These notes were taken by Fred Harris at the Scientific Detectectors workshop held in June 2002 in Waimea, Hawaii, USA. Please refer to the conference proceedings for more definitive information. Fred adds: You can talk faster than I can type; my apologies for any errors or omissions I may have introduced. Such "nicknames" as I have included are intended purely in the spirit of the fun atmosphere of the workshop. My heartfelt thanks to the organizers of this workshop for the enormous labors that went into its creation and operation. -Fred Harris Scientific Detectors Workshop FHH 2002-Jun-17 Morning Session: -1. Jim "Kaumanua" Beletic: how and where and when it works. 0. Fred Chaffee: Welcome - the early days of electronographic cameras (at Flagstaff): 15% QE for an area detector! (Scribe's note: this equipment is still in storage...) - And from there, we have now: 1. Ian McLean: Advent of CCD and IR array detectors - Can find "the book" (Electronic Imaging in Astronomy) by checking Ian's web site at UCLA. - Q: Why do IR arrays cost so much? A: The manufacturers want to get paid... - Manufacturers, remember the poor optical designer with regard to pixel size. 2. Mike Lesser: Back Illuminated 4Kx4K (Lesser and larger) CCDs - Differences in plotted QE are usually due to the AR differing coatings. - Fairchild CCD486: 80K e- linear range for 15-micron pixels. - JPL 4Ks: 5-nines-7 typical CTE. - STA 4Ks: 1.9 e- rms read noise found for best device so far. Many smaller low-noise devices hiding in wafer chords. - Kodak 4Ks: frontside, but many devices cosmetically flawless. Backside parts made in his lab on AlN substrate; 6 made so far. CTE at -100C for backside parts is excellent (versus older parts). - Thinning of 4Ks with 10-micron epi now possible. - Large devices have a lot of stress inherently in the crystal; a lot of work into keeping devices flat during packaging. - Q: Thinning process? A: Very similar to small devices; mechanical grind, then chemical etch, then final etch; agitation process is important. - Q: CCD486 readout speed? A: 200KHz/amp is fastest tested; read-noise numbers done at 100KHz/amp. - Q: Availability of backside 4Ks? A: all done with some partner, so must ask the partners. 3. Paul Vu: Scientific detectors at Fairchild Imaging - Fairchild Imaging now a separate company after prior ownerships. - CCD595: 9216x9216 CCDs with 8.75-micron pixels; frontside only at present. Is available. Readout rate 1 frame per second demonstrated. - Monolithic as large as 107 mm. - Incoming wafer inspection and characterization. - Backside passivation and AR coating similar to Lesser's. - Alternate CCD486 packaging; Invar package is smaller than the die for 4-side buttability. - For CCD486 non-butting Invar package: fine at -90C, starts bowing at -100C, so material needs to be replaced. - Custom processes: - window thinning for frame transfer, thicker in storage area; - metal bus to strap channel stops (minimizes ground bounce). - Q: CCD486 sold thinned? A: 80K USD for quantity 1. - Q: Flatness peak to valley? A: 20 microns typical; 10 microns possible. - Q: Intention to build deep-depletion devices? A: Not at present. - Q: Flatness at overhang with 4-side-buttable packaging? A: Rim is thick, so "window frame" of thick material is minimum of 300 microns. 4. Barry Burke: Back Illuminated CCD imagers - 3Kx6K with 10-micron pixels. - 2Kx4K OTCCD. - 150-mm wafers with 6 of 2Kx4K 15-micron-pixel CCDs. - MBE process for far-UV down to 200 nm; about 37 atomic layers thick, must be on the thinned wafers, so thinned except for rim. Almost 40% QE at 200 nm. Ion- implant laser anneal is the standard process, much lower QE at these wavelengths. - HfO2 AR coating, stable over >1 year time scale. - 2D scanned (intrapixel) photoresponse: - 36x reflective objective, 3-micron light spot; - to test for fully-depleted condition - illuminating center of pixel, 85% to 90% of electrons in just that pixel. - At longer wavelengths (near IR), "red halo": reflections from channel stops and undersides of topside gate overlaps are postulated. "IR mirror" technology enhances the problem. - Q: Studies on MBE in ionizing radiation environment? A: No. - Q: Made measures of QE in the UV at low temps? A: Have started; unchanged down to -75C, then a fall-off. - Q: Time scales for MBE part availability? A: 2 wafer lots completed with MBE, more in process, so lot of devices available. 5. Steve Holland: CCDs at LBL - "Fully depleted" is **not** "deep depleted"! - SNAP want a half-billion-pixel CCD. - Working with commercial vendor to transfer technology. - LBNL can do 4-inch wafers. - 60nm ITO AR coating (biased backside contact). - 80V backside bias for full depletion. - Pin-hole projection system allows 4-micron illuminated spot to test charge spreading; spreading minimization improves with increasing bias volts. - Read more at 2001-Sep NOAO newsletter (on line). - Current R&D effort: 150-mm wafers at Dalsa, back-end at LBNL. - Q: What is future availability after SNAP? A: Hope that technology is transferred. - Q: Packaging buttable? A: Working on 4-side buttable. - Q: For backside contact, how achieved? A: High-temp step, poly-Si doped as it is grown. Q: Optical qualities of poly-Si? A: 20nm thick, more loss than from single-crystal. Q: AR coating? A: HfO2 on top of ITO. - Q: Projected cost? A: More challenging with wafers >4-inch, but still possible. 6. Paul Jorden: (A lack of) Secrets of Marconi CCDs - Dither clocking: for non-inverted CCDs, modulation of parallel clocks can suppress surface dark current (from Burke & Gajar 1991). - Modulate clocks during "clear" cycle prior to integration. (Scribe's note: "Dither clocking" = Clocked Antiblooming.) - Deep depletion CCDs are 40-microns thick. Decrease in fringing is dramatic. - L3Vision CCD87 (512x512 with 16-micron pixels) soon. CCD60BI by end of year? - Graded-index AR coating for spectrographic work possible. - Q: CCD87 to be available thinned? A: Yes. - Q: What noise/stability levels on clocks on L3Vision? A: Need good stability of volts for stable gain; not definitive. - Q: Dither clocking in TDI clocking? A: DIVA observes x40-x50 dark current reduction (Klaus will post). - Q: Flying L3Vision? A: No radiation testing yet. 7. Jim Janesick: CMOS vs. CCD; pixel vs pixel - Sarnoff doing 50% CMOS, 50% CCD. - CMOS problems: - Many pixel flavors to choose from, so performance dependent. - $5 per megapixel CMOS detector, in quantity; half that in 18 months. - CMOS needs **major** development to equal CCDs in most performance areas. - Superior to CCDs in high speed (CDS per column). - DNs come out directly, so only see intermediate signals if specific tests points are included in test structures. - Fill factor: 0.18-micron and smaller design rules to improve fill factor. - Maybe 80% fill factor for a 10-micron pixel with present design rules. - Minimum of 3 MOSFETs per pixel. - For small pixels, pixel size might be less than the metalization stacking height, so light arrives through tunnels (bad for fast beams). - Low thinning yield (? due to 50-angstrom gate oxides getting mechanically stressed ?). - Sense-node cross talk, and don't want charge to diffuse directly into the MOSFETs, needs more work for thinned devices. - Built on 8-inch wafers, but Sarnoff can only thin a 4-inch region. - Runs on 3.3V or lower; less than 20 ohm-cm material, shallow depletion depths. - Fringing fields are weak (small) with such low voltages, making potential bumps. - Read noise limited by kTC reset noise. - About 30K e- full well with 4-micron pixels. - Comparable noise levels (4 e- rms) available with CMOS. - On-board ADCs can have bit-weighting problems; 12-bit ADCs are state of the art (16-bit ADCs mostly not to be found). - Changes of sense-node capacitance cause non-linearities. - Can have luminescence problems (mostly due to ground bounce). - Studying different geometries and designs. - Q: Rad hardness? A: Proton damage is less than CCDs. - Q: Hybrids? A: Customers are entertaining hybrids (CCD to CMOS); typically bumped, not monolithic. 