe
measurement are used to determine the IT0 thickness and
sheet resistance, respectively. If the sheet resistance is
found to be excessively high (1,000~40,000C&Q. the ITOcoated cell is placed in a Technics Planar Etch II plasma
etching system and exposed to a pure-H2 plasma (200 mtorr
HZ, 20 W, 2 min). This procedure reduces the sheet resistance of the IT0 to -600-800 R/o while still maintaining
optical clarity. It is believed that this process removes
excess 02 for the 1TO. thereby creating vacancy-related
carriers (8).
After IT0 deposition, top grid electrical contacts were
patterned using an additive lift-off procedure involving
chlorobenzene (9,lO). Following photolithography, but
prior to metallization, the cells were plasma cleaned in the
sameTechnics Planar Etch II systemalready mentioned (200
mtorr Ar, 100 W, 0.5 min). This promoted the adhesion of
the the subsequent metallization. Metallization was performed in an electron-beam vacuum system and involved
successive Cr/Pd/Ag layers of 80 nm, 40 nm, and 5 pm,
respectively (11). This top grid contact is an optimum
design which, in addition to very high aspect-ratio grid lines
(5 l.rrn wide [minimum] and 5 pm thick), employed tapered
bus bars and fingers. The grid also included two relatively
large contact pads (2.0 mm x 0.8 mm each) and an interconnect between the pads, a design conforming to the requirements of semiautomatic mounting equipment currently used
in the space industry (See Figure 1,a). After metallization
lift-off in acetone and cleaning in Shipley 1112A photoresist
remover, the active cell area was defined using photolithography and HCl chemical etching. Following cell definition,
the second layer of the ARC was formed using resistively
evaporated MgF2 (nominally 75 nm thick) (6).
As a final process step before cell measurement,a postdeposition heat treatment (PDHT) at 125’C for 30 min is performed. This treatment increased the short-circuit current
density (Jsc)of the cells by -2% without adversely affecting
other device parameters. This PDHT was found to be
necessary because currently used photolithographic processesare of lower temperature (~100°C) than those used in
previous research (-12O’C).
Thus, PDHT occurs
automatically
if higher-temperature photolithographic
processing is used. Although the PDHT does add an additional step to the process, it also yields the opportunity to
isolate and study an aspect of the ITO/InP cell fabrication
that has not been previously observed. After fabrication, the
cells were characterized using quantum efficiency measurements and light and dark current-voltage measurements
using standardized methods (11).
for IJoSAT-5
Although most of the fabrication procedures used for
the cell flown on the UoSAT-5 satellite were identical to
those used for the pilot-production project, somedifferences
exist becausethe UoSAT cells were produced before the the
pilot production. The fit of these is that the substrate
material was purchased from Nippon Mining and had a
carrier concentration specified at 2.6-2.7 x 10th cm-s.
Additionally, becauseof photolithographic limitations (that
were later resolved), the top grid design and metallization
pattern was different. Most of these differences resulted
from the fact that, although generally the sameline width and
metallization stack was used, the Ag deposition thickness
had to be limited to -1 pm. This necessitated the use of a
central (tapered) bus bar, which was photolithographically
redefined after the E-beam deposition and thickened with
-20 pm of plated Au. These differences in grid
configurations are illustrated in Figure 1,b.
a) Pilot-Production Grid
b) UoSAT 5 Grid
T
1
2cm
b---2cm
+
+
2cm -d
Figure 1. Schematic plan view of grid design used on 4 cm2
ITO/InP solar cells. a) Grid design used for pilot-production
project. Modeled losses of this grid am: Resistance -2.86,
Shadowing -3.3%. yielding a total grid loss -6.1%. b)
Grid design used on the ITO/InP cell flown on UoSAT-5
satellite. Modeled lossesof this grid are: Resistance-3.4%,
Shadowing -4.2%, yielding a total grid loss of -7.6%.
RESULTS AND DISCUSSION
Pilot N
The project began with 38 l-in.2 InP substrates of the
low-doping density range (low 1016 cm-j), and 20
substrates of the higher doping range (low lOI7 cm-j). At
this time, all of the 10th cm-3 substrateshave been fabricated
into solar cells, but only one cell has been fabricated for the
1017 cm-3 material. Thus, most of the results presented
hem involve performance characteristicsof cells made on the
1016cm-3 material, although somepreliminary, yet insightful results from the cell made on the higher-doped material
will also be discussed.
Of the 38 (1016 cm-s) substrates, four were broken or
damaged during back contacting procedures, one was
broken during chemical etching, and one suffered grid
adhesion loss. Shown in Figure 2 and Figure 3 is the range
of demonstratedAM0 performancefor the remaining 32 cells
(SERI AM0 Measurements). From these data, the average
cell efficiency is determined to be 15.5% with a standard
deviation of 0.35%. The highest cell performance obtained
is 16.2% AM0 (NASA measurement). Dark I-V data
analysis indicates that the cells demonstrate near-ideal
characteristics, with a diode-ideality factor and reversesaturation current density of 1.02 and-1.1 x lo-l2 n&cm-*,
respectively.
6
6
r6
@
84
P
As mentioned previously, the PDHT was found to
increase the Jsc of the cells. However, as indicated by
quantum efficiency analysis shown in Figure 4, the effect of
the PDHT is not completely beneficial. Indeed, although
during PDHT the central and short-wavelength response is
enhanced, the long-wavelength response is noticeably
reduced. A plausible explanation for this is that the PDHT
tends to reduce the extent of type-conversion throughout the
junction region, with the overall effect of shifting the
effective depth of the sputter-formed junction nearer the
surface. This is consistent with earlier observations, in
which higher-temperature heat treatments (2OO’C) resulted
not only in increased current density but in severely reduced
Vcc (12). However, in this earlier work, a reduction in the
long-wavelength quantum efficiency (QE) was not observed,
probably because the substrates and processes used at that
time resulted in a much poorer long-wavelength response.
L
a
$4
$
d
-8
g
$2
t2
oPIou)tnlnm(D*
CCC-L,-eC”
8) EMciancy
g
n
100.0
I
I
”
(W)
b) V,
OW
12
10
a
6
4
2
0
d) Fill Fador
z
t
t!
(56)
C) J,(mA-cd)
I
0.0
300 400
I
500
I
I
600 700 800
900 1000
Wavelength (nm)
Figure 2. Histograms illustrating the AM0 (1367 Wm-2)
performance parametersof the 32 4cm2 ITO/InP cells fabricated during the pilot-production. a) Efficiency. b) Open;in-c;it voltage. c) Short-circuit current density. d) Fill
Figure 4. External quantum efficiency of 4-cm2 ITO/InP
cell. a) IT0 only. b) ITO/MgFz. c) ITO/MgFz & PDHT.
Figure 3. AM0 performance characteristics of the 32 4-cm2
ITO/InP solar cells as a function of fabrication experience.
Note that the only performance parameter that indicates a
slight progressive improvement is the V,.
Shown in Figure 5 is the light current-voltage characteristics of one of the two best 4-cm2 cells made on the 1016
cm-3 material, demonstrating an AM0 efficiency of 16.2%
(NASA measurement). By comparing these data with those
taken from the best (bulk) small-area cell produced (16.5%
AMO, 0.1 cm*, SERI measurement), one notes that the Jsc
and the FF values are nearly identical (13). This not only
suggests that the junction-formation mechanism is spatially
very uniform, but also that the grid design/metallization are
nearly optimal for this 4-cm2 cell. Only the Voc is lower (by
-10-20 mV) than that previously measured on the best
smaller ITO/InP cells made from previously used bulk
material. At present, the reason for this is not apparent.
Past observations from these ITO/InP cells have indicated a
trend of decreasing V, with increasing substrate doping
(i.e., increasing NA) (13.14). However, the recent results
from the cell fabricated on the 1017cm-3 substrate,as will be
discussed, indicate that the substratesand processesused for
this pilot-production demonstrate the opposite (but more
classical) behavior of increasing Vcc with increasing
substrate doping.
Although the efficiency spreadof the cells made on the
1016 cm-3 substrates is auite small, it should be noted that
several process-related aspects strongly affected the
measured performance of the individual cells. Perhaps the
most important of these is the amount of time during which
22-3
the cell is exposed to air between IT0 and MgF2 deposition.
Indeed, it was observed that a cell will degrade by up to -5
mV per week if it is not capped with MgF2. A possible
explanation for this is that the sputteredIT0 is believed to be
relatively porous, allowing 02 diffusion and subsequent
reaction at the emitter/IT0 interface. Here, the 02 may neutralize the passivating effect of the Ha. The evaporated
MgF2, however, may be much less porous, reducing 02
diffusion. Other parameters that were initially difficult to
control were the sheet resistance, transparency, and thickness of the ITO. As observed in earlier work, care must be
taken to maintain an optimum combination of electrical and
optical properties of the IT0 as the sputtering source erodes.
Although this can be accomplished through small adjustments in the @/II2 ratio of the sputtering ambient of the @rich IT0 layer, considerable variation is still observed.
Luckily, the effects of this problem (FF reduction) were
virtually eliminated once the post-deposition HZ plasma
exposure procedure was developed and implemented. The
final area of noted weakness in device fabrication was the
back contacting procedure. It was during this part of cell
production that the majority of cell breakageoccurred. The
underlying reason for this appears to be that, although the
two-step back metallization procedure gives a reliably lowresistance ohmic contact, it involves many stepsin which the
substrate is physically handled (e.g., during wax mounting,
chemical etching, annealing, etc.). Because most of this
handling results from the requirement to remove the Be0 that
forms during sintering, (6) it has been suggested that other
contacting procedures could be developed that would make
use of either different metals or entireIy in-vacua process
techniques.
‘.‘1”‘.,.‘..1”“I-,.‘i’--‘I’--.I.
r-
is that, instead of a reduced Voc, as was always observed in
past research, the Vet is 12 mV higher than than that from
the best of the 32 cells made from the lo16 cm-3 material
(nearly 20 mV greaterthan the averup Vet measuredfor the
32 cells). However, becausethe long-wavelength response
of the QE is reduced, the Jsc of this cell is -3% lower than
that of cells ma& on the lOI6 cme3 material. Presently,
studies are ongoing to determine if the grid design can be
modified to function without the benefits of the IT0 (lower
sheet and contact resistance). If this can be done, better
optical matching of the ARC may be possible. For example,
if a material such as ZnS replaces the ITO, modeliig studies
indicate that the Jsc of these (1017cm-3) cells would increase
to -33.8 mA-cm-z. If this can be done while maintaining
current values of FF and Voc. then the efficiency would
increase from 15.8% to 16.3% AMO. In addition, because
the ZnS is less absorbing than the ITO. modehng results also
suggest that Jsevalues up to 36.5 mAcm-2 may be possible
(assuming.4% shadow loss); this would result in a cell with
an efficiency of -17.7% AMO. Finally, because these largearea cell results indicate that the junction formation is
relatively insensitive to surface irregularities, investigations
are ongoing in collaboration with researchers at NASA
Lewis Research Center to determine what effects deliberate
surface texturing (V-Groove) may have on the junction
parameters (15). If these parameters are insensitive to
texturing, further increases in current collection may be
possible.
z
4 cm2 MgF2/lTO/lnP
AM0 1372 Watts me2
V,, 787 mV
J sc 34.09 mA cm-2
E 10.0)
Fill Factor 82.7%
g 5.OI
Efficiency 16.2%
s
00 ~~~~~.~..l~..~~..~.~~.~. . . . . ,...‘I,.,600
800
‘0.0
200
400
0”
2 15.0:
Efficiency = 15.8%
5.0 7
0
200
400
600
800
Voltage(mV)
Voltage (mV)
Figure 6. Light I-V characteristics of a 4-cm2 ITO/InP cell
produced from 101’ cm-3 substratematerial. Note that the
V, is higher than for the 10t6cm-3 material but that the .I%
is slightly reduced.
Figure 5. Light I-V characteristics of one of the best 4cm2 ITO/InP cells produced from 1016 cm-3 substrate
material.
ITO/InP Cell Flown on the UoSAT-5
The UoSAT-5 (micro)satellite was launched aboard
Ariane 4 from French Guiana on July 17. 1991. The orbit
of the satellite is 775 km, 98’ Inc. (i.e., sun synchronous,
Earth orbit). This orbit will bring the satellite through the
polar zones, and thus it will experience more radiation than a
satellite would in a similar low earth orbit placed with a mote
equatorial inclination. In addition to its primary cargo load
(which provides communications for medical teams and
disaster information services in remote areasand developing
countries), the satellite contains a single panel on which
several solar cells flight experiments are mounted. These
experiments constitute a joint project between the University
As mentioned previously, 20 substrates with a higher
doping density of l-2 x 1017cm-3 were supplied by AT&T
Microelectronics. Becauseearlier researchhad indicated that
the best cell performance has always been achieved on 1016
cm-3 material, only a single device has beenfabricated so far
on the 1017cm-3 material. Although it was thought that the
performance of this device would, as in the past, be much
poorer than that of the 1016cmm3material (due to reduced
Vet and Jsc), Figure 6 shows the surprising result that an
efficiency of 15.8% AM0 (SERI measurement) was
achieved. Perhaps the most noteworthy feature of this result
22-4
observation is the Jsc. loss that occurred after back
contacting. This loss was unexpected because, for all
ITOAnP cells ever tested, the sort of heating that was
immed during back contacting has tended to increuse the
Jsc (and reduce the VA: However, it should be not&i that;
ofthethreeITO/inPcellsmountcdbySpectrolabforthis
project (1 test coupon, 1 backup coupon, atid 1 flight
coupon), only the celI mountedon the flight coupon demonstratedanymeamable3~n2d~duringthis~.
Unlike the Jsc losses, only relatively small changes in
both Voc and FF were noted. Indeed, although it was
believed that the V, and FF on thest plasma-formed junctions might be susceptible to the mounting processes, the
acttd changes wexe found to be of the same order as those
notedontheadjacentepitaxially-grownInPcel&andinsome
instames, they wm smaller.
of Surrey (Surrey, England) and the R&al Air Force
Establishmentat Famborough (Hampshire, England). On a
small portion of tl& panel (a coupon), NASA has placed
three solar &Is, one of which is’ an ITO/InP 4-cm2 cell
provided by the authors’ group. Along witb ddsSERI cell,
thetwoothcr4-cmzceIlsconsistofancpitaxialIy-grownIllP
homojunction cell, provided by Spite Corporation, and a
GaAs CLBFI’ cdl, provided by Kqin Corporation (See
Figure 7). These three cells are CDnnectedsothat tekme&y
datacalllYereceivedfnnneachceIlthrougilouttheexpected
2- to 3-year working lifetime of the satellite. The total
radiation dose that will be expezienced by the solar cells is
being monitored by a separate, on-board experiment, that
uses seven specially-designed RADFETs (field effect
transistors that are sensitive to radiation). The solar cell
experiment was commissioned on July 23, 1991, and
indications are that early experimmtal (baseline) data have
been successfolly .received. However, at the time of &is
writing, the authors have not received additional post-launch
data. Nevertheless, a considerable amount of pnlaunch data
has been assimilated that lends insight into the ability of the
ITOAnP cell technology to withstand the processes
necessaryfor the assembly of space-flight experiments.
CONCLUSIONS
AND FUTURE STUDIES
This project has demonstrated that the sputtering
orocess
used to form small-area ITO/InP solar cells can
.
readily be scaled to produce large-area (4-cm2) devices.
These large-area cells demonstrate nearly identical performance to simk small-area cells, suggesting that the spatial
independence of the junction-formation mechanism may be
exploited further to productions involving larger batches.
These results also suggest that this method of junction fabrication is not as sensitive to the same predeposition surface
irregularities that tend to have devasiatingeffects in other
solar cell technoloeies. The highest resultant solar cell
efficiency from thg 32 cells pr&uced was 16.2% AhO
(NASA measurement), which is comparable to the highest
efficiency reported from another production method.
