Reliable InP-based Geiger-mode Avalanche Photodiode
Arrays
The MIT Faculty has made this article openly available. Please share
how this access benefits you. Your story matters.
Citation
Smith, Gary M. et al. “Reliable InP-based Geiger-mode
avalanche photodiode arrays.” Advanced Photon Counting
Techniques III. Ed. Mark A. Itzler & Joe C. Campbell. Orlando,
FL, USA: SPIE, 2009. 73200R-10. © 2009 SPIE
As Published
http://dx.doi.org/10.1117/12.819126
Publisher
Society of Photo-optical Instrumentation Engineers
Version
Final published version
Accessed
Thu May 26 07:55:43 EDT 2016
Citable Link
http://hdl.handle.net/1721.1/52685
Terms of Use
Article is made available in accordance with the publisher's policy
and may be subject to US copyright law. Please refer to the
publisher's site for terms of use.
Detailed Terms
Reliable InP-based Geiger-mode Avalanche Photodiode Arrays
Gary M. Smith, K. Alex McIntosh, Joseph P. Donnelly, Joseph E. Funk,
Leonard J. Mahoney, and Simon Verghese
MIT Lincoln Laboratory, 244 Wood Street, Lexington, MA USA 02402
ABSTRACT
Arrays as large as 256 x 64 of single-photon counting avalanche photodiodes have been developed for defense
applications in free-space communication and laser radar. Focal plane arrays (FPAs) sensitive to both 1.06 and 1.55 μm
wavelength have been fabricated for these applications. At 240 K and 4 V overbias, the dark count rate (DCR) of 15 μm
diameter devices is typically 250 Hz for 1.06 μm sensitive APDs and 1 kHz for 1.55 μm APDs. Photon detection
efficiencies (PDE) at 4 V overbias are about 45% for both types of APDs. Accounting for microlens losses, the full FPA
has a PDE of 30%. The reset time needed for a pixel to avoid afterpulsing at 240 K is about 3-4 μsec. These devices
have been used by system groups at Lincoln Laboratory and other defense contractors for building operational systems.
For these fielded systems the device reliability is a strong concern. Individual APDs as well as full arrays have been run
for over 1000 hrs of accelerated testing to verify their stability. The reliability of these GM-APDs is shown to be under
10 FITs at operating temperatures of 250 K, which also corresponds to an MTTF of 17,100 yrs.
Keywords: single photon detector, photon imager, Geiger-mode, avalanche photodiodes, reliability
1. INTRODUCTION
Single photon sensitive detectors and arrays have substantial applications in laser radar and free-space optical
communications as well as other uses. Devices sensitive to near-infrared (NIR) photons are not as well developed as
those made of silicon and sensitive to visible photons. NIR sensitive arrays can be fabricated in the InP-based material
system and the wavelength sensitivity can be tailored by changing the bandgap of the InGaAsP absorber layer while
maintaining a lattice-match to the InP substrate. Unlike linear-mode avalanche photodiodes (APDs), Geiger-Mode (GM)
APDs are temporarily biased above their breakdown voltage so that a single absorbed photon results in a large current
pulse. These current pulses are sufficiently large to switch CMOS digital logic, hence allowing arrays of these devices to
be directly bonded to readout integrated circuits (ROICs).
This paper describes the reliability of GM-APD arrays. Unlike linear-mode APDs that have low-level currents, GMAPDs can sustain current pulses of many milliamperes through small-diameter devices (15 μm) resulting in current
densities of over 5000 A/cm2 and corresponding power densities of over 300 kW/cm2. Even with very short pulse widths
(~2-20 ns) these currents can lead to degradation of GM-APDs over time if the device is not properly designed and
fabricated. We have previously demonstrated that mesa-etched GM-APDs can be made reliable with proper passivation
and care during fabrication [1,2]. In this paper detailed reliability studies are used to establish a failure rate model and
determine the fitting parameters to extrapolate failure rates at operating conditions.
