STScI Calibration Workshop 1
An Overview of Detectors(with a digression on reference pixels)
Bernard J. RauscherNASA Goddard Space Flight Center
22 July 2010
JWST NIRSpec HAWAII-2RG
STScI Calibration Workshop 2
Introduction
• After the diameter of the primary mirror, no component affects the performance of an observatory more than the detectors
• Detectors imprint a signature (e.g. dead pixels, hot pixels, QE variations etc.) onto the data
• Calibrating out this signature is critical to getting the most out of these observatories
• Need to understand how detectors function and why specific signatures occur
• In this talk, I present an introduction to detectors with an emphasis on JWST’s HAWAII-2RG (H2RG) sensor chip assemblies (SCA)
22 July 2010
STScI Calibration Workshop 3
Common detector types for the visible through mid-IR
CCD
•
• Intrinsic Si photoconcuctor
• Photons collected and charge read out in same piece of silicon
• During readout, charge physically moves from one pixel to the next
• Usual readout is correlated double sampling
• Because charge moves, on-chip binning is possible
Near-IR array
• Photon collection separated from readout– Optimized detector layer collects charge– Optimized readout integrated circuit senses charge in place (it does not
move like in a CCD)
• Multiple non-destructive reads typically used to beat down read noise and integrate through cosmic ray hits
Mid-IR arrayWFC3 CCDs
JWSTNIRSpec H2RG
JWST MIRI
Hybrid detector arrays
• • Intrinsic HgCdTe or InSb
photoconductor• NICMOS, IRAC, WFC3,
JWST NIR instruments
• • Extrinsic (intentionally doped)
Si:As photoconducor (other dopants are possible for longer wavelength response)
• IRAC and JWST/MIRI
22 July 2010
STScI Calibration Workshop 4
Photon detection in semi-conductors
• Photons are absorbed in the semi-conductor creating electron/hole pairs
• For photon energies less than the bandgap, the photo-conductor does not respond to light unless it has been doped (e.g. MIRI’s detectors)
– No calibration issues –light is just not detected
• For photon energies greater than about 1/3rd of the bandgap (very blue light), multiple carrier creation becomes increasingly likely
– Probable calibration issue for JWST. Both NIRSpec and FGS use 5 micron cutoff detectors at 600 nm
– RQE > DQE!• To see how MIRI’s Si:As detectors work,
compare the diagram of crystal structure (above) with the band gap diagrams (below). To free an electron in intrinsic material (1) requires a certain energy indicated by the band gap. It takes less energy to free charge carriers from impurities (2) and (3).
Rieke, G.H. 2010, Elixir School22 July 2010
“p-type” “n-type”
STScI Calibration Workshop 5
How a JWST near-IR array works
Readout Integrated Circuit
(ROIC)
Simplified structure of a hybrid IR array detector. In real arrays, there is often an epoxy backfill between the indium bumps.
22 July 2010
STScI Calibration Workshop 6
For WFC3 and JWST, the HgCdTe is graded to sweep charge (actually holes) to the depletion region
n-HgCdTe
p-HgCdTe(implant)
HgCdTe Buffer
Cap
Surface Passivation
CdZnTe substrate (removed)
In bump interconnect
H2RG ROIC
HAWAII-2RG pixel architecture. Photons enter from the bottom.
AR coating goes on at the dotted green line after the substrate has been removed
22 July 2010
n-HgCdTe
p-HgCdTe(implant)
HgCdTe Buffer
Cap
Surface Passivation
CdZnTe substrate (removed)
MBE Growth Direction
Detector Band Diagram
Valence bandholes
Conduction bandelectrons
STScI Calibration Workshop 7
What happens in a JWST NIR pixel?