Additionally, since the sputter-deposition technology can be
configured for in-line (rather than only batch) production
modes, this process may possess additional economic
advantages.
The pre-flight data from the UoSAT-5 project indicate
‘that relatively small parameter variations were incurred
during cell mounting procedures. This is encouraging
because the experiment reoresentsone of the few times that
ITO/InP cells ‘have been ‘subjected to the rigors of spacequtiified cell mounting techniques (See also Reference 16
concerning ITO/lnP cells launched on the LPPS-III exuerimerit). Th;: preliminary results also indicate that there’ is a
good chance that the flight experiment will provide very
useful data concerning the radiation hardnessof this particular InP technology.
Future research on this type of solar cell will focus on
ways to increase the Voc, primarily through studies of the
junction-formation process. To this end, solar cell junctions
have been formed by non-deposition plasma exposure techniques using pure& or pore-H2 gas as the plasma species.
Preliminary analysis of these experiments has indicated that
exposing p- InP to either Ar or H2 plasmas results in junction parameters similar to those observed after an IT0
deposition. This is surprising because, in the case of an
ITO-deposition, the addition of HZ (to an otherwise predominantly Ar plasma) was found to improve the junction
characteristics. Presently, a similar pure-gas plasma study is
being undertaken that will use pure oxygen as the gas
source. Finally, since the above results indicate that the
junction formation is relatively insensitive to surface
-irregularities, investigations are ongoing to determine what
effects deliberate surface rexturine. such as V-Groovine.
may have on the junction pammete; If these parameters&
insensitive to texturing, increases in current collection may
be possible.
Figure 7. Plan view of flight coupon used for the UoSAT-5
experiment showing three 4-cm2 solar cells (note that
UoSAT-F became UoSAT-5 after a successful launch).
The ITOAnP cell was mounted onto the coupon at
Spectrolab, Inc. (Sylmar, Califqnia) in December of 1990.
Before this mounting, the cell demonstraW an efficiency of
15.7% (AMO, NASA measurement) which was essentially
the same value as measured at SERI immediately after ceil
production, attesting to the stability of the-cell.
At
Spectrolab, light I-V data was taken after each aspect of the
cell mounting, indicating how the cells responded to the
individual mounting processes. The resulting data for the
ITO/InP cell are listed in Table 1. After the module fabrication processes were completed, the NASA remeasured efficiency of this cell was 13.7% AMO.
From Table 1, it can be seen that this reduction is primarily due reduction in Jsc. Additionally, it can be seen that
the majority of this Jsc loss occurred after filtering, and is
believed to be caused by an increase in reflection. This increased reflection was anticipated, and is due to the optical
mismatch between the top layer of the ARC (MgF2) and the
DC 93-500 adhesive/(X$X cover glass filtering. When
future cells are made for flight testing, the ARC will be
adjusted to reduce or eliminate this effect. A more curious
225
5.
ACKNOWLEDGEMENTS
.~
The authors wish to thank Keith Emery and Paul Phelps of
NREL for assistance with efficiency measurements, Brian
Keyes of NREL for photoluminescence measurements,
Brian Smith of Spectrolab for providing detailed information
on the UoSAT-5 cell mounting and performance data, and
Mike Piszczor of NASA Lewis Research Center for background information and photographs of the UoSAT flight
project. This work was supported by NASA Lewis
Research Center under Interagency Order No. C-3000-K
and by the U.S. Department of Energy under Contract No.
DE-ACO2-83CH10093.
.,
6.
7.
8.
9.
REFERENCES
1.
2.
3.
4.
10.
C.J. Keavney, V.E. Haven, and S.M. Vernon, Proc.
21st IEEE Photovoltaics
Specialists Conf.,
Kissimmee, FL, May 21-25,199O (IEEE, New York,
1990) p. 141.
M. Yamaguchi, T. Hayashi, A. Ushirokawa, Y.
Takahashi, M. Koubata, M. Hashimoto, H. Okazaki,
T. Takamoto, M. Ura, M. Ohmori. S. Ikegami, H.
Arai, and T. Orii, Proc. 21st IEEE Photovoltaics
Specialists Conf., Kissimmee, FL, May 21-25, 1990
(IEEE, New York, 1990) p. 1198.
I. Weinberg, C.K. Swartz, R.E. Hart, Jr., and T.J.
Coutts, Proc. 20th IEEE Photovoltaic Specialists
Conf., Las Vegas, NV, September 26-30, 1988
(IEEE, New York, 1988) p. 893.
T.J. Coutts, X. Li, M.W. Wanlass, K.A. Emery and
T.A. Gessert, Proc. 20th IEEE Photovoltaics
Specialists Conf., Las Vegas, NV, September 26-30,
1988 (IEEE, New York, 1988) p. 660.
11.
12.
13.
T.A. Gessert, X. Li, M.W. Wan&s,
and T.J. Cows,
Proc. Second Int. Conf. on Indium Phosphide and
Related Mat., Denver, CO, April 23-25, 1990 (IEEE
Cat. No. 9OCH2895, IEEE, New York) p. 260.
T.A. Gessert, X. Li, T.J. Coutts, M.W. Wanlass, and
A.B. Franz,. Proc. First Int. Conf. on Indium
Phosphide and Related Mat. for Adv.‘El&onic
and
C&$xl Devicyts. Notyan, OK, March ?O-22, 1989,
PmceedmgsVol. 1144 (SPIE, Bellmghaq WA,
1989) p. 476.
T.A. (dessert, X. Li, and T.J. Coutts, &t&&&,
30
(1991) p.459.
T.A. Gkssert, D.L. Williamson, T.J. Coutts, A.J.
Nelson, KM. Jones, R.G. Dhcrc, H. Aharoni, and P.
Zurchq, JVST-A, A 5 (4) (1987) p. 1314.
M. Hatzakis, B.J. Canavello, and J.M. Shaw, IBM J.
Res. Develop., 24 (4) (1980) p. 452.
T.A. Gessert and T.J. Coutts, MRS Proc. 181 (MRS,
Pittsburgh, PA, 1990) p. 301.
“Standard Test Methods for Electrical Performance of
Non-Concentrator Photovoitaic Cells Using Reference
Cells,” ASTM StandardE948.
T.J. Coutts, X. Wu, T. A. Gessert, and X. Li, J. Vat.
Sci. Technol. A. 6 (3) (1988) p. 1722.
T.A. Gessert, X. Li, M.W. Wanlass, A.J. Nelson,
and T.J. Coutts, J. Vat. Sci. Technol. A, 8 (3) (1990)
D.
1912.
14. ?.J. Coutts and S. Naseem, Appl. Phys. Lett., 46 (2)
(1985) D. 164.
15. 3. Ba&y, N. Fatemi, and J. La&s, Published at this
conference.
16. N.M. Pearshall, C. Goodbody, N. Robson, I. Forbes,
and R. Hill, 21st IEEE Photovoltaics Specialists
Conf., Kissimmee. FL, May 21-25, 1990 (IEEE, New
York, 1990) p. 1172.
Table
1
Parametersfor the 4-cm2 lTO/InP solar cell used on the UoSat-5 flight coupon as it progressed through
the stages of .mounting. Data taken at Spectrolab, Inc. by calibrating the simulator to the as-received
NASA Jscmeasurement.
*Fill factor calculatedfrom available data.
Qrom Spectmlabcommentthat cell showed no degradation after top contact ultrasonic bonding.
**NASA remeasuredcell efficiency after mounting 13.7% AMO.
22-6
MODELED PERFORMANCE OF MONOLITHIC, 3-TERMINAL InP/Ca&~sSAs
CONCENTRATOR SOLAR CELLSAS A
FUNCTION OF TEMPERATURE AND CONCENTRATION RATIO
C.R. Osterwald,
M.W. Wanlass, J.S. Ward, B.M. Keyes, K.A. Emery, and T.J. Coutts
National Renewable Energy Laboratory
(Formerly the Solar Energy Research institute)
Golden, Colorado USA
ABSTRACT
InP m
Using measured device parameters from a monolithic,
3-terminal InP/Galn*
(0.75 ev) tandem concentrator solar
cell, a numerical model has been constructed that calculates
efficiency as a function of temperature and concentration
ratio. The device measurements indicate that the series
resistance in the InP top cell severely limits the maximum
efficiency at high concentration
ratios. Results from the
model in which a single-junction
InP concentrator solar cell
that has a lower series resistance by a factor of six was
substituted for the top cell show that peak 20°C air mass zero
&MO) efficiencies should approach 30% at concentration
ratios greater than 100. At 8O”c, this tandem cell should
exceed 24% AM0 efficiency.
-
Entech cover
Top contact
AR coating
w
Top cell
Middle contact
Stop-etch layer
Conduction layer
Bottom cell
Buffer layer
Figure 1
INTRODUCTION
3-terminal
A monolithic, 3-terminal InP/GalnAs (0.75 eV) tandem
solar cell exceeding 30% efficiency under concentrated direct illumination,
and 27% under concentrated extraterrestrial (AM01 illumination, has recently been reported (1). Asthe
primary application of this cell design is expected to be in
high-efficiency
space power systems, performance at elevated temperatures and high concentration ratios is of particular importance.
The efficiency peaked in the relatively
low concentration ratio range of 20-30, and it was believed
that the drop at higher concentration
ratios was due to
excessive series resistance in the InP top ceil. The primary
purpose of this work is toverifytfte cause of the&idency
drop
and provide pe&nmance
projections of the tar~dem cell, a
cross-section of which appears in Figure 1.
MODEL CONSTRUCTION
The performance project*ions reported here use results
from a numerical model constructed from quantum efficiency, dark current-v&age
O-V), and photoluminescence
measurements of the tandem cell asa function of temperature.
Asimpleequivalentcircuitofasolarcell,
consistingofadiode
in parallel with a light currentgenerator
and both in series with
Cross section
tandem solar cell.
of InP/GalnAs
monolithic,
a resistor, allowed the illuminated
I-V parameters to be
calculated as the temperature f7I and concentration ratio (cl
were varied. Output from the model was then compared
iteratively with available illuminated I-V data to obtain an
estimate of the series resistance for both the top and bottom
cells.
Ouantum
Efficiencv and Photoluminescence
Absolute quantum efficiency (QE) measurements at
25°C and 80°C (Figure 2) were obtained to allow calculation
of the light-generated short-circuit current density, Ib from
integration of the AM0 spectral irradiance. Because the QE
could not be measured at high concentration ratios, the model
assumes that the quantum efficiency does not vary as C is
increased. Using this assumption, the calculated h is then
multiplied by C to obtain the light-generated current at any
concentration ratio.
Figure 2 shows that, within the resolution limits of the
measurements, only the band gap cutoffs of the InP and
GalnAs cells change with temperature, while the shape and
level of the quantum efficiency remains unchanged.
The
23-1
magnitudes of the band gap temperature coefficients were
determined
from photoluminescence
measurements
of the InP and GalnAs at a series of temperatures from
2OT to lOO”c, as seen in Figures 3 and 4. A value of
-0.344 meV”C-r was obtained for the InP band gap temperature coefficient and a value of -0.238 meV“CY for the GaInAs.
25
,. 0.8
E
.$
E
w 0.6
E
E
5 0.4
7%
E
i?
w 0.2
1.20
1.25
1.35
1.30
1.40
1.45
1.50
Photon Energy (eV)
Photoluminescence
versus photon energy and
Figure 4
temperature of InP film grown on InP by MOVPE. Excitation
was 647 nm Kr laser.
0.2
0.4
0.6
0.8
1 .O
Wavelength
Figure 2
of InP/GalnAs
8O’T (dashed
concentrator
1.2
1.4
1.6
1.8
fpm)
External quantum efficiencyversus wavelength
top and bottom cells at 25°C (solid line) and
line); quantum efficiency of single-junction InP
cell (*I.
10’
c”
100
6
5
lo-9
0.4
0.70
0.75
0.80
0.85
0.90
0.95
1.00
0.6
0.7
0.8
0.9
1.0
1.1
1.2
Voltage 0
Photon Energy fev)
Photoluminescence
versus photon energy and
Figure 3
temperature of GaaA71no.s3As film grown on InP by MOVPE.
Excitation was 647 nm Kr laser.
0.5
Log dark current versus voltage of InP/GalnAs
Figure 5
tandem top cell (solid lines) and single-junction InP concentrator cell (dashed lines) at several temperatures.
23-2
Dark Current-Voltaee
Comoarison
Dark I-V measurements of the InP top cell over a 60°C
temperature range are presented in the lower series of curves
of Figure 5. A numerical fit, using the ideal diode equation,
to each of these curves gave the saturation current density 1,
the diode quality factor n, and an estimate of the series
resistance R,. The temperature dependence of these parameters allowed the diode I-V functionality to be calculated at
any temperature.
In order to obtain model results as close to the actual
tandem cell I-V data as possible, the pm-exponential terms of
the diode equations were adjusted to match the open-circuit
voltages of the top and bottom cells measured at 25°C. These
adjustmentsweresmall
andamountedtoafactorof
1.5 forthe
InP cell and 1 .l for the GalnAs cell. The quantum efficiencies
were scaled by small amounts, about l-3%, so the shortcircuit currents would match the measured one-sun values.
For the 0.75 eV GaInAs bottom cell, similar measurements are shown in Figure 6. These curves show nearly ideal
diffusion current characteristics with n = 1.03 until about
0.4 V, at which a transition to high injection begins. Also, no
bending due to series resistance can be seen in these curves,
in sharp contrast to the InP top cell. Because of the onset of
high injection, the single exponential diode model could not
be used to describe the I-V relationship. The dark I-V curves
were instead fit to a four-parameter function that modeled the
observed behavior (2):
Varying the temperature and concentration ratio in the
model produced a series of efficiency curves for the top and
bottom cells. Initial comparison of the curves with measured
efficiency versus C at a single temperature showed that the
series resistance obtained from dark I-V was too low. This fact
was evident because the roll-off of efficiency with increasing
concentration ratio did not match measured data. The series
resistances were therefore adjusted until the roll-offs matched
the measured results. A value of 39 mQ ems was used for the
InP top cell, and a value of 7.7 mQ crr$ for the GalnAs cell.
The low series resistance for the bottom cell is attributed to the
low-resistance InP conduction layer immediately above the
CalnAs emitter. Figure 7 illustrates the modeled AM0 efficiency versus C and T, plotted along with the measured
efficiency at 25°C after the series resistances were adjusted.
Table 1 presents a comparison of modeled t-V parameters
with measured parameters at one-sun and 25°C and 80°C
under direct illumination.
After fitting the curves, functions describing the temperature dependence of the four parameters were used with
Equation 1 to provide the diode characteristics of the bottom
cell for the model.
with Measured I-V Data
1
InP TOD Cell
a
5
GalnAs Bottom Cell
100
Concentration
Ratio
Figure 7
Modeled AM0 solar cell conversion efficiency
versus temperature and log concentration ratio of InP/GalnAs
3-terminal tandem top and bottom cells after series resistance
adjusted to match high concentration
roll-off of measured
25°C conversion efficiency (01.
Figure 6
Log dark current versus voltage and temperature of lnP/GalnAs tandem bottom cell.
I
10
23-3
PERFORMANCE
PROJECTIONS
Improved Top Cell
Addition of the top and bottom cell efficiencies gives the
tandem cell efficiency, which is plotted as the lower series of
curves in Figure 8. The efficiency at lower temperatures peaks
at about 30 suns, while at higher T, the peak shifts to about 50
suns.
Following analysis and modeling of the InP/GalnAs
tandem cell, a single-junction
InP concentrator cell was
modeled in a similar fashion to determine if this design might
be more suitable as the top cell of the tandem. The singlejunction cell has a thicker emitter, 31 vs. 24 nm, and because
there is no middle contact, the grid finger spacing is half the
tandem top cell spacing (both devices use the same prismatic
cover design). Figure 2 shows thatthe single-junction cell has
a reduced blueresponseduetoitsthickeremitter,andthedark
I-V for this device, included in Figure 5, indicates a much
reduced roll-off because of series resistance and an improved
diode quality factor.
