From fhh@nofs.navy.mil  Wed Oct 16 17:59:12 1996
Received: from sun.nofs.navy.mil (SUN-GW.NOFS.NAVY.MIL) by uwila.cfht.hawaii.edu with SMTP id AA12388
  (5.65c/IDA-1.4.4 for ); Wed, 16 Oct 1996 17:33:48 -1000
Received: from ccder.nofs.navy.mil ([192.68.148.181]) by sun.nofs.navy.mil (4.1/SMI-4.1)
	id AA21264; Wed, 16 Oct 96 14:28:43-070
Received: by ccder.nofs.navy.mil (5.65/DEC-Ultrix/4.3)
	id AA21793; Wed, 16 Oct 1996 14:39:50 -0700
Date: Wed, 16 Oct 1996 14:39:50 -0700
From: fhh@nofs.navy.mil
Message-Id: <9610162139.AA21793@ccder.nofs.navy.mil>
To: ccd-world@cfht.hawaii.edu
Subject: notes from ESO detector workshop 96
Status: RO

Disclaimer: many notes here will be repetitive with respect to the published
contents of the proceedings; all have been filtered by my personal view points.
Apologies for missed and/or misspelled names; comments and corrections to
Fred Harris at fhh@nofs.navy.mil

ESO DETECTOR CONFERENCE, FIRST DAY:

1. Guy Monet: Introduction to the conference

Opening Session:

2. Albert Theuwissen of Philips Imaging Technology:
- ref to Janesick's sandbox paper from SPIE
- not part of Philips Research, but shares same facilities at Eindhoven
- processing capability beyond 0.5um with 6-in wafers, going to 8-in in 1998
- most devices are 4-phase, with two poly-Si layers
- n-type substrates, one flat gate oxide
- gradients vs depth flatten out by 7um depth
- +10v gate potential; +2V barrier-gate potential will still collect light;
  at 0V gate potential then drains vertically to substrate => collect electrons
  only in first 2.5um depth, deeper drain to substrate; also implies an
  electronic-shutter capability
- unit fab block for image area is 1Kx1K; can repeat as needed to fab larger
  array sizes, up to 7K (H) x 9K (V) to fill 6-in wafer; block layout (modular)
- full frame or frame transfer
- single output up to quad output
- image array split vertically, serial regs split horiz
- triple poly (2 used in image area, 3 used in horiz reg) plus 1 metal
- amps are 3-stage, first 2 stages have on-chip loads, 3re stage load off chip
- >200K e- for 12um pixel
- <0.5 nA/cm^2 dark current at +60C, non-MPP
- QE of 13% at 420 nm; 27% at 530 nm, 7% at 800 nm (losses due to vertical
  antiblooming)
- 6 uV/e-, 30 e- rms at 5 MHz; 110MHz bandwidth; summing gate and floating
  diffusion capacities >600K e-; random pixel non-uniformity <1.0%; fixed
  pattern noise <50 e- (at 60C, 20ms integration time)
- Lesser's tests: cooled to -150C with no problems; >10 e- at 33Kpxl/sec
  (results on web page)
- typical anticipated wafer delivery time is 2 months
- would eliminate vertical antiblooming if thinned for backside operation
- noted by Mackay that amp very similar to Kodak amp
- estimated cost $150K (for lot run?)
- guess that dark current would increase x2-x3 in p-type substrate (IR
  sensitive without antiblooming)

3. Morley Blouke: SITe 3-side buttable arrays
- primarily speaking to 2Kx4K CCD (SI002A)
- 15um pixels, frame-transfer image area organization
- 2 amps
- 25 extended pixels
- no transfer gate
- serial pixel well x2 that of image area, summing well same as serial pixel
- three phase, triple poly (3 level) buried channel
- 63.43 mm x 31.464 mm
- 37.5 mm chip center-to-center butted spacing
- phase 3 is mpp gate in image area
- can do frame transfer in MPP only
- can do clocked antiblooming alternating phases 1 and 2
- 2 devices per 4-inch wafer
- non-, visible- and UV-AR coatings available; peak QE 60% for UV coating;
  85% peak for visible coating; QEs drop with increasing cold temperatures
- packaging effort to provide means for flattening the die
- 32 pads total for pinout including temperature sensor (AD590)
- ribbon cable to external connector
- edge of die recessed 25 um from edge of package for protection in H-dir
- chip carrier is Invar 36 with 4 blind holes on back for mounting
- CTE 0.999995 H and 0.999999 V; Fe55 testing done by Elliott at JPL
- SITe test system limited to 1Kx2K image storage for testing
- 2.9 e-noise at 50Kpxl/sec
- <30 pA/cm^2 dark current in MPP
- SI008A: 4K H x 2K V, buttable on one edge to make 4Kx4K arrays, for the HST
  ACS (Advanced Camera System); H reg longer for improved radiation tolerance
- Al gnd barrier around periphery and over amp to make amp light-tight
- Invar package is tied to package substrate (= GND)
- bow of parts has not been measured as a function of temperature
- Rich Reed has measured increased bows in 2Kx2K parts versus increasing cold
  temperature

