Ninth International Conference on Solid State Lighting
Conference 7422 - Proceedings of SPIE Volume 7422
Green LED development in polar and non-polar growth orientation
Christian Wetzel, Mingwei Zhu, Yufeng Li, Wenting Hou, Liang Zhao, Wei Zhao, Shi You,
Christoph Stark, Yong Xia, Michael DiBiccari, and Theeradetch Detchprohm
Future Chips Constellation, Rensselaer Polytechnic Institute, Troy NY 12180 U.S.A.
Department of Physics, Applied Physics and Astronomy, Rensselaer Polytechnic Institute, Troy NY
12180 U.S.A.
ABSTRACT
The green spectral region provides a formidable challenge for energy efficient light emitting diodes. In metal organic
vapor phase epitaxy we developed GaInN/GaN quantum well material suitable for 500 – 580 nm LEDs by rigorous
defect reduction and thrive for alloy uniformity. We achieve best results in homoepitaxy on polar c-plane, and non-polar
a-plane and m-plane bulk GaN. By the choice of crystal orientation, the dipole of piezoelectric polarization in the
quantum wells can be optimized for highest diode efficiency. We report progress towards the goal of reduced efficiency
droop at longer wavelengths.
Keywords: green LED, GaInN-GaN, MOVPE, homoepitaxy, non-polar, efficiency, green gap, threading dislocation
1. THE CHALLENGES OF THE GREEN GAP
Worldwide trade has triggered the unprecedented hope of lifestyle improvements for billions of people. Drastic resource
and energy efficiency gains in all aspects of our daily life are urgently required to avoid an ecological collapse. Artificial
lighting takes a preeminent role in the global savings potential by technological development. The human eye is most
sensitive to light in the green spectral region, yet, state-of-the art artificial lighting is particularly inefficient in this range,
aptly termed the "green gap".
Visible light emitting diodes (LEDs) can provide a spectrally narrow (~30 – 50 nm) band of light emission and
so suppress the wasteful generation of a huge portion unwanted infrared emission in the case of incandescent lamps. The
blue spectral region can most efficiently be generated in AlGaInN-based LEDs,[1,2] while the yellow and red spectral
region currently is best addressed with AlGaInP LEDs.[3] Yet, in between for green, the best-most candidates are
GaInN/GaN heterostructures of the former type even when their efficiency for photon emission still ranges not higher
than some 20%.
For white light generation, the combination of various colors is required which can readily be obtained by the
combination of a narrow-band blue LED and a rather broad band yellow phosphor photoexcited by the same blue
LED.[4] Such approaches meanwhile can achieve efficiencies beyond 100 lm/W [5] and are the workhorse of all solidstate lighting effort. Yet this approach frequently fails in consumer acceptance due to the incomplete spectral emission
pattern of such sources. Similar to compact fluorescence lamps, the absence of green and red spectral portions is poorly
received by the selective customer resulting in only a poor penetration of such lamps in the residential market. Recent
efforts have made big gains by adding stronger red-emitting phosphors to the lamps yet at a significant cost of lamp
power efficiency. These losses are incurred as an inherent problem of the "single-color excitation plus broadband
phosphor emitter" approach. The quantum mechanical nature of the photon requires such Stokes losses and amounts to a
minimum of 20% for green light obtained by down-converting blue light in even a 100% effective phosphor. Realistic
phosphor conversion efficiencies further lower the system performance.
The only feasible alternative is to implement LEDs that directly emit at the desired wavelength, e.g. green
GaInN/GaN LEDs. The direct bandgap of Ga1-xInxN spans the entire visible spectral range and beyond from 370 nm
(GaN) to 1800 nm (InN) and can be expressed in terms of Eg = 3.42 eV (1-x) + 1.89 eV x - b x( 1 – x) for 0 ≤ x ≤ 0.2,
where b = 3.2 eV for luminescence in strained layers and b = 3.8 eV for the absorption bandgap extrapolated to the
relaxed.[6] Such findings have later been expanded to the entire composition range to InN and can be expressed as Eg =
3.42 (1-x) + 0.77 x - 1.43 x(1-x) for the luminescence gap.[7] We find that the actual InN fraction is best determined
using x-ray diffraction of the averaged multiple QW layers well under the condition of pseudomorphic growth.
Composition values in the literature vary by up to a factor of two and should always be dealt with caution.
Quantum wells (QWs) of typically 3 nm Ga1-xInxN embedded in GaN barriers raise the transition energy and
assure a strong wavefunction overlap of heavy holes and electrons for high dipole transition matrix elements and high
radiative recombination rates. Even under strong piezoelectric polarization of the QWs, equivalent to electric field values
of 1 MV/cm, we find that the recombination rate should not drop below half of its value without electric field.[8] In the
case of significantly wider QWs, e.g. 10 nm, a relevant separation of electron and hole wave functions must be expected.