2. DEVICE DESIGN AND ARRAY FABRICATION
Many of the details of the design and fabrication of GM-APD arrays have been published previously [1,2]. Devices
described here use a separate absorption and multiplication region (SAM) structure to absorb the photons in a narrow
bandgap material then multiply the carriers in a larger bandgap avalanche region (see Fig. 1). A highly doped field-stop
layer is used between the absorber and avalanche region to control the electric field in various parts of the device.
Depending on the intended application, the absorber is either lattice-matched InGaAsP (1.06 μm) or InGaAs (1.5 μm).
The APD mesa is defined by etching through the epitaxial layers into the highly p-doped InP contact layer. Edge
passivation is then applied and metal contacts deposited. The edge passivation process is critical to the reliability of the
devices. Polyimide coated with silicon nitride has been found to be a very stable passivation for these devices [2].
Advanced Photon Counting Techniques III, edited by Mark A. Itzler, Joe C. Campbell,
Proc. of SPIE Vol. 7320, 73200R · © 2009 SPIE · CCC code: 0277-786X/09/$18 · doi: 10.1117/12.819126
Proc. of SPIE Vol. 7320 73200R-1
Downloaded from SPIE Digital Library on 15 Mar 2010 to 18.51.1.125. Terms of Use: http://spiedl.org/terms
n-Contact
Edge Passivation
10-nm n+ InGaAs
Contact
n+ InP
≈ 4 μm
p-Contact
n-
InGaAsP Absorber
n- InP Avalanche
n+ InP
Field Stop
p+ InP
n InP substrate
AR Coating
Back Illumination
Fig. 1. Schematic of a mesa-etched InP-based Geiger-mode APD.
7.5 mm
Fig. 2. Hybridized 128x32 GM-APD array.
Indium bumps are deposited and patterned on top of the mesas for bonding to the ROIC. The substrate is then thinned
and an anti-reflective coating is applied to the back side. After arrays are cut from a fabricated wafer, they are
individually flip-chip bonded to ROIC die and epoxied together. Finally microlenses are actively aligned to the array and
epoxied in place. A picture of a hybridized 128x32 array is shown in Fig. 2.
3. ARRAY AGING AND FAILURE RATE DETERMINATION
Lincoln Laboratory has established the capability for simultaneous accelerated aging of 120 GM-APDs in three different
aging chambers. The DC and AC overbias voltage can be independently controlled on each device while the pulse width,
repetition rate, and temperature are variable for each chamber. These tools have been used to characterize a large number
of arrays from many fabrication lots. Analyzing these results has led to the development of a failure rate model that fits
the data and allows extrapolation of expected failure rates for fielded devices. Some insight into the degradation
mechanics is also described in this section.
Proc. of SPIE Vol. 7320 73200R-2
Downloaded from SPIE Digital Library on 15 Mar 2010 to 18.51.1.125. Terms of Use: http://spiedl.org/terms
10000
Sam ple A
30 um Diam eter APD
Polyim ide Passivated
9000
8000
1000+ Hrs, GM Aging
4 V Overbias, 23 C
300 nsec pulse, 1 kHz
DCR (kHz)
7000
6000
5000
0-1000 Hrs, Linear-mode Aging
45 V, 85 C
2 V Below Breakdow n
4000
3000
2000
1000
0
0
200
400
600
Hours
800
1000
1200
Fig. 3. Dark count rate as a function of time while aging several GM-APDs. The first 1000 hrs were aged in linear-mode at
85 C and no degradation was seen. Post-1000 hrs devices were aged in Geiger-mode and degradation was seen within
several hours.
3.1 Linear-mode versus Geiger-mode aging
Avalanche photodiodes used for telecommunications applications have been shown to be very reliable [3-5]. However,
these devices are aged below breakdown in the linear-mode regime and have fairly low currents flowing through them.