4. Barry Burke: MIT-Lincoln Labs
- U of HI consortium: 2Kx4K buttable imager:
  - architecture similar to SITe 2Kx4K
  - output amp <2 e- rms for same-style amp as AXAF
  - back illuminated
  - flat to <15 um (+/- 5 um)
  - enhanced near-IR QE with 7000 ohm-cm bulk wafers
  - Aluminum Nitride package substrate of 3 layered pieces
  - 2 JFET source followers off-chip on-package to improve drive capability
  - full well >125K e- (not yet optimized)
  - 10-15 uV/e- output
  - packaging of devices to start Nov 96
- blooming control experimental chips:
  - 512x512 frame-transfer CCDs, 15 um pixels
  - blooming drains hidden in center of channel stops:
    - take away conventional LOCOS channel stops
    - place n+ blooming drain surrounded by p+ barrier underneath with n-
      blooming barrier on each side
  - blooming can be electrically turned off
  - optical overloads tested at x1000 and x1000000
  - QE tests still im progress
  - this structure is wider than standard channel stop, full well should be
    diminished
- Tonry's chips:
  - should be 4 poly levels; 1 is Al to allow no changes on rest of wafer
  - structure with 1 metal layer has excess of pockets
- 4-inch wafer capability to date
- 1960x2560 CCD frame-transfer CCD fills wafer:
  - package is Molybdenum with die attached; pc board attached to metal plate;
    bond wires direct from Si to pc board (10-layer)
  - 5 back-illuminated devices delivered to Air Force (GEODSS program)
  - no bright columns, just a few pockets
  - AR-coating is UV opaque
- going to 6-inch wafers, fab line conversion this month (Oct 96); driven by
  lack of good 4-inch wafers
- may not make chips commercially available; do take on research programs
  instead (no fixed-price contracts for deliverables, best-effort only):
  - possible to perform technology transfers to other manufacturers

5. Peter Pool of EEV:
- CCD42-80 is 13.5 um pixels 2Kx4K
  - 210 um butting loss on sides, 100 um on top
  - 3 devices max per wafer
  - could do 2 of 2Kx5K plus 1 of 2Kx7K on wafer max
- CCD44-82 is 15.0 um pixels 2Kx4K
  - 500 um butting loss on sides, 150 um on top
  - 2 devices max per wafer
- backsides are implanted, with laser anneals
- in-lab QEs can peak over 90%
- high-resistivity substrates show red enhanced QE, peak 88% at 800 nm at room
  temp
- output amps are two-stage, internal load on first stage, external load on
  second stage:
  - 400 ohms output impedance
  - electrically equiv to single stage
  - DC restoration available on-chip via extra drive-input connection
  - 3 e- at 100Kpsl/sec; 6 e- at 1Mpxl/sec
  - with 10pf load capacitance, 35ns settling time
  - should operate well at 1 Mpxl/sec
- two output gates:
  - stagger 1st to second output gate by 1 volt, good low-noise performance
  - bias second output gate to 10V, triples output node capacitance so larger
    dynamic range at output amp is realized
- CCD bonded directly to Invar; Invar bonded only near center to Al substrate
  to avoid thermal mismatch and maintain planarity over temperature
- CCD before packaging has 4 um peak-to-valley flatness; maintained after
  packaging at room temp
- at -100 C, actually gets flatter if starting shape is domed
- peak deflection at -75C where Si and Invar characteristics cross over
- picking the CCD doming vs Invar thickness can flatten combined system over
  wide temperature ranges
- future: package concept the pin grid UNDER part of Invar to minimize hit
  in image plane for 4-side butting; places for off-chip JFETs for buffering;
  either 13.5 um or 15 um CCDs will be so available
- 128x128 frame-store chip for AO with 1K frames/sec (for Beletic)
- 15 um 2Kx4K chips available end of 97

6. Dick Bredthauer of Lockheed Martin (/Loral/Ford):
- 5-in line now
- 6-inch line in Shanghai China to come on line next year
- 4Kx4K 15 um CCD for Dicomed / Hasselblad Camera ("BigShot")
  - almost 1K fabbed so far
  - Milpitas also fabs the whole back end to bolt directly to back of camera
  - outputs are NOT designed for low-noise applications
  - flat to about 25 um
  - "those who pay the most complain the least"...
  - Helin using on GEODSS scope at -5C with 10 e- noise @ 50Kpxl/sec
  - Table Mountain running colder
  - for astronomy, need to remove wire bonds and protective epoxy and rebond;
    this process not yet firmed
- 4Kx4K 19.5 um CCD on 5-inch wafer (79.9 mm x 79.9 mm)
- thousands of CCDs already fabbed for dental X-ray work
- China and India SPOT satellites use Loral CCDs
- Mars Global Surveyor mission uses backup CCDs from Mars Observer
- 1.5 um design rules, 3-phase 3-level poly process
- $2500 - $3500 for grade 1 2Kx2K CCDs
- 9216x9216 8.75 um CCD:
  - 50 Ke- charge capacity
  - 4 of 20 MHz outputs
  - 80.6x80.6 mm
  - Navy recon instrument (on F18?) (NRL program to demonstrate system)
  - 3 or 4 working arrays from R&D program; lot to be fabbed later this year
    for Navy program; first devices not cosmetically good enough for astronomy

