`
`L. KOZLOWSKI et al., “2.5µm PACE-I HgCdTe 1024x1024 FPA,”
`
`Proc. IRIS Detector Specialty, August 1994
`
`
`
`
`
`TRW Automotive U.S. LLC: EXHIBIT 1015
`PETITION FOR INTER PARTES REVIEW
`OF U.S. PATENT NUMBER 8,599,001
`IPR2015-00436
`
`
`
`2.5 µm PACE-I HgCdTe 1024x1024 FPA
`
`L.J. Kozlowski, K. Vural, S.C. Cabelli, C.Y. Chen, D.E. Cooper
`G.L. Bostrup, D.M. Stephenson, W.L. McLevige and R. B. Bailey
`Rockwell Science Center
`Thousand Oaks, CA 91360
`
`K. Hodapp and D. Hall
`University of Hawaii
`Institute for Astronomy
`Honolulu, Hawaii 96822
`
`W.E. Kleinhans
`Valley Oak Semiconductor
`Westlake Village, CA 91362
`
`
`
`ABSTRACT
`
`Rockwell Science Center and the University of Hawaii have developed a short
`wavelength infrared (SWIR) 1024x1024 focal plane array (FPA) for the U.S. Air Force
`Phillips Laboratory supporting their Advanced Electro Optical System (AEOS) 3.67 m
`telescope project on Haleakala, Maui. First light in the University of Hawaii’s 2.2 m
`telescope was achieved two days ahead of schedule; performance highlights include
`read noise of 8.5 e-, FPA dark current <0.1 e-/sec, pixel yield >99%, quantum efficiency
`>50%, and BLIP-limited sensitivity at low-109 photons/cm2-sec background and
`operating temperatures to 120K. Though specifically developed for infrared astronomy,
`the device is extremely useful for surveillance applications. Proprietary hybridization
`and mounting techniques are being used to insure hybrid reliability after many thermal
`cycles. The hybrid methodology has been modeled using finite element modeling to
`understand the limiting mechanisms; very good agreement has been achieved with the
`measured reliability.
`
`
`1.0 INTRODUCTION
`We report the development of the world’s first high performance megapixel IR FPA,
`which is named the HgCdTe Astronomical Wide Area Infrared Imager (HAWAII). The
`University of Hawaii and Rockwell team succeeded in integrating the first photons two
`days ahead of schedule in time to observe the comet Shoemaker-Levy 9 collisions with
`Jupiter. Figure 1 shows photographs of Jupiter generated during the first-light effort.
`
`
`Figure 1. Images of Jupiter showing Shoemaker-Levy 9 comet collision sites using
`HAWAII 1024x1024 FPA.
`
`
`1015-001
`
`
`
`The HAWAII FPA is a hybrid consisting of a HgCdTe detector array fabricated on an
`alternative Al203 substrate that is flip-chip bonded to a CMOS silicon multiplexer via
`indium interconnects. Figure 2 shows the hybrid in its 84-pin leadless chip carrier
`package. Its 18.5 µm pixel pitch accommodates both SWIR to MWIR optics and
`fabrication of the large readout (20 x 20 mm2) using world-class submicron
`photolithography. Defect-free multiplexer yield >12% was achieved in the first lot made
`by Rockwell’s commercial production line in 0.8 µm CMOS; the yield exceeded our
`conservative estimate for the low defect density process. Growing the HgCdTe
`detectors on sapphire substrates is unique to Rockwell; the sapphire-based (PACE-I)
`photovoltaic detector arrays are fabricated on 3 wafers and have mean detector RoA
`product >106 -cm2 at 150K with dark current much less than 0.1 e-/sec at 78K.
`
`
`Figure 2. Photograph of SWIR 1024x1024 HAWAII FPA
`
`Though specifically developed for infrared astronomy, the FPA’s features and
`performance make it useful for surveillance applications. Table 1 lists its key attributes
`including 8.5 e- read noise, FPA dark current <0.1 e-/sec, pixel yield >99% and >50%
`quantum efficiency. The specifications translate to BLIP-limited sensitivity at low-109
`photons/cm2-sec background to operating temperatures approaching 120K.