Conversely, Geiger-mode APDs are DC biased just below breakdown and AC pulsed over breakdown for a short period
of time. High overbias voltages can generate several mA of current through very small device diameters that can cause
tremendous localized stress. To draw a direct comparison between aging in below-breakdown linear-mode, and Geigermode, several GM-APDs were aged for 1000 hrs at 2 V below breakdown in an 85 C environment and periodically
tested in Geiger-mode to determine the dark count rate (DCR) [Fig. 3]. Then the same devices were Geiger-mode aged
with a DC bias of 2 V below breakdown (the same as linear mode) but with a moderate 4 V AC overbias for 300 nsec at
1 kHz in a 23 C environment. Within hours the DCR rapidly increases and does not recover. The stark contrast of the
degradation of APDs in Geiger-mode where the field in the APD is only slightly higher than it is in linear-mode suggests
that the degradation mechanism is not one accelerated by the field magnitude. Follow-up experiments with varied
overbias, pulse width, and repetition rate pointed to the charge flowing through the device (i.e. current) as the
accelerating mechanism. Surprisingly, in another experiment, DC current was found to not accelerate failures;
degradation is only seen when operated in Geiger-mode. A group of GM-APDs that were from a fabrication lot known
to degrade under Geiger-mode aging were operated with DC currents of 10’s of uA with no signs of degradation. Hence,
all aging must be done in Geiger-mode, which somewhat complicates the aging system design since both AC and DC
sources must be coupled to each channel.
3.2 Aging of Geiger-mode APD arrays
The majority of the controlled aging done at MIT-LL has been accelerated aging. This aging is done by mounting GMAPD arrays to fan-out boards then mounting the fan-out board in a package and wirebonding a sampling of the APDs for
subsequent testing (up to 40 in each round of testing). The GM-APDs are then operated at various conditions of overbias
voltage, pulse width, repetition rate, and temperature. The customized test board used for this aging has independent DC
and AC power supplies for each APD. The device is operated in Geiger-mode for some time then the DCR is
periodically measured. An example of this type of aging sampled from a 32x32 array is shown in Fig. 4. This plot shows
Proc. of SPIE Vol. 7320 73200R-3
Downloaded from SPIE Digital Library on 15 Mar 2010 to 18.51.1.125. Terms of Use: http://spiedl.org/terms
aging of 37 GM-APDs for over 300 hours with a very high overbias of 10 V. Six failures are evident in this aging
sample.
A limited number of unaccelerated aging tests have also been performed. These tests are performed with arrays bumpbonded to ROICs and more closely operated as they would be in a fielded system. With the ROIC and associated
electronics, a large number of GM-APDs can be aged and evaluated. However, there is limited ability to increase the
acceleration for this type of aging since the ROIC limits the amount of charge flowing through the APD substantially.
An example of aging on a 32x32 GM-APD array on a ROIC is shown in Fig. 5 which plots both the average and
standard deviation of the DCR for the full array. The average DCR and its standard deviation initially increases then
decreases to a relatively stable value over the period of 1000 hrs. No failures were measured.
4.0E+05
3.5E+05
3.0E+05
DCR (Hz)
2.5E+05
2.0E+05
1.5E+05
1.0E+05
5.0E+04
0.0E+00
0
50
100
150
200
250
300
350
Hours
2
3
4
5
6
8
9
10
11
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
37
38
39
40
Fig. 4. Dark count rate as a function of time for 37 GM-APDS aged at 23 C, 10 V overbias, 30 kHz, 300 nsec pulses.
6000
1.06 μm GM-APD Array
15 um Dia, 32x32
Poly/Nit Passivation
5C
Dark Count Rate (Hz)
5000
Average DCR
DCR St Dev
4000
3000
2000
1000
0
0
200
400
600
800
1000
Time (Hours)
Fig. 5. Dark count rate as a function of time for 1024 GM-APDS aged continuously on a ROIC with a 4 V overbias. The
triangles indicate the average DCR while the boxes are the standard deviation.