7. Richard Stover of Lick:
- thick CCD back illuminated: Lawrence Berkeley Lab:
- high-resistivity Si
- 300 um thick, completely depleted Si wafer
- good QE response at 350 nm; about 85% QE out to the near-IR
- spreading of e- in substrate, must apply backside bias (best for 18-24 V)
  to minimize spreading
- very thin ITO on backside for bias application
- can't use in fast beam due to thickness of CCD; f10 is probably slow enough
- 200x200 pxl size for these first test devices

8. Christoph von Zanthier of MPI: Fully depleted CCDs for X-ray, UV and IR
detection:
- for XMM-Satellite mission
- if have pn junctions on both sides of the wafer, and apply voltages to gates
  on both sides to fully deplete the thickness of the wafer
- clock lines on the front side are pn junctions
- epi layer 40 ohm-cm, 12 um deep
- n Si 2.5K ohms-cm, 270 um thick
- 2x6 array of CCDs on a 4-inch wafer
- pixel size 150x150 um; image area size 10 mm x 30 mm => longest dimension is
  200 pixels
- QE better than 65% from 200 to 800 nm; wavelengths outside this range now
  under test
- each column has on-chip amplifying JFET (? 5 e- noise ?)
- present technology could go down to 50 um pixels
- full not measured because amplifier saturates first; >500K e- known, about
  1M e- estimated

9. Paddy Oates of RGO: EEV CCD42 test and evaluation
- camera rear defines CCD in dewar
- serial reg has U-turn at end to minimize butting-area losses
- expect <2 e- noise at 50KHz and 5 e- at 1MHz sampling, BUT noise spectrum is
  flatter than expected => lower noise at the higher frequencies (read rates);
  determined with photon transfer
- output gate 2 run either at +3V or +13V, producing about x3 gain change
- 128K e- < full well < 190K e-
- Vcte 0.999997; Hcte very good
- clocking Verts 35 usec/line without noticeable spurious charge
- 3000 sec exposure, cosmic rays very point-like
- RGO QE measures: in good agreement with EEV now; QE uniformity within 10%
  at 350 nm
- flatness: one corner is low, middle flat to <10 um, periphery <15 um;
  measured with technique similar to Hartmann testing
- Dump drain along serial reg: V shift each row, activate dump drain, read out
  serial reg twice => 0.4 sec to clear chip
- if V binned x50 and H reg read out plus read serial reg twice at end => 2.0
  sec and results comparable to using dump drain

10. Gerry Luppino's open session:
- Jim Beletic: wants high QE in UV
  - question for Morley Blouke: can SITe UV and visible QEs be both realized:
    answer is maybe?
- H2 poisoning of Lesser-thinned surface by ion pump dropped QE from 90+% to
  4% (Geary)
- note that Loral devices seem not to perform at better than 7 e- noise;
  basically confirmed by Bredthauer
- comments by users of Lesser CCDs: they can hold up in an observing
  environment, although their QEs change as a function of handling procedures
- comments on Loral CCDs: blurring of PSF: thickness of epi changed from 15 um
  to 20 um for Lesser's convenience, 20 um epi kept since and this increased
  thickness results in the PSF spreading; Rich Reed reports one 800x1200 at
  KPNO was overly thinned to maybe 13 um thickness, reduced PSF spreading from
  2.6 pixels to 2.1 pixels
- one Si supplier has a stiffened lattice because the molten Si is in a quartz
  beaker which dissolves some of the beaker, loading the Si with oxygen to
  concentrations of 10^17 per cubic cm; on this epi is grown
- comment that wafers thinner than 150 um won't withstand processing
- Loral once made p-channel CCDs for radiation hardness, but lost out on CTE
- fringing issue: for some spectrographs fringing is THE limiting problem; can
  this be addressed by the vendors?  Is the CCD a 10 micron-thick etalon
  with an index of 4 or 5 or 6?  Spectroscopists want it all (except perhaps
  readout speed) i.e. a very small part of the astronomy market which is
  already small compared to the (paying) commercial market...

11. Kirk Gilmore of Lick: Noise tests:
- Loral: about 10 nV/root Hz at 100 KHz
- Orbit: still in progress?
- Lincoln Labs: about 15 nV/root Hz at 100 KHz; knee at about 1MHz
- Lawrence Berkeley Labs: about 10 nV/root Hz at 100 KHz

12. Roger Smith of CTIO: How linear are CCDs?
- his DC-coupled preamp has clamp diodes which changed from forward to reversed
  bias when VOS offset was improperly subtracted; some turn-down of linearity
  is still found with the diodes removed
- Burr-Brown OPT211 integrated photodiode& as LED intensity reference to
  guarantee constant flux rather than sending through a constant current
- Mackay says using switching FETs work best because they have a lower
  impedance; two resistors are still needed to bias the FET?; only a single
  resistor from source to gnd with gate to gnd seems to be enough
- Chris Stubbs wrote up a paper while at LBL on making such active loads

13. Richard Stover: CCD flatness measuring system
- fixed laser
- video camera on pen-plotter x-y stage
- CCD on milling-machine x-y stage
- laser up to mirror, down to CCD, up to larger mirror, down to video camera
- changing tilt on CCD translates the laser spot on video camera

14. ?: 
- grounding SITe amp gnds at the chip gives lower noise than if that gnd is
  returned to the preamp