8. Vyshnavi Suntharalingam: Silicon on insulator-based (SOI) single-chip sensors - Demo device: 128x128 CCD, 8-micron pixel; on-chip SOI-CMOS clocking, CCD ADC. - 50 nm of silicon on 200 nm oxide insulator, to make CMOS transistors; fully depleted. With transistors on discrete islands, no latch-up or crosstalk problems. Si layer too thin for optical absorption. - Backside thinned devices are 45-microns thick; 100% fill factor. - With CCD section at 3.3V, 100K e- full well with 8-micron pixels. - 640x480 frame-store CCD, 8-micron pixels, with on-chip timing and direct 12-bit charge-to-digital (QDC) converter. - Scaling down standard CCD output circuitry (20 nV/Hz^.5), can FET noise be low enough? - CMOS processes not optimized for analog performance. - Q: Low-light level multiplication possible? A: Should be possible. - Q: Commercially available? A: Should/will be. - Q: On 640x960, what is area of clock drivers? A: Less than 1mmx1mm. - Q: What clock driver output levels? A: 3.3V 9. Morley Blouke: CCDs at SITe - Tech updates: 12-micron pixels, flatter devices, broader spectral bandwidth with higher QEs, some new amplifier designs (not yet demonstrated). - Dual-layer coatings are HfO2 and MgF2, temperature stable (better than before). - 1Kx1K, 2Kx2K, 4Kx4K at 12-micron pixels to come. - Going to dual-row DIP instead of 4-side pin-outs. - 160K e- full well with 12-micron pixels; 4.7 microseconds for vertical transfer for thinned device; 2.5 microvolts/electron outputs. - Large range of volts for output amp for "sweet spot". - Q: Are new chips available? A: yes for 1K and 512. - Q: Read noise testing? A: Performed at 50KHz pixel rate. - Q: ESD on CCDs? A: No. ---------------------------------------------------------------------- 2002-Jun-17 Afternoon Session: 1. Ken Ando: Overview of Raytheon IR Operations - NASA plans to down-select between InSb and HgCdTe for NGST. - Hybridized Si PIN array to Aladdin mux; nearly 100% fill factor; drops in red and blue response at 30 K operation; this is the path for much larger formats (with smaller pixels, down to 8 microns). - Si:AS IBC: 15.5 e- rms read noise (Fowler 4) for region beyond 5 microns; also 412x512, 1024x1024 produced. - Four of Orion arrays fabbed to date (includes 1 science-grade array). 12-month lead time for orders placed now. - VIRGO-2K designed for J,H,K imaging; 9 dies per 8-inch wafer; 2 clocks to operate; designed to go down to 60 K, and with mods may go down to 30 K for InSb; 3-side buttable. - 3 of NGST dies per wafer, 2Kx2K, 25-micron pixels, buttable with 2mm gaps between detectors. - Future: InSb 3Kx3K soon, 4Kx4K monolithic someday (end of decade?); HgCdTe 2Kx2K; HgCdTe on Silicon with up to 2Kx2K with 4-inch wafers. - Q & A to follow Al Fowler. 2. Al Fowler: Orion Array - Reference pixels added since prior designs. - AR coating cuts off short-wise at 1.7 microns. - Invar 36 is substate material. - InSb is only 10-microns thick. - Different electrical interface: 37-pin MDM connectors, very reliable at cryo temperatures. - AlN motherboard between Si and Invar, patterned with gold traces, carries filter caps; output current mirrors for highest gain from source followers, with shut-down via single wire (to reduce dark current during integrations). - 5 mm dead space between butted modules. - Tile butted focal planes as you can afford detectors. - Mux has 2-micron design rules, avoids noise issues found in smaller design rules; Pwell CMOS has less glow at 30 K range; Double-metal single-poly improves readout speed compared to double-poly. - Hope to go to 8-inch wafers some day, to get more than 3 muxes per wafer. - Settling to 1.5 microseconds/pixel, so 64 outputs/device (effectively half the number of outputs of Aladdin); every output deals with 32 columns (not interleaved as in prior devices); can get rid of 50Hz or 60Hz noise. - 3-transistor unit cell. Can get references in several different positions. - Have PEDs: Photon Emitting Defects; are a nuisance rather than a problem (?), that can be removed by precision laser after covering with resist; removal blocks 5x5-pixel area typically. - Expect read noise to be about 25 e- rms or less with Fowler 1 sampling. - Future runs of Orion mask-set will use NGST mods offering lower noise. - Determined with 25 cool-downs, only 1 small crack, so satisfied with thermal stability. - Q: Problems with deep sub-micron CMOS design rules? A: Concerns at 30 K, are a little bit noisier; a lot is Al's "gut feel". - Q: What flatness when cold? A: Don't know. - Q: What alignment within package? A: Sub-micron with respect to alignment pins. - Q: Part to part piston variations? A: Want to hold to +/- 5 microns. 3. Lester Kozlowski: Rockwell progress - CMOS better than CCDs; read noise 5 e- rms at useful video rates. - Low CMOS amplifier glow with 0.25-micron design rules. - 2Kx1K devices with pixels about 3.3-micron pitch. - CMOS read noise equals or beats CCD at 10 MHz or faster read rates; at 75 MHz, CMOS < 25 e- rms, CCD > 80 e- rms. - These CMOS will end up in camcorders next year. - Hawaii-2RG and NGST device 25M transistors (per 2Kx2K). - Next year: 4 devices per 8-inch wafer. - Hawaii-2RG: Interleaving of full frame and guide-mode data enabled; internal gains selectable in 16 steps; in 32-output mode, 64x2K pixel blocks; guide areas arbitrarily selectable. - Q: CMOS fill factor makes CCD desirable? A: Not a problem with microlenses, if blue response not a problem. - Q: HDTV chips cost? A: $1100/each, kit is $10K. 4. James Garnet: Hawaii-2RG - As MCT absorber layer grown molecular layer by layer on HgZnTe substrate, so can monitor chemical composition; 2 layers per day per MBE machine. - QE 75% with no AR coating, will AR coat for NGST. - Growing on lattice-matched substrate (HgCdTe on HgZnTe) gives lowest defect rate. - Essentially non-existent persistence after over exposure. - At 60 K, diffusion-limited performance (J vs. V curve); bias V ranges -0.2 to +0.3 Volts; no evidence of mid-gap band traps. - Slight windowing around edge of Hawaii-1RG: due to mux. No glow with output-amps running. - Mean QE 82.6% with AR coating. - Q: When no glow, how many non-destructive readouts? A: Not seen with 10 reads when dark current less than 10 e-/1000 seconds; significantly below 1 e-/pxl/read. 5. Don Hall: UofH MBE HgCdTe NIR FPA - 2RGs have Schmidt triggers added to the clock inputs. - 2 e-/hour mux dark current. - Gains 0.86 with pull-up resistor, 0.95 with current mirror. - Thermal coupling to package done via mounting feet; pseudo-kinematic mounting: one foot fully constrained, second in slot, last two fully unconstrained. Piston tolerance about 10 microns. - 3-side buttable, pseudo-4-side buttable. Loose about a 2.5-mm gap on 3 sides, about 5-mm gap on fourth side to another's first 3 sides. - Carriers are Molybdenum (Mo) rather than AlN (Mo is a perfect CTE match to HgZnTe substrate down to 30 K). - All wiring under "shadow" of package. - Test head pulls down to less than 10^-4 milliTorr in less than 4 hours. Can accommodate 8Kx8K mosaic. Want to do 4Kx4K characterization with 1-week turn- around, of which 4 days cool-down, 2 days test, 1 day warm-up. Bottom section can be placed on back end of UofH 88-inch scope (goal by end of year). - Running off ASICs should eliminate Leach controllers. - Q: After test system goes on scope, can it come off? A: yes, maybe via spare parts. - Q: PEDs seen? A: Not in 20 parts. 