Fitting the series resistance to the
measured efficiencyvs. C gave avalue of 6.0 rnQ cm2, which
is a factor of six smaller than the top cell.
Existing Top Cell
5
0
Concentration
A study of the power loss due to the grid finger spacing
showed the tandem top cell loss to be three times that of the
single cell (3). The characteristics of the single-junction cell
were then inserted into the model in place of the InP top cell,
and the results appear in Figure 8. The initial efficiency is
higher than the existing tandem cell because of the improved
top cell, and at 2O“C, the AM0 efficiency peaks at nearly 30%
for a concentration
ratio of 130 suns. At 80X, the peak
efficiency drops to 24.5%, while the peak position shifts to
nearly 200 suns.
Table 1
Comparison of measured InP/GatnAs tandem
cell illuminated
I-V parameters with results of numerical
model.
Pmax
I
T
Voc
0
(mAk-2)
(mW)
PC)
InP top cell
Measured
Model
Measured
Model
25
25
80
80
0.864
0.865
0.751
0.726
27.47
27.47
27.86
27.83
1.298
1.339
1.098
1.091
GalnAs bottom
Measured
Model
Measured
Model
cell
25
25
80
80
0.333
0.333
0.233
0.208
22.10
22.10
22.69
22.73
0.347
0.347
0.216
0.184
,
-
Ratio
Figure 8
Modeled AM0 conversion efficiency versus
temperature and log concentration
ratio of existing InP/
GalnAs tandem cell (solid lines) and improved cell where
single-junction cell was substituted for top cell (dashed lines).
cell has been performed.
The results have been used to
construct a numerical model that can predict tandem cell
efficiency as a function of temperature and concentration
ratio. This model predicted an improvement of the efficiency
to nearly 30% (100 suns AMO, 2O“CI if the series resistance of
the InP top cell can be reduced by a factor of six.
ACKNOWLEDGEMENTS
Support for this work was provided by the U.S. Department of Energy under contract no. DE-AC02-83HC10093,
the
Naval Research Laboratory under interagency order RU-llW70-AD, and the NASA Lewis Research Center under
interagency order C-300005-K.
..
REFERENCES
M.W. Wanlass, J.S. Ward, K.A. Emery,T.A. Gessert,C.R.
1.
Osterwald, and T.J. Coutts, “High-Performance
Concentrator
TandemSolarCellsBasedon
Infrared-SensitiveBottomCells,”
Solar Cells, Vol. 30, pp. 363-371, 1991.
R.S. Muller,T.l. Kamis, Device Electronicsfor Integrated
2.
Circuits, 2nd Ed., p. 324, John Wiley & Sons, New York, NY,
1986.
SUMMARY
A comprehensive
3-terminal InP/GalnAs
3.
JS. Ward, M.W. Wanlass, T.J. Coutts, K.A. Emery, and
C.R. Osterwald, “InP Concentrator Solar Cells,” Proc. 22nd
IEEE PV Spec. Conf., Las Vegas, NV, 1991.
measurementstudy ofthe monolithic,
(0.75 ev) tandem concentrator solar
23-4
K. A. Bertness,B. T. Cavicchi*,SarahR. Kurtz,
J. M. Olson,A. E. Kibbler, andC. Kramer
National RenewableEnergyLaboratory(formerlySolarEnergyResearchInstitute),Golden,Colorado
*SpcctrolabInc., Sylmar,California
ABSTRACT
Effects of l-MeV electron irradiation damage are
mported for thirteen GaAs n-p single-&unctionsolar cells
with basedoping levels from 2 x 10 to 3 x 10” cma3.
The short circuit current densities before and after
irradiation to a fluence of 1015cm-2 show a downward
trend as a function of base doping, and the extent of
radiation damagealso increasesas basedoping increases.
Radiation damage coefficients extracted through
modeling of the spectralresponsecurvesare seento vary
from 2.5 x lo-* to 1 x lo-’ over the rangeof p-type base
doping covered. The estimatedAM0 efficienciesof the
cells after irradiation, however, are nearly constantas a
function of basedoping owing to simultaneousincreases
in the open circuit voltagesand fill factors.
doesindeedincreasewith doping. Valuesfor the damage
coefficient, however, are not in quantitative agreement
with previouswork on GaAs. Other effectsof radiation,
most notably the changesin cell open circuit voltages
(V,), complicatethe optimizationof end-of-life solarcell
efficiency, which is found to depend very little on base
dopingovera broadrange.
SOLAR CELL GROWTH AND PROCESSING
All the cells used in this study were grown using
atmospheric-pressure
organometallicvapor phaseepitaxy
(OMVPE), as describedelsewhere (5). A typical cell
structure. is illustrated in Figure 1. Layer thicknessesare
accurateto HO%. The emitter is dopedwith seleniumto
about 2 x 1018 cme3,while the zinc doping in the base
isvaried from 2 x 1016 to 3 x 10” cme3. Processing
consistsof electrochemicallydepositing gold front grid
INTRODUCTION
Radiation damage in solar cells occurs primarily
through the creation of deep trapswithin the energyband
gap which reducethe minority canier lifetimesand hence
the minority carrier diffusion lengths. The degmdationis
usually characterizedby a radiationdamagecoefficient,K,
tWined by the following equation:
lc4 = l/L2 - l/L*2 ,
.
where Q, is the radiation fluence, and L and Lo are the
post- and pre- irradiation minority carrier diffusion
lengths, respectively. The resulting degreeof efficiency
degradationdependsbothonthesolarcellmate&landits
structure (11, with GaAs falling between silicon and
indium phosphide due to its combination of a strong
absorption coefficient and moderately high defect
introduction rate. Theoretical and experimentalstudies
(2-4) of GaAs imhcatethat the damagecoefficientitself is
an increasingfunction of mate&l doping, possiblydue to
the active traps being derived from dopant-defect
complexes. In this study we have looked at radiation
damagein severalsolar cells with different basedoping
levels,and found that radiation-induceddegradationof the
short circuit current density (Jr& the device parameter
most closely tied to the minority carrier diffusion length,
24-1
FRONT CONTACT
\-
I=(
250 A
I
(1)
I
I ’
0.1 pm
n-Gao.sIno.s P
n -GaAs
ARCOATING
,
WINDOW
EMITI-ER
BACK CONTACT
Figure 1. Solarcell growth structure.
I
I
I
Table 1.
Solarcell materialsparameters.
lines and solid back contacts,etchinga mesathat defines
the cell areasto 0.252 cm2 or 0.280 cm2, dependingon
the busbar size, and then evaporating a double
antireflection coating of zinc sulfide and magnesium
fluoride. Energy conversion efficiencies and device
parameterswere measuredusing a solar simulator with
the short circuit current densitiescalibrated against the
responseof a silicon reference cell, which is in turn
calibrated against a GaAs cell at AMl.5 global
illumination. These efficiencies were converted to
estimatedAM0 efficiencies based on changesin short
circuit current in a GaAs cell calibrated under both
illumination conditions.
Thirteen devices taken from five substrateswere
irradiated with l-MeV electronsat a fluence of 1 x 1015
cme2. As can be seenin Table 1, the cells from different
substratesvaried primarily in the basedoping used,but
variationsin substrateorientationare alsonoted. Material
grown on 511B orientedsubstrateshashigherzinc doping
levelsthan materialgrown on the standardsubstrateswith
an orientation of 2’ off (100) toward the (110). This
effect can be seen by comparing cells from substrates
317-2 and 317-B, which were grown at the sametime on
hvo adjacentsubstrates.Finally, cells from substrate385
had a O.l-pm-thick GaInP2 emitter with an AlInP2
window layer in place of the GaAs emitter and GaInP2
window found in the other cells. This heterojunction
structure generally leads to slightly higher open circuit
voltages but also had a poorer blue responsethan the
standardcell. It had been included becauseits radiation
perfomanceis similar to that of the homojunctionGaAs
cells.
RESULTS
. .
. .
mud Current and Ra&@on Ds
As previouslymentioned,the mostsignificanteffect
of reducedminority carrier diffusion length is a reduction
in the short circuit current density. In Figure 2(a) we see
that Jsc valuesdecreasemonotonically with basedoping
both before and after irradiation, implying that in both
casesthe minority carrier diffusion length decreaseswith
base doping. Furthermore, the relative amount of
radiatrondegradationin Jsc increaseswith basedoping,
as can be seen in the ratios of post-/pre- irradiation Jsc
values in Figure 2(b). This result is in qualitative
agreementwith earlier work (3.4) on radiation damagein
G&S.
In order to be more quantitative, the spectral
responsecurvesof the cells were modeledusing analytic
expressions(6) for externalquantumefficiency basedon
materialsparametersincluding minority carrier diffusion
lengths, interface recombination velocities, absorption
coefficients, doping, and carrier diffusion constants.
Examplesof modeling with parametersthat gave good
agreementwith the experimentare givenin Figure3. ‘Ihe
majorsystematicdeviationsarisemostlyfrom insufficient
data for absorptioncoefficientsand the assumptionthat
reflectivityis uniform asa functionof wavelength,despite
I
0.96
,
I
I
I
.
I
I
I
I
I
I
”
‘(b)
18
0.80
0.0
I
I
me
I
I
I
2.0
3.0
basedoping(1017cma)
1.0
Figure2. (a) Estimatedshort circuit current densitiesfor
GaAs solar cells under AM0 illumination before (closed
symbols) and after (open symbols) 1-MeV electron
irradiation,and (b) ratio of post- to pre-irradiationvalues.
Heterojunction cells (sample 385) are indicated by
Sq-.
24-2
a known sharpincreasein reflectivity seenabove2.75 eV
for similar cells. Most of the quantumefficiency curves
could be fit well with a small recombinationvelocity of
2 x 104 cm/s at the window-emitter interface, but the
broad negative slope for samples 317-B and 385 (not
shown) could only be reproduced by increasing the
recombination velocity to 1 x lo6 cm/s and 5.5 x 105
cm/s,respectively.
The best estimatesand acceptablerange of damage
coefficients plotted in Figure 4 were derived by varying
the emitter and base radiation damagecoefficients and
calculating external quantum efficiency curves. Also in
Figure4 are data points from two earlier reports on
radiation damagein GaAs. While all three setsof data
indicate that the damage coefficient for p-type GaAs
increasesas initial doping increases,the new dataindicate
higher damage coefficients than previously measured.
The error bars are taken from the set of damage
coefficients that produce a curve within 3%-4% of the
experimentaldata throughout the whole spectral range
exceptfor the reflectivity inducedroll-off in the blue end.
In generalthe modelingshouldbe accurateto this level if
the layersare thicker than or comparableto the diffusion
lengths,which is true for the post-irradiationspectra(L=
l.O-1.9 pm) and nearly true for pre-irradiation spectra
(&=3.2-8 pm).
One might think that the difficulty in extractingthe
longer pm-irradiation diffusion lengths from quantum
efficiencycurveswould strongly affect the determination
of the damagecoefficient, but the technique becomes
1.0 0.8
ii
s
Q 0.6
c
;
9
0.4
0.2
(d)‘
-
-
0.8
8
E
$ 0.6
92
8
Q
0.4
0.2
5
2.0
2.5
3.0
energy (eV)
1.5
2.0
2.5
energy (eV)
Figure 3. Spectralresponsecurves(solid lines) and externalquantumefficiencymodelingresults(broken lines) for five
solar cel&sbefore and after irradiation. The modelingresultsare broken down into contributionsfrom the emitter,base
and depletionregions.
24-3
0
Ref. 3
00
0
/
0
overall device efficiency (8). The insensitivity of
efficiency to base doping arisesfrom the increasingfill
factorsand the lower radiationdamageto the open circuit
voltageas basedoping is increased,with the exceptionof
sample3 17-B whose substantialdeviationsin efficiency
are discussed in the next subsection. The relative
degradation in the fill factor (Figure 6[b]) induced by
radiationis constantas a function of basedoping for the
threelowest values. This constancyindicatesan absence
of radiation-induced loss of free carriers, which is
confirmedto within 10% of the origin doping levels by
frequency-dependent
capacitance-voltagemeasurements.
Previousstudies(3) suggestthat free carrierlosswould be
about 1 x 1015cme3for this radiation fluence level, and
thusbelow the levelof detectionin this study.
As shownin Figure 6, a noticeablefill factor decline
alsooccursin the cells with the lowest basedoping. The
resistanceof the baselayer at its nominal doping is not
large enough to affect the fill factor. Specifically, the
estimatedvoltage drop at Jsc for a base layer doped
uniformly to 2 x 1016cm-3 is 7 uV. Thesecells havea
significantseriesresistance,which is probably an artifact
due to the presenceof a high resistivity layer formed
because of failure of the dopant massflow controller to
work consistentlyat the extremelow end of its rangeor to
a seriesresistanceintroducedduring the processing. The
estimatedAM0 efficiency of thesecells would havebeen
1
0
PG-
n Ref. 3 0 :
+ Ref. 4 A HJ cell
Ref. 3 @eorY)
tr
_. I
I
I
I
1
10-g 1
lo’*
1o19
10% 10”
base doping ( cmS3)
This work
Figure 4. Radiationdamagecoefficientsfor GaAs.
moreaccurateand lessdependenton initial valuesof Lo as
the damage increases because of the inverse square
dependencein Eq. (1 . Even for the worst case,Lom2is
less than 20% of L- 2’, so the uncertainty in the damage
coefficientsdue to pre-irradiationdiffusion lengthsis less
than the error bars shown in Figure 4. The damage
coefficient differencesmay arise from nonlinearitiesin K
asa function of fluence,becausethe valuesin Reference3
aremeasuredfor fluencesfrom 2 x 1013to 2 x 1014cme2,
and from 3.2 x 1015to 1 x 1016cme2in Reference4, as
comparedwith this work at 1 x 1015cme2. Electronflux
variations and unintentional heating during irradiation
may haveaffectedthe damagecoefficients. The material
growth and diffusion length measurementtechniquesalso
vary in the three studies, with Reference3 apparently
using liquid phase epitaxy material and electron beam
induced current (EBIC) to measure minority carrier
diffusion lengths, while Reference4 uses a quantum
efficiency modeling (7) of cells grown with chloridetransportchemicalvapordeposition.
. . Volm
Efficiencv.Fill l&@r. and GpenCltcult
Despite the distinct decline in short circuit current,
the estimatedenergy conversionefficienciesunder AM0
illumination, both before and after irradiation, are almost
constant as a function of base doping until the highest
doping, as seen in Figure 5. Cells with a lower base
doping may still be desirable for tandemcell designs,
however, where current matching strongly affects the
16
8
0.5
0.0
I
0
I
I
J
1.0
2.0
3.0
base doping (10’ ’ cm3)
I
I
Figure 5. (a) EstimatedAM0 efficiency of GaAs solar
cells before (closed symbols) and after (open symbols)
I-MeV electron irradiation, and (b) ratio of post- to preirradiationvalues. Heterojunctioncells (sample385) are
indicatedby squares.
24-4
0.75t
0.0
,
I
1.0
,
I
2.0
,
recombinationcurrent dark current decreasesprimarily
becausethe volumeof the depletionregion getssmalleras
basedoping increases.From Equation (2) it can be seen
that Voc will thereforetend to increaseas a function of
base doping due to the decreasing dark saturation
currents.Acting in the oppositedirection to reduceV,, is
the doping dependenceof the short circuit current, which
also decreasesas base doping increasesdue to shorter
minority carrier diffusion lengths. The doping
dependenceof the dark current must have changed
significantly as a result of irradiation, since the observed
changesin the base doping dependenceof JSc cannot
explain the shift in the maximumof Voc vs. doping to a
higher value of basedoping after irradiation. This anticorrelationbetweenradiation effectsin Voc andJSc as a
function of basedoping is consistentwith previous work
(79), althoughin thosestudiesthe small numberof cells
and simultaneousvariation of other growth parameters
makesthis effect lessobvious.