For 470 nm blue LEDs, we typically employ InN alloy fractions up to x = 10% and for 530 nm green and 555
nm deep green LEDs, we do not need more than x = 20%. Yet, there are substantial differences between blue and green
emitting LEDs. While for the shorter wavelength blue LEDs, we find a tremendous resilience towards high densities of
structural defects such as threading dislocations and resulting alloy fluctuations in the QWs, we find that for the green
emitting structures, it is of utmost relevance to reduce the density of structural defects, in particular threading
dislocations and any alloy composition fluctuations in the active light emitting region.[9]
2. AVOIDING DEFECTS
Grown in heteroepitaxy on readily available c-plane sapphire, GaN-based structures always show high densities of
threading dislocations (109 – 1010 cm-2) as initiated in the interface of both materials.[2] Various efforts of complex
growth, processing and re-growth concepts have been developed over time to reduce those densities, yet it remains a
fact, that threading dislocations will penetrate the active QW region at a density of at least 107 cm-2 and most-likely at
significantly higher density. Upon their interaction with the GaInN layers of the QWs, threading dislocations frequently
induce the generation of V-shaped defects (Figure 1) causing higher index facet growth at reduced growth rate (Figure 2)
and resulting pit formation.[10]
Figure 1 Bright field TEM image of 549 nm green LED on sapphire showing the typical formation of Vdefects.[10]
Figure 2 High resolution TEM images of (a) QWs on the sidewall of the V-defect in a highly defective sample;
(b) Normal QW on the c-plane in a sample with low V-defect density. The same holds in the sample of (a) in the
areas between the V-defects.[10]
3. HOMOEPITAXY ON A NATIVE SUBSTRATE
Naturally, the best possible approach to avoid the generation of such detrimental V-defects is the reduction of the density
of threading dislocations all together. Here we make use of the capability of hydride vapor phase epitaxy to produce
thick freestanding bulk GaN templates with a typical density of threading dislocations along the c-axis and growth axis
in the mid 106 cm-2.[11]
We therefore developed epitaxial growth processes that, after appropriate surface preparation, allows the
replication of the low dislocation density crystal quality into the MOVPE grown GaN layers, such as the n-GaN layers in
the 420 nm blue LED shown in Figure 3.[12]
b)
a)
p-layers
growth
direction
MQWs
n-epi GaN
GaN substrate
well
barrier
growth
direction
Figure 3 Cross-sectional TEM (a) and high resolution TEM (b) images of a blue LED homoepitaxially grown on
the c-plane surface of bulk GaN. The low threading dislocation density of the substrate is well maintained and
replicated in the MOVPE grown layers. From TEM micrographs reaching over several µm, a maximum defect
density of 1x108 cm-2 was derived. The high resolution image reveals very clear transitions from the lighter GaN
barrier areas into the darker appearing GaInN well layers.[12]
The availability of such bulk-like GaN boules also allows the growth on one of the side planes of the wurtzite layers. By
growing GaN along the c-axis into layers of several mm thickness, wafer slicing can then be performed along cutting
planes that include the crystal growth axis. In this way, all threading dislocations are limited to the substrate and cannot
propagate as such into the new overgrowth direction. We succeeded in such homoepitaxial MOVPE growth of GaN
along various directions including the non-polar a-axis (Figure 3) and the non-polar m-axis (Figure 4) of wurtzite
GaN.[13, 14]
a)
11 -20
b)
0002
0000
p-layers
growth
direction,
a-axis
p‐GaN
MQW
m
MQWs
GaN substrate
n‐GaN
100 nm
Q1 0 4 0 9 B1 L ED o n a -Ga N.p
Figure 4 Cross-sectional TEM micrographs of (a) an a-axis grown green LED and (b) an m-axis grown LED.
Both are grown homoepitaxially in MOVPE on the side planes of c-axis grown bulk GaN. In larger field scans,
virtually no threading dislocations can be observed penetrating the active QW regions, respectively. Furthermore,
no in-plane defects can be observed in the surface morphology of the resulting layers (< 105 cm-1).[13,14]
This is in contrast to the achieved quality in direct heteroepitaxial growth of a-plane GaN material on r-plane
sapphire. Figure 5 shows a typical case of such heteroepitaxy in cross-sectional TEM. Dislocations densities of the order
of 1010 cm-2 are commonplace.
200 nm
p-layers MQWs
n-GaN
growth
direction,
a-axis
Figure 5 Cross-sectional TEM micrograph of an a-axis grown green MQW sample on r-plane sapphire. By this
continuous growth along a single axis, a high density (~1010 cm-2) of threading dislocations is observed similar to
conventional c-axis growth.[13]
4. DEFECTS IN THE ACTIVE REGION
Epitaxy on even the most perfectly dislocation-free GaN may lead to the generation of misfit and threading dislocations
during the growth of quantum wells. We find that the transition from MOVPE grown GaN to MOVPE grown GaInN
layers can induce dislocation densities of the order of 109 cm-2 (Figure 6).[15] This is in spite of MOVPE
homoepitaxially grown n-GaN layers perfectly replicating the low dislocation density of HVPE-grown GaN substrates
with typical threading dislocation density of 107 cm-2. This propensity is strongly dependent on the achieved emission
wavelength of the QWs. Virtually no new defects may arise in 420 nm material (see Figure 3), while structures of 530
nm may exhibit dislocation densities of the order of ~4x109 cm-2. In recent work we find indications that the generation
of such defects coincides with the relaxation of biaxial strain induced by the QW growth.[15]
[11 2 0]
Figure 6 Cross-sectional TEM micrograph of green LED on bulk GaN taken along
zone axis. Pairs
(1,3,4) and individual branches (2) of inclined dislocations were observed to be generated within the first three
QWs. All of them are tilted a certain angle off [0001], the growth direction.[15]
Nevertheless, we have succeeded in suppressing the generation of V-defects at the intersection of threading
dislocations with QWs to a very large extend in green LEDs up to at least 570 nm. Figure 7 shows the cases of a 545 nm
(a) and 565 nm (b) deep green LEDs on sapphire. While a high density of threading dislocations reaches the QWs from
the n-layers and additional dislocations are initiated in misfit defects in the QWs themselves, no generation of V-defects
and associated disturbance of the QW growth can be observed. Atomic force microscopy of the last barrier of the active
QW region shows substantially reduced surface roughness when V-defect generation can be suppressed (Figure 8).
b)
a)
c
Figure 7 Cross-sectional TEM micrograph of (a) 545 nm and (b) 560 nm deep green LEDs on sapphire. In both
cases, threading dislocations densities in the QWs reach up to some 4x109 cm-2 without generating any V-defects.
a)
b)
1m
1m
Figure 8 Surface morphology in atomic force microscopy of the last barrier in V-defect-free (a) and defective (b)
GaInN/GaN MQW for green LED. The roughness in the better sample is (a) 0.56 nm (RMS) compared to (b) 4.54
nm (RMS) in the defective one.