6. Mark Clampin: ACS - STIS 1Kx1K CCD thinned by Lesser for HRC. - In WFC, no correlated noise but there is channel-to-channel crosstalk (ghosting). - In SITe CCDs, 100-200-pixel-wide halos in initial CCDs; metal coat over frontside prior to thinning reduced the effect, but introduced a diffraction-like effect in one CCD axis (presumed due to clock lines and overlaps). - Annealing to +20C makes 80% of hot pixels go away (then they start creeping up again); annealing happens about once a month. - Q: Annealing higher temperature on ACS than WF/PC-2? A: No. - Q: Can higher annealing temperatures be achieved? A: No, due to concerns for higher outgassing. Special Presentation by Cyril Cavadore: - Picture: Alain Maury at the console... This talk is given in his name. - Asteroid Blouke has a less eccentric orbit than asteroid Bredthauer... - Asteroid Veillet is much closer to home (a lot closer to Mars...) - Asteroid Cuillandre will be given when he wakes up. - Special thanks to 3780 Maury for providing the "rocks". ---------------------------------------------------------------------- Round (in a linear way) Table of detector manufacturers: - Q: Some have been burned by fab runs with bad epi, that ruin most devices. How do we get better epi in the future? A: Paul Vu screens incoming wafers: firm fixed-price contracts puts the burden on the manufacturers. Qualify several different silicon vendors, so not locked to a bad source. - Q: Is there better qualifier other than making a CCD? A: Vu: You have to qualify ALL wafers in a lot. - Blouke: Possible to qualify the vendor and the machine of the vendor for the best epi, but it takes a lot of effort. - Q: Current limits on focal plane flatness and metrology? CCD, CMOS and IR: A: Test system with a large window to be able to watch an IR 2Kx2K as it cools down, shooting for peak-to-valley flatness of 12 microns. Can measure it warm, and for some technologies might expect it not to change as the device goes cold. - Some groups test via Zygo interferometer as device cools down; for IR arrays, flatness translates into yield successes. - For indium bumps, flatness variations greater than 10 microns turn into defects. - For Fairchild 4Kx4K CCDs, after assembly, center of array is always curved (high) by up to 50 microns. Cure is care during fixturing for die-attach process; this yields flatness to better than 20 microns. - Marconi numbers similar to Fairchild; Invar 36 is not the best match for substrates. - Q: Not so happy about flat-field variations at the extremes of wavelength range. With fringing problems in mind, what about the blue end? Image on screen shows what is presumed to be a laser-annealing residual; 30-50% modulation in the extreme blue, very difficult to calibrate this out. What to do to get less pattern? A: This is similar to the "brick-wall" effect. One way to eliminate this is the MBE process, but $1.5M quipment investment. Not convinced that this can be eliminated with any laser anneal short of a laser able to do the entire anneal in a single flash. The present laser step-and-repeat needs overlaps, leading to inevitable patterning. Sarnoff/RCA alternative: do the backside implant prior to the frontside metalization (which is a trick with the already thinned wafer, even when "edge thinned". A special mask with distortions to match the CCD's distortions has been tried, with some success. - A flash lamp has been tried, but it has blown up the CCD. - The heating MUST be confined the the back 0.5 micron. Multiple passes are already being made with the laser. - Q: Lesser's process, and MBE, don't have this effect? A: SITe CCDs and Lesser's process don't show this. - Lesser process has no annealing step. - Q: Stability of Lesser's process? A: With new coating using silver, no discerned effect. - Also, another solution is to go to hybridized devices. - Q: CMOS versus CCD: a lot of parameter space where CCD is more desirable. Do we stop paying attention to CCD vendors, and go CMOS? A: 17 years ago, CCDs were intractable as production devices. What has made the CMOS emerge is the current maturity from manufacturers. Adding more transistors can fix some of CMOS' present problems. CMOS will be more prevalent in 10 years. - Q: Limit to spacing of bumps for CMOS hybridization? A: For hybrids, last year was 18 microns, now working on 13 microns, trying for 8 microns in 2 years. - Q: Talking about monolithic or hybrid CMOS? A: For highest performance, the hybrid is best, but not cost effective if budget driven. HDTV has been won over with CMOS. - Q: Will hybrids ever be less expensive / more affordable? A: Stacked devices by IBM and similar developers will push the technology. - Devices must be done in lot-run quantities. Monolithic CMOS images are cheap. - Silicon-PIN hybrid will offer performance; reasonable in terms of total cost of the instrument. Buy the development lot, not just one detector. - Q: Application for 300 devices with 4 billion pixels. Can be done with hybrids? A: Do it with hybrids, can eliminate the big racks of support electronics. - Paying $10K versus $300K for a hybrid works. - The $10K/hybrid-device with high pixel counts is realizable in the visible, but not the IR. - At present, can't get such optical hybrids, so hard to plan for presently funded projects. - Q: Can we buy an optical hybrid? A: For classical CCDs, 4Kx4K devices can be bought off the shelf. - Manufacturers can/should put out more info on these products. - Make collaborations, they will provide advertising in this community. - Q: In space-based instruments, what about redundancy? CMOS versus CCD? Which is better for redundancy? A: For rad hard, CMOS is generally better. For CMOS, must invest more work to protect against single-point upsets. CMOS makes no hot pixels. - Silicon-PIN hybrid will win in rad hardness over monolithic. CCDs will always be soft compared to CMOS. - On space shuttle, CMOS is replacing HDTV CCD due to rad hardness. - CCDs are adequate for ionizing radiation (to a megarad) (but not neutrons). - Most imagers in space **are** CCDs. - CCDs in space, some more than 10 years. Highly orbit-dependent for damage versus lifetime. - SOHO CCDs up since 1993 in LaGrange point, not suffered much CTE or hot pixel degradations. - End of life specs may make CCDs look bad, but CCDs can do well for much of mission life prior to the end. - Q: Flight heritage for CMOS? A: Use IR detector heritage. - Q: Are all hybridization issues solved? A: Not safe to say; not production yet, hybrids are hand made. - Q: CCDs hybridized versus temperature cycling? A: LN2-and-back cycling is done now, but drives the price up. Monolithic is where the prices drop. - In IR hybrids, different problem: different coefficient of thermal expansion (CTE) between mux and detector. You can force one to match the other; tricks must be used; forces the silicon readout to match the detector, without warping. - In past, delamination at edges (WFC 3). Torture chamber built to cycle between reaching LN2 and 300-degree heat lamps, and the 1K with 28-mm diagonal did survive after many cycles. 2K has not been through torture chamber. Failures usually show up in the first 10 thermal cycles. ---------------------------------------------------------------------- Evening Session: Dick Bredthauer: High-resolution, high-speed CCD for cinematography - Critical: alias-free resolution, 24 fps, no chromatic aberrations => 4112 x 3138 pixel CCDs (IMAX = 5464 x 4096), 15-micron pixels. - 16 outputs at 40MHz (3-stage source followers); 12 bits/pxl => 1.3G Bytes/sec data rate. - 3 CCDs, R + G + B for color camera. - Use of standard lenses (so how does the larger-than-film CCD see the same-as-film image scale?) - It is harder to master MicroSoft than build a digital camera... - Rotating shutter to replicate blurring, a conventional shutter would produce images too sharp for the historical expectations of the film industry. - Lack of temperature stabilization in the electronics makes for "tiling artifacts" from the 16 ADC channels/chip. Jean-Charles Cuillandre: Digital photography - Gets his asteroid. - Took some pix, did some tricks, showed some flicks. - Q: How many batteries? A: A motorcycle battery, which must be kept vertical... - Next for panning, a motorized equatorial mount. - Q: What elevation for Dolomiti? A: 2700 meters. - Q: Time-lapse acquisition frame rate? A: Digital camera controlled by serial link from the computer, converted to USB; 8 seconds between frames. - Q: 3D illusion in stellar zooming? A: Artifact of the projector? - Q: Video available? A: Not yet, still in the planning stage. ---------------------------------------------------------------------- 2002-Jun-18 Morning Session: 1. James "Kaumanua" Beletic: Keck Detectors - Doing laser guide stars, effective 9th mag, good enough 700 frames/sec for AO. - Only large scope going out to 20 microns? Not for long. - Older Tek2k detectors should be changed out to MIT/LL or LBNL 2Kx4Ks. - Need new guiders. - Q: Upgrades to more focal positions to avoid connector cycling? A: More will happen than has been presented here. 2. Tetsuo Nishimura: Subaru Detectors - Detector industries in Japan not much interested in astronomy applications. - Only scope of this aperture class with a prime focus. - COMICS will be 6 of Si:AS 320x240 IBC detectors go out to 20 microns; Kataza control system. - Messia 3 electronics "falling apart" with time, needs replacement. - Q: How "falling apart"? A: Communications ICs failing, hard to find replacements. 