Transcript OTW: Detectors

Detectors for Optical and
Infrared Astronomy
Kyler Kuehn
Instrument Scientist, AAO
Observational Techniques Workshop
01 April 2014
The Dark Energy Camera
62 2k x 4k full-frame, fully-depleted, backside-illuminated, 4-sidebuttable, dual-readout, p-channel LBNL CCDs with 250 micron x 15
micron x 15 micron pixels, possessing 130ke- fullwell, 2.5 ADU/e- gain,
“five nine” CTE, totaling 520 Megapixels with >50% QE from 375 nm
to 1000 nm, operating at 173K, with an overall FOV of 3 sq deg.
Motivation
• For some of you, astronomical images and spectra are
things that from time to time show up on your screen
and/or in your hard drive as FITS files.
• If the materials scientists and optical engineers (and
software engineers, and…) have done their job
properly, you generally won’t have to worry about
most of the details in this presentation.
• But it can be useful to know what’s “under the hood”,
so you can (at least) understand the motivations
behind each data analysis step, and (possibly) identify
and contribute to the resolution of issues arising from
your own observations.
Detectors: a History (in one slide)
Unaided human eye:
Area ~40 mm2, “Integration time” ~0.04 s
cf. Rudolphine Tables (data acquired by Tycho
Brahe up to 1601, published by Kepler in 1627)
First telescopes (circa 1608): <700mm2
(though with significant aberration)
Galileo’s drawing of Jupiter’s “Medicean Stars”:
Cosimo, Francesco, Carlo and Lorenzo
Sources: Brahe’s Mechanica, Galileo’s telescope from Florence’s Museo Galilei, Image from Sidereus Nuncius
(OK, make that two slides)
Photographic Plates:
• Gelatin film (usually) containing
light-sensitive silver salts, activated
by incident light then chemically
“fixed” to prevent further change
• Glass plates do not deform like
photographic film does
• Area, resolution, & integration time
limited primarily by the optomechanical precision of telescope
and instruments
• Drawbacks: low QE (2% vs. 90%),
non-linear response, processing of
data is challenging
Source: UKST Southern Sky Survey (IoA, Cambridge)
Charge-Coupled Devices
• Photons converted to electronic signal via the
photoelectric effect
• CCD comprises an array of pixels, acting as “light
buckets”
• Shift register reads out pixel values by “clocking” the
gate voltage, “bucket bridgade”-style
• For sensitivity to optical wavelengths, detectors are
silicon, doped with B or P
• Contain a lattice of overlapping Si e- orbitals
• Individual e- can be dissociated from the lattice by an
incident photon and conducted to the pixel electrode
Smith & Boyle shared (half of) the Nobel
Prize in 2009 for the invention of the CCD.
CCD Fabrication
•
•
•
•
•
•
Source: LBNL
•
•
•
•
Start* with bare Si wafer
Slicing, polishing
Thinning
Doping
Plasma etching
Chemical vapor
deposition
Photolithography
Dicing
Packaging
Testing
* There’s a whole talk hidden in this one word too.
A Single CCD Picture Element (Pixel)
Fully Depleted
Deeply Depleted
Source: Dark Energy Survey/LBNL
CCD Quantum Efficiency
GMOS-S (Hamamatsu Red/Blue CCDs)
Source: LBNL
Non-Imaging Regions of a CCD
Source: LBNL
Voltage Bus
Imaging Array
Vertical
Clock
Voltages
Channel Stop
Shift Register
Source: FNAL Silicon Detector Facility (Gordie Gillespie)
Source: Simon Ellis (AAO)
Charge Transfer (In)Efficiency
Standard CCDs usually have 99.999% (“five
nines”) charge transfer efficiency in the shift
register. That sounds pretty good, but…
(0.99999)(4096+1024) = 95%
(0.999999)(4096+1024) = 99.5%
Six nines is significantly better!
Noise Characteristics: Bias Voltage
GMOS-N CCDs
A baseline voltage is applied to all pixels so that all random noise is by definition positive.
But this requires subtracting this “base level” read noise from all science-related exposures.
Bias frames can also be used to identify the effects of some physical defects in the CCD.
Depletion Fraction/Voltage Effects
650 μm thick CCD
Not fully depleted at 100V
Source: LBNL
200 μm thick CCD
Fully depleted at 100V
CCD Cold-Probe Testing at FNAL
Source: FNAL Silicon Detector Facility
Errors Arising from CCD Fabrication
•
•
•
•
•
Tape bumps (from physical assembly process)
Glowing edges (e-  γ in amplifier near pixels)
Hot spots (substrate voltage leakage)
Bad pixels (the “bucket” has a hole…)
Bad columns (“traps” – the voltage or shift
register connection is broken)
• And many more….
From Pixels to CCDs: Choices
•
•
•
•
•
•
•
•
•
•
•
Will you be taking images or spectra?
Do you want a full frame CCD (needs mechanical shutter) or a frame transfer CCD (faster readout,
but 1/2 active Si area)? Do you want front- or back-side illumination?
How thick is your photoactive region (i.e., how red-sensitive)? Do you want deep/full depletion?
