Commentary : JWST near-infrared detector degradation-finding the problem . fixing the problem . and moving forward

Bernard J. Rausche") NIRSpec Detector Scientist, Observational Cosmology Laboratory, NASA Goddard. Space Flight Center, Greenbelt, MD, 20771 , USA Carl Stahle NASA Detector Degradation Failure Rev;ew Board Chair, Instrument Systems and Technology Division, NASA Goddard. Space Flight Center, Greenbelt, MD, 20771, USA Robert J. Hill NASA Detector Degradation Failure Review Board Assistant Chair, ObsenJational Cosmology LabomtoT'Y, NASA Goddard. Space Flight Center, Greenbelt, MD, 20771, USAb ) Matthew Greenhouse Integmted Science Instruments Module Project Scientist, Observat-:,onal Cosmology Labomtory, NASA Goddard. Space Flight Center, Greenbelt, MD, 20771, USA James Beletic Teledyne imaging Sensors, 5212 Verdugo Way, Camarillo, CA, 99012, USA Sachidananda Babu Detector Systems Bronch, NASA Goddard Space Flight Center, Greenbelt, MDJ 20771, USA Peter Blake Optics Branch, NASA Goddard. Space Flight Genter, Greenbelt, MD, 20771, USA Keith Cleveland Mission Assurance Bmnch, NASA Goddard. Space Flight Center, Greenbelt, MD, ~0771, USA Emmanuel Cofie Mechanical Systems Analysis fj Simulations Bronch, NASA Goddard. Space Flight Center, Greenbelt, MD, 20771, USA Bente Eegholm Optics Branch, NASA Goddard. Space Flight Center, Greenbelt, MD, 20771, USAo)

The James Webb Space Telescope (JWST) is the successor to t he Hubble Space Telescope.JWSTwill be an infrared-optimized telescope, with 8...."1 approximately 6.5 m diameter primary mirror, that is located at the Sun-Earth L2 Lagrange point.Three of JWST's four science instruments use Teledyne HgCdTe HAWAII-2RG (H2RG) near infrared detector arrays.During 2010, the JWST Project noticed that a few of its 5 I'm cutoff H2RG detectors were degrading during room temperature storage, and NASA chartered a "Detector Degradation Failure Review Board" (DD-FRB) to investigate.The DD-FRB determined that the root cause was a design flaw that allowed indium to interdiffuse with the gold contacts and migrate into the HgCdTe detector layer.Fortunately, Teledyne already had an improved design that eliminated this degradation mechanism.During early 2012, the improved H2RG design was qualified for flight and JWST begBJl making additional H2RGs.In this article, we present the two public DD-FRB "Executive Summaries" that: (1)  determined the root cause of the detector degradation and (2) defined tests to determine whether the existing detectors are qualified for flight.We supplement these with a brief introduction to H2RG detector arrays, some recent measurements showing that the performance of the improved design meets JWST requirements, and a discussion of how the JWST Project is using cryogenic storage to retard the degradation rate of the existing flight spare H2RGs.

I. ItlTROOUCTION
The James Webb Space Telescope (JWST) will be a 6.5 1:"1 class infrared optimized telescope located at the Sun-Earth L2 Lagrange point.As the successor to the Hubble Space Telescope, it is designed to enable a broad science program that includes studies of the first suI>galax)' sized clumps of stars to light up after the Big B ang, galaxy formation and evolution, the birth of stars and planetary systems, and the search for planets that can support life.It is infrared optimized because the cosmological expansion shifts the spectroscopic features that Hubble sees at yisible wavelengths in the nearby universe into the infrared.Infrared wavelengths are also able to penetrate the heavily dust obscured regions of the universe where stars and planets form.
To enable this broad science program, JV{ST carries a suite of four science instruments: (1) a Near Infrared CamEra (NIRCam), (2) a Near Infrared Spectrograph (NIRSpec), (3) a Fine Guidance Sensor (FGS) with Near-Infrared Imager and Stilless Spectrograph (NIRlSS), and (4) a Mid-Infrared Instrument (MIRl).The three "nearinfrared" (NIR; 0.6 <::: A <::: 5 I'm) instruments use the Teledyne HgCdTe HAWAII-2RG (H2RG ' ) detector arrays that are the focus of this paper.The cutoff wavelength of an HgCdTe detector is tunable by varying the mole fraction of cadmium in the HgCdTe.The NIRSpec and FGS/NIRISS use 5 I'ID cutoff H2RGs, whereas NIR-Cam uses 2.5 I'm and 5 I'm cutoff H2RGs.The MIRI uses a different detector technology for the 5 -29 I'm "'avelength range, Raytheon Si:As blocked impuritv band detectors.There has been no indication of any dpgradation IT. 1\1IRJ's detectors.We refer the interested reader to Gardner 2 and Greenhouse et al. 3 for more information about the JWST mission and its science instruments.
During April, 2010, the NIRSpec team noticed degrada.tion of pixel operability in one 5 I'm cutoff H2RG that had been stored for about 18 months at room temperature.By the end of the year, the NIRCarn team had noticed similar changes in four more 5 I'm cutoff H2RGs that had been stored for about two years at room temperature.One sign of degradation in JWST's H2RGs was an increase in the number of inoperable "warm pixels" (Fig. 1).JWST defines a warm pixel as a pixel having dark count rate 0.1 < rate < 60 e-8 -1 , where the count rate is measured using a linear 2-parameter fit to the up-the-ramp samples spanning 1000 sec.Under ITAR, we cannot legally publish information that would facilitate duplicating the H2RG technology.
As is described in detail in Sec.n, the DD-FRB assigned root cause to a design flaw in the pixel's barrier layer that allows indium from the interconnects to interdiffuse with gold in the contact structure and migrate into the HgCdTe detector material.Based on destructive physical analysis (DPA) of hybridized detectors and process evaluation chips (PEC 4 ) from every flight detector, the DD-FRB concluded that ''there is the potential for degradation in every pixel."The DPA of PECs is described in more detail in Sec.III.
Although it was clear that the performance of the 5 IJffi parts degraded more rapidly than the 2.5 I'In parts, the DPA of PECs made it equally clear that the underlying physical process was active in both.Within the DD-FRB, this was referred to as a "dead pixel walking" scenario, the implication being that the performance of the 2.5 I'm parts would eventually degrade if exposed to room t emperature for a sufficient period of time.:Moreover, because of the large number of variables that modulate tbe degradation rate, it was not possible to project tbe degradation rate for individual parts without performing destructive testing on a larger number of detectors than was available (see Sec. Ill). .
For many applications, a reasonable work-around would be to keep the detectors cold as the indium interdiffusion degradation process scales exponentially with temperature.Assembly of the J\YST necessitates that the detectors withstand several years of room tempera-tUre storage prior to launch in 2018, For JWST, the plan is therefore to make additional flight detectors using an improved barrier layer design.To ensure maximum flexibility, the JWST Project is storing the existing flight spare detectors, which use the old barrier layer design, at cryogenic temperature to slow the degradation as mt:.ch as possible, The rationale for t his is given in Sec III.This article is structured as follows.In Sec.I A, we provide a brief int roduction to JWST H2RG detector arrays.Readers who are familiar with this technology and its application to JWST may wish to skip over this background material and go directly to Sec, II.The act"al DD-FRB Executive S=maries comprise the bulk of the article in Sees, II and III.These are presented in as close to their original ~orm as is practical in this journal.Because the two public Executive Summaries did not describe the specific design improvements that were made to eliminate degradation, we provide a high level overview in Sec [V.In this section, we also discuss how the new design was validated for JWST.Finally, in Sec.V, we discuss some of the lessons learned and how the JWST project is storing its existing flight spare detectors under crvogenic conditions to slow the degradation as much as possible.
A. Brief Introduction to JWST H2RG Detector Arrays