3. Doug Simons: Gemini Detectors - Only 2 fully-accepted instruments. - IR wavefront sensor. - 5 separate optical ports on back of Gemini north. IR-optimized scope. - Detector costs are now only a few % of cost of instruments; as they get pixel hungry, this may grow to 50% of instrument cost. - Q: Standardization of support electronics? A: Not standardized, similar problems to Keck multi-vendor situation when spares needed. - Q: In what readout mode the Hawaii-2RG? A: Not known yet, could be 4-channel to ASIC or otherwise. 4. Derrick Salmon: CFHT Detectors - 60-micron focal plane spec beaten by a fair bit by Luppino's measurement and lapping skills. - Build new scope upper end to mount big prime-focus camera. - IR camera optics typically all refractive. - Q: Pulse tube cooling: No problem with orientation? A: Has been tested, did have problems initially; reoriented tube, no problems between zenith and 70- degrees (? from zenith ?). 5. Dietrich Baade: VLT Detectors - Want detectors with wavelength resolution. - FIERA will be upgraded to PCI bus. - ESO second generation detector controller? - Q: Why next workshop in 2005? A: Is what has been done in past, could be sooner if developments come quicker. 6. Gert Finger: ESO IR Detectors - Hawaii-1 detector odd-even effect: caused by the clock in the fast shift register. - SI:As BIB array goes to 28-microns; 320x240 array size. - Q: Have the transputers been designed out of the IRACE system? A: Transputers are not the most obsolete component that needs to be changed. 7. Barry Starr: NOAO Observatory Plans - 8Kx8K OTCCD focal plane planned: 8x8 unit cells. - New image-acquisition system is "community-wide" = MONSTEROID. - Monsoon electronics interface close to focal plane. - Flex-PCB vacuum interface electrical path. - LSST: 1400 of 1Kx1K optical detectors => 1.4 Gigapixels; want readout in less than 3 seconds... 200Kpxl/sec/amp. - Q: LSST first light? A: Costed proposal in 1 year from now. - Q: Why flex-PCB from Monsoon backplane? A: Don't want to work with discrete wires ever again; can control crosstalk and impedances; can test as assemblies before installing in the dewar. - Q: How fit 50cm filter at LSST? A: Not trivial, not yet addressed. 8. Ricardo Schmidt: CTIO Status and plans - With SI002AB CCDs, 100 seconds readout time (16 channels for 8 CCDs); 70Ke- full well at 6 e- rms read noise; digitizing to 15Ke- range gives 3 e- rms noise. - SOAR dewar holds 6 liters LN2. - 37-pin Micro-D connectors from AirBorn easier to use than Nanonics, but may be harder to find. 9. Derek Ives: UKATC Detectors - Ultracam: GPS system for absolute time stamping of images. - IR Array camera: - CCD47-10 guider for fast tip-tilt guiding. - 4 detectors, 90% detector spacing, 5% overlaps with 4 exposures to make 4x4 effective image array. - CCD mounted in midst of IR arrays, in stainless-steel package to Faraday-shield IR detectors from CCD drives. Can this be done while keeping the IR detectors clean? 10. Ralf Kohley: GTC CCD cameras - www.gtc.iac.es for project office. - Guiders developed by OCIW, modified by project office; 11 guiders systems. 11. Jason Griesbach: AAO Detectors - CCD parallel clock drivers are fast-DAC driven for custom wave shaping. - Serial clocks with programmable ramp transitions (trapezoidal?). - SW and R clocks are low noise. - Controller may be available to the community in the future. - Q: Readout rate? A: CCD controller: 500Kpxl/channel is speed limit. - Comment: MIT/LL CCDs can be read out at 1Mpxl/sec. ---------------------------------------------------------------------- 2002-Jun-19 Morning Session: 1. John Tonry: Orthogonal Parallel Transfer camera - 3K-10K shifts typical in an exposure; minimum PSF degradation. - "Guide" area is also OPT so guide star can be moved in the serial direction while still in its image area, moved to just above its readout amp, then make parallel shifts to get to the output. - Image motion nearly as big as intrinsic image size. - As function of image separation, image improvement of .86 out at 12 arc-min from guide star (at Mauna Kea). - At Mauna Kea, all but 0.3arc-sec of seeing is at **low altitude**. - Correction at 10Hz is usually adequate. - Q: What bands? A: R and I. - Q: Quantify image quality degradation versus CTE? A: 10^-5 e- left behind per transfer. - Comment: Near is not Kolmogorov; agreed. - Q: Finer pixel scale? Pixel shape changing as you move around? A: The paired triangles is preferred by Tonry. Barry Burke has alternative intrapixel shapes. - Q: Don't stop at intermediate pixel phases (therefore, always move by integer pixels)? A: Yes, so shape of integrating phases always same in this clocking scheme. 2. Martin Roth: Ultra-Deep Spectroscopy with PMAS using nod-and-shuffle - Using same slit and CCD detector pixels, do integrations of spectra; move charge off of CCD area (vertical transfers) and beam switch between object and sky; reregister object and sky spectra by suitable vertical transfers: 2001, PASP 113,197. - Factor of x10-20 improvement in sky subtraction. - PMAS: lens array to fiber bundle to fiber-fed spectrograph: no slit effects. - Can have only 1/2 number of spectra on sky as fit detector, to allow room for the shuffled "beam-switched" spectra in between. - Q: Dispersing across the gap between butted chips? A: Yes, since spectra are shuffled via vertical transfers and the CCDs are narrower in the horizontal. Want to change to a monolithic 4Kx4K CCD. Using SITe SI002A CCDs at present. 3. Philippe Feautrier: NAOS visible wavefront sensor - Marconi CCD-50 used. - Tip-tilt at 250 Hz with 185 actuators on deformable mirror. - IR and visible wavefront sensors up to 670Hz frame rate; CCD-50 at 500Hz, 7 e- noise; only 14 of 16 outputs used. - Shack-Hartmann wavefront sensing; 2 microlens arrays in cryostat, close to CCD, can be exchanged, at -110 C. - Aligning microlenses: moved at room temperature, then camera cooled and alignment checked; iterate if necessary to establish repeatability and accuracy. - V mag 17 AO loop closed on VLT, up to V mag 18. - Q: Is IR wavefront sensor working? A: Yes. Only problem, VLT secondary mirror has movement. - Q: Why bandwidth limited to 440Hz? A: Decided to limit bandwidth because faster increases readout noise; at this limit, 6 e- noise, faster goes up to 9 e- noise. - Q: 3 e- at 50Kpxl/sec readout not impressive; what timing? A: Clamp and sample with no summing well on the CCD, older version amplifier than newer Marconi CCDs. 4. Jean-Charles Cuillandre: CFHT's SkyProbe - V filter in front of camera lens. - Integrated Linux box rides telescope 3 meters from camera, then onto network. - 35 arc-sec/pixel. - 30 second exposures => V mag 11.5, exposure every minute. 0.7% accuracy after removal of intra-pixel QE variations. - Undersampled data with front-side CCD (KAF0401). - Due to one phase poly and other phase ITO, variation of response within pixel. - Q: How stable CCD camera system? A: Sky noise much larger than camera noise and and stability. - Q: Kodak is changing to CCDs with microlenses. Can this system still work with such a CCD in the camera? A: Yes, it is expected that the effect of the microlenses can be calibrated similar to the present intra-pixel variations. 5. Fernando Pedichini: Prime focus CCD cameras for LBT - The Marconi CCDs are sufficiently flat for the fast input beam: peak-to-valley less than 7-8 microns. - Custom mask on the last 200 rows of the MAT42-10 to act as a frame-store region. - Beam splitter with reflector to give simultaneous inside-and-outside focus images for AO system. - Spherical LN2 container made by electroforming in one piece, copper inner with nickel outside for strength; spherical since ratio of volume to surface area is maximized with sphere. Fill pipe to center of sphere. - Outer Sphere machined aluminum with electroformed nickel outer, polished. - CCDs mounted to Invar plate. - 4 getter containers: 1 with zeolite, 3 with coconut charcoal. - Marconi says 2-degree K shift in cold temperature does not effect CCD characteristics. - 15-sec readout (single-amp)/8Mpxls with 15 e- noise. - In-vacuum PCBs on Kapton substrates. - Q: Relative cost of this cryostat? A: 20K USD; built in Italy. 