Does your telescope track (non-)sidereal objects, or do you observe in drift scan mode? For the
latter, your readout must be aligned precisely with the scan direction.
Do you want to cool your instrument? To what temperature -- do you need a cryogenic system or
will thermoelectric cooling provide sufficiently low dark current?
What wavelength region will you observe? What detector material and A/R coating will you use?
What size pixels do you require in microns, and in arcseconds on the sky (the “plate scale”) given
your telescope optics? Are you over- or under-sampling (2 * pixel scale > minimum seeing)?
What field of view do you want to cover in an exposure? How many CCDs will be in your
detector? How close do they need to be? This determines whether to use picture frame or 2-side
buttable or 4-side buttable devices.
What level of cosmetic defects will hamper your science? What voltages can the silicon
withstand?
Your readout likely saturates at 65535 (216 – 1 for a 16 bit system); what fullwell capacity (between
say, 100ke- to 300 ke-) is optimal? What is the requisite gain?
At what stage does saturation/non-linearity begin to affect your measurements?
Source:
Dark Energy Survey/
LBNL
Fabricating Devices is Tricky!
Device Yield
Fabrication: 80%
(“Mechanical CCDs”)
Cosmetic Defects/
Cold-Probe Tests: 60%
(“Engineering CCDs”)
“Science Grade”: <25%
Source: DES/FNAL
The finished product (sort of)
Source: DES/FNAL
CCD Focal Plane Installation
Source: DES/FNAL
Instrument Installation
Source: DES/CTIO
Data Acquisition (DAQ)
IRAF/IDL (and even cfitsio) are still far away.
Between the shift register and the image is
the DAQ
For example: Monsoon crate containing Clock
& Bias Boards, Master Control Boards
Also, the electronic signal may be turned
back into photons via S-Link fibre optic cable
to the data PC.
Then telescope/observation control and
telemetry data are added
Then you have a (FITS) file with pixel array
data and header information
Then you can test your on-sky performance
ASIC Readout for SNAP
Source: LBNL
Fullwell/Saturation
(~120ke- = 65535 ADU?)
Fullwell degradation from
over-exposure?
Gain (ADU/e-)
Linearity (fraction of FW?)
Shutter Vignetting
Source: DES/PreCam
Shutter Vignetting, Saturation
Source: DES/PreCam
Dark Current
• Integrate for the same amount of time as a
science exposure, but with the shutter closed.
• This determines the (thermal) noise rate that
is then subtracted from each pixel.
• Cryogenic cooling to ~173K is common for
many astronomical instruments, some can get
away with 183 or warmer without swamping
the detectors with thermal noise.
Flat Fielding
Pixel-to-pixel variation in
(quantum) efficiency means
uniform illumination will not
result in uniform counts
from every pixel
Thus we divide each science
image by a flat field image
so that a pixel with, say, 2x
the efficiency is weighted by
½ relative to the others.
Dipoles?
AAT Wide Field Imager +
Prime Focus Unit in z band
Fringing
Scattered light from within the CCD, especially noticeable in redder wavelengths
GMOS-S CCDs
Cosmic Rays
Source: DES/PreCam
What’s the source of this noise?
Spoiler: readout cable/pin connector micro-fractures
resulting in intermittent additive (common mode?) noise
Source: DES/PreCam
What’s the source of this noise? II
Sources of Counts in a Pixel
•
•
•
•
•
•
Read Noise (Hopefully subtracted properly)
Dark Current (Hopefully subtracted properly)
Sky Noise (Ambient and Anthropogenic)
Scattered light/ghosting (LEDs in instrument)?
Photons from a source (Signal + Shot Noise)
Signal to Noise calculators take all of these (as
well as system throughput) and determine
exposure times
• This feeds into observing proposals…
TAIPAN: A Case Study
35.00%
Total System Throughput
30.00%
25.00%
20.00%
15.00%
10.00%
5.00%
0.00%
375 400 425 450 475 500 525 550 575 600 625 650 675 700 725 750 775 800 825 850 875
Includes:
Spectral Cross-Contamination
CCD QE
Spectrograph Optics/Coatings
Fibre Output/Connector
Fibre Length
Fibre Injection Loss
Fibre Alignment
Interface Thermal Aberration
Focal Plane Thermal Aberration
Focal Plane Optics
Vignetting
Telescope Optics
Atmosphere
TAIPAN: A Case Study
From Survey Requirements through Exposure Time
Calculator to CCD Specifications
• Magnitude 18 galaxy with S/N ~10 in 2.5 hours
• Magnitude 12 star with S/N ~100 in 2.5 minutes
– Long exposures for faint objects
– Short exposures for bright objects
– Medium exposures for medium objects
• Under what conditions do Read Noise, Dark
Current, Sky Noise, or Shot Noise dominate?