H2RG Detector Arrays
The H2RG5,6 (Fig. 2) is a hybrid NIR detector array.Light is collected in a 2040x2040 pixel array of HgCdTh photovoltaic diodes.The 2040x2040 pixel ph<r tosensitive area is surrounded on all sides by a 4 pixel wide border qf reference pixels.For JWST, the entire 2048x2048 pixel area (reference pixels plus regular pixels) is read out using four analog outputs.The reference pixels, which have been engineered to electronically mimic a regular pixel, can be used to suppress correlated noise. 7,8 Electronic readout and control are accomplished.in a silicon readout integrated circuit (ROIC).When it was first introduced,• HAWAII-2RG name refered only to the ROle, although it is now common usage in the as-tronornical community to refer to entire • detector arrays as HAWAII-2RGs or H2RGs.The ROIC is permanently attached to the HgCdTe detector array by indium bump bond, .There is one indium bond per pixel.In the case of a JWST H2RG, a low ;iscosity epoxy is backfilled into the bumps to increase mechanical strength.
Unfortunately, the indium that is used to make the bump bonds does not interact well with some of the other materials that are used in the H2RG for their electrical or optical properties.As is described in Sec.II, indium interacts readily with gold in the contact structures to form In-Au int-ermE%allic compounds. 9 ,lO Indium is an ntype dopant in HgCdTe, and in the JWST H2RG architecture, it is :n close proximity to a p+ doped implant.
Sec. II includes an extensive discussion of how indium penetrated the barrier layer in the H2RG design to interac! with both gold and p+ doped HgCdTe to degrade the detectors.

JWST NIR Performance Requirements
Compared to many other appiications, JWST places a premium on ultra low dark current and low read noise.JWST's H2RGs are read out at a 100 kHz pixel rate and biased to provide a well depth of about 10' e-.JWST's NIR detector requirements derive principally from the need to observe extremely faint astronomical sources.In pract:ce, this means that the quantum efficiency ;s required to be as good as is practical and the dark current noise is required to be low compared to noise resulting from t he background Zodiacal flux.Tab.I lists an illustrative subset of the requirements that were flowed down from JWST's science program for NIRSpec.In practice, these NIRSpec requirerner.tsare slightly more challenging than what is needed for NIRCam and the Fine Guidance Sensor (FGS), but they give a good high level impression of the performance that is needed.For most science observations, JWST will use an upthe-ramp readout scheme.Although each JWST instrument differs sorr:ewhat in the details I the basic idea is to destructiyely reset the detector arra~' and then sequentially rcad it out to build up an integration one frame at a time.When the 2048 x 2048 pixel detector array is read out using four outputs at 100 kHz, the frame time is 10.73 s .This includes 12 pixels of overhead at the end of each row and one row of overhead at the end of each frame.Across JWST, approximately 10' s long integrations are taken as the baseline.For NIRSpec, this readout pattern produces one s8r.lpleevery 10.73 s and the baseline science integration contains 88 samples.The integrations are called ujrthe-ramp because when pixels are illuminated, the 88 ui>'the-ramp samples follow an approximately straight line with positive slope.For more information on JWST detector readout modes and noise models, the interested reader is referred to Rauscher et al.ll