15: general discussion:
- more observers should have means (and intent) to monitor CCD/camera
  performance characteristics
- more instruments should have diagnostic stimuli built in
- cryogenic warmup alarms to prevent the warmup and outgassing of the getter
  followed by the detector cryopumping the outgassed species (i.e., operator
  error in cold vacuum maintenance)
- urge for trend of adding telemetry of in-camera CCD excitations to new
  camera designs
- detector systems with critical alignments cannot be easily removed from the
  telescope environment, so desire testing capabilities internal to the
  instrument

ESO DETECTOR CONFERENCE, SECOND DAY:

1. Satoshi Miyazaki of National Obs of Japan:
- Large format and high QW CCDs for Subaru Telescope
- need 14 of 2Kx4K devices for instruments
- prime focus camera expects f1.9 beam
  - needs less than +/-30 um flatness for 15 um pixels
  - EEV 2Kx4K anticipated
    - measured EEV flatness +/-2 um
  - readout 10 usec/pxl, really need 1 Mpxl/sec rate
- high resolution spectrograph has f3.5 camera
- in-vacuum clock driver and preamp board per CCD
  - single board from CCD via FPC, and FPC out to 41-pin hermetic conn in dewar
    wall
  - board folds in half to make preamp compact with clock drivers
  - preamp heat removed via copper gnd plane
- bias generators and signal-chain+ADCs outside of vacuum
- tests of Hamamatsu CCDs, backside
  - 2Kx4K 15 um pxl intended, frontside
  - peak QEs near 90% measured

2. Johannes Andersen of U of Copenhagen:
- CCD detector program for NOT and 1.5-m Danish on La Silla
- circa 1991: Loral wafer run, thinning by Lesser; camera to be built in-house
  - 7 e- read noise for Loral, 4 e- for SITe
  - runs on PC with Linux
  - fiber optic connection between instrument and data system
- coming soon: 2Kx2K CCDs, single chips for most instruments, 2x2 mosaic at
  cass
  - Preben Noerregaard is contact for controller hardware development
- circa 1994: Loral run with 88 chips, 66 alive; 2nd run had >40 alive
- SITe chip is still working great

3. Jim Beletic of ESO: ESO's plans for optical detectors
- VLT and support for telescopes on La Silla
- optical imagers, and wavefront sensors for AO
- one Lesser CCD operating for 1.5 years with no problems
- 3.6 e- noise at 1 Mpxl/sec from Lincoln Labs CCD; this speed needed for
  direct cameras
- faint spectrographs can run at 100Kpxl/sec
- Toshiba paper describes <1 e- noise at 1 Mpxl/sec, hopes this can be realized
- FIERA CCD controller: in development/test
- will get 2 of 7Kx9K Philips CCDs; 2Kx2K 24 um with 4 tracker CCDs on sides;
  UH/MITLL chips, want 3; EEV 2Kx4K 15 um want 10, also EEV wavefront CCD
  128x128 24-um; need CCDs for 8Kx8K mosaic
- note that no mfr can test CCDs at high speed, will loan FIERO controller for
  speed tests
- ACE controller going to NTT next month

4. Russel Cannon of AAO: AAO detectors, present and future
- Tek 1Kx1K used almost exclusively
- f3.3 prime focus needs more area coverage
- 90% of scope time used for spectroscopy
- Schmidt FLAIR system is fiber coupled
- want to do:
  - 8Kx8K for prime, 2Kx4K elsewhere; UH/MITLL and EEV usage expected
- circa 1983 controller: 2.3 e- noise at scope, 1.7 e- in lab; does
  charge-shuffle readout (CSR):
  - CSR: image in middle 1/3 of CCD, mask each end for storage areas
- Leach controllers as interim solution, in-house next generation continues
  - first prototype on new sys 1 yr, in use 2 years

5. Gerry Luppino of IfA: U of HI plans for CCDs
- existing 2Kx2K TEK camera, Loral/Lesser 2Kx1K camera, mosaic 8Kx8K  of Loral
  thick CCDs, travels around
  - UH/MITLL thinned to replace thick CCDs in 8Kx8K camera
- will build 8Kx12K camera for CFHT (2x6 array)
- 3.6-m AF scope on Haleakala to get high-res spectrograph needing 8Kx8K cam

6. Rick Murowinski of DAO (for Todd Boroson): Gemini CCDs
- wants high-speed mode and slow-speed mode, hopes calib shots in high speed
  can be matched to data taken in low speed
- GMOS wants: <3 e-/hr dark current, flat <25 um over WHOLE focal plane,
  temperature stability <0.18C/hr to maintain velocity resolution
- vendors for 2Kx4K buttable parts to be contacted
- controllers: Int'l bid for integrated controllers
- Shack-Hartmann sensors are EEV39, 3-5 e- noise, 200 frames/sec, CCD
  integrated with CCD package

7. Richard Stover of Lick Obs: development of CCDs for Keck
- DEIMOS will use 8Kx8K of 2Kx4K 15 um CCDs for Keck II
- ESI spectrograph with one of 2Kx4K um CCD (echelle)
- 4Kx4K upgrade of HIRES via 2 of 2Kx4K 15 um
- will use SDSU-II controller
- for Lick, need 2Kx2K CCDs (1 of SITe and rest of 15 um pxls)
- hoping for SITe CCDs flatter; don't think EEV will be ready for Keck timeline