Dark Currenta&Q&al w of m
Basem
The cells from the sample with the highest base
doping, 317-B, have undergone an abrupt and large
degradationof the fill factor and open circuit voltage
during irradiation, although the short circuit current
densities are in logical progression with samples of
l*.j
3.0
base doping (10’ 7 cma)
Figure 6. (a) Fill factors for GaAs solar cells before
(closed symbols) and after (open symbols) 1-MeV
electron irradiation, and (b) ratio of post- to preirradiation values. Heterojunctioncells (sample385) are
indicatedby squares.
around 20.7% before in&&ion if they had the samefill
bctor as morehighly dopedcells.This valueis essentially
equal to the efficiency of the cells with the next highest
basedoping.
The opencircuit voltagesbeforeandafter irradiation
are displayedas a function of basedoping in Figure7(a),
with the ratio of post-&e-irradiation values given in
Figure 7(b). As indicated above, the relative radiationinduced degradation of V,, actually decreasesas a
function of basedoping for all but sample317-B, leading
to a degradationof efficiency that is almostconstantwith
base doping despite increasing short circuit current
degradation. The dependenceof Voc on basedoping is a
competition between two effects, namely the
simultaneousdecreasesin Jsc and in the dark currentas a
function of base doping. In general,near any particular
voltage the cell current can be modeled with a single
ideality factor n, so the open circuit voltage could be
expressedas
Voc = (nkTlq) ln(J&J& - 1),
(2)
where k is Boltzmann’s constant, T is the absolute
temperature,q is the chargeof an electron,andJon is the
diodesaturationcurrent for the dark current.
The diffusion dark currentdecreasesas a functionof
base doping because the diffusion constant and the
equilibrium concentration of minority carriers at the
depletion region edge become smaller, while
1.1
E ‘.O
::
’ 0.9
0
0.8
I
0.92
I
e!
E
m
CD
a
00
I
.
I
I
I
I
I
0
I
I
l(b)
a
-
"B 0.88
0
::
> 0.84 -.:
'L
0,
a
base doping (10" cm3)
Figure 7. (a) Open circuit voltage of GaAs solar cells
before(closedsymbols)and after (open symbols) I-MeV
electron irradiation, and (b) ratio of post- to preirradiation values. Heterojunctioncells (sample385) are
indicatedby squares.
24-5
lower basedoping. The dark 1-V curvesfor representative
cells after irradiation,displayedin Figure 8, showthat the
change in shape of the I-V curves comes from the
prevalenceof a new dark currentmechanismwhich is not
of the same functional form as either an ohmic shunt
resistance,a recombinationcurrent, or a diffusion (shunt
diode)current. This excesscurrent,seenas a tilted hump
in InO-V from about 0.45 V to 0.75 V, was not present
before irradiation. Dark I-V measurementsat liquid
nitrogen temperatureshow that the current persistsbut
decreasesaboutone orderof magnitude.
The small temperaturedependenceof the excess
current suggests a conduction mechanism in which
electronsfrom the degeneratelydopedemittertunnelinto
radiation-induced trap states in the depletion region,
where they then recombine with holes from the p-type
region. Higher basedoping enhancesthis mechanismby
decreasingthe depletion width and hence the distance
through which the electronsmust tunnel. The degreeto
which the excesscurrent appearsis correlatedto the base
doping in Figure 8; the apparentlysuddenonsetof large
efficiency loss occurs when the new current mechanism
overtakes the recombination current. GaAs cells with
high basedoping have previously been seento undergo
unusually large degradationin fill factor after irradiation
(9). The excess current does not appear in the
heterojunctioncells, perhapsbecausethe conductionband
offset at the p-n junction increasesthe electrontunneling
barrier height.
CONCLUSIONS
Radiation damage to the short circuit current in
GaAs n-p solarcells is seento increasein thep-type base
as the doping level of the baseis increased.The minority
carrierdiffusion length damagecoefficientsderivedfrom
modelingof externalquantumefficiency are about 4-10
times greater than values reported in earlier work, but
agreequalitatively with trends as a function of material
doping level. Other radiation-inducedeffectsinclude an
increasein radiationresistanceof the open circuit voltage
as basedoping increases,which effectively counteracts
the short circuit current degradation so as to produce
efficienciesthat are almostconstantas a function of base
doping. An abrupt degradationin fill factor and open
circuit voltage for cells with the highest base doping is
alsoobserved,and their uniquedark I-V featuressuggest
that the degradationonsetis causedby tunneling-assisted
recombinationthrough radiation-induceddefect statesin
thedepletionregion.
ACKNOWLEDGMENTS
We wish to thank D. J. Friedmanand H. Branz for
critical readingof the manuscriptand B. E. Anspaughof
the Jet Propulsion Laboratory for irradiating the cells.
Work at the National RenewableEnergy Laboratory was
performedunder ContractNo. DE-AC02-83CH10093to
the U.S. Departmentof Energy(DOE).
REFERENCl3
T. Markvart, J. Materials Sci. 1, 1 (1990).
M. Yamaguchiand C. Amano. J. Appl. Phys.. 54,
5021(1983).
3. C. Amano, M. Yamaguchi, and A. Shibukawa,
Technical Digest of the First Intl. Photovoltaic Science
and Engineering Conf., Kobe, Japan,Nov. 1984 (Japan
Times,Tokyo, 1984),p. 845.
4. J. C. C. Fan, GaAs Shallow Homojunction Solar
Cells,Final Report,NASA CR-165167(1980).
5. I. M. Olson and A. Kibbler, J. Crystal Growth, 77,
182(1986).
6. H. J. Hovel, Semiconductors and Semimetals, Vol.
II, Solar Ceils (AcademicPress,Orlando,Florida, 1975),
pp. 17-20,24-25.
7. J. C. C. Fan, C. 0. Bozler, and B. J. Palm, Appl.
Phys.Len., 35,875 (1979).
8. B. T. Cavicchi, D. D. Krut, D. R. Lillington. S. R.
Kurtz, and J. M. Olson, Conf. Rec. 22nd IEEE
Photovoltaic Specialists Co@ (IEEE PublishingServices,
New York, 1992).in press.
9. K. A. Bertness,M. Ladle Ristow, M. E. KlausmeierBrown, M. Grounner,M. S. Kuryla, and J. G. Werthen,
1.
2.
1o-g
0.0
0.2
0.4
0.6
0.8
1.0 1.2
voltage (V)
Figure 8. Post-irradiation dark current versusapplied
forward voltage for severalcells at 300 K, and for the
most highly doped sampleat both 300 K and 77 K. The
excesscurrent startsat about 0.4 V, and persiststo low
temperatures in cells with higher base dopings after
irrad&ion.
Conf. Rec. 21st IEEE Photovoltaic Specialists Co@.
(IEEEPublishingServices,New York, 1990),p. 1231.
24-6
InP CONCENTRATOR
SOLAR
J.S. Ward, M.W. Wantass, T.J. Coutts,
National
Renewable
CELLS
K.A. Emery and C.R. Osterwald
Energy Laboratory
(formerly the Solar Energy Research Institute)
Golden, Colorado,
USA
ABSTRACT
The design, fabrication, and characterization
of highperformance,
n+/p InP shallow-homojunction
(SHJ)
concentrator
solar ceils is described.
The InP device
structures
were
grown
by atmospheric-pressure
metalorganic
vapor phase epitaxy
(APMOVPE).
A
preliminary
assessment of the effects of grid collection
distance
and emitter
sheet
resistance
on cell
performance is also presented. At concentration ratios of
around 100, cells with efficiencies of 21.4% AM0 (24.3%
direct) at 25°C have been fabricated.
These are the
hi hest efficiencies yet reported for single-junction
inP
soBar cells. The performance of these cells as a function of
temperature
is discussed,
and areas for future
improvement
are outlined.
Ap lication of these results
to other InP-based photovoltaic Bevices is also discused.
INTRODUCTION
Recently, one-sun InP solar cell performance
has
begun to ap roach the high conversion
efficiencies
predicted by tRe early modeling work (1). However, until
now, experimental studies of InP concentrator cells have
not been pursued.
In 1988, Goradia,
Geier and
Weinberg
modeled both rectangular
and circular InP
concentrator
cells (2).
They concluded
that high
efficiencies were possible for these devices and pointed
out that their potential radiation resistance should make
them attractive for space applications.
Advances in space
photovoltaic
concentrator arrays have demonstrated that
excellent power densities and power-to-mass
ratios are
achievable with these systems (3). Although concentrator
arrays have been designed to minimize
the effects of
radiation, notably for &e Strategic &t&se
initiative, they
do so at the expense of pe&rmance.
Radiation effects
are considered to be problematic
for the array designs
that exhibit stated-the-art
power to mass ratios.
Deep
level transient spectroscopy IDLTSB studies have indicated
that at the elevated
temperatures
and high current
densities associated with
ation under concentration,
the sombinatTon of therma 9p”and injection annealing may
render If4P solar cells practically
impervious to radiation
damage (4. Thus, tr@ may serve as a radiation-resistant
alternative to GaAs,
At this point, one of the major obstacles to a more
widespread use of IMP is the price of high-quality, single-
25-l
crystal substrates.
A number of strategies have been
suggested to limit the impact of this cost. They include
the development of multijunction cells to boost efficiency
and heteroepitaxial
(HE) growth techniques, which would
eliminate the need for InP substrates. The work done on
InP concentrator cells should yield information that will
be directly
applicable
to the emerging
HE-cell
technology.
We have already
seen dramatic
improvements
in the performance parameters of HE InP
cells when measured under solar concentration
(5). At
higher concentration
ratios, the erformance parameters
of HE InP cells approach those of I: omoepitaxial cells.
Improving the performance of the InP/Gae.471no.ssAs
monolithic
concentrator
tandem cell requires advances
in the design of the InP top cell (6). The Gaa.47Ine.ssAs
bottom cell is exhibiting
near theoretical performance
levels, but the InP top cell is showing evidence of series
resistance problems at concentration
ratios above 40
suns. Minimizing these series resistance losses may allow
the tandem efficiency to approach 35% under the direct
spectrum at high concentration
ratios (7). At 31.8%
under 50 direct suns at 2S°C, this is already the most
efficient
monolithic
photovoltaic
device
yet
demonstrated.
The points discussed above have motivated
the
present work. In this paper we describe our initial efforts
to fabricate
and characterize
high-performance
InP
concentrator cells designed to operate under 100 AM0
suns concentration.
DEVICE DESIGN AND PROCESStNG
A schematic diagram of the InP concentrator solar
cell structure is given in Figure 1. The device structures
are grown by APMOVPE on Zn-doped, p+-InP substrates
oriented in the (100) direction. Growth is carried out in a
vertical reactor vessel at a temperature of 620°C and in a
purified hydrogen ambient.
The primary reactants are
trimethylindium
and phosphine.
The dopants consist of
hydrogen sulfide and diethylzinc.
A p+- back-surfacefield layer, grown to a thickness of 0.38 pm, is followed by
a p-base layer that is doped to - 10’7 cm-3 and grown to
a thickness of 3.8 Pm. A thin n+- emitter layer, doped to
3.7 xl 018 cm-J, completes the growth.
Table 1: Performance parameters for an InP concentrator cell at 25°C under
the direct spectrum.
A rectangular cell geometry was chosen to simplify
photomask development
and to limit the unilluminated
junction
area.
This rectangular
design incorporates
double bus bars which allow probe placement at both
ends of the grid-lines, thereby limiting the electrical losses
in the fingers.
The area of our concentrator
cells is
determined
by measuring
the total mesa area and
subtracting the area of the bus bars. In this case, the
computed area is 0.0746 cm*.
After etching the back surface in a bromine and
methanol solution (1% by volume) for 5 minutes, an
ohmic contact is formed by electroplating
0.1 pm of Au,
0.01 pm of Zn, and 3 pm of Au onto the back surface and
then annealing on a graphite strip heater at 37S” C for 90
seconds.
The rid pattern on the emitter surface is
defined by stan d ard photolithography.
Pure Au is then
electroplated
to a thickness of 5 pm. Cell isolation is
accomplished
by etching in concentrated
hydrochloric
acid after a photolitho
raphic mesa definition.
The
devices
are complete f by depositing
a two-layer
ZnS/MgFz anti-reflection
coating and applying Entech
prismatic covers (Figure 1J.
The Entech prismatic
cover (8) is an essential
component
of the cell design.
With the resistivity of
electroplated
gold often in excess of five times the bulk
value (91, metallization
schemes designed to handle
current densities of 2.9 A /cm* necessarily entail a high
grid coverage (-20%).
We have found that with a
properly designed antireflection coating, the optical losses
associated with the use of the prismatic cover are less
than 5%. The major limitation associated with using the
cover for this device is that the grid-line spacing is fixed.
This aspect of the device design is discussed in more
detail in the next section.
operation under concentration,
we decided to examine
the effects of grid fin er spacing and emitter layer sheet
resistance on cell per Pormance.
In previous work (101, we performed an empirical
investigation of the InP shallow homojunction (SHJJ solar
cell designed to operate at one sun.
A thin (25 nm)
emitter was found to be essential to minimize the roll off
in the blue response attributable to the unpassivated InP
surface.
For concentrator
cells, the benefits of this
enhanced blue response must be weighed against the
high sheet resistance associated with thin emitter designs.
At one-sun current densities f-29 mA cm-*), the negative
effects of the high sheet resistance can be minimized by
adjusting
the grid finger spacing.
However,
our
concentrator
cells incorporate
Entech prismatic covers
originally designed for GaAs concentrator cells operating
at 100 suns. This aspect of our concentrator cell design
results in the grid-line spacing being fixed at 127 pm.
Therefore, it is reasonable to expect that the optimum
concentrator cell structure may differ from the optimum
one-sun structure.
Hall measurements of the n+-InP layers provide
values of 1200 cm* V-t s-1 for the electron mobility and
Sunlight
I
m-
Entech
Cover
q
--------------
Our primary
objective
in this work was to
demonstrate
the potential
of InP concentrator
cells.
However,
the development
of the single-junction
InP
concentrator cells is important as a basis of comparison
with
the
HE InP cells
and
the
monolithic
InP/Gaa,47lne.saAs
tandem. As a starting point in our
attempt to optimize
the tnP SHJ cell structure for
n-InP emitter
p-InP base
0.38 flrn
EXPERIMENTAL
2-layer ARC
/
It
Top
Contact
InP
SW
p-InP BSFL
_
Back
Contact
Figure i. Cross-sectional shematic dia ram of the InP
shallow-homojunction
concentrator so7ar cell
structure.
25-2
3.7 x 10’8 cm-3 for the free electron density. This results
in a resistivity value of 1.4 x 10-X R-cm. Previous work on
one-sun cells indicated that the emitter thickness should
be limited to between
200 and 400 A.
For these
thicknesses, the sheet resistance of the emitter will be
between 350 and 700 Q/square,
respectively.
The
expression for the fractional power loss due to the grid
finger spacing is given by
The performance of these cells was characterized by
absolute
external
quantum
efficiency
(AEQE)
measurements,
illuminated
current-voltage
(I-V)
characteristics as a function of the concentration ratio and
temperature, as well as temperature-dependent
dark I-V
measurements.
Efficiencies reported here are referenced
to either rhe direct (ASTM E891-87) (1 I) or the AM0 (12)
spectra.
The cell performance
is discussed in the
following section.