II. DD-FRB EXECUTIVE SUMMARY 2A: ROOT CAUSE DETERMINATION
This section presents DD-FRB Executive Summary 2a, the root ca.use finding.It is presented in as close to original form as possible, wit h only minor formatting and stylistic changes.Although we have tried to minimize duplication, some redundancy was unavoidable because each Executive Summary was intended as a stand alone document.The original Gocument (JWST-RPT-017457; dated 29 April 2011) is available from the JWST public web site. 12

A. Context and Statement of the Problem
The James Webb Space Telescope (JWST) science instrument payload contains four science instruments and a fine guidance sensor.Three of the science instruments and the fine guidance sensor utilize HgCdTe detectors that are designed to achieve high responsivity to light over the O.~5 11m spectrum.One instrument also utilizes HgCdTe detectors that are designed for the 0.6-2.5I'm spectrum.Se;-en of the 5 /lorn cut-off detectors and 8 of the 2.5 I'm cutoff detectors are required for flight as shown in Tab.II.
Flight model integration has begun on all of the instruments listed in Tab.II.Teledyne Imaging Sensors produced all of the JWST 'HgCdTe detectors during the 2007-8 timeframe.The JWST assembly and test sequence requires that the science instrument detectors have an ambient temperature shelf life of several years prior to launch and an operational life of at least 5.5 years after launch.
Instrument team test data obtained over the past year has revealed degradation of pixel operability impacting several of the 5 and 2.5 /lorn cut-off detectors.There is a In the time since the DD-FRB completed its work l the FGS-TF was replaced with the Near-Infrared Imager and Slitl€SS Spectrograph (NIRISS) .The NIRlSS uses the same detectors that were intended for the FGS-TF.To maintain.consistency with the earlier DD-FRB reports, we retain the previous FGE-TF name here.
a strong concern that the degradation will continue with time and many of the flight arrays will be out of speci.'icationby the time of launch.The key detector degradation observed was an order of magnitude increase in the dark count rate of individual pixels to levels in the range of 0. 1 to 60 electrons per pixel per second (e-/pix/sec).Fig. 3 shows an example of this increase in dark count rate for one pixel in a flight spare NIRSpec detector (S060) .Other performance anomalies were also observed and are listed in Tab.III.

Root Cause Determination
The DD-FRB finds that the detector degradation is caused by a design flaw in the barrier layer of the pixel interconnect structure.The flawed harrier layer design makes the deteCtors vulnerable to migration of indium from the indium bump interconnect into the detector structure, degrading its performance.
The most obvious effect is the formation of an indium (In) gold (Au) intermetallic that is highly visible in Scanning Electron Microscopy (SEI\I) images taken during Here raw signal is measured in analog to digital converter units (ADU), and dark count :-ate is equal to the fitted slope.The blue data arc for a good pixel and the reci data are for the same pixel that has degraded with time.
destructive physical analysis.The electrical da.ta of degraded pixels reyeal curved , "RC" shaped dark ramps that are indicative of parasitic capacitance, reactance, and shunting in the HgCdTe side of the interconnect.Typically a few hundred seconds atter reset, true leakage currents become dominant.These effects cause pixels to fail to meet operability requirements.Fig. 4a shows a cross-section of the pixel contact structure design.In this sensor design, each HgCdTe pixel is connected via the In hump to a. source-follower amplifier in a silicon Read-Out Integrated Circuit (ROIC).The critically important barrier layer .isintended to pre-,-ent In bump material from reading with the Au pad a.nd Au contact material such that it can not diffuse into the HgCdTe detector material.Figs.4b and 4c show cross-sectional microg::aphs obtained with SEM of a nondegraded pixel from a 2.5 ,..m NIRCam detector array (0105) and a degraded pixel from a 5 /lm NmOam detector array (0094).The cross-section of the pixel structure was generated by destructive physical analysis (DPA) using a iocused ion beam (FIB) to cut through a line of pixels in the arrar.Fig. 4c shows the formation of an AuIr.2 intermetalHc as well as a crack in the left corner of the pixel contact structure propagating into the HgOdTe detector.The intermetallic expands upon forma.tionand most ilkel!' created a pocket of stress in the ;>ixel.
Fig. 5a shows a diagram depicting failure of the barrier layer.Poor sidewall coverage of the layers over the step of the passivation layer or porosity of the barrier layer can allow In to inter-diffuse with the Au contact and Au poo metals to create In-Au intermetallies.Fig. 5b illustrates some potential degradation mechanisms; the intermetall!cexpansion may cause strain and lattice dislocation damage to tbe HgCdTe and/or enable In to diffuse i:J.to the p+ HgCdTe of the implanted junction layer.Apart from productior. of charge traps in the semiconductor band gap, dislocation damage can also allow In or Au to diffuse more rapidly into the HgCdTe resulting in a dark current performance degradation ra.te that can be non-linear and difficult to reliably estimate.
Fig. 6 shows the flow diagram of the degradation mechanisms.
A degraded detector ;>ixel can be modeled by an electrical circuit (Fig. 7), which produces an integration ramp signal with an "RC"-like curvature early in the ramp (see Fig. 3).More extensive damage or indium diffusion will produce additional leakage currents through the photodiode.Although this circuit model approximately captures the essential behavior of degraded pixels (an "RC" at early tim~s and leakage at later times), the actual circuit elements are far from ideal.
Formation of the In-Au intermetallic was confirmed by Energy Dispersive x-ray Spectroscopy (EDS) to provide a direct measure of the elemental composition.Fig. 880 shows a SEM image of a corner of another detector pixel in detector array C094 with a corresponding elemental map for Au, In, and the barrier layer in Fig. 8b.For these samples, the cross-section was prepared by cutti!Ig through the sample with a wire saw followed by mechanical polishing.The data show the formation of the In-Au intermetallic with a break in the barrier layer at the side• wallo: the contact opening.
Additional EDS data were taken on another pixel in detector 0094 as well as the Process Evaiuation Chip (PEC) for C094.Fig. 9a shows the SEl\I and the x-ray analvsis area (red box) from the PEO and Fig. 9b shows the x-ray spectrum.Quantitative analysis of the weight perceI:tage of the volume measured shows that the In-Au compound is Au1n2.
Fig. 10 shows a SEl\! image and a backscatter electron image of a cross-section of a pixel in detector array C094.Combined ,,•ith EDS analysis on the different regions, the results show that there is interdiffusion of both In and Au past the barrier la),,, with the formation of AuIn2 and Auln intermetallics that expand in volume.