`Table 1. HAWAII FPA Characteristics
`Parameter
`Value
`10242
`Array Format
`Pixel Pitch
`18.5
`Pixel Yield
`>99
`Detector Type
`HgCdTe/Al2O3
`Quantum Efficiency
`>50
`Cutoff Wavelength
`2.5
`Detector Dark Current
`<0.01
`105
`Detector RoA
`Optical Fill Factor
`>95
`FPA Dark Current
`<0.1
`1014
`Mean D* at T<120K
`Maximum Frame Rate
`3.8
`Minimum Read Noise
`8.6
`104
`Charge-Handling Capacity
`Maximum Usable Dynamic Range
`0.91
`Responsivity Nonuniformity
`<10
`Output Channels
`4
`Power Dissipation
`<3
`Maximum Data Rate per Output
`1
`
`Units
`
`µm
`%
`
`%
`µm
`e-/sec @ 78K
`-cm2 @ 150K
`%
`e-/sec @ 78K
`cm-Hz1/2/W
`Hz
`carriers
`carriers
`104
`%
`
`mW
`MHz
`
`
`
`1015-002
`
`
`
` 2.0 INFRARED DETECTOR ARRAY
`Large area, high performance FPA development is a multidisciplinary challenge
`encompassing detector physics, IR substrate growth, detector material growth and array
`fabrication, readout integrated circuit design, silicon multiplexer wafer fabrication, and
`hybrid assembly. Highest detector performance is normally accomplished by
`processing bulk material or growing the photosensitive layer on a lattice-matched
`substrate to minimize the potential for intrusion of yield- and performance-limiting
`defects into the photosensitive layer. This is normally accomplished in the HgCdTe
`material system via epitaxial growth on Cd(Zn)Te substrates using liquid phase (LPE) or
`molecular beam epitaxy (MBE).1 Large substrates are, however, not readily available
`and expensive; substrate availability is a key issue shared by many high performance
`detector materials including InSb. Since the problem is pervasive, fortunate ultimate
`detector performance is not necessary for SWIR astronomy (c 2.5 µm) since
`cryogenics for cooling to 80K are readily available at most observatories. SWIR FPA
`operation at 80K is compatible with extremely long exposure times due to very low dark
`current.
`Another key issue is the mismatch in coefficient of thermal expansion between the
`detector substrate and the readout integrated circuit; mitigation often requires that the
`active detector material be grown on alternative substrates. Such substrates can have
`properties better matched to silicon, which is by far the best material for the readout
`multiplexer. Alternative substrates also provide dramatically larger processing area,
`enable batch processing for increased throughput, and can reduce fabrication cost.
`Most significantly, the ability to place several die on each wafer greatly reduces the
`impact of random variables affecting each lot’s success.
`Rockwell has successfully developed large MWIR and SWIR HgCdTe FPAs by
`fabricating photovoltaic detector arrays grown epitaxially on sapphire (Al2O3)
`substrates. Figure 3 compares the chronology of HgCdTe/Al2O3 FPA development to
`that of the rest of the IR industry including HgCdTe/Cd(Zn)Te, InSb and PtSi. InSb is
`often believed to be the most mature second generation FPA material since it is a III-V
`compound semiconductor. The sapphire-based HgCdTe technology has nevertheless
`yielded the first 256x256, 640x480 and 1024x1024 FPAs. The HgCdTe/Al2O3
`detectors are referred to as PACE-I detectors; PACE is an acronym describing the fact
`that the devices present a Producible Alternative to CdTe for Epitaxy. InSb FPAs use
`bulk material which does not directly benefit from the burgeoning commercial photonics
`market using III-V materials.
`
`
`Figure 3. Chronology of mosaic IR FPA development (SC.3699CS.071594)
`
`
`The PACE wafers are prepared by first growing CdTe on the sapphire substrate by
`metal organic chemical vapor deposition (MOCVD). The photosensitive HgCdTe is then
`grown via liquid phase epitaxy (LPE) from a Te–rich melt on to the buffered substrates
`to produce 2 or 3 HgCdTe wafers. Epitaxial layer thickness, as measured by the
`
`1J.M. Arias, J.G. Pasko, M. Zandian, L.J. Kozlowski, R.E.DeWames, “Molecular beam epitaxy HgCdTe
`infrared photovoltaic detectors,” Optical Engineering 33 (5), May 1994, 1422-1428.