Proc. of SPIE Vol. 7320 73200R-4
Downloaded from SPIE Digital Library on 15 Mar 2010 to 18.51.1.125. Terms of Use: http://spiedl.org/terms
3.3 Degradation characteristics and mechanism
Some insight into the degradation mechanism of these devices can be gained by more closely examining the
characteristics of the device while it is degrading. Figure 6 shows the DCR as a function of time and the current-voltage
(I-V) characteristics at several different points during the aging period. The DCR remains relatively constant for this
device from 0 to 20 hrs. However the dark current during this time increases four orders of magnitude. Then at 20 hrs of
elapsed aging the DCR begins to increase rapidly with only modest continued increases in the dark current. It is likely
that the increasing dark current is only a surface leakage along the sidewalls of the APD. If this is the case then electrons
could conduct along the perimeter of the device without ever going through the very high field avalanche region of the
device where they would cause an avalanche event (i.e. a dark count).
Figure 7 is a schematic of a biased GM-APD and illustrates a possible origin of the degradation in these devices. Ions or
charges may be accumulating on the perimeter of the GM-APD while it is operating. Over time these ions could
accumulate to a sufficient quantity that they start acting as a pathway for charge to flow along the perimeter of the device
and increasing the leakage current dramatically. Initially the surface conduction does not interfere with the Geiger-mode
operation since the carriers never penetrate into the avalanche region of the device to initiate a breakdown event.
Eventually the charge on the surface accumulates so much that it begins to bend the electric field in the device and starts
drawing the carriers into the avalanche region leading to a sudden large increase in DCR despite only very small further
increases in dark current.
3000
23.1 hr
23 C, Dry N2
4 V Overbias
300 nsec pulse, 1 kHz
2500
↓
DCR (kHz)
2000
1500
↓ 0 hr
↓ 7.2 hr
1000
↑ 2.7 hr
500
0
0
5
10
Hours
15
20
25
1.E-04
10-4
1.E-05
10-5
1.E-06
10-6
Current (A)
1.E-07
10-7
23.1 hr
1.E-08
10-8
1.E-09
10-9
7.2 hr
1.E-10
10-10
1.E-11
10-11
2.7 hr
1.E-12
10-12
0 hr
1.E-13
10-13
50
55
60
65
Voltage (V)
70
75
Fig. 6. Dark count rate as a function of time (top) and dark current as a function of voltage at a number of times during the
aging (bottom). The arrows in the top plot indicate the points at which the dark current was measured. The dark current
curves indicate the leakage increases several orders of magnitude before any increase in DCR is measured.
Proc. of SPIE Vol. 7320 73200R-5
Downloaded from SPIE Digital Library on 15 Mar 2010 to 18.51.1.125. Terms of Use: http://spiedl.org/terms
+
-
E↑
Fig. 7. Schematic cross-section of a GM-APD accumulating charge along the sidewall of the device. As the surface charge
accumulates it eventually leads to charge injection into the avalanche region.
Enlarged
4.0E+07
4.0E+07
1.06 μm GM-APD
Polyimide Passivated
200 um Dia
Vob = 4 V
300 nsec, 1 kHz
3.5E+07
3.0E+07
3.5E+07
3.0E+07
2.5E+07
DCR (Hz)
DCR (Hz)
2.5E+07
2.0E+07
2.0E+07
1.5E+07
1.5E+07
1.0E+07
1.0E+07
5.0E+06
5.0E+06
0.0E+00
16 hrs @ 175 C
43 hrs @ 190 C
0.0E+00
0
20
40
Hours
60
80
100
87
87.5
88
Hours
88.5
89
89.5
Fig. 8. Dark count rate as a function of time for a degraded GM-APD. The right plot is zoomed in around 88 hrs when the
device was baked at high temperature for long periods to diffuse the surface charge accumulated on the sidewalls away
from the device.
To further investigate this degradation mechanism, a high temperature bake was performed on a degraded GM-APD to
see the effect on the DCR of the device (Fig. 8). A GM-APD was aged and degraded until the firing rate was saturated
(i.e. it fired every gate, indicated in the plot by the DCR plateau at 70 hrs). The GM-APD was removed from the aging
chamber and baked for 16 hrs at 175 C then placed back in the aging chamber for more testing. The GM-APD returns to
its original pre-aged DCR and maintains that for a short period of time then begins to degrade very quickly again. The
test was stopped yet again and another bake was performed, this time at a higher temperature and longer time. After the
bake, the GM-APD again returns to its original DCR and maintains that performance for a bit longer but finally degrades
rapidly. This experiment further supports the theory of ions accumulating on the sidewall of the device and they diffuse
away from the mesa somewhat during the high temperature bakes. They do not move very far away as the onset of the
degradation on these subsequent rounds is much more rapid than the initial degradation but the charges do move far
enough away that their effect on the fields in the APD are negligible – recovering to the initial DCR before degradation.