RESULTS AND DISCUSSION
where PI is the sheet resistance of the emitter, S is the
spacing between grid-lines, Jmp is the current density at
the maximum
power point, V,, is the voltage at the
maximum power point, and PlorJPmp is the fractional
power loss at maximum power. Using Jmp = 2.8 A. and
V
= 0.9 V at around 100 suns concentration,
the
fr$ional
power loss for the minimum grid finger spacing
of 127 pm will range from 1.4% to 2.8%, depending on
the thickness of the emitter.
The lop cell of the
previously
mentioned
three-terminal
tandem device
requires a grid finger spacing that is twice the minimum.
Due to the 52 dependance of the power loss term, this
design results in a 5.6% to 11.2% power loss at 100 suns.
Solar cell grid designers generally strive to limit this
fractional power loss term to around 2%. The plor of
efficiency vs. concentraGon in Figure 2 demonstrates the
consequence
of allowing this term to dominate.
For
optim.um performance
at 100 suns, it is therefore
necessary to use the minimum finger spacing afforded by
the cover material.
A systematic empirical
study is
planned to explore the performance of these devices in
the range of emitter thicknesses from 200 to 400 A.
However, as a first alternot. an intermediate thickness of
312 A was chosen.
’ .
A comparison
of the efficiency
as a function of
concentration
ratio for two grid-finger spacings (127 pm
and 254 pm) in Figure 2 reveals a clear performance
advantage for the closer spaced grid design. Devices using
the 127 pm grid-finger spacing and very thick emitters
(24OOA) were fabricated and their fill-factors as functions
of current density were compared with those of the
devices with 312 A emitters.
Since there was no
significant
improvement in the thick emitter devices we
conclude that lateral current spreading iti the emitter is
not a dominant power loss mechanism for the 127 pm
&id finger spacing.
The single-junction
InP concentrator
cdl design
using an emitter that is 312 A ihick has achieved high
efficiency levels at concentration
ratios of around 100
Peak efficiency
at 25°C was 21.4% at a
suns.
concentration ratio of 106.5 AM0 suns and 24.3% at 99.4
direct suns
This represents a ain of 2.3 efficiency
percentage
oints compared to tP;e best reported onesun result o P 19.1% AM0 at 25°C (13). Current-voltage
data for an tnP concentrator
ceil at peak efficiency is
provided
in Figure 3. Dark I-V measurements
as a
function of temperature provide a value of 0.0025 R cm*
0.35
Single-junctipn
. . . . . . . . . . . . . . . . ..*...........*
l.
cell (51 pm)
0.25
I
.
..
.
L
3
18~
20
40
60
80
100
AM0 ConcentraGon Ratio
-
. . . . ..*8O”C
E 0.15 5
”
0.1 r AM0 C (suns):
0.05 1: v,, w:
:_ jsc (Acm-2):
0 : FF (“/a):
:
_ AM0 ?.t(%I:
Tandem top cell (102 pm)
0
0.2
-0.05 f
-0.2
120
Figure 2. AM0 efficiency versus concentration ratio
data for tnP concentrator cells with similar structures
but different grid-line spacing.
’
’
0
*
’
0.2
2!X
106.5
0.978
3.73
85.7
21.4
’
{.
8OOC.
125.6
0.876
4.53
82.5
19.1
’
’
’
0.4
0.6
Voltage (VI
’
I
0.8
Figure 3. Current-voltage data for an InP shallowhomojunction cell at peak AM0 efficiency under
concentration at 25°C and 80°C.
25-3
.
i
i
:
‘:
i
:
:
*
1
for the series resistance fR,). This value of R, may allow
the device to reach peak efficiency
at concentration
ratios of close to 300 suns. However, the measurement
system at NREL is currently
incapable
of generating
intensities of over 100 suns.
Figures 4-6 show the
measured
performance
parameters
vs. direct
concentration ratio at 25°C. These data indicate that the
efficiency is stift increasing at t 00 suns.
Similar devices have been sent to Sandia National
Laboratories
for flash testing at higher concentration
ratios. Data from these tests will be available in the near
future. As computed from the dark I-V measurements,
the diode quality factor as a function of temperature
ranges from 1.035 to 1.068 near V,,, illustrating
the
excellent
characteristics
of the APMOVPE
grown
junctions.
81:
211
8Ofrrr7rr~~..,....,-.,....,..
000000000
rolObV)colcOO
CONCENTRATION RATIO (Suns)
Figure 4. Efficiency versus concentration ratio data for
an InP concentrator cell under the direct spectrum
at 25°C.
960
CONCENTRATION RATIO (Suns) .Figuie 6. Fill-factor vesus concentration ratio data
for an InP concentrator cell under the direct
spectrum at 25C.
The temperature
coefficients for the performance
parameters is given as functions of concentration ratio in
figure 7. These data show a loss in efficiency
of
approximately 0.17% PC at concentration ranges of 40 to
100 suns. A good estimation of the efficiency at any
operating
temperature
may be obtained
with this
information and the efficiency value at 25°C. Analysis of
the AEQE data in figure 8 indicates that improvements in
the performance of these cells will likely be achieved
fabricating devices with even thinner emitters, which wi7 I
enhance the blue response and increase the short-circuit
current density fJrJ. Although R, will increase with a
thinner emitter, it should not significantly
impact the
device performance at concentration ratios of 100 or less.
Development of a passivating window layer should also
add to llc as well as providing a possible increase in the
open-circuil voltage.
970-f
970’
960{
960{
zgm:
zgm:
- 9409408
’ 9301)
9301)
9201
9201
910:
0 07 s a%s%gssg
88
<‘I’
CONCENTRATION RATIO (Suns)
Figure 5. Open-circuit voltage versus concentration
ratio data for an InP concentrator cell under the
direct spectrum at 25’C.
SUMMARY
As part of an ongoin effort to make InP-based solar
cells a realistic option Por both space and terrestrial
applications, InP concentrator cells have been fabricated
and characterized as functions of concentration ratio and
0
o^
g
90 -
;
.-
80 -
H
F-v-loo0
g
70 -
5
60 -
E
G
iz -1500
5z- 50 -
$
Q
-500
2
40 -
8
30 -
$
20 -
3
lo-
ki
? -2000
%
E
1
.-
-2500~
0
CONCENTRATION RATIO (Suns)
Wavelength
Figure 7. Temperature coefficents of the
performance parameters
versus concentration
ratio for an InP concentrator cell under the direct
spectrum.
Figure 8. Absolute external quantum efficiency
curve for an InP concentrator solar cell at 25°C.
temperature.
This is the first report of high-efficiency
InP
concentrator
solar cells. Devices have been fabricated
that exhibit conversion efficiencies of 21.4% at 106.46
AM0 suns and 24.3% at 99.4 direct suns at 25°C. These
are currently the highest efficiencies reported for InP solar
cells. The
wer loss due to lateral current spreading in
the emitter p”ayer was found to be within acce table limits
using a grid design that incorporates an avai Pable Entech
prismatic cover. These results indicate that the necessary
technologies
presently
exist for fabricating
highperformance InP concentrator solar cells.
The InP concentrator
cells described in this paper
have attained high levels of performance
using welldeveloped growth and processing techniques.
Areas for
further research include a more detailed look at the
optimum emitter thickness for the present SHJ design.
Higher efficiencies are expected for devices with slightly
thinner emitters.
Experimental evidence suggests that at
80°C and at the current densities observed at 100 suns,
these devices may become self-annealing.
The power-tomass ratios of space concentrator
systems may be
improved if radiation Foferance can be efiminated as a
design conslraint.
The resufts given here have importarrt implications
for other
I&-based
PV devices
currertFJy under
considerafion.
The fnP concentrafor ceff comprises the
Fop cefl of Fhe most efficient monoiithic
device yet
demonstrated fthe fnP/Gan.4ztne.s&
tandem). This work
has shawit FhaFthe I27 pm grid-finger spacing used in the
single junction
Concentrator
destgn results in a much
lower value of R, than the Fop cell of the tandem (7).
When
this
design
is incorporated
into
the
InP/Gae.&te.saAs
concentrator
tandem, conversion
fnm)
efficiencies approaching
are anticipated.
35% under the direct spectrum
ACKNOWLEDGEMENTS
Support for this work was provided by the U.S.
Department
of Energy under contract
DE-ACOZ83HC10093,
the Naval Research Laboratory
under
interagency order RU-1 l-W70-AD
and the NASA Lewis
Research Center under Interagency order C-3OUOO5-K.
REFERENCES
1
1.1.Loferski, J. Appl. Phys. , 27, 777, 1956.
2.
C. Gordia, I.V. Geier and I. Weinger. Conference
Record of the 20th IEEE Photovoltaic Specialists
Conference, New York: Institute of Electrical and
Electronic Engineers, 1988, pp. 695-701.
3.
M.J. O’Neil and M.F. Piszczor, Conference Record
Space Photovoltaic Research and Technology
Conference (SPRAT) 1989, 443, 1991.
4.
R.J. Waiters and G.P. Summers, J. Appl. Phys. ,69,
1991, pp. 6488-6494.
5.
M.W. Wanlass, T.J. Coutts and J.S. Ward.
6A, these proceedings.
6.
M.W. Wanlass, J.S. Ward, K.A. Emery, T.A. Cessert,
C.R. Osterwald and T.J. Coults, Solar Cells, 30,
1991, pp. 363-371.
25-5
Session
7.
C.R. Osterwald, M.W. Wanlass, J.S. Ward, B.M.
Keyes, K.A. Emery and T.J. Coutts Session 9A.
these proceedings.
8.
M.J. O’Neil,
1987.
9.
U.S. Patent No. 4,711,972,
11.
” Standard for Terrestrial Solar Direct Normal
Solar Spectral irradiance Tables for Air Mass 1.5,
ASTM Standard E891”, American Society for
Testing and Materials.
12.
C. Wehrli, “Extraterrestrial Solar Spectrum,” Physical
Meteorological
Observatory and World Radiation
Center, tech. rep. no. 615, Davos-Dorf, Switzerland,
July 1985.
13.
C.J. Keavney, V.E. Haven and S.M. Vernon,
Conference Record of the 2 1st IEEE Photovoltaic
Specialists Conference, New York: Institute of
Electrical and Electronics Engineers, 1990, pp.
Dec.
T.A. Gessert and T.J.Coutts, “Requirements of
Electrical Contacts to Photovoltaic Solar Cells,“.
Materials Research Society Symposium
Proceedings.,Vol 181, Pittsburgh Pa.: Materials
Research Society, 1990, pp. 301-312.
10. M.W. Wanlass, T.A. Cessert, K.A. Emery and T.J.
Coutts. Conference Record of the 20th IEEE
Photovoltaic Specialists Conference, New York:
institute of Electrical and Electronics Engineers,
1988, pp.491 -495.
141-144.
BACK SURF&F
FIELDS FOR GalnP:, SOI AR CElJ&
D. J. Friedman,S. R. Kurtz, A. E. Kibbler, andJ. M. Olson
NationalRenewableEnergyLaboratory
(formerly the SolarEnergyResearchInstitute)
Golden,CO
ABSTRACT
We studied back surfacepassivationof the GaInP2
top cell in GaInP2/GaAs two-terminal tandem cells.
Becauseof the requirementof currentmatchingof the top
and bottom cell, the top cell must be madevery thin (on
the order of 1 pm), and thusproper passivationof the top
celI back surface is important in achieving high open
circuit voltages.In this paper,we comparetwo candidate
top-cell back surface fields: (1) an AlGaInP quatemary,
and (2) GM-2 grown to give a band gap higher than that
of the baseof the cell.
INTRODUCTION
The tandem combination of an optically thin
Gao.51Ir~49P top cell and a GaAs bottom cell has
achieveda one-sun, air mass 1.5 (AM1.5) efficiency of
27.3% (1). The Gao51Ir~49R(hereafterGal.@) top cell,
with a band gap of 1.85eV, mustbe -1 pm thick in order
to achieve current matching (2). At this thickness, the
surfacerecombinationvelocity at the back of the cellwill
n-AMP
n-GalnP2
thickness
dop9
(pm)
)
(
5c:10’7
0.025
0.1
2 x 1018
pGalnP2
1 x 1Ol7
layer
.
I p+ GaAs
0.8
purpose
window
emitter
base
substrate
1. Structure of typical GaInP2.top cell (not
to scale).The contacting layer and contactsare not
shown.
File
1
significantlyaffect the open circuit voltage(Voc). Hence,
the Voc of the 27.3% device, which contained no back
surface field (BSF), was about 100 meV less than the
expectedvalue.To remedy this situation, we studied the
efficacy of two BSF structures. The first was an
Alo.~~G~.45h4)4gP (hereafter AlGaInP) alloy with a
band gap of -1.95 eV. The second was Gab@ grown
under conditions that yield a band gap of 1.88 eV. (The
bandgapof GaInPz,at constantcomposition,is a function
of numerousgrowth conditions and can be varied from
1.8 to 1.9 eV (3,4)). This paper examinesthe differences
betweenthe two candidatematerialsfor passivatingthe
backsurfaceof the GaIn?2 top cell.
The.cells.which areof the n-on-p configuration,were
grown by atmospheric-pressuremetal-organic chemical
vapor deposition,lattice matchedto GaAs. The structure
of a typical cell is shown schematicallyin Figure 1. A
GaAscontactinglayer doped-1019/cm3 n-type is usedto
provide an ohmic contact for the front contact. The
contacting layer provides a tiont contact resistanceof
-10e4 Q-cm2 without the needfor sintering. The process
of metallixation and mesa etching produced individual
cells with an areaof 0.25 cm2,A rangeof characterization
techniqueswas applied to these cells. The techniques
discussedin this paperare dark and light current-voltage
(IV) electrical measurements;and secondary ion mass
spectroscopy(SIMS) profiling.
RESULTS AND DISCUSSION
Figure 2 showsdark IV curvesfor two typical cells,
one with a GaInI’2 BSF and one with a quaternaryBSF.
For thesecells, the behavior of the dark current can be
divided into two regimes. Above about 1.1 V, the cells
exhibit n = 1 dark currents,Le., I = exp(-eV/nkT) with n
= 1, while below 1.1 V, there is a transition to n = 2
behavior. While n = 2 behavior is most frequently
attributedto generation/recombinationin the junction, in
practice, this current can be dominated by perimeter
currents (5). For our cells under one-sun AM1.5
illumination, the maximum power-point voltage was
8
1.1
6
0.4
1.0
1.82
0.6 0.8 1.0 1.2 1.4
Forward bias (volts)
Figure 2. Dark current J1 as a function of forward
bias voltagefor selectedtop cells.
I
I
I
I
‘I’ll
1.88
Figure 4. V,, vs.bandgapfor variouscells.The type
typically 1.25 V or greater; thus Figure 2 shows that
perimeter and junction recombination currents are not
significant for theseceils.
The n = 1 dark currentsJ1 for a numberof cells are
summarizedin Figure 3, wherefor eachcell, Jl(1.3V) is
plotted against the correspondingVoc for that ceii. The
BSF usedfor eachcell is indicatedin the figure.The cells
with the high band gap GaInP2 BSF have a lower dark
current (and hence a higher Voc) than the cells with the
quaternaryBSF. Note, however,that Voc increaseswith
the cell band gap, and that there is somevariation in the
latter, due mostly to variationsin the compositionx of the
Gal-dn,P. To confirm that the improvedVoc is due to
the superiority of the GaInP2 BSF and not merely to
variations in the band gap of the emitter/btie, Figure 4
l-1 “I
1.84
1.86
cell band gap(eV)
of back surface field used in the cells is indicated.
The solid line shows the expectedslope of V,, vs.
bandgap whenall otherfactorsare held constant.
displaysVoc againstbandg@ for the cells of Figure 3, as
well as for othercells that did not have,aclearly definedn
= 1 region. The figure confirms that for any given band
gap, the cells with the GaInP2 BSF have higher Voc
values than the cells with the quaternary BSF. The
expectedvariationof Voc with bandgap is easilyseenfor
the GaInP2 BSF cells: for the quaternaryBSF cells, the
scatterin Voc masksthis dependence.