`
`1015-003
`
`
`
`nondestructive beta–backscatter technique, typically varies +2 µm across a 3 wafer.
`Both thickness and uniformity are controlled by optimizing melt temperature control to
`make arrays having high and uniform photoresponse. Figure 4 shows a 3 wafer
`populated with five 1024x1024 and four 256x256 NICMOS3 detector arrays. Each
`wafer hence provides both five chances for 1024x1024 success and also enables
`material characterization using the fully characterized NICMOS3 vehicle. This capability
`for making large, high performance detector arrays is unique to Rockwell.
`
`
`Figure 4. Processed 3 HgCdTe/Al2O3 (PACE-I) wafer.
`
`The photovoltaic detectors are formed by boron ion implantation at room temperature
`followed by annealing. The detector architecture is n–on–p. The typical goal is to
`demonstrate pixels with >90% fill factor and <2% crosstalk. The junctions are
`passivated by ZnS or CdTe. CdTe passivation has been demonstrated to be more
`radiation hard. After metal pad deposition for contact to the junction and ground, indium
`columns are evaporated to provide interconnects for subsequent hybrid mating. Recent
`development has focused on the 18.5 µm pitch of the 1024x1024 detector arrays.
`Detector illumination is through the sapphire, which is transparent for visible and
`infrared radiation to beyond 5.5 µm. Transmission can be extended to 6–7 µm by
`thinning the sapphire to <7 µm. Sapphire is commercially available in large wafers, its
`intermediate value (1.73) of refractive index acts to reduce reflection to a few percent
`over the broad spectral range of the detectors.
`
`
`3.0 CMOS READOUT
`The HAWAII FPA is structured in four independent quadrants having four outputs. Six
`CMOS-level clocks, two 5 V power supplies (one analog and one digital), and two fixed
`dc biases are required for basic operation. The multiplexer architecture has been
`optimized to minimize glow; lowest glow can be achieved by lowering the voltages
`below 5 V. The simple architecture also maximizes the fabrication yield.
`The multiplexer is an array of cascaded source follower stages separated by MOSFET
`switches. Each pixel’s signal voltage is first read through the first stage source follower
`consisting of a pixel-based driver MOSFET and a current source FET shared among the
`elements in a column. The column bus output then drives the output source follower or
`can be read directly to eliminate output amplifier glow in trade for reduced pixel data
`rate. The various MOSFET switches are appropriately enabled and disabled to perform
`the functions of sequential pixel access and row-based pixel reset. A maximum of six
`externally-supplied clocks is required. The CMOS-level clocks do not require precise
`adjustment for optimum performance.
`Direct Detector Integration. Direct detector integration (Figure 5), also known as source
`follower per detector (SFD), is used for interfacing the detector in the HAWAII readout.
`This simple scheme works very well at low backgrounds and long frame times. The
`frame rate used for testing HAWAII FPAs is typically about 1 Hz.
`
`1015-004
`
`
`
`
`
`Figure 5. Schematic of HAWAII Detector Interface circuit.
`
`IR Signal Integration. Photocurrent is stored directly on the detector capacitance, thus
`requiring the detector to be reverse-biased to maximize dynamic range. The changing
`detector voltage modulates the gate of a source follower whose drive FET is in the cell
`and whose current source is common to all the detectors in a column or row. The
`limited size of the FET in the cell constrains its drive capability and thus the maximum
`readout bandwidth to about 1 MHz. The overall gain reduction through the two source
`follower stages of the readout is 0.78 under nominal bias. Figure 6 is an oscillogram
`showing the outputs from six pixels under “light off” and “light on” conditions (upper and
`lower traces, respectively) at 312.5 kHz data rate. The double exposure shows the data
`rate capability at nominal bias. Data rate can be increased to 1 MHz in exchange for
`additional source follower amplifier glow, i.e., larger FPA dark current. The next
`oscillogram (Figure 7) shows the output of several rows including two complete rows.
`The incident light creates a signal approximately one-half the full-well capacity at
`detector bias of 0.5V. The full-well capacity is approximately 0.4V at 0.5V bias and
`>0.7V at 1 V bias. Both correspond to maximum charge-handling capacity of slightly
`over 100,000 electrons.
`
`
`Fig. 6. Output signal for several pixels under “light off” and “light on” conditions.