These characteristics of the degradation of the device and the baking experiment agree well with a theory of charge
accumulation around the perimeter of the device. The evidence suggests that the failure is not being caused by an
epitaxial defect which could not repair itself at the modest temperatures used for the bake cycle. It also is probably not
the formation of conductive surface oxides since these also would not have a recovery of performance at modest bake
temperatures. However, we have not yet determined the physical origin of the ions that are accumulating on the
sidewalls or what causes the substantial variation in reliability of different fabrication lots which nominally are
fabricated with the same process. More work is ongoing to determine the origin of the degradation and improve the
fabrication process to further improve the reliability of these devices.
Proc. of SPIE Vol. 7320 73200R-6
Downloaded from SPIE Digital Library on 15 Mar 2010 to 18.51.1.125. Terms of Use: http://spiedl.org/terms
3.4 Failure rate model and aging statistics
With the accumulation of a substantial database of aging results at multiple conditions, it becomes possible to develop a
failure rate model for the devices. Typical semiconductor devices have failure rate models that take into account both
electrical stress on the device as well as the operating temperature. These models are exclusively experimental in origin
as the physical processes that cause these failures are very difficult to determine and then impossible to translate into a
mathematical model. MIT-LL has experimentally developed a failure rate model that has good agreement with the
measured aging data and is similar to models used for traditional photodiodes and laser diodes. Equation 1 is this model
which is simply that the failure rate of the device is proportional to the Geiger-mode current through the device times an
Arrhenius factor for thermal acceleration. The remainder of this section will discuss the application of this model to GMAPD array aging data to determine the fitting parameters such that extrapolations can be made of the failure rate at real
operating conditions.
⎛ E ⎞
FailRate ∝ I × exp⎜ − A ⎟
⎝ kT ⎠
(1)
When applying a failure rate model to aging data, the failure rate being measured should be fairly constant with time. If
the devices are experiencing a high rate of infant mortality there will be many failures early in the aging cycle with a
decreasing failure rate over time. If the failure rate is substantially increasing with time then the population is said to be
in the “wear-out” regime and is also not a very good data set to which to fit a model. A way to characterize the failure
rate as a function of time is using a Weibull plot and calculating the Beta of the data. Beta is proportional to the slope of
the Weibull plot and if it is unity then the failure rate is constant. Shown in Fig. 9 is an example Weibull plot of a
particularly poor aging lot in which all of the devices under test failed within 50 hours. The failure characteristics of this
lot are typical of other lots with failures, although most lots only have a handful of failures and are not sufficient to
produce a good Weibull plot. This Weibull indicates that the failure rate remains constant as a function of time with a
Beta of 1.1. No evidence of infant mortalities has been seen during any of the aging performed.
Table 1 lists several examples of how the charge flow acceleration factor is calculated for each aging test. The pulse
height when the GM-APD fires is used to calculate the current flow through the device per firing. The charge per firing
assumes a firing pulse width of half of the applied AC overbias pulse width (i.e. randomly distributed). All of the aging
to date has occurred in the dark and relies on random dark firings to age the device. The DCR can be easily translated
into a fraction of the applied pulses that actually fire. Finally, using the pulse repetition rate, a charge flow in electrons /
sec can be calculated. The ratio is then taken with the charge flow rate when operating on a ROIC to result in the
acceleration for that particular test just from the current acceleration. The highest acceleration factors are for high
overbias voltage (the ROIC is typically 4 V) and for larger diameter devices which have high current flow which is
proportional to the area of the device.