22 July 2010
e-
photonHoles are collected in the depletion region where p-type HgCdTe meets n-type HgCdTe and electrical fields (arrows) are strong
e+
Outside the depletion region, E fields are weaker and charge can diffuse between pixels
QE depends on wavelength. Blue light is absorbed near the surface and red light is absorbed deep in the detector
If an anomaly is strongest in the blue, it might be a surface effect. If it is strongest in the red, it might affect deeper detector layers
8
JWST’s H2RGs are part of the Teledyne HxRG family
H: HAWAII: HgCdTe Astronomical Wide Area Infrared Imager x: Number of 1024 (or 1K) pixel blocks in x and y-dimensions
R: Reference pixels G: Guide window capability
Substrate-removed HgCdTe for simultaneous visible & infrared observation
Hybrid Visible Silicon Imager; Si-PIN (HyViSI)
NameFormat
(# of Pixel)Pixel Pitch
(mm)# of Outputs
H4RG-15 4096 × 4096 151, 4, 16, 32,
64
H4RG-10 4096 × 4096 101, 4, 16, 32,
64
H2RG 2048 × 2048 18 1, 4, 32
H1RG 1024 × 1024 18 1, 2, 16
Institutions, Observatories, and Programs Using HxRG Arrays
Wide-field Infrared Survey Explorer (WISE)
Orbiting Carbon Observatory (OCO)
Development Programs in Astronomy & Earth Science
James Webb Space Telescope (JWST) - NIRCam, NIRSpec, FGS
Joint Dark Energy Mission (JDEM)Astronomy institutions and observatories: Calar Alto, Caltech, CFHT, ESO, ESTEC, Gemini, GSFC, IRTF, ISRO, IUCAA, JHU-APL, Keck, LBNL, LMU, MIT, MPIA, MPS, OCIW, Penn State, RIT, SALT, SAO, Subaru, TATA, U. Arizona, UCLA, UC Berkeley, U. Hawaii, U. Rochester, U. Toronto, U. WisconsinSpace surveillance applications
Joint Milli-Arcsecond Pathfinder Survey (J-MAPS)
Development Programs in Astronomy
In Development, first on sky telescope test in 2011
• Source follower per detector (SFD) architecture is not unique to Teledyne. Raytheon has also used an SFD with their astronomical detector arrays
9
VresetReset
Voltage
Cell drainvoltage
DsubDetectorSubstrateVoltage
PhotovoltaicDetector
ReadSelect
ResetClock
IndiumBump
Source FollowerMOSFET gate
Photocharge integrates here
3-T ROIC Pixel CellColumn
BusDetector Pixel
ColumnSelect
BufferedOutput Disable
Output Bufferdrain voltage
Output
Horizontal Read Bus
Buffered OutputGreen: ClocksPurple: Bias voltages
HxRG Pixel in the ROIC
STScI Calibration Workshop 10
Some calibration “gotcha’s” and where they might originate in the sensor chip assemblies (SCA)
Open Pixels
Random Telegraph Noise (RTN)
Hot Pixels
Flux Dependent Linearity
Ghosts
Inter-Pixel Capacitance (IPC)
Electronic crosstalk
Fringing(only if substrate removal
was not complete)
Persistence and latent images
Inter-pixel sensitivity
variations (IPS)
Non-linear response(also electronics)
22 July 2010
Charge diffusion
Flatfield structure
STScI Calibration Workshop 11
An example of how understanding the device can aid understanding a calibration issue: reciprocity failure
• For NICMOS, strongest in the blue– Suggests surface trapping is important
• According to U. Michigan group, cooling helps– Suggests traps are shallow
22 July 2010
Courtesy Bob Hill
STScI Calibration Workshop 12
Some “gotcha’s” originate in the readout electronics
• 1/f noise (more on this later)– Particularly evident with SIDECAR ASIC in JWST ultra-
low-power & temperature operation– Also seen in ground based controllers
• Bars & bands– Happen when one part of the system pulls down the
biases for another• Tails
– Caused by settling time issues in the readout electronics and harnesses
• Pedestal drifts– Caused by unstable biases
22 July 2010
STScI Calibration Workshop 13
Schematic of a MIR IBC Detector
Rieke, G.H. 2010, Elixir School22 July 2010
STScI Calibration Workshop 14
Readout• For CCDs, charge is moved to the
output, sensed, and discarded– Nevertheless, noise performance
of CCDs is outstanding. JWST’s NIR and MIR arrays do not yet match them
• For NIR and MIR arrays, charge is sensed in place by the ROIC– Can use multiple non-destructive
reads to average down noise and integrate through cosmic ray hits!