`
`Fig. 7 Output signal for several rows under “light off” and “light on” conditions for
`illumination half-filling the integration capacity.
`
`
`Elimination of Reset Anomaly. A performance-limiting issue with prior high density
`devices using a source follower per detector input with capability for individual pixel
`reset was a deleterious side-effect most often referred to as the “reset anomaly.”
`Charge redistribution amongst the two FETs in the reset structure of the NICMOS3, for
`example, generated random offset, which upset the reset level, thus degrading the
`FPA’s ultimate noise capability. While normally the uncertainty in the reset level should
`be limited by kTC noise, the charge redistribution raised the noise. The prior
`oscillograms visually suggest reset anomaly suppression. In Figure 6, for example, the
`reset, signal and delta voltages are consistent from element to element. The read noise
`data convincingly suggest that the reset anomaly has in fact been eliminated; we are
`hopeful for corroborative data from the astronomical community.
`Conversion Gain. Figure 8a is the transfer characteristic beginning at 4x109
`photons/cm2-sec background to slightly beyond saturation at 8x1010 photons/cm2-sec
`background (0.839 sec integration time). As expected due to detector capacitance
`modulation, the transfer characteristic has slight nonlinearity. Figure 8b shows the
`deviation from linearity expressed vs. percentage of the full-well capacity. The
`
`1015-005
`
`
`
`nonlinearity is fully correctable as evidenced by the excellent imaging performance of
`the various PtSi Schottky barrier detector FPAs widely available.
`
`
`Fig. 8 Photocurrent transfer characteristics: a) Signal vs flux at FPA
`b) Nonlinearity vs. percentage of full-well capacity.
`
`Figure 9 compares the readout conversion factor for 0.5 V and 1.0 V detector bias. The
`former corresponds to a conversion factor of 293 e-/mV and the latter 160 e-/mV. Since
`the total source follower gain reduction is 0.78, the detector capacitance is 20 fF at 1.0
`V bias in this case. The bulk of the measurements reported in this paper used other
`material with even lower 1.0 V capacitance of 18.2 fF.
`
`
`Figure 9. Gain determination for 0.5V and 1.0V detector bias.
`
`Read Noise. The SFD circuit is capable of very low read noise due to the small detector
`capacitance and concomitant high photoconversion gain. Assuming that correlated
`double sampling is employed off-chip, the percentage of BLIP, BLIP, is
`b
`g1 2
`A Q
`
`
`det
`kT
`2
`kT
`
`N
`
`qR
`qR
`
`where Rrst is the off resistance of the reset FET. The noise of the source follower
`amplifier is2
`b
`z2
`cos
`f t
`2
`1
`
`b
`g
`S
`f T
`2
`1
`
`
`D
`where Vn(f) is the MOSFET noise as a function of frequency, TD is the CDS time
`constant given by TD=t/2, and Sv is the readout conversion gain in V/e-. Depending
`upon the detector and its C-V characteristic, noise can also arise from nonlinear
`dependence of capacitance on voltage (and thus charge level). This second order
`noise can be roughly approximated by kTCdiff where Cdiff is the differential capacitance
`stemming from the bias shift.
`Though our determination of typical MOSFET Vn(f) has not been completed, very low
`read noise has been achieved using simple correlated double sampling invoked by
`clocking a reset frame and then two read frames. Using a low-pass electronic filter to
`match the noise bandwidth to the signal bandwidth, a mean read noise of slightly under
`15 e- was achieved at 0.5 V bias. When the detector bias was subsequently increased
`to 1.0 V, the mean read noise was reduced to 8.6 e- as shown in Fig. 10, a histogram of
`
`OQP
`
`
`
`int
`
`rst
`
`/
`
`2
`
`2
`amp
`
`
`
`B
`
`int
`
`int
`
`
`
`det
`
`A Q
`
`det
`
`B
`
`
`
`int
`
`LNM
`
`
`
`BLIP
`
`
`
`/
`1 2
`
`OQPP
`
`g
`
`df
`
`
`
`V f( )
`
`
`2
`n
`
`f
`
`
`LNMM
`
`V
`
`N
`
`amp
`
`
`
`
`2J.R. Janesick, T.Elliott, S. Collins and H. Marsh, "The future scientific CCD," SPIE Vol. 501, (1984).