MIT-LL has aged over 1100 GM-APDs from 31 arrays from 12 fabrication lots accumulating over 140,000 actual device
hours. The failure rate of these devices tends to vary from lot to lot with some very poor reliability lots even though they
had very good performance at the beginning of life. The lots with good reliability tend to have similar failure rates to
each other, suggesting that there is a baseline failure rate for our fabrication process that produces uniformly good
material, with the occasional bad lot. Data thus far indicates that arrays on good fabrication lots have similar failure rates
and arrays on bad fabrication lots can have largely varying failure rates, but all substantially higher than failure rates on
good lots. Hence the reliability of any particular GM-APD array can be assured by aging sample arrays from each lot to
verify the failure rate is below a certain threshold. This lot verification is a routine procedure used in commercial
telecommunication laser diode manufacturing and is now being applied to our GM-APD array programs.
Device reliability can be described by the hazard rate which is the number of failures divided by the number of hours
aged. Frequently quoted is the mean-time-to-failure (MTTF) which is the inverse of the hazard rate (i.e. hours aged /
failures). A useful metric is also the failure rate in FITs (Failure-In-Time) which is the number of failures expected in
1E9 device hours. Since failure rates are based on sampling with fairly small populations and relatively few failures, a
statistical confidence factor is frequently applied to the failure rate. This factor effectively increases the reported failure
rate for aging data with few failures and has no effect when there are a large number of failures.
Proc. of SPIE Vol. 7320 73200R-7
Downloaded from SPIE Digital Library on 15 Mar 2010 to 18.51.1.125. Terms of Use: http://spiedl.org/terms
Examples of the reliability of good GM-APD array fabrication lots are shown in Table 2. This table indicates the
conditions during aging for the devices, the number of APDs aged, the number of failed devices, the maximum time on
test for the lot, the total accumulated device hours, the failure rate with 60% statistical confidence, and the raw failure
rate (point estimate). The italics rows take into account the current acceleration of that population to extrapolate a failure
rate at room temperature (23 C). The extrapolated failure rates for these three lots operating on a ROIC at 23 C are 1122,
475, and 376 FITs.
ReliaSoft Weibull++ 7 - www.ReliaSoft.com
Probability - Weibull
99.000
Probability-Weibull
Data 1
Weibull-2P
RRX SRM MED FM
F=34/S=0
Data Points
Probability Line
90.000
Unreliability, F(t)
50.000
10.000
5.000
Gary Smith
MIT-LL
2/12/2009
8:19:57 PM
1.000
0.100
1.000
10.000
100.000
Time, (t)
β=1.1108, η=14.1386, ρ=0.9693
Fig. 9. Fraction of failed devices as a function of time; a Weibull plot from the aging of GM-APDs. The fit to the data
indicates the failure rate is constant as a function of time (i.e. Beta = 1.1).
Table 1. Example calculations of the acceleration factors from aging of GM-APDs. The acceleration is calculated from the
peak current, pulse width, repetition rate, and DCR. Acceleration ranges from 10’s to over 1000.