C. Key Physical Observations that Support Root Cause
To avoid focusing on a single aspect of the observed degradation, the DD-FRB developed a list of key observations that any root cause analysis would have to explain.This list began at 14 items and has since grown to 25 items, with each new observation adding or reinforcing the list (Tab.III).There are some common elements for all explanations: 1) formation for an RC circuit element, most likely an nip or Schottky barrier that completely intercepts the circuit after the contact; and 2) defects which increase the detector junction leakage current.These common elements are likely caused by damage (dislocations, displaced ions) induced by the intermetallic formation itself due to an inadequate barrier layer.The damage is further increased in its effect by enhanced" diffusion of indium, now present at or in the HgCdTe from the proximate In-Au intermetallic.Beyond this, every diode will have its own story, and there are millions of them in a detector array.

A. Introduction
The DDFRB has released its findings for the root cause determination of the degradation of pixel operability impacting several of the 2.5 and 5 /lm cutoff detectors used in the NIRCam, NIRSpec, and FOS instruments of the James Webb Space Telescope (JWST-RPT-017457, http: //j wst.nasa..gov/ resources/017457.pdf).The key finding is that the detector degradation is caused by a design fla.w in the barrier layer of the pixel interconnect structure.The flawed barrier layer design makes tbe detectors vulnerable to migration of indium from the indium bump interconnect into the detector structure, degrading the performance of the detector.The fraction of pixels that are out of specification due to degradation over three years sir.cemanufacture ranges from 0.2% on some 2.5 /lIT! arrays to 1-2% for the affected 5 /lm arrays.Although these detector arravs as a whole are not yet out of specification, there is a strong concern that the degra,.dationwill continue with time and that many of the flight arrays will be out of specification by the time of launch.
Likewise, there is a concern that there may be a latency period for the onset of measurable degradation once the

B. Destructive Physical Analysis on Process Evaluation Chips for Flight Detectors
The DD-FRB finds that every pixel in every flight detector most likely has an In-Au inte .... metallic so there is the potential for degradation in every pixel.This result is based on Destructive Physica.iAna.iysis (DPA) using a Focused Ion Beam (FIB) to cross-S€Ction an individual pixel and then Scanning Electron Microscopy (SEM) to image the pixel structure.FIB/SEM was done on 3 pixels in a 400 pixel mini-a..'Tay from a Process E,-a.Iuation Chip (PEC) for every ffight detector.A total of 72 pixels from 24 PECs were tested and every pixel showed the formation of In-Au intermetallics from the failure of the barrier layer.Fig. 11 shows the SEM images of individua.ipixels from a 2.5 I'm detector (NIRCam) and 5 I'm detectors (NIR-Cam, NIRSpec, and FGS).