• TAIPAN Requirements:
– Read Noise < 3 e-/pix
– Dark Current < 0.01 e-/pix/s
Wavelength
Galaxy mag at wavelength
Line Flux in SRE
Line Width
Source Area
Fraction of light per fibre
Mag on fibre
Fnu from object cuum
4380 Angstroms
15 mags
40 1e-17 ergs/cm2/s
120 km/s
167748.5232 nJy/m^2
Flambda from object cuum/fibre
2.6232E-19 W/m^2/A
Photons from object cuum/fibre
Photons from object line/fibre
Line width
Photons from object line/fibre/angstrom
Sky brightness
18.46 mags
6.3536E-18 W/m^2/A
1.65147E-13 ergs/m^2/s/fibre
6.3536E-15 erg/s/cm2/A
1.13 m^2
13.5 %
Exposure time for one integration
1600 seconds on target
4063 Jy (Vega mags)
18.5 mags/arcsec^2
Fnu from sky
161750.9434 nJy/m^2/arcsec^2
Flambda from sky
2.52941E-19 W/m^2/A/arcsec^2
Photons from sky
ph/m^2/A/sec/arcs
0.557005115 ec^2
Fibre Area
7.068583471 arcsecs^2
262.3 10-18 erg/s/cm2/A True sky/Fibre
1.65147E-17 erg/s/cm2
0.578 ph/A/sec/fibre/m^2
0.036 ph/sec/fibre/m^2
1.752 Angstroms
0.021 ph/A/sec/fibre/m^2
Telescope area
System efficiency atm->detector
4380 Mag zero point
1.75 A
171.21 arcsec^2
0.04129
Total Flambda from object cuum
Flambda from object line/fibre
5384
3.93723715 ph/A/sec/fibre/m^2
Pixel spectral size
0.952173913 Angstroms
True sky per pix
Pixels per SRE
Instrument PSF
Dark count rate
Scattered OH rate
Readnoise
Gain
3.748934503 ph/sec/SRE
2 pixels
2 pixels
0.003 electrons/sec/pix
0 electrons/sec/pix
3 electrons/pix
1.788 electrons/ADU
Detected sky
electrons
1830.079867 per SRE
11300 cm^2
Det .back. electrons 1839.679867 per SRE
Detected cuum electrons
269 per SRE
150.1695991 ADU/SRE
Back. noise
Detected line electrons
5 per SRE
2.702175787 ADU/SRE
Sky subtraction fac.
Total Detected Line electrons
8
Cuum Signal/noise per integration
Line Signal/noise per integration
Total Line S/N per Integration
Number of integrations
5.8 per SRE
0.1 per SRE
0.2
3
Sky/Object cuum
10.0 per SRE
1 sqrt(2) or 1
23.46183742 per SRE
Spec Resolution R=
2,300
Dark Current Sky Noise
Read Noise
Object Noise
3.098386677
42.77943276
6 16.38606857
Spec Res final =
Cuum Final Signal/noise
43.30911991 per SRE
575
Other Detector Technologies
• HgCdTe (“MerCadTel”): IR-sensitive (1.5-12 um) detectors.
– Hawaii 2-RG detectors (e.g. in the Gemini Planet Imager) have
a (non-destructive!) readout layer made of Si.
– Thermal noise (infrared photons) is more of an issue, requiring
high-performance cryogenic cooling.
– Much of what has been described here for Si is applicable to
HgCdTe.
• InSb (e.g., ESO’s ALADDIN sensor arrays on VLT), or InGaAs
• Photomultiplier tubes (“avalanche” detectors)
• Electron Multiplying Charge-Coupled Device
(Si CCD with increased Gain like PMT, cf. NuVu Cameras)
• Other wavelengths? (gamma, X, UV, microwave, radio)
• Other astrophysical messengers? (neutrinos, gravitational)
• Microwave Kinetic Inductance Detectors (MKIDs)…
The Future(?): MKIDS
LC Circuit gives energy and timing information of individual photons
but requires (even more) sophisticated readout and precise cooling
than a standard Si-based device
Source: Mazin Lab (http://web.physics.ucsb.edu/~bmazin/Mazin_Lab/MKIDs.html)
The Dark Energy Camera
62 2k x 4k full-frame, fully-depleted, backside-illuminated, 4-sidebuttable, dual-readout, p-channel LBNL CCDs with 250 micron x 15
micron x 15 micron pixels, possessing 130ke- fullwell, 2.5 ADU/e- gain,
“five nine” CTE, totaling 520 Megapixels with >50% QE from 375 to
1000 nm, operating at 173K, with an overall FOV of 3 sq deg.
Only after this entire process can your detectors
obtain results like this:
Coma Cluster; Source: DES
Or this:
NGC 1365; Source: DES
Thank You!
Source: Marcelle Soares-Santos, FNAL
Useful References
• AY257 Modern Observational Techniques
http://www.ucolick.org/~bolte/AY257/s_n.pdf
• Dark Energy Survey Instrumentation
http://www.darkenergysurvey.org/DECam/
(see also http://www.ctio.noao.edu/noao/content/DECam-Known-Problems)
• Mazin Lab
http://web.physics.ucsb.edu/~bmazin/Mazin_Lab/MKIDs.html
• Gemini Instrumentation
http://www.gemini.edu/node/10002
• Me (Follow up questions? Want more details?)
[email protected]
• Wikipedia/Amazon (no kidding)
APS Committee on International
Freedom of Scientists