Light output power in devices show substantial improvements as the number of V-defects can be reduced.
Figure 9 shows the light output versus drive current in 530 nm LEDs in 1 mm diameter scratch diode arrangement.[16]
Over the course of many of our LED fabrication runs, the light output power versus current density proves to be an
excellent indicator of final LED performance after die fabrication in (350 µm)2 devices.
Electro-optical characterization of the resulting 555 nm deep green LEDs on c-plane sapphire after mesa
fabrication is shown in Figure 10 as a function of current density in (350 µm)2 and (700 µm)2 devices.[17] At this stage,
no efforts have been made to enhance the light extraction efficiency by e.g. surface patterning or dielectric
encapsulation. The dominant wavelength shows the familiar blue shift as the current density is increased. At 35 A/cm2,
the typical operation condition of (1 mm)2 350 mA devices (1W lamp), a dominant wavelength of 555 nm is reliably
obtained (Fig. 10 a). The partial light output as measured through the substrate into a 1 cm orifice reaches 5 mW in the
smaller device and 20 mW in the larger one (Fig. 10 b). It is well established, that upon standard measures to enhanced
light extraction efficiency, a factor of three to five times higher output power can be achieved. The external quantum
efficiency (EQE) as measured under the same rudimentary extraction conditions is strongly dependent on current
density. Figure 10 c) shows EQE under pulsed LED operation conditions (pulses of 50 µs duration at 1% duty cycle).
Maximum values are reached at very low current densities, here up to 15%, while at 35 A/cm2, EQE drops to some 11%.
In individual dies, a low current maximum of EQE as high as 40% has been observed, possibly indicating some nonlinearities in either aspect of the radiative recombination, light extraction, or detection scheme.
Partial Light Output Power (mW)
no Vdefects
10
10QWs
5QWs
1
5QWs
10QWs
0.1
with Vdefects
0.01
1E-3
0.01
0.1
1
10
100
2
Current Density (A/cm )
Figure 9 Partial light output versus current density in 1 mm diameter devices of 520 nm LEDs as measured
through the substrate of the device. Samples without V-defects show higher output and output increases with
numbers of QWs. Samples with V-defects show the lower output and output decreases with higher number of
QWs.[16]
560
b)
80
c-axis LED
on sapphire
(350 m)
2
555
550
545
2
(700 m)
540
70
60
c)
20
bare, fabed,
wafer level
(700 m)
2
50
40
30
20
10
2
(350 m)
535
10 100 1000
0
Partial External Quantum Efficiency (%)
565
Partial Light Output Power (mW)
Dominant Wavelength (nm)
a)
10 100 1000
2
Current Density (A/cm )
15
pulsed 50
duty cycle 1%
(700 m)
2
10
5
(350 m)
0
2
10 100 1000
Figure 10 Partial light output versus current density in mesa-fabricated deep green LED devices of (350 µm)2 and
(700 µm)2 size grown in c-plane epitaxy on sapphire . a) The dominant wavelength shows the well-known blue
shift yet still reaches 555 nm at 1W lamp typical 35 A/cm2. b) Under pulsed operation conditions, at partial light
output power of the bare fabricated devices reaches 70 mW in the larger die and 45 mW in the smaller ones. c)
The corresponding external quantum efficiency drops from values of some 15 % to some 3 %.[17]
5. PIEZOELECTRIC POLARIZATION
The uniaxial nature of the wurtzite group-III nitrides in combination with particularities of the strong and partially ionic
bonds lead to strong piezoelectric polarization in those materials.[18,19] Combining layers of different polarization
and/or biaxial strain condition can lead to discontinuities of polarization which can be expressed by planes of sheet
charges of the order of +/- 1014 cm-2 at the interfaces.[20] For the typical case of pseudomorphically strained GaInN well
layer embedded between relaxed layers of GaN barriers, grown along the positive c-axis of GaN, a negative sheet charge
is formed at the interface from barrier to the GaInN well, paired by a positive sheet charge of identical density at the
ending of the well, i.e. at the transition from the well to the GaN barrier (Figure 11).[21] This pair of charges forms an
electrostatic dipole that strongly interacts with the injected non-equilibrium carriers. An immediately apparent
consequence is induction of an effective bandoffset between both GaN barrier layers, given by the energy of the
piezoelectric dipole.[22]
Ga0.88 In0.12/GaN QW
Electron energy (eV)
cb
_
3.4
cb+
e1
3.2
Polarization
dipole
replicates in
well
0.4
0.2
hh 1
vb
_
vb+
0
0
30
c-Axis growth direction (Å)
Figure 11 Realistic electronic bandstructure of a GaInN/GaN quantum well as used in polar c-plane LEDs. The
polarization dipole seen in the bandoffset across the well, apparently translates into a Stokes shift in the quantum
well between absorption edge and peak emission (schematically shown).[21]
This quantity also proves to be a very good approximation of the frequently observed huge Stokes shift between
optical absorption bandgap and the energy of the luminescence peak.[21, 23] This discrepancy can amount to the large
value of 300 meV in green LEDs, i.e. some relevant 15% of the photon energy.[24] A tentative explanation of the
mechanism can be found in the consideration of an electrostatic charging of the local piezoelectric dipole in the instant
of radiative recombination of the pair of screening electron and hole (figure 11). It can be argued that the mechanism of
a well-defined Stokes shift in the recombination process of GaInN/GaN QW structures is a relevant advantage for the
efficient light generation in group-III nitride heterostructures.