8. Paul Jorden of RGO: RGO and INT CCD developments
- at present mostly using Tek 1K and 2K CCDs
- working on Loral/Lesser 2Kx2K 3-side buttable, EEV 2Kx4K CCD42
- INT to get 4Kx4K camera:
  - Loral non-buttable CCDs as close as possible plus autoguider CCD
  - no circuitry in vacuum
  - one controller for science CCDs, separate controller for autoguider
  - rotating sector shutter, 6-position filter wheel coaxial

9. Rich Reed of NOAO: CCD plans
- 20 of 2Kx4K CCDs ordered from SITe (? thinned grade 2 ?)

10. Craig Mackay of U of Cambridge: High speed wide dynamic range controllers
- no trimpots on boards, DACs instead
- for interferometer, 10Hz response in telescope autoguiding CCD controller;
  readout rate up to 150 frames/sec of guiding subarray (interferometer 
  detectors are APDs)
- 64x64 15 um pixel frame-transfer CCD by Loral, 8M pxl/sec (12-bit) (100 e-),
  5M pxl/sec (14-bit), up to 865 frames/sec
- using this controller for 2x2 mosaic of HgCdTe 1Kx1K IR arrays

11. Claudio Cumani of ESO: ACE and FIERO
- ACE is transputer based; libraries used for hardware dependent features,
  rest of code (80%) in common
  - 100K pxl/sec per port
- FIERA:
  - analog bias board, video board, clock driver board, common board per
    CCD detector head
  - 880 Kpxl/sec/port, 1.3 e- noise at 1 Mpxl/sec, "negligible" crosstalk
- circa 1997: 2 Mpxl/sec/port limited by 16-bit ADC (could change to 10 Mpxl/sec
  with 12-bit ADC board), 16 Mpxl/sec total

12. John Geary of SAO: preamp designs
- likes to put preamp in vacuum near CCD => should be small and clean
- JPL/Jamesick preamp deemed complex, V noise less than AD745
- AD745: low V noise, low I noise (less important), slew rate & settling time
  modest; 2 in parallel slightly lower V noise than JPL
- OPA627 (or OPA637): V noise x2 of AD745, faster slew rate
- Fred Harris' tests: AD745 too slow at 100 Kpxl/sec; OPA637 slews fast enough
  to pass too much feedthru
- Philip MacQueen's use of OPA637: works well with gain foldback to knock down
  reset feedthru, must heatsink parts
- 5-volt-rail opamps will probably have dynamic range problems

13. Philip MacQueen of McDonald Obs: 18-bit analog signal processing
- full well < full ADC scale
- read noise > 3 ADUs
- analog signal processor linearity >> CCD linearity
- Analogic 18-bit ADC
- integrator is major weakness in system:
  - multiple integrator reset pulses for best linearity (even if total reset
    on-time is comparable?)
  - ? charge stored in reset switch, leaking back into integrator
    capacitor ?
  - jitter noise in integrator switch windows major noise contributor
  - make zero signal level to be 0 volts, then gain changes from jitter have
    minimal effect on small-level signals
- speed changes via multiple selectable integrators

14. Roger Smith of CTIO: readout speed optimization
- doesn't yet run I+ directly after I-; waits for video to settle between
  I- time and I+ time
- OPA627 saturation recovery takes 10-15 usec
- analog switch to switch preamp gain to unity seems not to add noise

15. Maki Sekiguchi of NAOJ: Messia CCD controller
- no CPUs => no code maintenance; waveforms stored in SRAM, loaded with DSP
  (DSP is Motorola 16-bit fixed-point)
- fast data transfer to host workstation: fiber-channel hardware between front
  end and workstation