Fab
Variant
Reference ROIC
482G
1.06 um, Poly/nit, n+ sub
516A
1.06 um, Poly/nit, p- sub
516C
1.06 um, Poly/nit, p- sub
521B
1.06 um, Poly/nit, p- sub
583G
1.06 um, Poly/nit, p- sub
583G
1.06 um, Poly/nit, p- sub
583G
1.06 um, Poly/nit, p- sub
583G
1.06 um, Poly/nit, p- sub
583G
1.06 um, Poly/nit, p- sub
583G
1.06 um, Poly/nit, p- sub
583G
1.06 um, Poly/nit, p- sub
APD
Rep Pulse
Pulse Sense
Peak
Array Dia Temp. Rate Width Vob Height Resistor Current Charge / DCR % firing Charge Flow
Size (um)
C
(kHz) (nsec) (V) (mV) (ohms)
(mA)
firing
(kHz) / gate
(e-/sec)
Acceleration
15
3.00E+06
50
2.25E+10
1
32x32 15
23
30
300
8
250
50
5
4.69E+09 63
1.9%
2.63E+12
117
32x32 15
23
30
300
8
260
50
5.2
4.88E+09 200
5.8%
8.52E+12
379
32x32 30
23
30
300
8
675
50
13.5
1.27E+10 400 11.3%
4.29E+13
1908
32x32 15
23
30
300
8
260
50
5.2
4.88E+09 100
3.0%
4.32E+12
192
128x32 20
23
30
300
8
130
25
5.2
4.88E+09 90
2.7%
3.90E+12
173
128x32 20
77
30
300
8
155
35
4.4
4.15E+09 1100 28.1%
3.50E+13
1556
128x32 20
57
30
300
8
155
40
3.9
3.63E+09 400 11.3%
1.23E+13
548
128x32 20
47
30
300
8
155
35
4.4
4.15E+09 255
7.4%
9.17E+12
408
128x32 20
3
30
300
8
210
40
5.3
4.92E+09 40
1.2%
1.76E+12
78
128x32 20
-3
30
300
8
165
35
4.7
4.42E+09 30
0.9%
1.19E+12
53
128x32 20
23
30
300
4
50
25
2.0
1.88E+09 30
0.9%
5.04E+11
22
Proc. of SPIE Vol. 7320 73200R-8
Downloaded from SPIE Digital Library on 15 Mar 2010 to 18.51.1.125. Terms of Use: http://spiedl.org/terms
Table 2. Summary of aging populations from several good fab lots of GM-APDs sampled from several arrays from each fab
lot. The failure rate is calculated with the number of failures on each test and the accumulated hours of all devices on
test from that sample. The failure rates with a statistical confidence of 60% are circled and are around 500 FITs.
Diameter
Rep Rate Pulse Width
(um)
Temp C
(kHz)
(nsec)
15
23
30
300
Acceleration
Vob (V)
8
379
Sample
Size
107
Max
Fail Rate Fail Rate
Time on
(kFITs,
(kFITs,
Fails
Test
Hours
60% CL) point est.)
1000
13
34384
425
378.08283
13
13031536
1.122
0.998
Fab
516A
Variant
Poly/nit, p- sub
Size
32x32
Array
Sum
521B
Poly/nit, p- sub
32x32
Sum
15
23
30
300
Acceleration
8
192
84
0
0
665
10047.5
1929120
91
0.475
0
0
516C
Poly/nit, p- sub
32x32
Sum
30
23
30
300
Acceleration
8
1908
118
11
11
286
17503
33399533
717
0.376
628
0.329
1.E+05
Failure Rate Scaled for Charge Flow (kFITs)
583G 0109
583G 0204
1.E+04
680G 0206
Ea = 0.8
1.E+03
Ea = 0.45
1.E+02
1.E+01
1.E+00
1.E-01
240
260
280
300
320
340
360
Temperature (K)
Fig. 10. Failure rate as a function of temperature for three different GM-APD arrays. The raw failure rate was scaled by the
charge flow acceleration factor then plotted on this graph. The dotted lines correspond to 0.45 and 0.8 eV thermal
activation energies.
The thermal acceleration factor is also quite strong for these devices. Figure 10 shows the failure rate as a function of
temperature for GM-APDs aged from the same arrays. Three arrays were aged at different temperatures ranging from
270 to 350 K. The failure rate was appropriately scaled at each temperature for the acceleration due to the charge
flowing through the device, then this result plotted in Fig. 10. With the current acceleration properly removed, the
remaining difference in failure rate as a function of temperature can be fit with the Arrhenius factor to extract the
thermal activation energy of the failure rate. The dashed lines in Fig. 10 correspond to activation energies of 0.45 and 0.8
eV, with two of the arrays aged falling along the 0.8 eV line and one falling along the 0.45 eV line. Due to the
uncertainty of the reliability measurements it is likely that the real thermal activation is the same for all of the devices
and the difference is due to sampling fluctuations. Hence a mid-range value of 0.6 eV is likely a reasonable
approximation for the thermal activation of these devices. Ongoing tests will provide more data and a more accurate
estimate of this activation energy.