In the literature, commonly the case is made that the Stokes shift results from a large non-uniformity of the GaInNalloy, essentially providing laterally separated zones of different bandgap energies. Such an interpretation can find
support in the fact that phase separation in In-rich GaInN alloys is frequently observed in growth processes not
optimized for lateral uniformity. We maintain that all of our above mentioned approaches to reduce the generation of
structural defects include the efforts to reduce spatial non-uniformity of the alloy constituents as far as possible.
Trying to establish the case of the relevance of piezoelectric polarization control in group-III nitrides, this group of
authors have found open ears by virtue of the argument, that in wide QWs in the presence of a large piezoelectric dipole,
the electron and hole wavefunctions may have only a minimal spatial overlap. As a consequence, the dipole transition
matrix element should be reduced, rendering associated recombination processes extremely inefficient. This could
potentially be considered a limiting factor for the observed reduced efficiency of green and deep green GaInN/GaN
LEDs when compared to their blue counterparts. In order to reach the green spectral region, one would either have to
increase the InN-fraction in the QW alloy and by these means increase the piezoelectric dipole, or widen the QW which
also increases the piezoelectric dipole. By the same token, and approximated by the same quantity, the dipole transition
matrix element would be reduced and could so explain the conflicting challenge of the green gap.
6. NON-POLAR GROWTH
To either circumvent the problem of the electron-hole separation within the electric dipole of a QW or at least to reduce
its magnitude and impact, epitaxial growth of QW structures along crystallographic axes of GaN with either reduced or
absent piezoelectric polarization is highly desirable.[25] Figure 12 gives a schematic of the polar c-plane, and non-polar
m-plane and a-plane of the wurtzite structure.[26] Shown also are the respective bandedge line-ups in QWs grown on
such planes resulting in either fast or slow recombination processes.
a)
Bandstructure
P
CBE
h
slow
CBE
VBE
E
30
0
Growth (Å)
e)
h
fast
VBE
E
0
c)
CBE
h
fast
VBE
E
30
0
Growth (Å)
30
Growth (Å)
Polarization only in-plane
b)
d)
c
f)
m
Crystal
planes P
z
c-plane
y
m-plane
P
a-plane
a
P
x
Figure 12 Schematic of the bandstructure of GaInN/GaN quantum wells grown along polar and non-polar crystal
axes (a, c, e) and the associated hexagonal crystal planes (b, d, f). In polar c-plane growth (b), radiative
recombination may be reduced due a lateral offset of the charges (a). Higher efficiency, particularly in the green
can be expected from non-polar m-plane (d) and a-plane (f) growth.[26]
The absence of a piezoelectric dipole in QW structures grown along non-polar axes has significant ramifications
also for the achievable emission wavelengths in LEDs. Figure 13 shows the photoluminescence spectra of a c-axis
grown polar and m-axis grown non-polar multiple QW LEDs. Both structures have been grown simultaneously under
identical offered growth conditions.[26] For this, a template of c-plane GaN on sapphire has been placed next to an mplane GaN template on bulk GaN. While the polar structure reveals a peak emission wavelength as long as 558 nm, the
non-polar one reaches a mere 488 nm. This corresponds to a large discrepancy of 320 meV. From our earlier analysis of
the Stokes shift in MQW samples leading to the interpretation of the piezoelectric dipole,[27] we extrapolate a dipole
energy of 250 meV for samples of the present case. Apparently, the observed discrepancy of emission wavelengths in
polar and non-polar structures can well be accounted for by the expected quantity of the piezoelectric dipole energy in
polar structures of such type.
It should be noted that due to the different thermal conductivity of both sample structures, different In incorporation
rates into both structures are conceivable. The same experiment performed with two polar sample structures, namely cplane GaN templates on c-plane bulk GaN and on c-plane sapphire result in peak wavelength variations by not more than
20 nm.[28] Such a component can be a contribution in the remaining discrepancy of some 70 meV.
In our interpretation, the shorter emission wavelength in such non-polar structures therefore should be a direct
consequence of the absence of the piezoelectric dipole controlling the radiative emission process in the c-axis polar
structures.[26] This observation provides supportive evidence of our related statement that the optoelectronic properties
of the group-III nitride structures are predominantly controlled by the piezoelectric properties of the material.[29]
MQW LED
T = 300 K
558 nm
Polar LED
on
c-GaN
30
Non-polar LED
on
m-GaN
20
488 nm
10
Polarization
Dipole?
p052507b.o Q10284A1&A2
CL Intensity (arb. u.)