– Achieving CCD like noise performance with JWST’s NIR arrays will require new readout approaches (yes, we are working on this!)
22 July 2010
STScI Calibration Workshop 15
How noise averages downwith multiple non-destructive reads
• Model does not include 1/f noise, will under estimate the noise of JWST’s SIDECAR based detector systems somewhat
• This differs slightly from what is shown in Rauscher et al., PASP, 119, 768 (due to a transcription error while finalizing the manuscript)
– Error caught by Massimo Robberto of STScI (Thanks!)– Massimo presents an independent derivation that expands this result somewhat in an internal
STScI memo (please speak to Massimo for details)22 July 2010
sread - Read noise per readn – Number of up-the-ramp groupsm – Number of frames per grouptf – Frame readout timetg – Group timef – Photonic current (includes dark current)
STScI Calibration Workshop 16
ADVANCED TOPIC: REFERENCE PIXELS
Richard G. Arendt1, Dale J. Fixsen1, Don Lindler1,Markus Loose2, Samuel H. Moseley1 & Bernard J. Rauscher1
1NASA Goddard Space Flight Center2Markury Scientific
22 July 2010
SPIE Telescopes & Instruments 17
Overview• Performance of 2kx2k Teledyne HAWAII-H2RG detectors and SIDECAR
ASICs is key to the success of the JWST mission• Broadband imaging is generally background limited. With QE ~ 80%,
only incremental improvement still possible• Spectroscopy & narrow band imaging are generally detector noise
limited –large improvements still possible even with NIRSpec’s 6 e- rms total noise requirement
• We have begun a program to analyze the noise characteristics of the NIRSpec detector subsystem, studying the correlations among the detector outputs and with the reference output, as well as the temporal correlations in a given detector section.
29 June 2010
• Using the measured characteristics of the noise correlations, we can determine the optimal coefficients for the removal of correlated noise as a function of frequency. By using all available reference sources, and by adding more frequent references, we can potentially reduce the noise by a factor of two
• We find that there is a frequency dependent gain and a frequency dependent correlation between the regular pixels and the reference pixels and the reference output
• In this talk we will1. present a demonstration of the analysis and mitigation techniques, and2. describe how to improve the next generation of detectors and readout electronics
JWST NIRSpec H2RG sensor chip assembly (SCA)
SPIE Telescopes & Instruments 18
Principal Components Analysis• Principal components analysis (PCA) puts noise studies on a firm quantitative
foundation• Computed the covariance matrix of vertical and horizontal cuts across the detector
array, as well as in ~ 64 x 64 pixel regions• Computed the eigensystem of the covariance matrix and sorted the eigenvectors by
descending eigenvalue• Major noise components of Flight NIRSpec detector subsystem are
1. 1/f noise2. Alternating column pattern noise
• These components are highly correlated with available references and can be removed using standard techniques
• Almost all of the correlation is temporal –there is little difference between pixels
29 June 2010
SPIE Telescopes & Instruments 19
Many references available for removing the extra noise
HAWAII-2RG Detector Array
4 rows of reference pixels along the “bottom” and “top” edges of each detector array
4 columns of reference pixels along the “left” and “right” edges of each detector array
A separate reference output that is always available for all pixels
Regular pixels (used as a reference) because they are vignetted and never see light
29 June 2010
SPIE Telescopes & Instruments 20
AN EXAMPLE OF USING MORE AND BETTER REFERENCES
29 June 2010
SPIE Telescopes & Instruments 21
Raw Test Data• Outputs 1-3 sample the
detector array, but single-ended (not differential which is the default)
• Output 4 samples the reference output
• For each frame, power spectra of the time-ordered data are calculated for each output. Results averaged over 88 frames of a single ramp.