`
`1015-006
`
`
`
`the noise for one quadrant. Since only sixteen frames were used for calculating noise,
`the 18.3% standard deviation is statistically rather than device-limited.
`
`
`Figure 10. HAWAII read noise at 1.0 V detector bias (single quadrant).
`
`Assuming that the HAWAII multiplexer has Vn(f) of roughly 4 µV/Hz1/2 at 1 Hz (i.e., the
`same as NICMOS3), a 1 minute integration time, 22 fF capacitance, and external noise
`of 35 µV rms,3 the predicted HAWAII read noise is about 3.6 e- with off-chip CDS and
`24 e- without. The 150Å dielectric used in fabricating the HAWAII multiplexer should
`reduce the source follower amplifier MOSFET 1/f noise to roughly 3.1 µV/Hz1/2 at 1 Hz
`based on extrapolations of test data from larger area MOSFETs. Assuming the lower
`MOSFET 1/f noise, the measured 18 fF detector capacitance, a 1 minute integration
`time, and 35 µV rms noise, the pre- and post-CDS read noise levels can actually be as
`low as 22 e- and 3.2 e-, respectively. Figure 11 shows the predicted read noise as a
`function of detector capacitance assuming the lower MOSFET 1/f noise. Included in the
`figure are measured HAWAII, NICMOS3 (256x256) and NICMOS2 (128x128) results.
`The measured values are in reasonable agreement with the predicted noise using the
`conservative NICMOS parameters. Table 2 lists the minimum and maximum readout
`characteristics including noise as have been described.
`
`
`Figure 11. HAWAII FPA read noise: predicted and measured values vs. detector
`capacitance using predicted MOSFET 1/f noise for 150Å dielectric of 0.8 µm CMOS
`process.
`
`
`3L.J. Kozlowski, K. Vural, D.Q. Bui, R.B. Bailey, D.E. Cooper and D.M. Stephenson, “Status and Direction
`of PACE-I HgCdTe FPAs for Astronomy,” SPIE Vol. 1946, pp. 148-160, (1993).
`
`
`1015-007
`
`
`
`Table 2. HAWAII Readout Characteristics
`Parameter
`Minimum
`Maximum
`1024 1024
`18.5
`84 Pin
`0.8 µm CMOS
`150
`SFD
`Off-chip CDS
`2.2
`5
`18
`35
`
`Format
`Cell Pitch
`ROIC Package (Optional)
`Custom ASIC Technology
`SiO2 Thickness
`Input Circuit
`Noise Suppression
`Supply Voltage
`Capacitance for 2.5 µm
`HgCdTe/Al2O3 detector
`Charge Capacity @0.5V
`Input Offset Nonuniformity
`Dynamic Range
`Data Rate
`MOSFET 1/f Noise at 1 Hz
`Measured Read Noise after
`CDS
`Predicted Read Noise after
`CDS
`Conversion Gain (Sv)
`
`10.2
`
`>0.1
`
`3.1
`8.5
`
`3.2
`
`3.4
`
`10.5
`<15
`1
`1
`4.0
`<15
`
`4
`
`6.85
`
`Units
`Pixels
`µm
`LLCC
`
`Å
`
`
`V
`10-15 F
`104 e-
`mV p-p
`104
`MHz
`µV/Hz1/2
`e-
`
`e-
`
`µV/e-
`
`
`
`
`4.0 HAWAII FPA CHARACTERIZATION
`The HAWAII FPA is the largest and most sensitive staring FPA now available. Figure
`12 shows a histogram of the measured D* at 3x109 photons/cm2-sec for the first hybrid
`fabricated. The histogram shows only 1/16th of the yielded population (98.9% yield)
`due to an earlier analysis software limitation which has since been circumvented.
`Another early limitation was the inability to analyze no more than four frames of data.
`The histogram width is hence limited by statistics and not device quality. The mean D*
`of 1.27x1014 cm-Hz1/2/W is BLIP-limited for the measured K-band quantum efficiency
`of 66%. The first FPA has total operability of 98.9% with quadrant operabilities of
`98.69%, 99.44%, 97.85%, and 99.64%. The first FPA was fabricated using lower grade
`assets to prove out the hybrid fabrication methodology. The two quadrants having
`<99% operability were hence limited by multiplexer column defects. The first
`multiplexer lot of 8 wafers yielded 11 defect-free readouts which were reserved for
`fabricating hybrids once the fabrication methodology was fully defined.