40
0
400
500
600
Wavelength (nm)
Figure 13 The luminescence spectra of simultaneously grown polar c-plane and non-polar m-plane LEDs. We
attribute the much longer emission wavelength in the c-plane structure to its large polarization dipole. In the nonpolar structures, higher In concentrations or wider well widths are needed to reach the same green spectrum.[26]
Under high optical excitation density, non-polar m-plane and polar c-plane LEDs also show substantially different
behavior. Figure 14 shows photoluminescence spectra under 337 nm pulsed excitation. While the polar c-plane material
shows the well-known blue shift of its maximum with increasing excitation level (Fig. 14 a), the non-polar sample
maintains its peak wavelength at 488 nm fixed up to the highest levels of excitation employed (Fig. 14 b). This blue shift
in the polar structure is commonly ascribed to the quantum confined Stark effect that is gradually being screened by high
densities of photo-generated carrier pairs. Upon detailed inspection, however, it becomes apparent, that the longer
wavelength peak at 560 nm does not disappear, but it merely loses in relative intensity and the spectrum becomes
dominated by the peak near 510 nm at higher excitation. This peak also will be the one that eventually shows stimulated
emission under sufficient high excitation. This is in agreement with other detailed reports in the literature.[30]
b)
c-axis M Q W
LED
Excita tio n 3 3 7 n m
Ed g e e m issio n
T = 300 K
m -axis M Q W
LED
E xc. d e n sity
(m J/cm 2)
1000
517 –
502 nm
2.9
2.3
1.5
100
0.3
2 .9
1 .5 4
0 .9 1
10
400
560 nm
488 nm
0 .0 3
2 .9
500
600
400
500
p040 708a Q10284a 1m- 2 cL ED
PL Intensity (cts/ms)
a)
600
Wavelength (nm)
Figure 14 Photoluminescence spectra 560 nm deep green polar c-plane (a) and 488 nm blue-green non-polar mplane (b) LEDs under variable high power optical excitation. In the standard polar case a significant blue shift of
the dominant peak is frequently interpreted. Upon closer inspection it becomes apparent that the line at 560
maintains its wavelength, while a second peak arises near 510 nm that further shifts to shorter wavelengths and
dominates the spectrum under high excitation. None of this shifting and intensity exchange is seen in the non-polar
case. Here, the dominant peak remains fixed in wavelength.[26]
An alternate explanation for the strongly differing peak wavelengths between polar and non-polar structures has
been given by proposing a substantially lower In incorporation rate on the non-polar m-growth surface when compared
to regular polar c-plane growth.[31] To verify this possibility, we analyzed the non-polar structures in reciprocal space
mapping by x-ray diffraction to derive the quantum well region's average alloy composition. By help of additional crosssectional TEM, actual thicknesses of wells and barriers were determined to separate the region's alloy average into an
actual alloy value for the well. While for c-plane grown structures commercial software is generally available to derive
the alloy compositions, no such is available for the various other growth axes of the wurtzite system.
The reciprocal space map around the (21-30) and (20-22) lattice diffractions for various m-plane structures is shown
in Figure 15.[32] Under the reasonable assumptions of pseudomorphic growth within the growth plane and absence of
shear stress, the growth axis may be assumed to remain strain free.
(2022)
(2130)
+1st
+1st
m-GaN
-
m-GaN
m-GaInN/GaN
0th
m-GaInN/GaN
0th
-1st
Lattice period along
a axis
--1st
Lattice period along
c axis
Figure 15 Reciprocal space map of m-plane GaInN/GaN QW structures revealing pseudomorphic growth along
the a-axis and c-axis. The QW-averaged alloy composition is determined from the separation of the 0th-order alloy
peak from that of GaN.[32]
By help of this analysis, the alloy composition on the wells was established as summarized in Figure 16 versus the
achieved photoluminescence peak wavelength. Unlike the c-plane polar material, in this non-polar case, a well width
beyond 3 nm can be used to increase the emission wavelength by a reduced quantum size effect. In the polar case, this
would have lead to reduced wavefunction overlap of electron and hole and rapidly decreased the radiative recombination
rate. Just like in the case of polar c-axis growth, the growth of thicker wells, however, is complicated by the GaInN
alloy's tendency to segregate In into InN-rich regions effectively spoiling the well's homogeneity. Substantial effort in
development of growth processes therefore is necessary to achieve well widths as large as 6.7 nm as shown here. In such
structures, with as low an InN fraction as 10%, luminescence peak emission wavelengths as long as 500 nm and 580 nm
have been achieved. For comparison, published data by Tsuda et al.[33] is included requiring some 17% of InN to
achieve 470 nm blue emission in a 3 nm well.
m-GaInN/GaN
MQWs
well width
(nm)
~ 5.4
~ 4.8
Tsuda et al.
APEX 1, 011104 (2008)
Figure 16 InN-fractions x of m-plane Ga1-xInxN/GaN MQW structures emitting at the indicated
photoluminescence wavelength. The well width is a parameter. Apparently, emission wavelengths as long as 500 - 580 nm can be achieved with as little as 10% of InN using a 7 nm well.[32]
The same analysis was performed for a-plane grown structures. Figure 17 shows the reciprocal space mapping for a
range of a-axis grown QWs and LEDs. The alloy composition averaged over QW and barriers is determined from the
separation of the 0th-order alloy peak from that of GaN. By help of additional cross-sectional TEM, the well widths and
barrier widths are determined to conclude on the actual composition of the well. In this, the absence of sheer strain is
reasonably assumed resulting in a strain free growth axis, the a-axis.
(2020)Space Map
Reciprocal
a-GaN
a-GaInN/GaN
0 th
--1 st
Lattice period along m axis
(1122)
+1stst
a-GaN
th
st
a-GaInN/GaN -1
th
-2nd
th
0 th
th
-3rd
Lattice period along c axis
Figure 17 Reciprocal space map of a-plane GaInN/GaN QW structures revealing pseudomorphic growth along
the m-axis and c-axis. The QW-averaged alloy composition is determined from the separation of the 0th-order
alloy peak from that of GaN.[32]
The luminescence peak wavelength as a function of so derived well alloy composition is shown in Figure 18.