IJ
1.38
z 1.36
Z
e 1.34
8
>
1.32
o4
’
10"
IO4
1 o9
J,(1.3V) (amps/O.25 cm*)
I
2
I
I I lllll
4 68
I
2
“1’1
4
6
loQ
basethickness
Figure 5. Calculation of J1 as a function of base
thicknessfor a model cell simulating the thick/thin
pair discussedin the text, for various valuesof the
backsurfacerecombinationvelocity.
Figure 3. Dark currentJ l(l.3V) vs. Voc for various
cells. The type of back surfacefield usedin the cells
is indicated.
26-2
In order to estimatethe back surfacerecombination
velocity provided by the GaInP2BSF, we comparedtwo
top cells with GaInP2 BSFs, identicai except for the
thicknessof the base,which was 0.6 pm for the thin cell
and 6 m for the thick celL The Jl currents for the two
cells were rehted by J~(thick)/J~(thin)= 4.1. Figure 5
shows a calculation (6) of J1 as a function of base
thicknessfor a model cell simulating the thick/thin pair,
for various values of the back surface recombination
velocity. For the thick cell to havea J1 current four times
that of the thin cell, the back surface recombination
velocity had to be about 5x1@ cm&c, a 10~ value
consistent with the conclusion that the GaInP2 BSF is
effectivein reducingrecombinationat the backof the celL
Jnordertoprovidesomeinsightintothenatumofthe
problem with the quatemaryBSF,Figure 6 showsa SJMS
scanthrough a top cell with a quaternaryBSF. The level
of oxygencontaminationpeaksat the locationof the BSF,
as might be expectedfrom the known oxygen-get&zing
Propemesof aluminum.It seemsreasonableto guessthat
the oxygen contentof the quammarylayer is responsible
for its poor performanceas a BSF, and that reduction of
the oxygen content might lead to a more effective
quatemq BSF.
-h
,05
SiMS Depth Profile
-; ..-----
‘1------
-----_
----;
)... -. - _-. - .- .-. -
lo4
i
iu
i;
“?
i
0
$
0’
2 103 -L \, --,, ~ _.... v--.-‘I..
As
L
’
*
.
.
’ .
0.5
-’ (
I
t
10*-
100
0
-. - .- -. -.-. -. -. -.
’
ir
* ’ IL .
1
L
’ a.. *
’ *
1.5
depth WW
Figure 6. SIMS depth proftie of a top cell with a
quatemaryBSE Note the peak in the oxygenlevel at
the location of the BSF.
SUMMARY
In summary,we have presenteddark-IV, V,, and
SJMSdata on GaJnP2top cells with quaternaryand with
high-band-gapGaInP2 BSFs. Although the aluminumcontaining BSFs were of mediocre quality, the GaJnP2
BSFsprovedhighly effective.
26-3
ACKNOWLEDGMENTS
l’his work was performed under Contract No. DEACO2433CH10093to tbe U. S. Departmentof Energy.
We thank S. Asher for the SIMS measurements,and K.
Bertness and P. Parilla for a careful reading of the
nlanuscripL
REFERENCES
1. J. M..Olson, S. R. Kurtz, A. E. Kibbler?and P. Fake.
AppL Phys.Lea., 56,623 (1990).
2 S. R. Kurtz, P. Faine,and J. M. Olson,J. AppL Phys.,
68,189O(1990).
3. S. R. Kurtx, J. M. Olson, and A. E. Kibbler, Appl.
Phys.Lat., 57,1922 (1990).
4. A. Gomyo,T. Suzuki, and S. Iijima, Phys.Rev. L&t.,
60.2645 (1988).
5. G. B. Stringfellow, J. Vat. Sci. Technol., 13, 908
(1976).
6. J. P. McKelvey, Solid State and Semiconductor
Physics (Ibiegex, Malabar, 1982)p. 422.
CONTROLLED LIGHT-SOAKING EXPERIMENT
FOR AMORPHOUS SILICON MODULES
W. I.&
B. von Roedern,B. Stafford, D. Waddington, and L. Mrig
National RenewableEnergy Laboratory
(formerly the Solar Energy ResearchInstitute)
Golden, Colorado. 80401-3393USA
ABSTRACT
Multijunction amorphous silicon (aSi) modules from
three manufacturerswere subjectedto light-soaking at 1-sun
intensity at WC for 200 hours, with annealing to 70°C in
the dark after loo0 hours. Characterization was done
periodically, both under a Spire solar simulator and
outdwrs. Aperture-areaefficiencies as high as 7.2% were
obtained after 1000 hours of light-soaking. The power
output after loo0 hours of light-soaking and subsequent
partial annealing ranged from 78% to 92% of the initial
power output. The recovery in power due to annealing was
2%4%. This recovery is not consistent with a thermally
activated process with a 0.9 eV activation energy. The
performance is fitted with a stretched exponential curve
obtained using the data obtained for the first loo0 hours of
light-soaking. For two types of modules, stabilized
performancewas reached before 1000 hours.
hours of illumination at 50°C. We also wanted to seehow
well a stretched exponential curve fitted to experimental
points would predict subsequentperformance,and whether
the 70°C annealing would affect the long-term performance
or only improve the performancefor a short time.
The average daily integrated time equivalent to an
intensity of the sun of 1,000 W/m2 is 5-5.5 hours for the
continental United States. With 300 days per year of full
sunshine, this would be the equivalent to 1.500 - 1.650
hours of 1.000 W/m2 illumination. Considering, however,
that the light-induced degradation of a-Si is more closely
proportional to the squareof the light intensity. the average
period per day of the equivalent to I,000 W/m2 would be
less than 5-5.5 hours per day. Consequently, the yearly
equivalent would be less than 1~500-1.650hours. Thus,
about 1,100 hours at 1,000 W/m may be a good estimate
for one year of outdoor exposure.
INTRODUCTION
SAMPLE DKSCRIPTION
The stabilized efficiency for amo@ous silicon modules
is important becauseof the impact efficiency has on the
cost. Outdoor tests of a-Si:H modules of 1988 or earlier
vintages have stabilized power outputs that correspond to
module efficiencies of 456-4.596(11. To assessthe progress
-made in module development, we have conducted a
cwtrolled experiment to obtain the stabilized efficiency of
a-Si &womype ssaddes frtrricrtsd 3rr 1990 by three
-a&e
iizzmatIoa*heip
typedUIeS~OUtdOOT~CWditiOiiS.The
objective w-as80timu#ate rtradute~ormanceXtnder actual
*
a5nwmmM in the United
!&&es. ts%ldaaw
WOUMresult in
.
~soOCmoduletem@rtiture&3rmostofthe
year~~~b~aurktg~~rwItths~2).
ca#mqaII
we std#aed~~ mt.aules Doam hours of
mm w/M ilhmlinatlon at 50°C ad thea raised the
temperature to 70°C to allow some level of recovery.
Subsequaaly, we subjected,$hemodules to another loo0
The following groupsof moduleswere subjectedto our
extended light-soaking test. in the subsequenttext, these
modules will be called “test modules.”
27-1
1.
Three 1xl ft modules, same-bandgapdual-junction
(a-St/a-Si) p-i-n/pi-n units deposited by radio
frequency (RF) glow discharge on glass with a
sputtered aluminum alloy back reflector. The p
layers are a-SiC:H for both cells in the stack The
i-layer thicknessesare approximately55 and 330 nm.
The moduleshave two glass sheets,Ethylene Vinyl
Acetate (EVA), no frame, and contabt 29 or 30
active cells. The aperture area is 879 cm2 for the
29-cell modules and 909.3 cm2 for the 30-cell
modules.
2.
Five 1xl ft triple-junction (Siii/SiGe) (1.7/l .7/l .45
eV) p-i-n/p-i-n/p-i-n modules depositedon glass by
direct current proximity discharge. The modules
have one glass sheet,plastic frames,and contain 30
cells. The back refiector is indium tin oxide/Ag.
The back encapsulation is polyurethane. All three players are a-SiC:H. The i-layers for the modules ‘we
75, 400. and 100 nm thick. The aperture area as
determined by the frame dimensions is 962.5 cm2.
3.
One 1x4 ft module with a metal frame and an
aperture area of 3,676 cm2 (consisting of 13 ceils).
It is a dual-junction (a-Si/a-Si) p-i-n/p-i-n unit with
1.8 to 1.7 eV band gaps deposited by RF glow
discharge on stainless steel, with ZnO/Al back
reflector and indium oxide/Ag-grid front electrode.
The p-layers are microcrystalline Si:H, not a-SiC:H.
The i-layers are approximately 200 and 500 nm
thick. T@ front encapsulation is Tefzel and EVA.
In addition, there was one control module for group I
and two control modules for group 2. There was no COIIVOI
module for group 3.
TEST CONDITIONS AND PROCEDURES
The initial module efficiencies were obtained by
measurements outdoors under prevailing conditions on B
clear day in Colorado in December 1990, and indoors on a
Spire solar simulator. Subsequently, the control modules
were kept in the dark at room temperature. The test
modules were light-soaked at loo0 W/m2 intensity using an
argon plasma light source at a module temperature of 50°C
and were operated near their maximum power point using
fixed resistors.
The light-soaking was done in an
environmental chamber with the chamber temperature
adjusted such that thermocouples on the back of the
modules measured SOOCwith the light source on. The
relative humidity in the chamber was very low (15%). The
light-soaking test started on January 2. 1991. Subsequent
efficiency measuretients at logarithmic time intervals were
done on the Spire solar simulator at a module temperature
of 25°C. After a total of 1000 hours of light-soaking, the
light-soaked test modules and-their controls, which were
kept in the dark, were again measured outddors on a clear
day in Colorado in April 1991, and indoors on a Spire solar
simulator.
After light-soaking, the test modules were exposed in
the dark to 60°C for a total of 82 hours and then to 70°C
Periodic efficiency
for an additional 34 hours.
measurements were taken on the Spire solar simulator at
module temperatures of 25oC. Following the tetiperature
exposure, the modules were remeasured outdoors under
prevailing conditions in May 1991, and indoors on a Spire
solar simulator. These data are shown as “recovered” in
Table 1. The control modules were only measured outdoors
after the test-module annealing exposure.
The test modules that had previously been light-soaked
for ioo0 hours and annealed to 70°C were light-soaked for
an additional ‘1000 hours at a module temperature of 500~
and were operated near their maximum power point using
fixed resistors. The controls were kept in the dark at room
temperature. Periodic efficiency measurements were taken
on the Spire solar simulator at module temperatures of
25oC. Following the light-soaking, the modules and their
controls were ‘remeasured outdoors under prevailing
conditions in September 1991, and indoors on a Spire solar
simulator.
RESULTS
The initial, IOOO-hour, “recovered,” and XIOO-hour
average aperture-area efficiencies, as measured indoors and
outdoors, and the difference in efficiencies as measured
under the solar simulator and outdtirs for the three sample
groups, are shown in Table l., For group 1 and group 2,
average efficiency values and their standard deviation are
reported. Shown in Table 2 are the performance losses in
percent of the initial power as measurea under indoor and
outdoor test condi;ions.
The partial recovery in power output after anneaIing in
the dark is indicated in Table 3. Much of the recovery
occurred within .lO hours at WC. After 84 hours at 60 OC,
no further recovery occurred from that observ&d ai 34 hours
and the temperature was increased to 70°C. Interestingly,
this increase in temperature caused no further ‘discernible
recovery from the 60°C annealing; Therefore, thi test *as
terminated after a total of 34 ho&“‘& .70°C. The
“recovered” efficiency reported in Table 1 reflects this state
in the test modules history. : Thereafter, light-soaking
continued under the condititis of 1000 W/m2 and 50°C for
_’
Table 2. Degradation in Percent o&t&l
Power under
Various Conditions as Measured.under the Spire Solar
Simulator and Out&urs
Indoor
Outdoor
.:.
%
Sample
Group
Condition
1
lOOO-hr
recovered
2ooo-hr
14.8
12.0
14.1 .
i2.4
7.7,
9.4
2
1000-hr
recovered
2Ocm-hr
21.6
18.6
23.1
18.9
15.4
19;8
3
IOOO-hr
recovered
2OtXl-hr
15.1
13.9
16.0
12.3
10.2
11.7
%
I
/ ,-
Table 1. Average Aperture-Area Efficiencies under Various Conditions as Measured
under the Spire Solar Simulator and Outdoors, and the Difference between these
Measurements
Sample
Group
Condition
Indoor
%
Outdoor
%
I
(average
of three
modules)
initial
WOO-hr
recovered
2000-hr
6.10
5.19
5.37
5.24
f 0.21
f 0.15
f 0.16
f0.16
5.88
5.15
5.43
5.33
f 0.18
f 0.16
f 0.17
f0.16
2
(average
of five
modules)
initial
lOOO-hr
recovered
2ooo-hr
8.84
6.93
7.21
6.80
f
f
f
f
8.18
6.63
6.92
6.56
f
f
f
f
fsingle
module)
initial
lOOO-hr
recovered
2000-hr
7.36
6.21
6.34
6.18
0.09
0.07
0.10
0.12
Difference
indoor/outdoor
in % of indoor
0.09
0.08
0.08
0.12
6.01
6.85
6.15
6.05
3.6
0.8
-1.0
-1.7
7.5
4.3
4.0
3.5
6.9
3.2
3.0
2.1
Table 3. Average Recovery after 1,000 hours of Light Exposure due to 60°C and 70°C Annealing
in Percent of WOO-hour Power Output as Measured with the Spire Simulator
Sample
Condition
Group
10 hr at WC
(d=W
%
34 hr at 60°C
(di-&
%
34 hr at 7oOC
(d=W
%
IlOOhrat5oOC
(light)
%
1
2
3
2.9 f 0.3
2.6 f 0.4
3.1
3.2 k 0.3
4.6 f 0.9
4.8
3.4 f 0.3
4.1 f 0.9
2.1
2.7 f 0.6
2.6 f 0.8
2.1
8.9
22.8
Expected (a) 2.6
(a) Calculated recovery for a defect annealing process with a 0.9 eV activation energy, in relative units (see
discussion).
an additional 1000 hours. After 100 hours of additititial
light-soaking, the module efficiencies still exceeded the
values obtained before annealing in the dark tinder the
above conditions. This status is indicated as “1100 hours”
in Table 3.
In Figure 1, we show the average aperture-area
efficiency of the three groups of modules as a function of
illumination time. The data points up to 1000 hours of
illumination have been used to generate stretched
exponential curves 131.The titting parametersare shown in
Table 4. Ihe modules underwent w recovery stepsin the
27-3
Table 4. Parameters for the stretched exponential fits of
the average aperture-area efficiency of each group of
modules as a function of light exposure. The model fit is
MO = (q,it -rl~,&ewWVPl
+ Inner
Sample
Group
tlioil
96
1
6.12
8.86
7.36
2
3
nlru,
10
P
96
hours
5.21
6.14
6.21
46.8
0.64
645.5
72.9
0.52
0.46
Figure 1.
8
6.0
ii
"w
5.5
0.1
Average Aperture-Area EfTiciency based on Spire Simulator Data versus
Light-Soaking Time for Three Groups of Modules. The inset shows the
lOOI%to 2000-hour data points compared to the stretched exponential
curve. (The error bars indicate the standard deviation for the modules of
groups 1 and 2, and an estimate for the random error for group 3. The
“initial” data are shown at 0.1 hours)
1
100
10
1000
TIME (HOURS)
dark after 1000 hours, so we are using the stretched
exponential fit as au indication of how the degradation
would have continued beyond 1000 hours without the
recovery steps [4]. The inset in Figure 1 shows the data
points obtained upon continued light-soaking.