C. Projecting Future Performance
Once the barrier layer fails, the subsequent degradation of a detector array is a complex and highly variable process.To understa!ld the issues, it is helpful to focus on a single degrading pixel.If the future degradation of an individual pixel cO:.lld be projected, the degradation of an entire SCA could be treated as the projection of an ensemble of pixels.
Detector degradation due to In diffusion is a multifaceted process that begins when the barrier iayer fails.The failed barrier layer a.ilowsIn and Au to inter-diffuse, forming In-Au intermetaUics.The formation of these in- For purposes of this discussion, we make a simp!ifybg assumption that the degree of degradation of a pixel is roughly proportional to the amount of In that has diffused in.The actual degradation process can be more complicated, but a simple scaling argument is sufficient to show the large number of unconstrained variables that must be included to model even one pixel.These yariabies include: (1) diffusion area, (2) diffusion permeahility, (3) number of dislocation defects that intersect the diffusion area, (4) diffusion coefficient for each dislocation defect, (5) depth to the first sensitive area in the pixel, and (6) the scaling factor between In concent:-ation and dark current.Moreover, the number of dislocation defects is likely to depend on -(7) the numher of thermal cycles, and (8) the diffusion coefficient is likely to be variable along a defect, By picking parameters in a l\lonte Carlo simulatioI!, it is possible to obtain any degradation The number of warm pixels'" increases with time in both the 2.5 J..Lffi and 5 p,m cutoff detectors that show degradation.2 l'In degraded detectors b I some warm pixels get better at the same time as a larger number get ,•:orse."3 The rate of degradation of t he detectors varies from part to part and is not necessarily linear with time.4 "Although clustered, the new warm pixels do not form a contiguous group." 10 5 'The spatial distribution of the warm pixels appears to be similar for all the NIRC~m 5 ,urn detectors.In addition, there are similarities in the spatial distribution of warm pixels " among the affected NIRSpec detectors, but the distributions are riifferent from those of the NIRCam parts.However, there is at least ODe small area near the edge of the detectors with a higher density of warm pixels that is common to both the NIRCam and NIRSpec parts." 6 '-No l\-arm pixels have heen observed in the reference pixels of any degraded detector, even though new warm pixels are seen in the immediately adjacent regions of some degraded detectors."7 Areas with an increased density of warm pixels also show a small decrease in flat field response relative to good regions.8 '•While some new warm pixels may be hot pixel c neighbors, most new warm.pixels are not reiated to hot pixels.lI9 The regions with high densities of new warm pixels are preferentially found near the edges of the detectors rather than at the centers.These regions are also where the stress-induced curvature o~ the detectors is at a minimum.10 A 12hr bake at SOC:: in a dry nitrogen environment resulted in an increased number of warm pixels, indicating an increased rate of formation while at elevated temperature in one of the degraded NIRCam 5 p.m detectors (C094)."11 'The new warm pixels that appeared after the 12hr-500 bake of C094 have a similar spatial distribution and electrical properties (dark count rates, "ramp shapes) as the pixels that had become warm during ambient storage." 12 "The character of the degradation of some ~'"FC3 detectors at their operating temperature of 14SK is veT!' similar to that of the JWST detectors at their 40K operating temperature, despite the differences in the long wavelength cut-off (107m vs. 5 ,urn), processing details, and subsequent storage and handling.It is possible that the same physical processes are at work in both instanc€"S, whiie the details may differ."13 "Eight of t he eleven tested 5 JLm detectors show degradation.However, only two out of thirteen 2.5 p.m detectors have degraded.In addition, two FOS 5 p.m detectors show no degradation but have been stored in ambient conditions for 1 year less than the other J\\•ST detectors."14 The slope of the dark signal ramps for most (80-85%) new warm pixels shows statistically significant curvature (RC.like beha.vior). .path from no degradation to all pixels simultaneously going out of specification.,However, it does not follow that all such possible distributions arise in reality.The only way out of this conundrum is to make on the order of tens of measurements to constrain these unknowns , with clear degradation between each measurement.In addition, it would also be necessary to degrade the SCA to at least the onset of rapid degradation to constrain some of the most important parameters.The DD-FRB believes that a research program like this is not practical for the current JWST flight detectors due to a very limited inventory of SCA samples that are available.In order to not get fooled b~' small number statistics, we estimate that a minimum of 10 SCAs of bo~h 2.5 and 5 I'm cut-off wavelength are needed for these measurements.We have currently identified 5 candidate SCAs at earn wavelength.

D. Bounding the Risk
The DD-FRB realizes that a fuil complement of better SCAs may not be available for JWST when they are needed.For t his reasor:, the DD-FRB considered what steps the JWST Project might take to bound the risks associated with using the existing SCAs if necessary.Because detector degradation depends on so many factors, and these factors differ widely from one detector to the next, the main risks are associated with sta.tistical outliers.
The most we can hope to do is to gain some informa-tion on the range of degradation paths (e.g. the shapes of the warm pixel percentage vs.time curves) that a pool of SCAs might experience in the future.Accelerated testing (through a combination of elevated temperature storage and thermal cycling) may, for example, may provide information about the prevalence of two potentially distinct degradation behaviors: 1) The pixel-destroying indium diffusion process has a. critical step with a characteristic latency period, such that little degradation is seen for an extended time period, and that rapid degradation then ensues, yielding a highly non-linear, "degradation cliff" behavior to the long-term bad pixel trend for a SCA.This degradation path is the worrisome case, for the situation where that cliff occurs within the ground storage period of the detectors.One could have SCAs whose performance s~ill looks perfectly acceptable at ~he last reasonable replacement opportunity-, but which would nonetheless degrade unacceptably before launch.
2) There are intrinsically wide distributions in the ;'clues 6f the physical parameters (the assorted defects) that determine a pixel's future behavior.These ~.'ide distributions convolve to yield a wide range in the latency period before serious degradation of individual pixels occurs.In this case, the composite degradation of a large ensemble of pixels would be more gradual, and a SOA, as a whole, would not exhibit a performance "degradation cliff" that will put it out of speCification by the time of launch.
The DD-FRB belie,"CS it would be useful to measure some information regarding the relative prevalence of these behaviors.However, we recognize that wjth only a small pool of devices available for accelerated life testing, such tests are quite limited with respect to their ability to make strong statements regarding the future behavior of current flight SCAs.If, in a small sample of devices, a significant number were shown to exhibit the cliff-like degradation behavior on an accelerated timescale that corresponds roughly to the required ground storage period, there would be reason for grave concern about even SCAs that still have acceptable performa.1ce.If, in contrast, all of the SCAs that are monitored or put through accelerated life testing show gradual, smooth degradation versus time throughout the relevant time period, then one could make a reasonable inference that parts with a smooth degradation projection will likely remain that way.
By subjecting a few of the existing SCAs to accelerated life testing, it should be possible to understand how the average SeA will evolve, and moreover to piace some loose bounds on the likely dispersion around that average.For purposes of this testing, the 2.5 I'm ana 5 I'm cutoff SCAs would need to be considered as separate populations.A JWST SCA could be baked at a maximum temperature of 50 C over an extended period of time to monitor degradation (50 C is the maximum temperature that occurs in the SCA fabrication process).Additional stresses, which will not be encountered in the flight application, may be introduced at higher temperatures.Once this testing is done, there will still be a residual risk of statistical outliers in the existing detector complement, and the J WST Project would need to properly account for the risk that a few flight detectors might unexpeetooly degrade more (possibly much more) than is projected even when the measured dispersion around the mean is taken into account to arrive at a worst case statistical projection.The JWST Project would have to evaluate the impact of this risk for each science instrume~t .
To lower t he risk that ~ full complement of replacement detectors are ava.ilablewhen needed to preserve the JWST mission schedule, the DD-FRB recommends storing t he current flight spare SCAs at cryogenic temperat ure to retard their degradation rate.Although we don't kno,,; the exact functional dependence of the degradation ra.te, previous experience with HgCdTe detectors fo~ diffusion and electricai activation processes s".lggests an exponential dependence 0" temperature so a SCA should be stored .at the lowest practical temperature to greatly reduce t he risk of further degradation.