29 June 2010
Op #1 Op #2 Op #3 Op #4(REFOUT)
Appearance of raw single-ended data. The horizontal banding indicates the presence of highly correlated 1/f noise
SPIE Telescopes & Instruments 22
Fourier analysis of the raw dataPower Cross Power
29 June 2010
Cross power is a measure of the power that is correlated between the two data sets (e.g. real output vs. reference output)
SPIE Telescopes & Instruments 23
Effect of different ways of using the references
• Traditional JWST differential feeds the H2RG’s reference output to the SIDECAR ASIC’s differential inputs
• Differential w/ Frequency dependent gain digitizes everything in single ended mode. A frequency dependent weighting is applied to the H2RG’s reference output before it is subtracted
– This weighting account for gain difference at low freuqncy– And lower degree of correlation at high frequency
• Interleaved references jump out to read 8 blanked off pixels every 128 pixels. Includes corrections for 1/f and alternating columns
29 June 2010
Traditional JWST Differential Differential w/ Freq. Dep. Gain Interleaved References
s = 13.5 e- s = 10.8 e- s = 9.6 e-
SPIE Telescopes & Instruments 24
Power at the Nyquist Frequency• Expanded view of power near and at the Nyquist frequency for one of the detector outputs• Symbols show results before and after optimal use of reference pixels and outputs
29 June 2010
SPIE Telescopes & Instruments 25
The Noise Floor: Traditional vs. Optimal
Traditional CDS Optimal CDS
29 June 2010
The input data in both cases are a set of one hundred 88 frame up-the-ramp sampled darks
Ignore the right hand output. It is looking at the reference output, not photo-sensitive pixels
σCDS ~ 13.5 e- rms σCDS ~ 9.6 e- rms
SPIE Telescopes & Instruments 26
Future prospects
29 June 2010
• This work highlights the importance of sampling low-noise references frequently and weighting the references by frequency
• In current generation H2RG detector arrays, reference pixels in rows and columns are: (1) too far away and (2) too noisy to suppress 1/f noise
• In current generation SIDECAR ASICs, there is no good way to weight the H2RG’s reference output by frequency in the differential input
• To be most effective1. References need to be sampled above the 1/f “knee” frequency2. References need to be significantly quieter than the regular pixels that
they are intended to correct3. Reference corrections need to take into account possible frequency
dependent weighting between the reference signal and the signal that is being corrected
• These goals can be met by many different ROIC and readout electronics designs
SPIE Telescopes & Instruments 27
To sum up: More & Better References
• Noise of JWST’s NIR detectors is much better than we thought!• SIDECARs are injecting correlated noise
– Can be removed by using more and better references– Almost all the correlation is temporal rather than spatial– Must work in Fourier domain; reference corrections must be
frequency weighted• Flight NIRSpec DS has total noise ~ 6 e- rms for 88 up-the-ramp
samples (EXPTIME ~ 900 s)• The techniques describe here should drop that to about 3.5 e-
rms without changing the hardware• Work is ongoing to demonstrate these improvements in
practice29 June 2010
STScI Calibration Workshop 28
Summary
• In this short talk, I’ve tried to present an overview of common astronomical detectors for the visible through mid-IR– Emphasis on JWST’s HAWAII-2RG near-IR arrays
• Briefly discussed some of the anomalies that are expected, and where they originate in hybrid near-IR arrays– Others will no doubt discuss many of these further at this
conference• Discussed how using more references more effectively can
significantly improve the performance of JWST’s detector limited instruments
22 July 2010