6. Bonner Denton: Astronomy impact on chemical analysis. - Given at 0.6 Dentons (Bonner pleads mitigating circumstances). - Prior Faraday-cup technologies require 6K ions/second for detection limit; using modern detector techniques, 50 ions definitely detectable, working down to performance of 10 ions. ---------------------------------------------------------------------- 2002-Jun-19 Afternoon Session: 1. Don Groom: Cosmic rays and other nonsense - (a microsquish-free talk...) - Short stubs of tracks: muons - 95% of what is seen at sea level - Tracks don't ever bend - Worms (wandering curving tracks): wandering electrons being Compton scattered off of other electrons. - Spots: also as worms - Potassium 40 is everywhere, major source of background. - 1 cm of Lead cuts such counts by x5. You don't win much with any more shielding. - Black die in IC sockets is bad as a radiation source. - Much potassium 40 (K40), 3-4%, in BK7 glass. - Q: Worms not seen in CMOS devices? If you don't have 100 microns of silicon ( CMOS has 5 microns) then harder to see. - Q: Iron in concrete? A: Not the problem. Also, Cobalt 60 put under fire bricks in steel production, to monitor when the bricks are wearing out, so much Co60 in modern steel. In concrete, Uranium and Thorium decay products, and K40 are the problems. 2. Klaus Hodapp: Hawaii-2 detector arrays - PACE-1: fringing, glowing; some show more residual image than others. - Hawaii-2: no electrical lines cross quadrant boundaries; pin-grid carrier, with external rows of pins for electrical, internal rectangle of pins for thermal connections. - ZIF socket makes electrical connections to the HI-2 electrical pins. - New reference pixels do OK, but not great, job of tracking temperature-dependent drifts/offsets. - Glow effects visible at amplifiers. - Q: Reference always buffered? A: Always buffered. - Q: Hawaii-1 problems solved in Hawaii-2 devices? A: PACE-1 material similar; sapphire layer still shows fringing; glow levels reduced to levels acceptable for spectrographs. ---------------------------------------------------------------------- Roundtable: Issues of device characterizations in general: Q: Optical CCD characterizations? A: Hard to measure items: UV response, red/IR fringing. Need to close the loop with vendors from lab test results (such as AR coatings). Measuring items harder to measure on the telescope. A: Other issues: crosstalk: we need to share expertise: not entirely under control, is it in the CCD or electronics?, (power-supply) bypassing questions. A: In Rockwell 1-R and 2-R, good to include reference pixels, also used with capacitors to simulate non-optical pixels. Q: What is the best configuration of reference pixels? Which version? We need to feed this back to the manufacturers for better structures or eliminating unneeded structures. A: Try to correct with reference output(s), but reference does not track with temperature drift. Using dark pixels, the corrections of low-frequency noise works quite well. About persistence, MBE on a PICNIC array, did not see any persistence; will be repeating measurement. Think that with MBE persistence will be better. A: Regarding reference pixels, references with capacitances mimicing the pixel capacitances work much better. The reference pixels at high precisions have their own problems. Reference pixel situation is under control in the Rockwells. In the HgCdTe's, persistence is down, and residual image is under control. Must try to get different labs to make consistent measurements (that agree); this is difficult; getting agreement on read noise is hard (different tau's, bandwidth). A: If you are going to compare numbers, you must have similar time-domain operation. A: Dwell times on CCD sampling must be reported along with the read noise found. A: Specify settling time and read rate. - Different panel members have different techniques. Some figure these temporally rather than spatially. A: Must set sampling method standards to differentiate between spatial and temporal sampling. A: 32 versus 4 outputs: time between reading video pixels and reference pixels is different. A: On Orion, get a reference every 32 columns of data, for reducing low-frequency noise and herringbone. Q: How do you do absolute QE? A calibrated photodiode for 1 to 2.5 microns? A: A Planck curve with measured filter (at LN2 temperature) is best; calculate how many photons on your detector; you must know the geometry. A: In CCDs, between 300-400 nm hard to get agreement with the manufacture. A: In IR, less-steep part of Planck curve, so IR is easier. A: Main uncertainty in IR comes if you are not shot-noise limited. A: In CCDs, many people use X-rays rather than photon transfer. Q: Do X-rays and photon-transfer curves differ? A: Yes, due to charge transfer issues. Q: Is there something similar to X-rays for IR detectors? A: Will be tried but answer not known yet. Q: When Lesser measures QE, is it done cold? How in UV? A: Custom-calibrated diodes. They degrade a few percent per year. Bad at 200-300 nm range. Calibrate at yearly basis. Measure both room temp and cold, and say at what temperature the measurement was made. A: QE can depend on temperature. At IR, direct QE measure is difficult, if you don't know what your geometry and fill factors now. With variable-area diodes, can determine some more parameters directly. A/Q: Incident photons on the focal plane are the issue. What if you don't collect them all? What if you scan a small spot in the IR? Do you see non-uniformities? A: IR QE "holes" are seen. Q: Has anyone done IR sub-pixel spot scans (intra-pixel variations)? A: IR Manufacturer(s) do not have a good technique, either for MTF or pixel-to-pixel crosstalk. MTF is a function of wavelength for backside-illuminated arrays. A: We can find the solution to these problems, where it counts. Line widths in spectrographs for instance. A: On Hawaii-1, have made a spot scan. Don't know exactly how large the spot is, but presume it is diffraction limited. Integrating across all of the pixels responding to the spot, and moving the spot, do not see variations in the TOTAL signal. Q: Is anyone testing? A: Stepping with 1/20 of pixel steps, takes a long time. A: Put the arrays on the scope, and see how they perform. How to make a spot measurement that people believe. A: Cd109 makes true event spots. Q: How do you disentangle when the particle goes through the readout rather than the front-end? A: You can get it with cosmic rays, but you don't know the illumination. With a monochromatic X-ray source, easier to do. Janesick wants Fe55 testing of IR. A: Do tests separately on bare muxes and hybridized arrays. A: X-rays deposit a ball of charge. Cosmic-ray muons have charge diffusion, can tell by pulse-height. Q: Using reference signals: should we be multiply-sampling reference outputs? A: First clock to end of fast shift register, then make 16 conversions and average them, so noise does not go up by root 2. At end of row, do this again, then linearly interpolate between beginning and end. Q: Does this help? A: Yes, this helps with suppressing 50Hz. Q: Problems with 50 or 60Hz? A: If integration time is multiple of 50 or 60Hz, then you don't see it; being synchronous with line frequency maximizes the suppression. A: With leach controller, problem of switching power supply... A: Could test by synchronizing the readout. A: Also with Leach, watch out for the fan. A: With Leach, put your tongue on the power supply (?case?); if you get a shock you have a problem. Q: Mechanical flatness of Hawaii-2 detector when cooled? A: Not yet measured by the panel. Q: Charge persistence in Aladdin arrays? A: Some with none, some with 1% or more. Thought to be in the processing. If you drop the bias