E. Summary
The DD-FRB finds from destructive physical analysis tests on Proceas Evaluation Chips for each flight detector that ever~.rpixel in every flight detector most likely has an In-Au interme~allic so there is a high potent:al for degrfl dation in every pixel -a "dead pixels walking" scenario.It may be possible to place bounds on the expected degradation wit h ambient temperature storage tine by subjecting a few of the existing SCAs t o accelerated life testing.The current flight spare SCAs should be stored at the lowest practical temperatme to reduce the risk of degr.dation.

IV. IMPROVED JWST H2RG DESIGN & CRYOGENIC STORAGE
As has already been mentioned, JWST is in the pmcess of making additional flight detector arrays using an improved design that eliminates t he degradation's root cause.In Sec III, the DD-FRB recommended storing the existing fiig~t spare H2RGs at cryogenic temperature to retard the degradat ion rate.We discuss these topks in the following paragraphs.

A. hnproved Barrier Layers
The details of hO~l the root cause of degradation was eliminated are ITAR sensitive Teledyne Proprietary information.AB such , t his discussion is very high level.Teled:' llle is working on an article that will explain the new design in more detail. 14xecutive Summary 2a (Sec.II) pinpointed the root cause of t he degradation as, "a design flaw in the barrier layer of the pixel interconnect structure."The corrective actions take!! included: (1) redesigning the barrier layer (see Fig. 5) and (2) process improvements aimed at achie\Oing more uniform and conformal barrier layer coverage.
The improved design uses different materials for the barrier layer.Component level la.boratory testing at Goddard and Teledyne .showed that the specific material that was used for the primary barrier in t he JWST R2RG design was permeable to indium.The improved design uses a different combination of materials that the same testing shows are completely impermeable to indium, even at temperatures ,.,armer than room temperature.
A second aspect of t he improved design is applying the barrier layer in a more conformal w.ay.SEM imagery, including Fig. 4, showed that the old barrier layer design did not achieve good step coverage, The improved barrier layer is applied using newer processes that achieve better step coverage.Also, some aspects of the interconnect structure were mod,fied to facilit ate achieving good step coverage.

B. Validating the New Design
Validating the new design requires: (1) tests showing that the degradation had been completely eliminated and (2) tests showing that the improved design meets JWST performance requirements.By : March, 2012, sufficient progress had been made in both areas for JWST to resume flight production.All testing to date has been successful, with no indication of any problems associated.with the new design.

Tests Showing that Degradation was Eliminated
In order to show t hat the new detector design is not susceptible to degradation, Teledyne built a demonstration lot of RIRG SCAs that incorporate the new barrier layer.An RIRG is a lK x lK pixel SCA that uses the same pixel design as t he R2RG.The DD-FRB recommended two kinds of accelerated life testing to demonstrate the validity of the improved barrier layer design.These were (1), showing through DPA that the new barrier layer remains intact after thermal cycling and a high temperature bake and (2) life testing of the new SCAs to show that no degradation occurs over the required JWST lifetime.
During the course of the DD-FRB, anecdotal evidence was acquired regarding the stability of t he new design through the baking of several samples followed by DPA.The barrier laver remained intact in all cases.However, while encouraging, t he results are not conclusive because the samples were not fully realized SCAs and hence were not subject to the same stresses (hybridization and thermal cycling) that the JWST detectors will encounter.
The high temperature bake test is a worst-case scenario that consists of taking two demonstration lot SCAs, subjecting them to 30 thermal cycles between room temperature and 35 K, a 100 C bake for 30 days, and another 20 thermal cycles prior to performing a DPA.The 30 d ay bake at 100 C provides the equivalent of a 10 year lifetime at room temperature for a thermal process with an activation energy of 0.58 e\'.The formation of the In-Au intermetallic has higher ac~ivation energy (~0.68-0.72 e V)9,1O and hence, the bake should provide sufficient margin against the room temperature storage time for future JWST detectors (as much as 7 years).The SCAs are currently undergoing the 30 day bake and the DPA resuhs will be available in Mav 2012.
The JWST flight detectors will never be subjected to temperatures above 45 C, and therefore a completely valid life test must be conducted at this temperature or lower to insure that there is no significant degradation mechanism with an even lower activation energy than 0.58 eV.Unfortunately, this relatively low temperature liinit means t hat the life test recjuires extended bake times to provide the equhalent of a 7 year room temperature storage lifetime.The bake tests will therefore r~n for 2.5 years, with performa.'1.ce te.,ts interspersed at roughly 6 month intervals.For each of the 5 ",m and 2.5 I'm cut-off wavelength detectors, two SCAs will be baked at 45 C and two more will be baked at 35 C, while others will be stored at room temperature.If any degradation is observed, the rate of degradation seen at each temperature can be used to determine the activatiO!lenergy of the degradation process.This test will not just be sensitive to the formation of the In-Au intermetallic, but will uncover any thermally activated degradation mechanisms with activation energies as low as rvO.35 eV.The initial performance testing of the SCAs to be baked in t hese tests are currently being completed.