After
annealing the following behavior is observed. For group 1,
the recovery is lost again after less than 600 hours of
further light-soaking. For gn~up 2, for which degradation
had not yet saturated after the first 1000 hours of lightsoaking, the degradation appears to continue on a delayed
time scale. After 600 hours of additional light-soaking, the
27-4
performance of the modules is about the same as after the
first 1000 hours prior to recovery. The data for the single
module of group 3 have too much experimental scatter to
conclude details how the recovered performance is lost with
further light-soaking.
DISCUSSlON
The data represent the best stabilized amorphous silicon
multijunction module performance measured to date at
exposure during their measurement. ihe control module
output over the test period varied in the range -0.3% to
+0.9% for the indoor measurementsand -3.0% to +4.0% for
the outdoor measurements. The change in the conttol
module power output is a measureof the measurement
error.
k.
We have established that light-soaking for 6001000 hours under realistic controlled conditions (1000
W/m’. PC) leads to stabilization in the caseof a-Si:H/aSi:H dual-junction modules. In the case of the ttiplejunction modules, stabilization did not ~ccut after 1000
houts of exposure. It has been previously reported thal aSiGe:H alloys may degrademore comparedto a-Si:H [51.
It has also been repotted that, in these modules, stabilized
performanceis futthet reduced by a light-induced shunting
effect [a].
After 1000 hours of light-soaking, annealing leads to a
tapid partial tecovety (less than 5% in power-output or less
than 20% of the total degradationexperienced). Mote than
50% of the recovery occutred within 10 hours after keeping
the test modules in the dark at WC. Increasing the
temperatureto 70°C did not lead to any futthet discemable
recovery. ?his is a surprising result, as it is commonly
believed that the partial recovery would be due to partial
annealing of dangling bond defects in the intrinsic layers.
me annealing process is commonly believed t? be
characterizedby an activation energy on the order of 1 eV,
which should lead to a significant enhancement in the
recovery at 700 C comparedto 60“ C for the sameanneal
times. In Table 3, we indicated in relative units the
tecovety expected if the annealing processwere governed
by an activation energy of 0.9 eV [7]. Out results suggest
that, if the recovery was conttolted by an activated process,
its activation energy would have to be much smaller than
0.9 eV. Similar conclusions were made from studies of
solar-cell degradation,where fast changescan be observed
in the petfotmance when small changes ate made in the
operating tempetatuteof the cell [8].
Compared to the time scale for the 60°C annealing
recovery (10 hours half-time), the low-temperature
annealingshowsa significant resilienceagainstfurther lightsoaking at 500 C, as even after an additional 100 hours of
lighht-soakingthe beneficial effects of the ptiot annealing
step ate noticeable(seeinset in Figure 1). This featuremay
provide fot a significant improvementin performanceunder
actual outdoot operating conditions, but it only seemsto
delay, rather than to arrest, the degradation unless the
moduie performancehad beenpreviously stabilized. While
it may be a coincidence, it is of interest to note that within
the experimental accuracy the loss of the recoveredpower
may occut on a time scale similar to the characteristic time
(to given in Table 4) during the initial 1000 hours of lightsoaking.
All degradation appearsto be related to light-induced
changes within the modules. We derive this conclusion
from the fact that control modules did not show any
significant long-term degradation, and environmental
&itions
in the test chamber ate genetally consideted
bet&u. ‘J’besecantrol modules did experience some light
What is the validity of the effkiency and degtadation
results? The petfotmanceunder the Spire solat simulator is
consistently higher than that under outdoor conditions. The
difference is particularly pronouncedfot groups 2 and 3 for
the initial condition. The absolute(ran&G and systematic)
error in efficiency measurementsis flO% [91. but the
random etrots from one measurementto the next ate quite
small for the simulator measurement(< fl%) as seenfrom
the consistencyin the periodic measutementresults. Thus,
for the purposeof this study (i.e. assessingdegradationand
determining when stabilized petformance is reached),
meaningful results were found despite as much as f5% of
systematicuncertainty in the aperture-areaefficiencies. The
degradationasmeasuredoutdoors (8%-20%)is consistently
lower than that measuredunder the Spite simulator (12%23%). Group 2 has the highest initial and 1CXKLhout
efficiencies, but it also has the highest degradation.
After degradation,the magnitudeof the indoor/outdoor
measurementdiscrepancydecreases.When individual solar
module performance parameters ate compared, the fillfactors measuredindoor ate lower than those measured
outdools. Yet, the efficiency is higher in the indoor
measurementsbecause of higher short-circuit currents.
Under outdoor conditions, the module temperaturemay
vary, which leads to changes in the open-circuit voltage.
However. we expect thesetemperaturefluctuations to have
a minimal effect on the power output, as it is known that
amorphoussilicon modules have a temperaturecoefficient
neat zero for the power output after the light-induced
degradation has stabilized [lo]. The open-circuit voltage
losses observedas a function of light-soaking time under
indoor measurementconditions eccoutitedfor 15%-24%of
the total power lossesrepotted in Table 1. The remainder
of total power loss arises predominantly from fill-factor
degradation, becauselosses in the short-circuit current, if
any. were less than f 3%.
CONCLUSIONS
1.
2.
27-5
The power output of some multijunction modules
stabilizes after 600 houts under conditions of 1000
W/m* illumination at 50°C.
Stabilized module efficiency of up to 6.2% (apettute
atea) has been confitmed. Values as high as 7.2%
after 10 houts of light-soaking were obtained in
modules that do not. appear to have completely
stabilized after 1000 hours of light+xxtking. These
stabilized eff&ncies are improved over 1988
3.
4.
5.
6.
7.
modules.
The magnitude of the light-induced degradation is
8%22% for 1000 hours of light-soaking under
controlled conditions. This is an improvement over
1988 modules.
There is some (2%-4%) recovery in the power
output upon annealing in the dark at temperatures up
to 70°C.
The recovery observed upon annealing appears to
delay, but not arrest, further degradation in modules
that were not stabilized.
The observed degradation is inferred to be lightinduced, No indications for any other degradation
mechanisms were observed.
The test results provide a realistic assessment of
stabilized efficiency and light-induced degradation of
modules made in 1990.
6.
Bennett, MS., J. Newton, C, Poplawski, and K.
Rajan, “Impact of Defects on the Performance of
High Efficiency 12” x 13” a-Si Based Three-Junction
Modules,” these Proceedings.
7.
Guha. S., et al, to be published in a semiannual
report for period ending June 30, 1991, SERI/l’PAn activation energy for thermal
214-4453.
annealing of 0.9 eV from an analysis of solar cell
degradation is reported. Values found in the
literature sometimes report a spread in activation
energies, e.g. W.B., Jackson and M. Stutzmann
[AUK& Phvs. Lett 49, 1986, p. 9571 deduced a
distribution centered at 1 eV with a full width at half
maximum of 0.3 eV. Even a value as low as 0.52
eV [reported by L. Chen and L. Yang, to be
published in Proceedings 14th Int. Conf. of
Amorphous
Semiconductors,
GarmischPartenkirchen. Germany, 19911 should have led to a
noticeable improved recovery at 7oOC compared to
60oC.
a.
von Roedem, B., “Fast Changes in a-Si:H Solar
Cells after Severe Light-Soaking,”
Materials
Research Society Svmnosia Proc.. 219, Amorphous
Silicon Technology - 1991, p. 493.
9.
A portion of the discrepancy between indoor and
outdoor measurements may be due to the difference
in the reference device, which is a filtered Si solar
cell for the Spire simulator and a pyronometer for
the outdoor tests.
10.
Townsend, T., P. Hutchinson, and S. Hester, “An
Update on Performance Trends at PVUSA,”
Proceedings Photovoltaic
Module Reliability
Workshop, Lakewood, Colorado, 1990, SERI
Publication CP-4097, p. 1.
ACKNOWLEDGEMENTS
The contributions of S. Rummel. K. Emery, Y. Caiyem,
and P. Longrigg to this experiment ate gratefully
acknowledged. This work was supported by the US,
Department of Energy under Contract No. DE-ACO283CHlOO93.
REFERENCES
1.
2.
Jennings, Christina, and C. Whitaker, “PV Module
Performance Outdoors at PG&E,” Proceedinas 20th
IEEE PV Soecialists Conference, 1990, p. 1023.
Catalano, A., et al., Research on Stable. High-
;Effici nc
Modules, Semiannual subcontract report Phase 1, 1
May 1990 - 31 October- 1990 by Solarex Thin Film
Division, SERI/IP-214-4271, 1991, p. 17.
3.
Redfield, D.. and R. H. Bube, “Comprehensive
Kinetics of Defects in a-Si:H,” Materials Research
Society Symnosia Proceedings. 219, Amorphous
Silicon Technology - 1991, p. 21.
4.
We verified the validity of using the projection of
the stretched exponential fit to provide a meaningful
prediction.
AS the light-soaking experiment
progressed, we found that the stretched exponential
fits to the data through 250 and 600 hours projected
the next efficiency measurement within the
experimental accuracy for each module.
5.
e
Catalano. A., et al. -search
Lame Area Amoruhous Silicon Based Solar Cells,”
Final Subcontract Report 1 February 1989 - 28
February 1990. SERI/I’P-21 l-3906, p. 2.
27-6
Bean Nann
Centre for Solar Enernv and Hvtltoero
Hessb%ehlstr. bl ‘7ooO Stuttgart g0. F.R.G.
Keith Emery
R~Y~YI~I+ N:t~i~)nal Renewable Enerav I;tboratorv
1637 Cole Blv3:
Golden. CO 80401, USA.
ARsTltACT
A computer
model has been~developetl
to simulate
solar cell power production from meteorological data and
solar cell mcaswemenrs.A pew featureof the model is that
it Ca&ulateS the &Q&Jrrradiance for clear
tidy
skies from readily available meteorol
ical dam.
The investigation includes a mono-crystalline
SI Icon and
P
GaAs cell and the thin-film cells CdS CdTe, CdS/(irinSe
a-Si:H, and a two-terminal a-Si:H/a- d i:H/a-SiGe:l I clcvic:
Compared to earlier studies the present one includes
spectra under cloud skies to study rn detail the effect of
variations in the soar
spectral irradiance on the device’s
r
efficiency. A second intention of this work is to analyse the
sensitivity of different power and energy rating methods 10
spectral irradiancc, total irradiance aMf cell temper:*tffre.
As a r$sult. a multi-value energy rating scheme nppiyinp the
concept of “Ciiti
al
e
is proposed ;~ntl
compared with the current single-value power raring procedure under “Standard Reporting Conditions”.
This discrepancy, however, should occut between
mensnrements and rating, but not between
rating and
customer. It is important that an agreement is reached
within the PV-community (customers, manufacturers, system
designers and scientists) on how to rate solar modules and
ce4ls closer to their real performance. This becomes even
ies will be applied.
more important if new cell lechnol
Compared to crystalline silicon cells,7 mearity in temperature and irradiance response is not valid for most of the thin
film devices and the s
ral sensitivity of the higher band
gap devices like a-!$ cp”dTe 01 GaAs is more pronounced. As
a contribution to this we investigate how the meteorological
environment influences cell/module efficiency at a cloudy
site (ShJttgart. F.R.G.) over the period of three years and
how spe&al variati&s
affect power and ene&y ratin
schemes. A customer-oriented
multivalue sDecif?cation o$
INTRODUCTION
APPROACH
Rating which is the common procedure IO nttrilwte
;I
“Name-Plate Rating” to a solar cell or module. The currcn(
standard power rating method consists of a single-value
specification of efficiency. This efficiency is measurctl for
nonsonccnlrator
lerrestrial
photovoltaic
cells under
Standard Heporting
Conditions (1000 W/m2 insolation,
25 “C Ceil temperature, AM 1.5 global reference spectrum).
Energy rating, on the other hand, is site and time specific
with one or several values characterizing the energy output
of a PV-technology for a given site and period,
It has been reported by field experimenrs that phocovoltaic modules do not meet their name-plate power rating
under actual operaling conditions by 10 76 to ,111% on :III
annual average. In this report we analyse three of several
effects causing (his discrepancy:
l).Modules
often operate at cell temperatures hi$cr
than the 25 “C standard.
2) The incident irradiances are often lower than the I000
W/m2 standard.
3) IIJ addi@,
there are changes in the relnt.ive spectral
dlstributlon which lead to higher or lower efficiencies
compared to the efficiency under the AM l.S reference
spectrum.
Consider a PV-device operating outdoors with the
ontical / electrical behaviour of the cells as measured in the
I& in&rs
and the thermal behaviour of a module exposed
to the actual weather. The operating mode is fixed latltudetilted flat-plate (non-concentrator).
Electrical losses due to
cell and module connection (mismatch, resistance), other
wiring losses, optical losses due to soiling or degradation are
not taken into account. The device is operated at the maximum power pdint. The weather is characterized by hourly
sums of global and diffuse irradiance, hourly averages of
ambient temperature, relative humidity and wind speed.
,&lar -.-.----_
Cell Model
.- .-- and Data
‘Ihis investigation
comprises typical production
mono-crystalline silicon (mono-Si), state-of-the-art GaAs,
thin film technologies including CdS/CdTe, CdS/CulnSeZ
and amorphous silicon (a-SJ:H) devices. The a-Si:H
material is kepresented with’ three different designs: a single
iunction cell. a multiiunction a-Si:H/a-Si:H/a-SiGe:H
two&minal device, whe;e the efficiency’& raised by inte ratin
a-Si:H with a-StGe:H alloy cells. and a four-termina B stat f
consistin of a-Si:H on CtdnSe, cells. The electrical characteristic o f the solar cell is computed with the so-called “twodiode equation: a composition of. dark and illuminated
characteristics. The calculaiions are based on measured
quantum efficiencies (Fig. 1). Other cell input data, which
Wavelength (nm)
Fig. 1:
-
.
Wavelenglh (nni)
are shunt and series resistance, two diode quality factors
and four parameters describing the temperature depcndence of the dark saturation current, were derived from lab
measurements of the open circuit voltage, fill factor and
maximum power and their dependence on ceil temperature
and total irradiance (11. ?he computer code calculating
P
uses a numerical search routme. It does not apply
oR!?r simplifying approximations lo locate P . For the
case of multijunction PVdevices the J-V c&y for each
junction is computed first and then the two terminal
multijunction J-V curve is reconstructed hy summing the
voltages at the same current for each junction.
Met
-Model
Our research is aimed at predicting the parameters
solar total and spectral irradiance and cell temperature with
readily available
meteorological
data. From
spectral
measurements we have learned that the glohal solar
s ectral irradiance can be predicted from hourly sums of
cf!Iffuse and global broadhand irradiance, solar geometry
and precipitahle water vapor (which can he estimated from
relative humidity and temperature or dew point temperature). These four input data are utilized by the semi-empirical model
SEDESl
Fig. 2:
Measured
quantum efficicncics for protltlction
mono-Si, and state of the art a-Si:ll, Cd’l’e and
a-Si:H/a-Si:l i/a-SiGe:l I
GtlAS,
3-junction
tandem, CulnSe, and tht hott~~ni ccl1 in :I 4terminal tandem matte of :I-Si:l I mech;lnic;~lly
stacked on CulnSe,.
(Fig. 2). SEDESl
The principal corn nents of the
spectral model SE I!?ES 1.
[4]. The re uired inputs include the global horizontal and
diffuse (or Y*erect beam) total irradiance. Combined with the
input data for the spectral model these are six parameters
needed to predict the three quantities 1, E(X) and T, (Fig.
3).
The chosen approach is summarized in F@ 4. A
similar software package has been developed by Heldler et
al. 15). The six inputs were taken from bourl observations
of the German Meteorological Network (DW ii ) at Stuttgart
(F.R.G.. 49”N, 9”E) a region with about 18OOsunshine hours
a year (40 % of daylight hours). Hours were taken into
was above 4 km, the lobal
account only, if the visibili
horizontal irradiance above Y 0 W/m* and the angle o$*mcidence on the 48” tilted plane less than 85”. As a result of
these restrictions, 40 % of all hours were selected out and
7635 reliable hours from three years were used for our calculations.