across the array to zero, then re-establish, you eliminate the residual image, but the array takes a while to settle down. By varying the bias, might eliminate the residual image and minimize the drift; maybe don't have to go to zero bias. A: With Aladdin-3, when you go to zero bias and then restore, dark current is highly elevated, takes 30 minutes to come down; so just read out the array for the 30 minutes. Q: CCD residual image: takes many hours to recover? A: Not only the CCD that can respond this way, some windows do this also; must buy UV-grade windows. A: In InSb, Raytheon has some handle on what causes persistence, but reproducing this via adjusting the process is not yet well defined. A: For HgCdTe on HgZnTe, after over saturation, no residual image detected. Also don't see the fringing. For space, the substrate is removed, no fringing seen. Q: Amplitude of fringing on IR arrays with transparent substrate? A: 5-7%, 7% typical fringing amplitude. A: With Aladdin, not until longer wavelengths (5-5.2 microns) where photon goes through and then can come back. At shorter wavelengths, where absorption length less than thickness, fringing not seen. Q: Coating on PACE material? A: Have to put it on after hybridization so it does not get damaged. ---------------------------------------------------------------------- 2002-Jun-20 Morning Session: 1. Mike Lesser: Fully buttable imagers (FBIs) - Want more standardized (packaging) process. - LSST focal plane is 55 cm in diameter (not known what the shutter is); possibly Xenon gas-filled focal-plane area (not a vacuum dewar). - Chip-on-board has 5% packaging gaps; ceramic 3-D I/O could have 1% gaps (100 microns). - AlN substrates allow efficient TEC cooling; .040-inch thick, flat to 3 microns; holes drilled via laser. - AR coating shows 1% uniformity over 150-mm wafer. - Edge chipping is most critical problem when cutting through entire hybridized stack. - Metal frame on underside provides thermal cooling path. - Experiment: bumping pads on top of (frontside) pixels on the silicon. - Q: Have you found radioactivity in AlN? A: Yes, in past; 3 vendors used over time but don't do qualifying tests. - Q: What is done to AlN surface? A: Ti-tungsten-gold; no optical processing. - Q: Why 1Kx1K for experiment? A: Devices made in chords of wafers. - Q: Work with thick hi-rho with 40-50-micron thickness? A: Nothing for resistivity higher than 150 ohm-cm. - Q: See any problems with hi-rho? A: No problems anticipated. - Q: Have looked at shorter wavelength for print-through from bumps? A: See it a little in the red, have looked shorter, but test-CCD CTE was a limitation. Don't know if print-through is from the bumps or the pads-over-pixels. 2. John Tonry: Orthogonal Transfer Array - Use OTCCDs to look for killer asteroids. - POI = Panoramic Optical Imager; 1-3 Gpixel detector focal plane per scope. 16-inch x 16-inch detector area. - Detector costs: CCDs $0.01/pxl; IR hybrids $0.10/pxl; OT array $0.001/pxl. - Decreasing pixel size: 10-micron pixels cost 44% of 15-micron pixels. - Isolated point defects is what kills large CCDs; how to isolate defects to yield more devices? - Less than 8-micron pixels not practical. - NEOs move more than one PSF in 20 seconds of time, so want that exposure time, so want 2 second readout time: 1 Gpxl/2sec at 1MHz/amp => 500 amplifiers. - One monolithic 4Kx4K OTCCD is 8x8 grid of OTCCD cells with 1-3 arc-minute view. - Kill dead cells with bad defects (so get holes in the focal plane, but that is OK). - Some cells will be guide-star cells. - 8 output channels per monolithic OTCCD. - Bloomed stars have blooming confined to that particular OTCCD cell. Bright stars become guide stars. - On-cell mux to select between active parallel clocks and static parallel clocks. - Current OTCCDs have 4-level poly overlaps; new design has 3 levels. - Q: What feature of OTCCDs allows decoupling bad pixels? A: Don't eliminate bad pixels, eliminate whole OTCCD cell. - Q: How to reduce power with huge focal plane? A: Ripple through readout sequence; way less capacitance per cell than 2Kx4K. Readout time is the worst time for power dissipation. ---------------------------------------------------------------------- Roundtable: Mosaics - Builders of mosaics on panel. - Issues: crosstalk within focal plane; flatness; testing; heat; gaps and alignment issues; device-to-device uniformity; dewars, filters, shutters, etc. - Power dissipation in CCDs: as manufacturers decrease bus resistance, power density in focal plane is not so much an issue. - True for IR ground-based imaging applications, but spectrographic or space-based applications have very important constraints which make packaging constraints. Q: For IR focal planes, how well is temperature controlled? For CCD focal planes, how isothermal must the plane be? A: Attach each IR element to Mo (moly) subplate; to thermal control in one part of the subplate, measure a little ways away. Can regulate to milli-K levels, stable over weeks. As long as plate held constant, don't need better control over individual detectors. A: Temp control requirements depends on detector. MIT/LL "brick-wall" requires tighter thermal control. A: Focal plane doesn't have to be isothermal, just have a stable distribution (gradient). A: IR need better temperature stability by several orders of mag compared to CCDs. Seems solvable. For CCDs, not necessary yet. Q: If we use IR reference pixels? A: Can back off several orders of magnitude of thermal regulation, so 0.1 C regulation. A. SAO Megacam isothermal to 5 C; local resistors provide tighter control. Goal is -120 C, stable to 0.1 C. Q: Are CCDs controlled to better than 1 C? A: Some want 0.1 C regulation. - Use Pt100 temp sensor. - Use thermox sensors. - In flight applications, how to get heat out the back end while maintaining alignment? Mistakes in arrays? - Big problem is not making mistake, but **fear** of making mistake (and not having any spares). - Or knowing the proper caffeine levels. Q: What are most comfortable gloves? A: Don't have to wear gloves on cold parts; cold fingerprints don't outgas. Wash hands well first. For warm parts, **do** wear gloves. Don't smoke cigars in the lab. A: Use handling tools on CCDs; then no gloves are needed, since the handling tool does not stay in the camera vacuum. - CCD with silicon over-hanging, devices have been broken over time. Q: Should silicon **not** overhang the package? Q: How much of an inter-CCD gap is tolerable? Q: What is the insertion versus alignment benefits of each piece of tooling? - Kudos for Marconi CCD packaging. Q: Do users of Marconi CCDs **use** the alignment pins? A: Yes. A: Have very tight tolerancing on the receivers for the alignment pins. - Many users do not need critical alignment, for stare mode. More critical for spectrograph or TDI applications. - Still have to do astrometric corrections due to optical distortions. - A tiled focal plane has been dropped. An impromptu comment, changing method, caused loss of concentration and the focal plane was dropped. Ears on focal plane substrate sit on 3 posts; focal plane rotated when detached, and one CCD was broken. - Replace attach bolts with studs or set screws, one at a time. Then lift up the focal plane. Connectors? - CTIO/NOAO eliminates some connectors with flex-PCB perforating vacuum wall. - This has been done with off-the-shelf flex circuits. These leak check down to the levels of hermetic connectors. - You can optimize your electrical length versus your thermal length. Assemble on bench, test, **then** put it in the dewar. - Only the NRE (non-recurring engineering) kills you, then it is inexpensive. - But these flex-PCBs are bad with shear forces. Q: Long-term reliability of flex-PCBs at 70 K temperatures? A: Some said problems with multi-layer flex (vias cracking), but single-layer flex OK. - On Nanonics connectors, you have to back-support connector so surface-mount pads don't pop off. - Never let a mechanical engineer design your CCD mount if **you** have to put it together. Other Issues: Q: Scattered light in focal planes? - Shiny stuff on the CCDs OK when no mosaicing, but kills when larger tiling. Q: Do people think about shiny stuff when they are doing a design? What about looking down the cracks between mosaiced CCDs? A: Some have strung black wire between gaps (both over and under the CCDs). - Others have raised above CCDs a mask of metal. - Done close at focal plane if not on focal plane. - Marconi CCDs can have a bridge to cover the shiny parts of quasi-4-side-buttable devices. Not standard yet. - Black-plastic mask? Might be radioactive. Q: Limits of focal-plane flatness that can be achieved? Over 55-cm diameter? - With beam 85%; few-hundred microns spacing on 3 sides, 4 mm on fourth side. Pixel size 9x27 microns. 