Tests Showing that Performance Requirements are Met
Teledyne and the JWST Project validated the new harrier layer design's performance by testing prototype H1RGs at Teledyne, the University of Arizona, and in the Goddard Detector Characterization Laboratory (DCL).All t hree labs agree that the prototype HIRGs have performance comparable to the current flight parts when they were new and had little or no degradation.In the following paragraphs, we briefly summarize some of tl:e test results for the first two flight-design 5 J1.m H1RG prototypes that were tested in the DCL.These tests showed that . he parts either met flight specification, or exhibited only minor non-compliances that were completely consistent with the current flight parts that use the old design.
The DCL is responsible for NIRSpec detector characterization.It is equipped with an ultra-low background cryocooled dewar capable of operating the detectors at T "' 40 K.This dewar routinely achieves dark currents < 0.01 e-8-1 and read noise per correlated double sample (CDS), creDs < 12 C rms when testing flight grade NIRSpec detectors.The NIRSpec dewar is equipped with an internal cryogenic integrating sphere and Judson InSb diodes for measuring absolute QE.For the measurements that are reported here, an external monochromator was used to provide R = AI b.A "' 100 illumination.We refer the interested reader to :f\.Iott 15 for more information about the NIRSpec test setup in the DCL.
Tab. IV summarizes the performance of two new H1RG sensor chip assemblies (SCA) vs. NIRSpec requirements when operated at T =38 K.These are the first two of five prototype H1RGs that will be tested for NIRSpec this spring.For reference, we provide the average measurements for t he NIRSpec flight (854 & S55) and flight "pare (S58 & S60) detectors that use the old barrier layer design.These are the best four detector arrays produced out of approximately 60 that were made during NIR-Spec's initial production.If the two prototypes appear be be slightly lower performing in some area (e.g.dark current), it is important to recall that the current flight parts that serve as the basis for comparison were cherry picked from a much larger sample.
Fig. 12 shows that the measured responsive quantum efficiency (RQE) generally meets requirements to within t he ±10% zero point uncertainty of t he measurement.The improved barrier layer parts do not meet specification at every wavelength, but the same could be said for the old design parts.At the shortest wavelengths, the RQE is strongly modulated by the AR coating.The new barrier layers are buried deep in the detectors where only long wavelength photons can reach.There is no evidence that the improved barrier layer design has lower overall RQE performance thar.the old design.
Overall, testing at Teledyne, in the Goddard DeL, and at the University of Arizona has shown that the new barrier layer parts are very high performing, and certainly no , worse than old design parts when the performance is looked at comprehensively.The NIRSpec team plans to present more test results for improved barrier layer NIRSpec parts when the prototype program finishes this summer.Teledyne plans to present more information on the improved barrier layer design and performance this summer .14The University of Arizona and Canadian Space Agency will provide more information on the performance of improved barrier layer parts for the NIRCam and FGS respectively as part of the normal work of building the instruments.

C. Cryogenic Storage of Existing Flight Spares
Executive Surmnary 2d did not specify c~-ogenic storage parameters, but colder is clearly better.JWST accepted the DD-FRB's cr~-ogenic storage recommendation, and the flight spares that use the design that can degrade are being stored at the coldest practical tem-  NIRCam plans to store six flight spares at the University of Arizona.The two 2.5 Jlm cutoff H2RGs and four 5 Jl cutoff H2RGs will be stored in a commercial freezer at T~'-80 C. At the time of writing, the details of how NIRCam will store the parts are still being worked out.
fur NIRSpec and FGS, five H2RGs will be stored in a dewar in the NASA Goddard Space Flight Center Detector Characterization Laboratory (DCL) at T~60 K.These are as follows; have the potential to degrade and recommends cryogenic storage to slow the degradation rate.As of IIlarch, 2012, JWST is making additional H2RGs that use an improved barrier layer design that elim~nates the root cause of degradation.Furthermore, the JWST project accepted the DD-FRB's recommendation to store detectors at the coldest practical temperature.The NIRCam flight spares will be stored in a commercial freezer at T~-80 C. The NIRSpec spares and an FGS spare will be stored in a cryogenic dewar at T~60 K.
There are two important lessons learned from this investigation.One lesson.is the need to use reliability engineering in the production of complex detector arrays such ."HgCdTe detectors to predict the performance reliability for a given lifetime under specific operating and storage conditions.Accelerated life testing on test coupons such as t he PECs and sample SCAs provides critical information for the reliability analysis as well as information on potential degradation mechanisms, If a degradation mechanism is identified and correlated to a physical property of the detector array, it is important to use manufacturing quality assurance procedures such as destructive physical analysis as a screening test to remove parts that have a higher probability of failure.
A second lesson that enabled the succ",," of this investigation is the need to create a team of engineers and scientists from the appropriate government agencies, industry, and universities with diverse skills in the physics, fabrication, testing, materials science, reliability, and specific application of detector arrays to resolve subtle and complex detector degradation issues.This enabled the DD-FRB to quickly find the problem, fix the problem, and recommend a pa.th to move forward within the cost a!ld schedule challenges of the JWST P!oject.
Warm pixels are most easily seen in dark :mages.A dark image is one for which the detector is -enclosed in a c) Also at Sigma.Space Corporation, 4600 Forbes Blvd.}Lanham, 1-1D , 20706, USA d) Also f-t MEl Technologies, Inc. I!I)Also at Ball Aerospace, Boulder, CO, 80301, USA 3 completely blanked off dewar and continually reset until aU memory of the previous warm or illuminated condition is erased.Although dark images should be very stable at constant temperature, changes are easily seen in degrading detectors like those shown in Fig. 1.Soon after it was realized that the 5 I'm detectors were degrading, NASA initiated a 'Detector Degradation Failure Review Board" (DD-FRB) to address the following items.( a) Determine t he root cause of the detector degradation (b) Determine manufacturing and/or postmanufacture handling/process changes to avoid it (c) Define tests that are needed to screen-out degradation prone parts and ensure the continued integrity of flight parts (d) Define tests to determine whether the existing detectors are qualified for flight Within a few short months, the DD-FRB had detere mined root cause, identified design changes to eliminate the degradation, and also mitigat~ons in case the existing detectors had to be flown.The DD-FRB wrote four Executive Summaries and a Final Report for items' a-d.This article provides details for (a) and (d), while the informa.tionfor (b) and (c) contains technical information that is proprietary to Te!edyne Imaging Sensors as well as export controlled and International 1l:affic in Arms RegulatiOn.(!TAR) restricted.The !TAR is a set of United States government regulations that pertain to specified defenserelated technologies including JWST's detector systems.