The well thickness is listed as a parameter. Available literature data is included in the graph. A good proportionality
between InN-fraction and peak wavelength and a general agreement with the literature data is seen. Interestingly, also in
this case, deep green emission as long as 550 nm can be achieved with a mere 15% of InN in a 5.4 nm thick QW.
PL Peak Wavelength (nm)
600
a-GaInN/GaN
MQWs
550
5.4
well width
(nm)
2.7
500
3.7
3.1-3.5
4.1
450
Chakraborty
et al.
Chakraborty
et al.
400
0
5
10
15
20
InN Fraction (%)
Figure 18 InN-fractions x of a-plane Ga1-xInxN/GaN MQW structures emitting at the indicated photoluminescence
wavelength. The well width is a parameter. Emission wavelengths as long as 550 nm can be achieved with 15% of
InN using a 5.4 nm well. Included also is data from the literature, for which well widths are not known. Overall a
linear trend can be seen.[32]
7. DEVICE PERFORMANCE
550
on a-GaN
1
on r-sapp.
0.1
0.01
400
500
600
Wavelength (nm)
Dominant Wavel ength (nm)
EL Intensity (W/nm)
a) a-plane LED
540
b)
c)
on r-sapp.
530
520
on a-GaN
510
500
0
5
10
Partial Light Output Power ( W)
10
15
150
on a-GaN
100
50
on r-sapp.
0
0
5
10
15
Partial External Quantum Efficiency (%)
LEDs emitting in green spectral region have been achieved, both in m-axis and a-axis growth on the bulk GaN template.
Figure 19 summarizes device performance of a-plane LEDs in a direct comparison of an a-axis LED grown on an a-GaN
template on r-plane sapphire and one on an a-GaN template on a-plane bulk GaN. The spectral performance is shown in
Figure 19 a) reaching 3 µW/nm in the bulk-based sample. The dominant wavelength reaches up to 5350 nm in the bulk
based sample and interestingly increases slightly with current density, which is opposite to the performance of c-axis
polar material (Fig 19 b). The light output power reaches some 160 µW in the same sample, three times as high as the
sapphire based one (Fig 19 c). The external quantum efficiency in these bare dies, as measured through the substrate
side, are still quite low and actually do drop with current density (Fig. 19 d). A contribution in this drop is certainly die
heating due to the low electrical conductivity of the substrate used here as an n-layer and possibly also due to early
development stages of p-type doping in a-axis growth.
0.25
d)
0.20
0.15
on a-GaN
0.10
0.05
on r-sapp.
0.00
0
5
10
15
2
Current Density(A/cm )
Figure 19 Electro-optical performance of a-axis grown LEDs on a-plane bulk GaN in comparison with r-plane
sapphire. Device area is 1 mm diameter. Spectral performance (a), light output power (c) and external quantum
efficiency as measured through the substrate side (d) show the superiority of the bulk-based sample. The dominant
emission wavelength shows slightly higher values for the sapphire-based sample. Interestingly, the wavelength
increases slightly with current.[13]
Electro-optical performance of a blue-green m-plane based LED is summarized in Figure 20. The polar c-plane
structure shows the well-known blue sift with increasing current density (Fig. 20 a+c), while the non-polar m-plane one
shows fixed emission wavelength, here ~490 nm (Fig. 20 b+c). Light output power as measured on bare fabricated but
unencapsulated (350 µm)2 dies through the substrate shows comparable achievement levels of several mW at 35 A/cm2.
1000
c)
m-plane
c-plane
Spectral Power (W/nm)
b)
2
j (A/cm )
100
2
100
j (A/cm )
38
63
10
10
1
1
0.1
0.1
0.3
p011509a.o
0.01
400
450
0.1
500
550
0.01
600
400
d)
520
LED epi
on GaN
c-plane
10
515
1
510
c-plane
505
0.1
m-plane
500
m-plane
495
0.01
490
p011809b.c
450
500
Light Output Power (mW)
a)
Dominant Wavelength (nm)
1000
550
600
0.1
1
10
100 1000 0.1
1
10
p011509a.o
1E-3
100 1000
2
Current Density (A/cm )
Wavelength (nm)
Figure 20 Electro-optical performance of m-plane LEDs on m-plane bulk GaN in comparison with c-plane LEDs
on c-plane bulk GaN. Fabricated dies are (350 µm)2 is size. Spectral performance of the polar (a) and non-polar
(b) structures show the stable emission wavelength in the non-polar m-plane one. c) Dominant wavelengths versus
current density. d) Light output power as measured through the substrate reaches values of several mW at 1-W
lamp typical current densities of 35 A/cm2.[26]
8. CONCLUSIONS
Defect reduction and thrive for utmost alloy uniformity prove to be successful guidelines for the achievement of
GaInN/GaN LEDs spanning the green spectral region. Reduction of V-defect decoration of existing threading
dislocations, reduction of total number for threading dislocations by homoepitaxy on bulk GaN, replication of the bulk
GaN quality in the epitaxial layers are the necessary steps towards this goal. Upon the availability of thick GaN bulk
pieces, slices of arbitrary crystal orientation can be prepared[11] and utilized as growth templates for homoepitaxy. By
tilting the epitaxial growth direction away from the original substrate growth direction, avoidance of all original
threading dislocations becomes possible. By careful optimization of growth conditions, it is possible to avoid the
generation of new defects including line defects. We demonstrate achievement of blue-green and green emitting LEDs
homoepitaxially grown along non-polar a-axis and m-axis of GaN. We argue that the utilization of non-polar growth
directions eliminates the help of the piezoelectric dipole in the Stokes shift of the emission wavelength seen in polar
growth orientation. Achievement of similar emission wavelength therefore requires higher InN fractions of the QWs or a
reduction of the quantum size effect by the use of wider wells. In absence of this polarization dipole, emission
wavelength proves widely unaffected by variation of LED drive current. Achievement of those milestones demonstrate
our progress towards closing the green gap of subpar light output efficiency in the green and deep green spectral region.