30KHz parallel transfer; row readout time 300 microseconds; requires 16-bit dynamic range. - Q: Radiation damage causes CTE-related data shifts; how addressed? A: Identifying traps, build trap maps, rely on redundancy to retain sufficient centroiding accuracy. - Q: Accidental solar transit damage to CCDs? A: Don't know if damage would occur, hope that damage is not permanent and the array suffers only a thermal transient which would recover in time. - Q: Different clocking rates to take care of optical distortions? A: At moment, want optical design to minimize distortions so all CCDs can be clocked synchronously. 4. Augustyn Waczynski: HgCdTe detectors for WFC3 IR - IR channel: TEC cooled, special 6-stage, 150 K. Stage 3 also drives thermal shield. - On some IR devices, dark-current instability attributed to: varying red cut-off, or reset instability or reset transient. Right after reset, dark can be higher than by the end of readout. - (single-mode) Fiber-based single spot intra-pixel response testing; spot size 10 microns out of focusing optics at fiber end. - 9 e- rms read noise from reference pixels, but higher from imaging pixels. - Q: TEC specifications? A: 20mW heat load, 6 Watts total TEC power. - Comment: Gert Finger has found similar read-noise concern; attributes this to the short-wavelength cutoff of this material. - Q: When doing spot scan, what is ratio of peak-pixel intensity to total signal? A: From data, total peak is >99%; image wings have not been well studied, so whether signal out of peak pixel is from optical source or from pixel crosstalk has not yet been studied. - Q: Did Hawaii-1R detector solve persistence problems? A: There is some persistence observed in those detectors, with 1.7-micron material. 5. David Lumb: Optical photon counting STJ activities at ESA - STJs being tested with X-ray sources. - XEUS spacecraft is two separate pieces. - Small Josephson current across Al barrier, to be suppressed with a magnetic field. - 6x6 array has been fabricated (Tantalum); conventional array would require 36 individual charge-sensitive amplifiers: too complex, so developed row&column mux: readout energy resolution only slightly degraded. - Distributed readout within a pixel; make charge-division measurement via multiple (paired) output taps. Instead of 20x20-micron individual pixel, can use a 20x200-micron pixel with effectively 11 pixels spatial resolution with only slightly poorer energy resolution over the 20x20-micron single pixel. - Should be able to do direct QSO spectra to determine redshifts with STJs; good correlation with known QSO redshifts. - Q: Do you expect evolution of simplification and/or higher temperatures? A: Looking for lower-temperature operation for higher energy resolution, rather than higher-temperature operation. - Q: What is required to get so cold in a spacecraft? A: Joule-Thompson coolers followed by adiabatic chamber(?); cooler to reach 300 milliKelvins almost space qualified. 6. Lester Kozlowski: Control ASIC for astronomy detectors with ADCs and bias generators - Attempt to make detector (Hawaii-2RG) plus ASIC to be complete electronics system. Initial goal is flight systems. - Challenge: 16-bit ADC at 100KHz, 1 to 1.5 mW per channel, 8 mW total power. Both SAR and sigma-delta approaches being investigated. - Commercial ADCs often have digital error correction, this not being done to reduce power; expect industry trend away from error correction in a few years. - 16-bit ADC could be clocked faster, up to 1MHz conversion rate, if additional power dissipation can be tolerated. - Can be used for visible arrays also. - Q: Intended to be run at cryogenic temperatures? A: Can be run at room temperature; normally 180-120 K. Can be debugged at room temperature. - Q: Can you guarantee synchronous operation? A: It is designed for synchronous operations. - Q: What is the coldest that this ASIC will operate? A: As cold as 30 K (tested to 28 K). - Q: Single-point failures? A: Some amount of reconfiguration for redundancy. - Q: Have ADCs been characterized for typical characteristics? A: SAR not yet characterized; predicated upon an 18-bit-DAC approach. - Q: Have ADC test system? A: Yes. 7. Barry Starr: Monsoon - Want to bring everybody in to this effort. - This just offers pixels, does not propose to do processing/pipelining which are data-reduction dependent. Sky subtraction would be outside Monsoon. - Q: National or international availability? A: Can be international. - Q: Web site? A: Go to NOAO.edu, follow instrumentation links. - Q: Power requirements? A: Trying to push them down; want 5 Watts total as a limit. 8. Gert Finger: Readout Techniques for drift and low-frequency rejection in IR arrays - Make dark-pixel references by masking 1 column in the array. This includes this new dark reference in the (Hawaii-2) video chain. - At dark-current levels of 4 e-/pxl/1000seconds, must remove drift effects or they can dominate. - Reference output does **not** suppress low-freq pickup on Hawaii-2. - If reference output has drift, then subtracing it degrades the data. - Reference outputs do not have the same temperature coefficient(s) as the video outputs. - Suspect that noise coupling within the array is capacitive, and dependent on capacitance. Matching to same as pixel capacitance works best. - On Aladdin: reference output not offered on mux, but can be created; it works. Reference with column clamp circuit operation. - Q: 50Hz noise from detector or electronics? A: Mixture of both; have to redo the grounding and shielding. Multiple sampling can reduce this pickup. 9. Guy Woodhouse: ASICs for ultra-low-power CCD cameras - Waveform generator has fewer problems/overhead when doing loops. Error correction built in (protects against single-even upsets. - Custom ASIC with preamp, CDS, 16-bit ADC. Comparing with off-the-shelf solution. Not satisfied with readout noise of first unit. - DAC-produced biases; not used/tested yet. - Q: Thought about model for ADC/CDS with fast 14-bit ADC (space-qualified) with digital filtering, rather than CDS in analog with need for customizing filter? A: Don't know if this has been considered. What is important is to have all preamp, CDS and ADC in one package. - Comment: must turn up signal-chain gain so you are not limited by quantization noise. - Q: Relative cost of these ASICs? A: About 0.25M pounds. ---------------------------------------------------------------------- 2002-Jun-21 Afternoon Session: 1. Roger "Most Pathetic Paniolo" Smith: Dungeons and Dragons - "He know the job was dangerous when he took it..." - Keck IR MOS uses slit masks 30-inches diameter, with cold mask exchanger. ---------------------------------------------------------------------- Roundtable: Hardware (and Software) Q: ASIC versus standard versus hybrid-thin-film approaches: risks, performances? Q: Who is actively evaluating ASICs in the next year? Others than Rutherford Appleton or SNAP or Raytheon? A: Essentially those only. - SNAP has plans for ASIC, submitted this year. ASIC for CCDs. Q: Total cost of effort? A: Big. - Number of $2M USD for total effort, elsewhere. Q: Who do we talk to? Is Rockwell working on ASICs? A: Yes. - To ease the interface, do the ASIC in the same process as the mux. Time scale for development is 6 months to a year. Q: How reasonable is expectation that first run will yield? A: True for a sequencer; the CDS took a second run. Q: Also a sigma-delta ADC on the ASIC? A: Carrying a number of ADC architectures during the evaluation. - Even if we have an ASIC, we still have to address data