Proc. of SPIE Vol. 7320 73200R-9
Downloaded from SPIE Digital Library on 15 Mar 2010 to 18.51.1.125. Terms of Use: http://spiedl.org/terms
4. CONCLUSIONS
With the determination of an expected failure rate at room temperature of about 500 FITs and a thermal activation of 0.6
eV, the predicted failure rate at any operating temperature can be calculated using the failure rate model in Equation 1
and the result is plotted in Fig. 11. At a typical operating temperature of a GM-APD array used in laser radar of 250 K,
the predicted failure rate is 6.5 FITs. This is a very low failure rate, similar to that reported at room temperature for
telecom-style linear-mode APDs. Other equivalent metrics to this failure rate are an MTTF of 17,100 years or an
expected pixel failure rate of 0.005% per year for an array of these devices. This failure rate of GM-APD arrays is now
in the regime that these devices can be strong candidates for fielded systems for both military applications as well as
commercial systems where reliability is of critical importance.
10000
Predicted Failure Rate (FITs)
1000
100
10
1
EA = 0.6 eV
Fail Rate @ 23 C = 500 FITs
0.1
0.01
200
220
240
260
280
300
320
340
Temperature (K)
Fig. 11. Predicted failure rate as a function of temperature for a GM-APD operating on a ROIC.
ACKNOWLEDGEMENT
This work has been sponsored by the U.S. Defense Advanced Research Project Agency under Air Force contract number
FA8721-05-C-0002. The opinions, interpretations, conclusions, and recommendations are those of the authors and are
not necessarily endorsed by the United States Government.
REFERENCES
[1]
[2]
[3]
[4]
[5]
S. Verghese, J.P. Donnelly, E.K. Duerr, K.A. McIntosh, D.C. Chapman, C.J. Vineis, G.M. Smith, J.E. Funk, K.E. Jensen, P.I. Hopman, D.C.
Shaver, B.F. Aull, J.C. Aversa, J.P. Frechette, J.B. Glettler, Z.L. Liau, J.M. Mahan, L.J. Mahoney, K.M. Molvar, F.J. O’Donnell, D.C. Oakley,
E.J. Ouellette, M.J. Renzi, and B.M. Tyrrell, "Arrays of InP-based Avalanche Photodiodes for Photon-Counting,” IEEE J. of Selected Topics in
Quantum Electron., vol. 13, pp. 870-886, 2007.
G.M. Smith, J.P. Donnelly, K.A. McIntosh, E.K. Duerr, D.C. Shaver, S. Verghese, J.E. Funk, N.R. Kumar, L.J. Mahoney, K.M. Molvar, F.J.
O’Donnell, D.C. Chapman, D.C. Oakley, and K.G. Ray, "Design and reliability of mesa-etched InP-based Geiger-mode avalanche photodiodes,"
in IEEE LEOS Annual Meeting 2007 Technical Digest, WL1, 2007.
G. Hasnain, W. G. Bi, S. Song, J. T. Anderson, N. Moll, C. Su, J. N. Hollenhorst, N. D. Baynes, I. Athroll, S. Amos, and R. M. Ash, “Buriedmesa avalanche photodiodes,” IEEE J. Quant. Electr., vol. 34, pp. 2321-2326, 1998.
I. Watanabe, M. Tsuji, M. Hayashi, K. Makita, and K. Taguchi, “Reliability of Mesa-Structure InAlGaAs-InAlAs Superlattice Avalanche
Photodiodes,” IEEE Photon. Technol. Lett., vol. 8, pp. 824-826, 1996.
S. Smetona, K. Klunder, S. An, R. Rousina-Webb, B. Emmerstorfer, E. Jamroz, M. Rouabhi, R. Mallard, “Reliability of InP-Based Avalanche
Photodiodes for High Bit-Rate Fiber-Optic Communication Systems,” InP and Related Materials Conference, pp. 373-376, 2002.
Proc. of SPIE Vol. 7320 73200R-10
Downloaded from SPIE Digital Library on 15 Mar 2010 to 18.51.1.125. Terms of Use: http://spiedl.org/terms