16. CCD controller round table:
- Rolf Gerdes: FIERA has 2 of 320C40 DSPs for camera head control;
  no software in the camera head, just in VME crate, to insure synchronization
  in multiple detector systems; in detector head small PLDs provide local
  sequencing (Lattice 1040C-70 ?)
- Gerry Luppino representing Bob Leach from SDSU:
  - timing board is 56002
  - clock driver board has 24 clock output via analog switches with DAC-set
    rails; no slope/slewrate control
  - Dual ADC board: 500 Kpxls/sec at 16-bits; dc-coupled preamp; 12-bit bias
    DACs
  - uses separate timing generation board
  - uses utility board
- ESO is using DAC per clock-driver output with ability to write to multiple
  DACs simultaneously so waveform synthesis is possible, viewed as desirable
  for edge control for tri-level clocks for clocked antiblooming; 80MHz DACs
  from Analog Devices
- Craig Mackay's clocks: 16 MHz in serials, perhaps a Teledyne part?
- Chris Stubbs: U of Washington CCD MACHO controller:
  - OPA627 with LM6321 in feedback loop for the preamp output to drive cables
    some distance from the dewar heads
  - read out everything in lock-step
  - opto-isolate everything
  - only use single amp per CCD
  - NOT one-size-fits-all controller
  - FORTH-interpreting micro as front end to DSP which generates timing
  - trimpots generate voltage levels
  - transversal filter DCS at 250 Kpxl/sec
  - the 2 output drains tied together with single-wire in => big mistake with
    the Loral 2-side-buttable 2Kx2K thick CCDs
- Craig Mackay: SITe devices with 4 outputs being used: drive the reset drains
  independently; output drains and reset gate drives entirely separate
- Richard Stover: reset connections seem to matter most in inter-amp crosstalk
- Mingzhi Wei: using an optical pattern to stimulate a multi-amp CCD makes it
  easier to look for crosstalk problems
- Jim Beletic asks: can the community build electronics lacking crosstalk?
  - Gerry Luppino says yes, already done in 8Kx8K camera; controllers must be
    entirely synchronous
  - Craig Mackay: must meet EM compatibility reqs as a mfr; asynchronous
    controllers are deadly
- Luppino asks: why don't people use transversal-filter signal chain?  Maybe
  because the totally-reflective implementation has not been published?
- Beletic asks: why not take multiple fast ADC conversions to digitally 
  synthesize the dual-slope integrator digitally?
  - noise goes down at root-N conversions, but:
  - do you lose that by the greater noise bandwidth needed by the faster
    electronics to run the multiple conversions?
  - advantage might be increasing the summed dynamic range beyond that of the
    single ADC conversion (i.e., slowing down a dual-slope integrator increases
    the gain which reduces dynamic range; summing multiple ADC conversions
    can increase the resultant dynamic range beyond that of the ADC
    digitization limit)
  - could use the multiple samples to calculate the asymptotic result of where
    the system SHOULD reach given enough time
- Craig Mackay uses dual-slope integration at 5.5 MHz: use RF components and
  RF circuit techniques; keep amplifier gains to x2-x3 only; use discrete FETs
  meant to be switches instead of integrated analog switches (? ref to SD5000
  by Geary ?); separate +1 and -1 feeder amps, use the FETs to GND the output
  of the path not being used => organize the circuit topology so that all
  switches tie a node to GND so that the drive to the switch is TTL-like;
  may want to have a switch from integrator input node directly to GND;
- Philip says: SD214DE FET switch for integrator input, but HI-201HS for
  conventional integrator reset; a single switch to the integrator input node,
  and a second switch on the input side of the integrator input switch which
  clamps the input source signal to GND
- Roger Smith uses a pulsed summing well drive, 2 usec long at present, to
  minimize the offset from a single summing well transition into the integrator
  path

17. Claude Trottier of EG&G Canada Ltd: Silicon Avalanche Photodiodes
- active quench circuit, separate from reset circuit
- quench ckt drops the voltage on the APD below the breakdown threshold when
  an output pulse is detected, waits briefly, then re-establishes the normal
  breakdown voltage with a rising exponential waveshape
- reset is when you recharge the APD
- correction factor applied to transfer curve (of input rate versus output
  when output turns over because input comes faster than APD can recharge to
  breakdown potential) which is a rising curve, curve rises slower for smaller
  area devices
- electrons get trapped in the avalanche process, must be allowed to relax
  before recharging the APD or would cause a prompt (next) avalanche
- when an APD fires it emits light, must be properly shielded in 4-quadrant
  APDs; another solution is when one APD quadrant fires, reset all quadrants
- quadrant module not convenient for assembly for mass production; crosstalk
  between cells still needs to be studied
- threshold, voltages, cooler, etc all preset for the user

18. Anthony Peacock of ESTED: Superconductivity Tunnel Junction (STJ) detectors
- program started in 1986 to evaluate these for X-ray spectrometry
- temp below critical temperature (operating temp 1.2 K)
- electrons in Cooper pairs; Cooper pair is broken by a phonon from photon
- when a Cooper pair is broken, generate 2 charge carriers (quasiparticles)
- two thin films 100 nm thick, separated by an insulator 1-2 um thick
- time from photon absorption to carrier production is a few nsec
- critical temp is 9.25 K for a Niobium device
- mechanical coolers operating at 1.4-1.5 K are available
- charge carriers are blocked from entering the leads by having niobium nitride
  at the leads
- apply a magnetic field parallel to the barrier of the device
- started working with niobium due to higher critical temp and mature
  technology; evolving to Hafnium via Tantalum
- by choice of materials can effect the temporal characteristics of the device
- each device has to have its own preamp, etc.
- in addition to Fano noise, have noise factor due to internal tunneling
- for niobium at 500 nm, resolve 14 nm spectrally
- devices 20 um on a side are realized now
- photo-absorption in the poly-crystalline niobium layer
- electronics are room-temp charge amplifier + shaping stage
- charge output per photon: = ? 1.74 x 10^4 e- per blue photon ?
- output linear with photon energy to better than a percent
- at 200 nm incident wavelength, see 60 nm spectral resolution; resolution
  drops with increasing wavelength; with backside illumination, get 15 nm
  resolution at 200 nm
- newer device has an epitaxial tantalum structure, but leads are still
  niobium, top is poly-aluminum followed by poly-tantalum; run at T = 0.3 K;
  at 200 nm, resolution 9 nm direct, 8 nm with noise correction; small
  nonlinearity better than 0.6%; reflectivity around 20% at 500 nm, going up
  to 80% at 1000 nm; IR response: ?
- to go into UV, change substrate from sapphire to magnesium fluoride
- 3x3 array of 25 um devices have been fabbed and tested, with 1.5 um gaps;
  test array of 4x4 coming out of fab; masks made up for 6x6 array;
- first large format device will be 18x50 of 25 um elements
- 3x3 devices have not experienced crosstalk problems
- count rates of 10KHz limit, may be limited by electronics
- going to the William Herschel telescope with 6x6 array at end of 1997