flow, data pipelines. - When you look at these systems as software, much of the work is the same as before. We could standardize, share libraries of code. In context of Asteroid, can't get too hung up on platforms and "glue"; bulk of the work is in the algorithms. - Focus more on the algorithms. For one or two camera systems, less labor to make custom programming rather than the labor of developing "generic" software. - Sometimes, have simpler modules that are easier to load and unload. - Harder to quantify software as compared to the hardware side. Q: What are software metrics? A: Idea of passing pixels in parallel to systems; come with something "cheap" so the little guy can use. May be choices that are only point-to-point, may be less costly. - Want clearly defined software interfaces (also clearly defined hardware interfaces); hard part is to make it happen. - Issue: reliability across the data link. Usually high-speed fiber. Systems where fiber links fail or get flaky: need cyclic-redundancy checks or single-bit error recovery. - Fast fiber links strip out protections, to get the speed, so error correction and/or recovery doesn't get done. A higher-level protocol dropped on that data link might be useful. - A high-speed data link might not drop bits for weeks; might get out of sync with the far end. A higher-level protocol could help. - Desirable to have a way to check that the link is fine. Q: Do we care that these links have single-bit errors once a month? A: No!!! - You can test a link for reliability. If the failure causes the system to halt, this is a problem. You can't put a protocol on a link that makes the link asynchronous and effects the data. - The detector should be read out so that the rate of the readout is preprogrammed and stable, and the array is not waiting on the protocol. - Thought that these are solved problems, that we have not applied. - We are using fiber for electrical isolation. Q: Anyone using LVDS to camera head? A: Yes, in one case, with LVDS from camera head to Linux box, then Linux box to fiber. Q: With this LVDS link, is noise seen from the Linux box's switching power supply? A: Don't know yet. - Crosstalk: we want to hear about crosstalk. There can be subtle ways to miswire detector component implementations. Q: What have we done wrong to get crosstalk? Q: How do we do our engineering? Are existing design tools exploited? Or are just the years of experience exploited? A: The word "expert" is thrown around too much. Everything is a transmission line. Yes, simulations are done, but the simulation tools are not good enough, and modelings are insufficient (including for getting to 16-bit resolutions/accuracies). - PSPICE won't cut it at 16 bits. If PSPICE tells you it won't work, don't do it. But if PSPICE says it works, it might not work in reality. - When we start doing complex systems, simulations say they work, remember Jurassic Park... Q: Mechanical design **has** changed due to FEA. Are electronic designers exploiting these latest tools? A: We haven't yet reached the point where we have thought of all of the things that we need to model. Outcome of modeling is only as good as you can represent your system. Example: power supplies with filtering; worry about safety so put tranzorbs on output; they blow up after a while, so put on a fuse; fuse has a finite resistance; fuse produced a lot of crosstalk when supply drooped. You become an expert when you do something wrong, and then remember to **not** do it again. - At the detector level, we get a substandard data sheet. - It is getting better. - Damn hard to model a cryocooler. Electronics fights are often with the mechanical engineer. - Need the integrated systems concept. Q: How much electronics characterization is done of the acquisition system without the detector? A: Done in one case will all components other than the detector. For IR, warm muxes are beautiful test devices; good confidence check on the electronics. - Dynamic test and characterization at rate are harder than read noise. Verifying AC line noise (50/60Hz) takes work sometimes. - Hand-wired detector fanouts have been done well when done carefully, but takes much attention. - Also, may be ASICs that will run one or a few CCDs, but not many. For CMOS, more likely to have common ASICs since CMOS drive capacitances are so much lower. CCDs run voltages that are much much higher, whereas CMOS is going to lower voltages. If you want to wait for ASICs for CCDs, the rest of us will build cameras. - Rockwell is willing to invest in ASICs. The rest of us should think of thin-film where we buy the ICs (ADCs, etc.) in die form, and assemble them ourselves. Considerably less expensive when you don't have to pay the ASIC designer. The thin-film technique is just a very small PCB; not a lot different in size (? as compared to an ASIC or PCB ?). - Rockwell has tried this; at low temperatures, these die may not work as advertised. - Working on circuits in close proximity to the detector, using flex PCBs, and standard off-the-shelf components. Use PCAD or OrCAD and send the board out to standard board house. - Idealy, a couple-of-IC solution and we can use the core, then add custom circuits such as the CCD clock drivers. Advantages is we could all buy the common product. - A lot of these things are process dependent, - SNAP's concern is the manipulation of high voltages. - SNAP is using multiple ASICs, so clock-generation ASIC uses a high-voltage process not needed in the CDS. - CCDs don't really need the high volts that we are accustomed to using. 3V clock swings can work. We are able to make the CCD/CMOS process work with 3.3V rails with 100K e- full wells. 8-micron pixels, 3.3V swing, 4-phase; tried a variation with an Arsenic buried channel with 140K-150K e- full well. Also went with 8V on the (output) drain. Q: Can you invert the clocks to get rid of the remnant image? A: Not with standard voltage, but yes with the substrate pulsed to a positive voltage. Q: Will the on-chip amplifier give the same performance? A: Still needs testing. Q: Do you need the higher voltages to fully deplete? Depletion depths for red response much shallower with lower voltages? A: You can get the best of both worlds, fully-depleted CCD process. Q: Can others do this? A: Yes, has been tried. - Charge transfer can work, 2 volts swing was the lower limit for good CTE. Q: Why haven't we been doing this? A: Hasn't been a need to explore this domain. But one can do lower volts and lower power. - Some CCDs used for guider cameras have been run with 4-6 volt clock swings, but for science cameras the swing is raised since it is known that it works. - Lots of people using SDSU controllers. Liked that you can test the analog clocks through a connector on the front, in a simple manner. Also, having a buffer is good for checking what the CCD is putting out. Having an extender board is not good. There needs to be a cable between controller and detector that may be 1/2-meter long. Q: Shielded cables? A: In big mosaics, 1/3-1/2 meter cable in box, no shielding, no preamp inside the dewar, noise is 5 e- rms. - Having an unshielded cable inside a Faraday box helps; having 15-20 microvolts/e- helps a lot. - Case of crosstalk within a dual-channel SDSU analog board; had to go to two separate analog boards. Level was 5 DNs out of 16 bits. - SDSU users mailing list; Tim Abbott may be hosting it. Q: Any other controllers available to the community? A: MKIR has a commercial solution. Monsoon should be an available solution. - Others: SDSU is out there because it is an entry-level solution, so many out there. It gets you going. - Enough people participating in Monsteroid that production should be in a single location. Somebody will be found. No desire to restrict access; unsolved who will make them. - Intended that Monsoon front-end could be used without back end. Can be used for single detectors. But it doesn't exist yet. NOAO will continue to support SDSU. - Doing fine running Linux off a $2K PC for a SDSU box. ------------------------------------------------------------------------------- Attendance from 14 countries, 6 continents (And Al Fowler has been to Antarctica).