`
`
`Fig. 12. FPA D* for first HAWAII hybrid (9-26R-2).
`
`
`1015-008
`
`
`
`Four additional hybrids have since been made and three shipped to the University of
`Hawaii for observing the Shoemaker-Levy 9 comet collision with Jupiter. A second
`hybrid, for example, yielded similar sensitivity, albeit with slightly lower quantum
`efficiency. Figure 13 is an array map for the best quadrant, #3, showing BLIP D* of
`1x1014 cm-Hz1/2/W (3x109 photons/cm2-sec background) and 99.73% operability. In
`addition to our continuing work on hybrid reliability and hybridization methodology, the
`array map also shows that we need to improve the engineering associated with the
`HAWAII hybrid including minimizing edge defects arising from the dicing process and
`detector cluster defects.
`
`
`Figure 13. D* array map at 78K for Quadrant 3 of Hybrid 9-26R-4.
`
`The detector arrays have dark current <0.1 e-/sec; we are currently working on
`optimizing bias conditions for lowest glow. Detector quality has been corroborated by
`measuring mean D* as a function of operating temperature as shown in Figure 14. A
`detector RA-limited trendline is included for comparison. Device sensitivity is BLIP to
`about 120K even at the lower background of 4x109 photons/cm2-sec. The mean D* at
`130K is still >80% of BLIP, with mean value 8.35x1012 cm-Hz1/2/W and only slight loss
`in operability.
`
`
`Fig. 14. HAWAII FPA D* vs. operating temperature.
`
`The low read noise and dark current imply FPA suitability for surveillance applications.
`The FPA Noise Equivalent Input is 13 photons or 4.55x106 photons/cm2-sec. Figure 15
`shows the predicted D* as a function of background flux for several detector dark
`currents at 1 minute integration time. Included on the figure are two measured data
`points showing excellent agreement with the theoretical trendline. Though Figure 17
`suggests that the <0.1 e-/sec dark current is well below that needed for BLIP-limited
`operability at the two test backgrounds thus far studied, the specification derives from
`the even longer integration times (20 to 30 minutes) used for faint-object astronomy.
`
`Fig. 15. Predicted and Measured D* vs. Background Flux.
`
`5.0. CONCLUSION
`The HAWAII FPA has excellent characteristics for astronomy and surveillance
`applications including 8.5 e- read noise after CDS, dark current <0.1 e-/sec at 78K and
`high broadband quantum efficiency (>50% from 0.8 to 2.5 µm). The device was
`developed ahead of schedule in an aggressive development program with the University
`of Hawaii for the United States Air Force. We look forward to continuing these
`
`1015-009
`
`
`
`successful interactions including a related program that is helping to expedite MWIR
`640x480 FPA deployment.4
`The HAWAII has yielded several improvements deriving from lessons learned on
`NICMOS3. These improvements will result in improved science and include the
`following:
` Elimination of the reset anomaly present in 256x256 FPAs.
` Read noise reduction. The minimum read noise now ranges from 8.6 to about 15 e-
`using simple correlated double sampling, suggesting that further reduction is
`possible.
` Readout glow reduction. Though this effort is ongoing, the preliminary data imply
`success and potential for detector-limited dark current at the FPA level.
`
`
`We are working to further improve detector yield and hybrid reliability on both the
`HAWAII and NICMOS3 devices. A NICMOS3 with nearly 99.9% operability was
`recently delivered. In addition, we are working to suppress or eliminate image
`persistence in the detector. This factor is exacerbated by the extremely low dark
`current and high sensitivity of these sensors.
`
`
`4R.B. Bailey, L.J. Kozlowski, D.E. Cooper, S.A. Cabelli, W.V. McLevige, Y.C. Chen, G.L. Bostrup, G.
`McComas, K. Vural, J. Duffey, A. Kalma, B. Hawkinson, T. Bradley, F. Piotrowski and A. Hahn,
`“Performance, Yield and Reliability of 640x480 MWIR PACE-I HgCdTe FPAs,” Proc. IRIS Detector
`Specialty, August 1994.
`
`1015-010