ESO DETECTOR CONFERENCE, THIRD DAY:

1. Tim Hardy of DAO: Charge transfer efficiency in proton damaged CCDs
- radiation damage via displacement: collisions with Si atoms in lattice
- affects bulk Si and buried channel
- most important is phosphorus-vacancy complex at 0.4 eV
- damage increases dark current, decreases CTE
- equations of model tested with TK512 damaged by 3 MeV protons, pure beam
  claimed, fit to equations good
- traps at 0.42 eV (phosphorus complex) and 0.223 eV (oxygen complex)

2. Alain Maury of OCA: converting a Schmidt scope to CCD operation
- more than 1/3 of discovered asteroids have been discovered with Spacewatch
- some films can be 3-4% QE, up to 8%; no commercial market for these products
- camera should be no larger than prior plate holder
- readout time SHOULD be short, even if noise is increased
- photon noise limited, so lowest readout noise not critical
- cosmetic quality not as important as some other astronomical applications
- current CCD controller at 500 Kpxl/sec, 15-bit data, PC-based acquisition,
  data reduction on Unix machine; current CCD is Loral 2Kx2K 15 um pxl thick;
  TEC cooled with glycol loop to remove excess heat
- limiting mag of 20th in 2-minute exposure
- next generation is Loral 4K CCD, 2.5 Mpxl/sec, 12-bits, controller from DLR

3. John Tonry of IfA: Orthogonal transfer CCD
- a number of science-grade devices already made
- pocket pumping does happen as charge is shifted around, may be a fundamental
  concern
- 512x512 orthogonal field next to standard 512x512 3-phase field next to amp
- pumped pockets often dig down right to the bias level
- pocket pumping in lab shows whole spectrum
- pockets are very sensitive to clock voltages
- kT about 50 e-; pockets shallower empty fast, deeper they do not drain
- 3-phase region shows fewer deep pockets, may be function of Al (fourth) gate
- most of power spectrum in atmosphere found to be at 1-sec timescale rather
  than 10-sec or 0.1-sec timescale; slower than Kolmogorov
- incredibly low degradation of PSF far from the guide star, suspects that a
  10 arc-min could be corrected (iso-kinetic patch)
- good cheap way to implement tip-tilt, with NO vibrations
- for (KPNO) 2.4-m aperture, see 0.2 arc-sec improvement (0.7 -> 0.5); pixel
  there is 1/6th arc-sec

4. Walter Kosonocky of New Jersey Inst of Tech: very high frame rate burst-
imager sensor
- captures frames at rates up to 10^6/sec by storing up to 30 images in frame
  store section
- 2x2 cm chip, 4x180x180 pixels, 50 um pixels, fill factor 13.5%, 30 memory
  stages per pixel, 3-phase BCCD, 4-level poly, 2-level Al
- each pixel contains detector, serial reg, parallel reg; single readout/quad
- 1st level poly forms channel stop for serial regs; 3 levels for 3-phase CCD
- blooming barrier gate per detector area into dump/overflow drain
- detector is pinned buried photodiode; N regions are graded (stepped implant
  potentials) to move charge toward readout structure
- each pixel has at least 10 contacts to aluminum busses
- for readout, pixel memory regs are configured into a large (standard)
  parallel-serial CCD
- image-area serial clocks separate from readout reg serial clocks
- TEC cooled to -30 C, 8 seconds to readout full image
- saturation signal 11 Ke-/pxl, readout noise 9 e- rms (dual slope CDS at
  120 Kpxl/sec), dark current 100-200 e-/pxl for 8 seconds of readout time;
  2.3 uV/e-
- used for studies of cavitation of propellers (such as for ships and subs)

5. Gerry Luppino of IfA: science with large CCD mosaics
- present: 8Kx8K, cost of $300K for initial effort (1 lot plus 1 backup lot),
  thick Loral 2Kx4K 15 um CCDs
- future 8Kx12K (2x6 array), thinned CCDs
- pincushion distortion at CFHT prime focus => no scan-mode operation
- 3 saddlebag system; 2 boxes of Leach I, each box read sequentially (don't
  know how to synchronize more than one such box); readout time 7 minutes =>
  twilight flats impractical
- flat fielding via superflats, precision 0.1%; minimum 60 exposures to
  generate the flats and biases; everyone's frames go into creating the
  superflats
- running at -70C to avoid serial CTE problems at colder temperatures, so
  dark current is non-negligible

6. Simon Tulloch of RGO: technique for determining coplanarity in mosaic CCDs
- going into f3.3 beam
- CCDs scanned cold within cryostat
- commercial scanners confused by reflections from surfaces of cryostat window
- Hartmann-like test with beam angles of 28.5 degrees
- centroids measured to about 1/50th of a pixel
- the CCD being scanned is the detector for the test; requires cosmetically
  good CCDs
- taking two exposures, one with one aperture and the other with the second
  (opposing) aperture, play Hartmann connect-the-dots and simple geometric
  optics to determine the depth (flatness) dimension; these are 1-dimensional
  scans
- flatness of individual CCDs and flatness inter-CCD in mosaics determined
- green LED for illumination