ACKNOWLEDGEMENTS
The authors thank E.A. Preble, T. Paskova, and D. Hanser at Kyma Technologies for providing the bulk substrate
material. This work was supported by a DOE/NETL Solid-State Lighting Contract of Directed Research under DEFC26-06NT42860 (Brian Dotson). This work was supported by the National Science Foundation (NSF) Smart Lighting
Engineering Research Center (# EEC-0812056).
1
2
Isamu Akasaki and Hiroshi Amano, "Crystal Growth and Conductivity Control of Group III Nitride Semiconductors
and Their Application to Short Wavelength Light Emitters," Jpn. J. Appl. Phys. Pt 1 36(9a), 5393–408 (1997).
Isamu Akasaki and Hiroshi Amano, "Breakthroughs in Improving Crystal Quality of GaN and Invention of the p–n
Junction Blue-Light-Emitting Diode," Jpn. J. Appl. Phys. 45(12), 9001–10 (2006).
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
G. Chen, M. Craven, A. Kim, A. Munkholm, S. Watanabe, M. Camras, W. Götz, and F. Steranka, "Performance of
high-power III-nitride light emitting diodes," phys. stat. sol. (a) 205(5), 1086–1092 (2008).
P. Schlotter, R. Schmidt, and J. Schneider, "Luminescence conversion of blue light emitting diodes," J. Appl. Phys.
64, 417 (1997).
U.S. Department of Energy, Solid-State Lighting Program, http://www.netl.-doe.gov/ssl.
C. Wetzel, T. Takeuchi, S. Yamaguchi, H. Katoh, H. Amano, and I. Akasaki, “Optical Band Gap in Ga1-xInxN
(0<x<0.2) on GaN by Photoreflection Spectroscopy,” Appl. Phys. Lett. 73(14), 1994-6 (1998).
J. Wu, W. Walukiewicz, K. M. Yu, J. W. Ager III, E. E. Haller, Hai Lu, and William J. Schaff, “Small Bandgap
Bowing in InxGa1-xN Alloys”, Appl. Phys. Lett. 80, 4741 (2002).
C. Wetzel, S. Kamiyama, H. Amano, I. Akasaki, “Optical Absorption in Polarized Ga1-xInxN/GaN Quantum Wells,”
Jpn. J. Appl. Phys. Part 1, 41(1), 11-4 (2002).
C. Wetzel, T. Salagaj, T. Detchprohm, P. Li, and J.S. Nelson, “GaInN/GaN Growth Optimization for High Power
Green Light Emitting Diodes,” Appl. Phys. Lett. 85(6), 866-8 (2004).
M. Zhu, T. Detchprohm, S. You, Y. Wang, Y. Xia, W. Zhao, Y. Li, J. Senawiratne, Z. Zhang, and C. Wetzel, "Vdefect Analysis in Green and Deep Green Light Emitting Diode Structures," Phys. Stat. Sol., 5(6), 1777 – 1779,
(2008).
A. D. Hanser, L. Liu, E. A. Preble, D. Tsvetkov, M. Tutor, N. M. Williams, K. Evans, Y. Zhou, D. Wang, C. Ahyi,
C. C. Tin, J. Williams, M. Park, D. F. Storm, D. S. Katzer, S. C. Binari, J. A. Roussos, and J. A. Mittereder, "An
Overview of Gallium Nitride Substrate Materials Developments for Optoelectronic and Microelectronic
Applications", CS MANTECH Conference, Vancouver, British Columbia, Canada, 2006.
M. Zhu, Y. Xia, W. Zhao, Y. Li, J. Senawiratne, T. Detchprohm, and C. Wetzel, "Structural Characterization of
Homoepitaxial Blue GaInN/GaN Light-Emitting Diodes by Transmission Electron Microscopy," J. Electron. Mater.
37(5), 641-645 (2008).
Theeradetch Detchprohm, Mingwei Zhu, Yufeng Li, Yong Xia, Christian Wetzel, Edward A. Preble, Lianghong
Liu, Tanya Paskova, and Drew Hanser, “Green Light Emitting Diodes on a-Plane GaN Bulk Substrates,” Appl.
Phys. Lett. 92, 24119 (2008).
Mingwei Zhu, Theeradetch Detchprohm, Shi You, Wei Zhao, Wenting Hou, Yufeng Li, Yong Xia, Liang Zhao,
Stephanie Tomasulo, Tanya Paskova, Edward A. Preble, Drew Hanser, Christian Wetzel, "Blue-Green GaInN/GaN
Light Emitting Diode on Non-Polar m-Plane Bulk GaN", Electronic Materials Conference, Pennsylvania State
University, June 24 - 26, 2009.