FIG. 1 .
FIG.1.These dark images show t he degradation versus time of several JWST 5 P.IIl cutoff H2RGs.Each panel shows a dark image in inverse grayscale, where pixels with high currents show up as black.A dark image is a map of integrated charge uncler dark conditions.Parts prefixed with a "0' are NIRCam 5 jl-m H2RGs and parts prefixed with an "S" are NIRSpec 5 JJ.m H2RGs.Panel a) shows degradation in four NIRCam parts, b) shows degradation in the NIRSpec "flight" parts, and c) shows the degradation of a NIRSpec 'ilight spare".Each nark image is taken with the detector enclosed in a. completely blanked off deW"af.The degradation manifests itself in the appearance of greater numbers of inoperable «Warm" pixels.
FIG. 2. a)This figure shows a NIRBpec H2RG detector array.The H2RGs used by NIRCam and FGS/NIRlSS differ only in the mechanica.lpackaging.The photosensitive area measures about 36.72x36.72 mm 2 .The H2RG hBB 2040 x 2040 photosensitive HgCdTe pixels that are surrounded on all sides by a four pixel \\"ide border of "reference pixels."b) Indium bump bonds are used to join the HgCd'Th detector array to the silicon readout integrated circuit (ROlC).

FIG. 3 .
FIG.3.Example of the increase in dark count rate for one pixel of a. degraded detector.Here raw signal is measured in analog to digital converter units (ADU), and dark count :-ate is equal to the fitted slope.The blue data arc for a good pixel and the reci data are for the same pixel that has degraded with time.
20: DEFINE TESTS TO DETERMINE WHETHER THE EXISTING DETECTORS ARE QUALIFIED FOR FLIGHTThis section presents DD-FRB Executive Summary 2d that defines tests to determine whether the existing detectors are qualified for flight.It is presented in as close to original form as possible, with only minor formatting and changes.The original document, which documents the situation as of July 2011, is available from the JWST public web site.13 FIG.';, a) Pixel contact structure; b) Scanning Electron ~licr08cope (SEM) image of a. non-degraded pixel in NIRCam detector C105; c) SEM of degraded pixel in NIRCam detector C094 FIG.6.Degradation process in a. pixel due to inadequate barrier layer FIG. 11.Scanning Electron 11icroscopy i~geB of pixels from a. Process Evaluation Chip of NIRCam, NIRSpec, and FGS flight detectors.The presence oCIn-Au intermetallics from the breakdown of the barrier layer, 88 indicated by lighter shading in the In and Au layers, is present in all of the images.

&FIG. 12 .
FIG.12 .The responsive quantum efficiency (RQE) of the new i~proved barrier layer detector arrays generaUy meets JWST requirements to within the ±10% zero point uncertainty for these measurements.This figure shows the RQE of two ir.1proved barrier la.yer HIRGs overlaid on NIRSpec requirements (Red) and the average of the four old-design NIR-Spec '"fIight"' and ''flight spare" H2RGs (Gray).The HIRG proto~ypes use a NIRCam AR coating that is optimized for longer wavelengths tha.n the J\'IRSpec coating that was used for th~ old design H2RGs.When this is taken into account, the performance of the improved barrier layer design is no worse ~han the old design.

•
one complete NIRSpec flight spare focal plane array containing two 5 Jlm cutoff H2RGs, • two individual NIRSpec 5 JlID cutoff H2RGs, and • one complete FGS flight spare focal plane array containing one H2RG.V. CONCLUSION In this article, we presented the JWST DD-FRB's two public Executive Summaries.The first explains the root cause for why some of JWST's 5 Jlm cutoff HgCdTe H2RGs degraded after 1.5• 2 years of room temperature storage.The second explains why all of JWST's H2RGs built up through 2009 (both 5 Jlm and 2.5 Jlm cutoff)

TABLE I .
Selected J\\TST NIRSpec Detector Requirements

TABLE II .
HgCdTh sensors in the JWST ISI~!

TABLE III .
Key Physical Observations 1