7. Florian Bauer of CEA Saclay: the two EROS 4Kx8K CCD mosaic cameras
- detection of massive compact objects, 10^-4 to a few solar masses
- 1 m RC scope; 450-650 nm blue passband, 650 nm onwards red
- Loral 2Kx2K 15 um 3-side-buttable CCDs
- coplanarity 10 um, alignment to 2 pixels
- variable thermal resistance between cold mass and detector head, varied with
  stepper motor, 900 copper contact points; one day to cool (15 Kg) thermal
  mass, 1 hour to cool CCDs
- CCDs have individual driver boards
- 12 usec/pxl (ADC limited), 50 secs to read both mosaics, one amp/CCD, 6-7 e-
  readout noise
- typical exposure 3-4 min (galactic bulge), will go to 80 fields/night

8. Rich Reed of NOAO: 8Kx8K mosaic at NOAO
- small gaps <1 mm between CCDs acceptable
- yield of 9 CCDs out 80 from wafer fab
- dewar base has dry N2 gas to keep dewar window dry
- radiative coupling of dewar cold source to CCD Invar mounting plate via nested
  concentric rings
- bimetallic strips (12 around periphery) connect dewar to CCD mounting plate,
  when the strips go cold enough the strips deflect and disconnect
- pneumatic-driven shutter (air cylinders) adds no heat load, 2 blades with
  bidirectional cylinder on each blade, can get photometric operation with
  either direction of shutter-blade throw
- filter motion via air cylinder "one-step stepper motor", one stroke advances
  filters by one position, bidirectional

9. Roger Smith of CTIO: experiences with Arcon
- multiple CCD controllers rather than a single large controller
- 3 transputers per controller
- the 4 controllers can be individually or individually addressed
- each controller can process 4 analog channels
- common microprocessor clock is distributed
- CCDs float with respect to their dewars and are referenced to their
  particular Arcon; grounds return to each Arcon; Arcons each have a star
  ground point on the backplane, this is connected to the shield; Arcon
  shields are connected by the dewar;
- clock distributed by coax between Arcons; 40 nsec change in delay between
  one Arcon to the next; changes in delay may produce small changes in overscan
  level
- test alignment of CCD pixel descrambling done with on-window copied
  transparency with LED illumination 1 meter away
- data on disk goes on as an IRAF file per amplifier
- only one amp per CCD used; electronics can support 2 amps/CCD

10. John Geary of SAO: MMT MEGACAM project
- 16Kx16K pixels
- 32 of 2Kx4K 15 um or 13.5 um CCDs; half are butted on 4 sides; purchase of
  CCDs not yet made; controllers not made yet; timetable of 3 years
- optical distortions will not allow TDI over the whole array
- Spectrograph: 300 fibers positioned via two identical robots, 150 seconds to
  completely reposition (hoped); CCDs at focus of Schmidt camera with cooling
  brought in from exterior tot he beam

11. Maki Sekiguchi of NAOJ: Survey camera for Sloan Digital Sky Survey
- 54 CCDs in focal plane; 30 of SITe 2Kx2K, 24 of SITe 2Kx400
- TDI mode, all CCDs have image areas clocked in unison
- all CCDs require individual rotational and tilt alignment
- preamp ICs are OPA627, not AD745 as stated in talk

12. Olivier Boulade of CEA Saclay: MEGACAM project
- want readout time <20 seconds for entire array, read noise < 5 e-
- want spectral response down to 350 nm
- driver board to drive 8 CCDs; readout board to read 8 CCDs
- budget for CCDs $2M for grade 1 devices (don't need grade 0); 40 of 2Kx4K
  CCDs to be purchased
- total budget $3.5M
- first light in 2000-2001?

13. Discussion session:
- organization of multi-chip data is not handled via a single organizational
  approach across the different conference participants
- appears that there is no standard for a compressed FITS format; proposed
  format from 2 years ago perhaps but not confirmed
- MACHO project stores each CCD image as a separate FITS file; will have 500GB
  of disk in the dome
- U of HI 8Kx8K, never look at full image at the telescope; suggest that you
  have enough disk space for any one night of observing
- experience to date that CCDs should be removable from a focal plane
- expensive accurate alignment of CCDs within a mosaic required ONLY for
  cameras running TDI modes
- discussion about fringing:
  - broad band (>100 nm bandpass): not too bad
  - narrow band: not uniform across field, gets you on faint extensions to
    galaxies, planetary nebulas, HII regions; hard on spectra
  - to fix, go to flat fielding: again, direct imaging not too bad; for
    spectroscopy, okay if fixed instrument format (i.e. no grating-angle
    changes)
  - question: do we ALWAYS want coated and/or thinned CCDs?  (some groups are
    gong back to frontside devices?)
  - note that no fringing was found in the LBL thick back-illuminated 
    high-resistivity CCD; but Fe55 testing difficult
- can manufacturers and users of CCDs agree on a simple set of tests for
  characterizing devices?
- DACs: bias voltages and clock voltages usually generated separated in
  different DACs; DACs can show 1/f noise so expect to heavily filter the DAC
  outputs; restrict rails, particularly for amplifier biases; provide reverse-
  protection diodes for amplifier volts with breakdown lower than the CCD;
  possibly gate the DAC outputs with analog switch per output and have the
  software verify legal voltages before connecting the DACs to the CCD
- setting of parallel clocks and full-well questions: some SITe CCDs seem to
  have a different full well in one direction than the other, is this effect
  sensitive/insensitive to clock overlap times?