Wind Speed
Solar Geometry
Temperature
consists of a clear-sky
approximation
code called SPCTRAU,
a normalization
procedure. and a “cloud-cover modifier” derived from statlstical analysis of measured spectra [2]. SEDESI converts
the four in ut data to a corres onding solar spectral distribution wit*I! a resolution of 11 nm. me overall standard
deviation is ahout 8 % in the visible. up to 15 % in the UV
and 20 % around 940 nm at the water vapor ahsarption
band. Beyond 1100 nm the standard deviation is about 25
%.
Global lrradiance
Diffuse lrradiance
The solar cell’s operating temperature
T of a
module is mainly affected by the plane-of-array (Pbh) irradiance, but also by the ambient tern erature and wind
ed. The model applied lo calculate P of a module was
“p
(eveloped by Fuentes [3]. The model asplied to calculate
the plane-of-array irradmnce 1 was developed by Perez et al.
I’iE. 3:
28-2
The six
Ill;Illcc.
p;trilnlefcK
influencing
W-plant
pcrfor-
Tahlc I:
Long teq hourly data
on global and dilluse
~-.-_-.
irradiance. temperature.
1
Mean,. maximum. minimum and standard
deviatmn of all 7635 ratios q : /‘I.
(cell temperature and total i#li;l”n”c’e fixed).
I )evice
Mean
a-Si:J-1
( We
a-Si:H/a-Si:lJ/
a-SiGe:l I
GIAS
mono-Si
CulnSe
a-Si/Cu 3 nSe2
Calculalion of
Parameters
Influencing
F’OA irradiance I,
POA spectrum E,, (k)
Stand.
llev.
0.05 1
0.035
0.037
0.043
0.025
0.015
0.025
Res .
WIJ th
Al!.!&
520
560
540
580
800
E
Cell temperature Tc
Fig. 4:
The principal components of the scuii-crul)il ic;ll
software package used to con1 ,utc the I’\‘-l>o\vtr
at hourly intervals
over extent 1cd pcriotl.~ OI‘ tinlc’
with measured site-specific niclcorologic:~l tl:~t;~
and technology
plane-of-array).
dependem
ccl1 tlzila (I’( ).I\
RESULTS
Svectraf Ekts
on EtXcienw
ting Conditions (SRC). The fraction r) ~ /rl
describes
what happens when a transition is made t%h SB’C lo contlitions where the spectrum E(X) becomes real, but the other
two influencing parameters T, and I are still fixed at their
standard values.
For all 7635 hours the fractions rl /qs c are plotted
in Fig. 5 against the broadband) inso&lon ! (4x” tilted).
For hi h insolation vaI ues (above 800 W/m*) there are, in
genera,f no clouds, and the zenith angle is less than OCP.
These conditions are close to the AM 1.5 reference
s ctrum conditions. Therefore, the ratio q
/qRC in Fig.
s” approaches one on the right hand side oP’&e kraph. For
low msolation values (below 200 W/m’) overcast skies are
predominant,
Under this condition the relalive spectral
distribution is shifted towards shorter wavelengths (21: as a
result the efficiency of the devices increases. III between ;sre
the situations with partly cloudy skies and / or with :I higher
turbidity than the standard atmosphere.
Fig. 5 shoti sortie characteristic spikes at about NH1
W/m*, 600 W/m*. 400 W/m*. Together with tail’s end at
IO00 W/m* these are the conditions under very clear skies
(no clouds and low turbidity) with high clearness indices K,.
(atmospheric transmission). As we calculate with hourly
averages, a total symmetry around solar nnon, and K, ?s the
main predictor variable for the solar spectral irradlance
mdel SEJXSl, there is a limited number of (1 - K,)- pairs
under these atmospheric conditions. As a result. there are
gaps inbetween these spikes.
Table 1 ives the statistical mean of the ensembles of
Fig. S along wit% the standard deviation from the mean the
maximum .and minimum ratio rlrc@
and the de&e’s
‘0 /n, according to
response width (quantum efficient ks sR!S
Fig. 1). As a general rule the standard deviation decreases
with increasing response width or decreasin band gap. This
is even the case for the a-Si:H/a-Si:H/a-St I3 e:H two-terminal device. An interesting observation is that mismatching
the current of the component cells actuallv increases the fill
factor of the finished device: this effect has been observed
experimentally
as well [6]. This mitigating effect makes
series connection an acceptable design option because the
current mismatch losses are partly compensated for. Thus,
energy delivery for multijunctron and single junction devices
with comparable vssRcare similar.
Fig. 5 and Table 1 give an idea on how im rtant
spectral effects are for the devices investigated. !rpectral
effects become an issue especially for higher band gap
materials where the efficiency can change by more than 20
% for hourly averages because of spectral variations. High
multijunction
devices.
The software package was also used to put the
spectral influence on efficiency into perspective with the
other two influences caused by cell temperature and total
irradjance. For each hour the efficiency under prevailing
cond!tlons (real I, !+$ T,) was calculated. Fig. 6 shows as a
function of total m latlon on the module the deviation
from the efficiency under Standard Reporting Conditions.
hnalysing Fig. 6 and the whole data set as a function of
s ectral and total irradiance and cell temperature we found
t Plat for the low band ap devices mono-Si and CuJnSe2 the
performance is mam
3 y affected by ceil temperature and
total irradiance; for the high band gap devices spectral
effects are important.
Etkiencv
under Critical
Oueration Conditions
Recause standardized terrestrial efficiency measurements are referenced to a fixed set of environmental conditions (SRC), they can only approximate the energy a specific
device would deliver at a site where the temperature. total
and speclral irradiance differ from the refelence conditions.
mono-Si
CuInSe2
a-Si/a-Si/a-Si:(
1.2
I .
'IW
..*.
*
'ISRC
t-
.
:
‘;
..
*4
;
. . . . . ..
0.8
200
400
600
Irradiance
Fig. 5:
800
1000
0
(W/m2)
Influence of the spectrum ‘ort’celt efficiency with
the total irradiance fixed at 1000 W/m* and the
cell temperature fixed at 25 “C. The efficiency
data are normalized to unity at Standard Reporting Conditions (SRC) and are plotted as a
function of plane-of-array total irradiance. Each
dot represents an hourly average of ‘I,:,,, / q%,,,..
Specifying erformance at a given location or environment
in terms o P the average energy pmduced over a given time
period is a desirable alternative to simply using the efficiency at SRC which is really an instantaneous value and is
rarely duplicated in the field. At present, no standards exist
for energy rating methods, although several laboratories are
working on the problem (see [I]),
Our proposal (see also Heidler et al. IS]) is to apply
the simulation technique illustrated in Fig. 4 based on
simulated time series specified by the six parameters of
Pig. 3. Depending on the specifications one can calculate
the average site-specific energy output for a given system
over a given time period.
400
200
600
Irradiance
Table 2:
800
1000
1200
(Wh2)
Critical Operation Periods for the most
important PV-system configurations.
PV-S stem
?
Cn~gyration
Critical Operation Period
Grid-conncctcd. fuel-saving
mode, hydrogen production
The whole year
Peak demand supply
During eak demand
(time o Pday, temperature)
At high temperatures
Remote system for cooling
Remote system with storage
During months with
low irradiance
Pump system for agriculture
During growth time
(time when water needed)
Table 3:
( ‘c~ntlitions (CCIC) normalized by the efficiencies under
averages of the power P and irradiance I over the
from Stuttgart. also give’t;“]n parenthesis is the
Icor
J car
Condition
a-Si:f 1
TF--:1 St:ll/
a-Si:l I/
C’IITC
<inAs
mono-Si
CulnSe2
a-Si:H/
il-SiCk:ll
.-_
Standard Reporting
Conditions
‘ISRC
The whole year q’, ~,
Month with hi@8
efficiency
Month with ]ow&~‘~’
efficiency
Month with low~$“t”’
irradiance
Hour with highe!!&Y”’
temperature
n-hi@ir(h)
10.0 %
13.4 9
._-. 2-terminll
--.-.-.-se
13.1 %
~
0.95
3 the following
12.9 %
12.3 %
16.1 %
0.94
0.94
0.95
3
0.06 (4)
0.96 (3)
0.99
1)
I.00 (1)
0.97 (2)
12)
0.89 ( 12)
0.93 (12)
0.9 I 7)
0.90 (8)
0.94 (8)
I-3
11.89( 12)
0.93 (12)
0.99 12)
0.99 ( 12)
0.95 (12)
0.8X (13)
0.93 ( 13)
0.85 (13)
0.81 (13)
0.89 (13)
Using this approach, different rating methods were
compared for each of the seven devices investigated. The
five normalized mean efficiencies r)* in Table 3 result from
an analysis of the most important PV-system configurations
and the time period when their operation is most critical
(see Table 2): For a remote system the important factor is
the output during the month with lowest irradiance; for the
peak-load applications or remote systems for cooling, temperature very often is the important parameter determining
efficiency. For water pumping systems used in farming the
months during the growing season are of interest.
From Table
25.1 %I
l-----t
0.95
observations
can be
The mean annual operatin efficiency of the
cells under the prevailing cf tmate at Stuttgart
is about 5 % less than expected from SK(:.
The values t)‘..nw, show, however, that for
producing
grid-connected
or
hydrogen
systems the current SRC do not hias the cells
investigated.
The spread between the month with the
lowestLefficiency and the hi hest efficiencies
is onlv 3% for GaAs and ta e four-terminal
tandem, but 10% for CuInSe,. The worst effioccur
in
for
ciencies
summer
lhe
temperature-sensitive
devices of our sample
mono-$
CulnSe, and the four-terminal
tandem.
If the stora e capacity of the system is
designed wtt3 monthly irradiance profiles
these numbers have to be considered very
carefully. While monoSi and CulnSe, perform very well this is totally different 6r the
other devices of our sample,
tl high(I)
It is obvious that the devices being less sensifive to cell temperature
perform relatively
better for periods with htgh ambient temperatures.
No general conclusions should be made about one
technology versus another upon this stud since the results
are critical1 dependent on the modelle J temperature and
irradiance CT
ependence, since the state-of-the-art is rapidly
improving and cost versus performance fi ures have not
been a part of this study. The values of Ta t le 3: however,
clearly demonstrate that the different matertals cause
pronounced differences in response to the meteorological
environment, which must be considered if one optimizes the
economics and reliability of PV-systems.
CONCLUSIONS
The PV-community
is in state of flux concerning
rating methods that are different from the SRC which are
25 “C cell or module tern erature. 1000 W/m2 total
irradiance and the ASTM ii 892 or IEC 904 global
reference spectrum. Before any specific recommendatton on
new rating methods can he made, the following points
should be considered:
1) Better comparison of PV-technologies and vendors
should be possible by a new rating scheme.
2) The new rating scheme will be used for (see also ]7]):
pricing, designtng, sizing, acceptance testing, warranty
discussions. renulatorv iustification, comparison of
initial and iongrterm operating parameters-to identify
system degradation
mechamsms and system problems, forecast of plant output, determinmg capacity
for qualifying facility’s capactty credit.
3) A meaningful rating scheme sets protlucriot~ p~;~lz Ior
the PV-manufacturers.
Furthermore, it enrollr:lvc~~
them to offer application- and site-specific l~~~~~l&.
4) An energy ratin scheme gives incentives to optinli7c
energy instead o$ power.
criteria
I) Rating methods should he developed from :I CIKtomer / user / system designer perspert ive.
2) There should be no technology or applic:lticlrl hi:Is.
3) One simple, fast and accurate ratmg like qnc. IviII
always be needed for research and big11 VOIIIII~~ IWIductton measurements.
Qoen Ouestions
1) Should a single rating method be adopted or SIMIIII~I;I
variety of ratings he adopted for different apl)lic:t[ions?
2) Should the rating(s) he performed at the m:tximum
power point only or should the rating(s) be
erformed as a function of voltage? This is important
Eecause many power trackers or inverters npcrate at
a fixed voltage and must be properly sizctl for IIIC
specific PV-technnlogy.
3) Should the rating be based upon power ;In(l / Or
energy?
4) Should the rating be reproducible by different groups
for the same module or array or should it be cites ecific?
5) J a Power rating is used, should a single reference
suectrum be used or a distribution of soectra river
6)
module itself is not independent of the system. Mismatch and load depend on the system and the module
temperature depends on the module position within
the array / system and the type of mounting.
7) If an ener rating is used, should the reflection losses
on the mo7 ule surface be taken into account?
8) How to include concentrating systems?
9) How much money should be s ent for the devclol~ment of an improved rating met Aod?
The present investi ation has not been ;Ihle to
address nil these questions. w few conclusions can bc tlr;lwn:
I) Five meteorological parameters (global and diffuse
irradiance, ambient temperature. wind speed awl
humidity) are sufficient to determine the nieteorological dnvironment in which a PV-system operates.
) For the new technologies with high b;~ntl g;qx
spectral effects influence the efficiency as much ;IS
cell temperature or total irradiance.
Snectral effects do not cause a-Si multiiunction twot&minal devices to be outperformed bi a-Si singlejunction structures not even for cloudy skies.
If the market will be shared by devices having different response to the five parameters, the numbers
from Table 3 clearly demonstrate the need for energy
rating schemes sup orting system designers to irk
the appropriate tee Rnology for their specific app F~ation and site-specific climate.
From an investigation of the most important PVsystem technologies critical operation conditions can
bc identified. Five efficiency values averaged over diffrrent time eriods are sufficient to determine the
solar ccll’~ e r*ftclency
for the critical operations condi.
lions.
New mctbods have to be developed to address the
clucsliolls in derail. Both, simulation techniques ;III~
long-term outdoor ex eriments are ap ropriate. To improve
:I simulation npprnnc rl based on a so Ptware package bke in
I@. 4 the following research areas should be intensified:
cqw11
I) Resource Assessment: It is still a prohlem to obtain
relinhle data on the solar resource for all the interesting sites.
2) Spectral Modelling:
SEDESI and other spectral
models have IO be verified and improved with data
from different climates.
3) Module temperature Modelling: It should be possible
to oredict the cell’s temoerature for different module
tec’hnologies and sites iore precisely with improved
models.
4) Solar Cell Modelling: The superposition princi le
assumed in our cell model is inappropriate for t If.m
film devices, which are non-linear In temperature and
irradiance.
5) Solar Cell Data: More data are needed for the cell
response as a function of total irradiance and cell
temperature. A statistically significant number of cells
should be measured and accurately modelled.
6) System Modellink: For the time being, the “system” in
our simulations Includes the cell only. A more comrehensive software package has to consider other
rasses. It would include simulation models which have
been developed for module, array and power control
unit losses as well as storage and back-up performance.
Both research centers, the National Renewable
Energy Laboratory
(NREL) and the Centre for Solar
IlnerRy and Hydrogen Research (ZSW) cover these six
areas-with the& co%mon research‘in aider to develop a
better rating scheme for PV-cells. arrays and systems. Right
now, no spe&ic recommendation’abo&what
rating scheme
should be adapted can be given. A more comprehensive
modelling study has to he erformed first, in which all the
different rating methods ( rIke SRC. NOCT. AM/PM and
the six parameters proposed here) will be corn ared fnr a
variety of PV-technologies. A software packa e ike the one
PVp,
presented here allows the performance of various
technologies to be directly compared and forecasted under
identical “real-world” conditions. This complex approach
could result in improved power and energy rating schemes
having :I relatively simple structure.
ACKNOWLEl)CEhlENTS
Illis work was supported by the German Ministry of
Research and Technology under contract numher 0329047A
and the 1J.S. Department of Energy under contract number
IX-ACO2-83CI 110093.
*U.S. -
F'W?X?G OFFICE: 1991-673-798
28-7