Mingwei Zhu, Theeradetch Detchprohm, Yong Xia, Wei Zhao, Yufeng Li, Jayantha Senawiratne, Shi You,
Lianghong Liu, Edward A. Preble, Drew Hanser, Christian Wetzel, “Structural Analysis in Low-V-defect Blue and
Green GaInN/GaN Light Emitting Diodes,” in Nitrides and Related Bulk Materials, edited by R. Kniep, F.J.
DiSalvo, R. Riedel, Z. Fisk, and Y. Sugahara (Mater. Res. Soc. Symp. Proc. Volume 1040E, Warrendale, PA,
2008), 1040-Q03-02.
T. Detchprohm, M. Zhu, Y. Xia, Y. Li, W. Zhao, J. Senawiratne, and C. Wetzel, "Improved Performance of GaInN
Based Deep Green Light Emitting Diodes through V-Defect Reduction," Phys. Stat. Sol. (c) 5(6), 2207 – 2209,
(2008).
C. Wetzel, Yufeng Li, J. Senawiratne, Mingwei Zhu, Yong Xia, S. Tomasulo, P.D. Persans, Lianghong Liu, D.
Hanser, and T. Detchprohm, “Characterization of GaInN/GaN layers for green emitting laser diodes,” J. Cryst.
Growth 311, 2942–2947 (2009).
A. Bykhovski, B. Gelmont, and M. Shur, "The influence of the strain-induced electric field on the charge
distribution in GaN-AlN-GaN structure," J. Appl. Phys. 74, 6734 (1993).
F. Bernardini, V. Fiorentini, and D. Vanderbilt, ''Spontaneous polarization and piezoelectric constants of III-V
nitrides," Phys. Rev. B 56, R10024 (1997).
C. Wetzel, T. Takeuchi, H. Amano, and I. Akasaki, "Piezoelectric Franz-Keldysh Effect in Strained GaInN/GaN
Heterostructures," J. Appl. Phys. 85(7), 3786-91 (1999).
Christian Wetzel and Theeradetch Detchprohm, "RPI starts to extinguish the green gap," Compound Semiconductor,
15(1), 21 – 23, 2009.
C. Wetzel, T. Takeuchi, H. Amano, and I. Akasaki, “Electric Field Strength, Polarization Dipole, and MultiInterface Bandoffset in GaInN/GaN Quantum Well Structures,” Phys. Rev. B 61, 2159-63 (2000).
C. Wetzel, M. Kasumi, T. Detchprohm, T. Takeuchi, H. Amano, and I. Akasaki, “Discrete Stark-Like Ladder in
Piezoelectric GaInN/GaN Quantum Wells,” Phys. Stat. Sol. (b) 216, 399-403 (1999).
24 C. Wetzel, T. Takeuchi, H. Amano, and I. Akasaki, “Quantized States in Ga1-xInxN/GaN Heterostructures and the
Model of Polarized Homogeneous Quantum Wells,” Phys. Rev. B 62(20), R13302-5 (2000).
25 T. Takeuchi, H. Amano, and I. Akasaki, "Theoretical Study of Orientation Dependenence of Piezoelectric Effects in
Wurtzite Strained GaInN/GaN Heterostructures and Quantum Wells", Jpn. J. Appl. Phys., Part 1 39, 413 (2000).
26 C. Wetzel, M. Zhu, J. Senawiratne, T. Detchprohm, P.D. Persans, L. Liu, E. A. Preble, and D. Hanser, "Light
Emitting Diode Development on Polar and Non-Polar GaN Substrates," J. Cryst. Growth 310, 3987-91 (2008).
27 C. Wetzel, T. Takeuchi, H. Amano, and I. Akasaki, “Electric Field Strength, Polarization Dipole, and MultiInterface Bandoffset in GaInN/GaN Quantum Well Structures,” Phys. Rev. B 61, 2159-63 (2000).
28 M. Zhu, Y. Xia, W. Zhao, Y. Li, J. Senawiratne, T. Detchprohm, and C. Wetzel, "Structural Characterization of
Homoepitaxial Blue GaInN/GaN Light-Emitting Diodes by Transmission Electron Microscopy," J. Electron. Mater.
37(5), 641-645 (2008).
29 C. Wetzel, Rump Session, The Second International Conference on Nitride Semiconductors - ICNS 97, 27-31
October 1997, Tokushima, Japan.
30 e.g., poster presentations by Hewlett Packard Research Labs, Yokohama, Japan, presented at The Second
International Conference on Nitride Semiconductors - ICNS 97, 27-31 October 1997, Tokushima, Japan.
31 A. Chakraborty, S. Keller, C. Meier, B.A. Haskell, S. Keller, P. Waltereit, S.P. DenBaars, S. Nakamura, J.S. Speck,
U.K. Mishra, "Properties of nonpolar a-plane InGaN/GaN multiple quantum wells grown on lateral epitaxially
overgrown a-plane GaN," Appl. Phys. Lett. 86 (2005) 031901.
32 T. Detchprohm, M. Zhu, Y. Li, Y. Xia, L. Liu, D. Hanser, and C. Wetzel, “Growth and Characterization of Green
GaInN-Based Light Emitting Diodes on Free-Standing Non-Polar GaN Templates,” J. Cryst. Growth 311, 2937–
2941 (2009).
33 Y. Tsuda, M. Ohta, P.O. Vaccaro, S. Ito, S. Hirukawa, Y. Kawaguchi, Y. Fujishiro, Y. Takahira, Y. Ueta, T.
Takakura, T. Yuasa, "Blue Laser Diodes Fabricated on m-Plane GaN Substrates," Appl. Phys. Express 1 (2008)
011104.