ASTR 511 (O'Connell) Lecture Notes
CCD'S IN ASTRONOMY
Subaru CCD Mosaic 8 x (2k x 4k)
Charged coupled devices (CCD's) have been used in astronomy
since the late 1970's. They are now nearly ubiquitous in observations made
at wavelengths between the near-IR (~1 µ) and the X-ray.
They have transformed the way astronomy is done.
I. REFERENCES FOR CCD'S
CCD Data Reduction and Applications
- Buil, CCD Astronomy
- Howell, Handbook of CCD Astronomy
- Kitchin, Astrophysical Techniques
- LLM Sec. 7.3.3.
MacKay, Ann Rev Astr Ap, 24, 255, 1986
- Rieke, Detection of Light : general introduction to detector physics
Timothy, PASP, 95, 810, 1983: discussion of many detector types
- Abbott, "In Situ CCD Testing"
Franx et al. AJ, 98, 538, 1989: sample application to
- Fruchter, "Imaging Dithering and CR Rejection in Undersampled HST
Data"; also see the HST MultiDrizzle
- Massey, A User's Guide to CCD
Reductions with IRAF (1997)
- Stetson: many papers on CCD applications to point source
photometry, especially of star clusters; author of widely-used
DAOPHOT program for point-source photometry
- Tyson, JOSAA, 3, 2131, 1986: high precision technique
Bertin & Arnouts, A&AS, 317, 393, 1996 SExtractor---source
ID and extraction software for digital images. Obtain software
II. GENERAL DETECTOR CHARACTERIZATION
- QUANTUM EFFICIENCY
- QE = percentage of photons incident on detector which produce
- Strong wavelength dependence (e.g. threshold activation cutoffs set
by workfunction/band gap)
- Typical peak values:
- Eye: 1-2%
- Photographic plate: 1-2%
- Photomultiplier tube: 20-30%
- CCD: 70-90%
- IR array (HgCdTe): 30-50%
Schematic QE curves for various classes of
- "Detective quantum efficiency" is defined as
[(SNRout)/(SNRin)]2, where "in"
and "out" refer to the input and output signals to/from the
detector, respectively. DQE combines basic QE with the noise
introduced by the detector. This quantity is really what
matters in comparing detectors, but it is so dependent on
specific details of operations/applications that it is rarely
- SPECTRAL RANGE
- Wavelength region over which QE is
sufficient for operation
- DYNAMIC RANGE
- Definition: ratio of maximum to minimum measurable signal
- E.g. maximum number of events in a CCD pixel is determined by
photoelectron "full well'' capacity or digitization maximum (typically 2 bytes);
minimum is determined by dark current/readout noise
- Applies to a single exposure; effective dynamic range
can be increased with multiple exposures
- Typical values: 100 (Pg); 104 (CCD); 105 (PMT)
- Related concepts:
- Linear Range: range of signals for which [Output] = k x [Input],
where k is a constant. Generally smaller than calibrateable range
- Threshold: minimum measurable signal. Influenced by detector
noise or other intrinsic characteristics (e.g. fog on Pg plates)
- Saturation point: level where detector response ceases to
change with signal
- NOISE CHARACTERISTICS
- PMT: ~0.0001 ms
- CCD: ~10 ms per pixel
- Spatial (array detectors)
- Pixel = minimum measureable area of detector surface. Typically
10-50 µ on Pg or semiconductor types. Pixels are not
necessarily independent of one another.
- Resolution cell: according to the Nyquist criterion, the
resolution cell is 2x the size of the minimum
independent measurable area. For proper sampling of image,
need at least 2 pixels per resolution cell. A lower pixel density implies
"undersampled" imaging. A significantly higher pixel density does not
provide more information and is a waste of pixels.
For an imaging application, the Nyquist criterion implies that
1 pixel should span ~ (FWHM)/2, where FWHM is the full width
half maximum of the point spread function.
- Most UVOIR detectors are broad-band; inherently poor spectral
resolution. Must use external elements (filters, gratings) to provide
- Exception: superconducting "3D" detectors, which measure photon energy
as well as position; current spectral resolution is
R ~ 100
III. BRIEF HISTORY OF ASTRONOMICAL DETECTORS
Photographic emulsions work by storing in
AgBr crystals the photoelectrons ejected following absorption of
photons during exposure to light. Chemical reactions during
development cause the crystals to precipitate grains of silver, which
form a permanent image.
Film was the detector of choice for almost all applications in
astronomy from around 1900 to 1960 and for imaging until about 1980.
It is impossible to exaggerate its importance to the development of
modern astronomy. Even with relatively low QE, photographic plates
offered decisive advantages over visual observations:
However, the photographic emulsion had serious limitations with which
astronomers had long struggled:
- Very long exposure times (up to a week in early applications),
meaning limiting fluxes thousands of times fainter
- A permanent,
objective record of astronomical images and spectra
- Geometric stability, important for astrometry and morphology
- Large formats for wide field surveys
- Extension of observations to EM bands beyond the
"visual" band (the classic "photographic" band is centered around 4500
For details on astronomical photography at low light levels, see Smith & Hoag
1979, ARAA, 17, 43.
PHOTOELECTRIC ASTRONOMY TO 1980:
- It is relatively insensitive,
with QE ~ 1%.
- Pg materials have non-negligible thresholds (a minimum of ~4 electrons
per grain), a strongly non-linear response function, and limited
dynamic range. These properties plus the use of hard-to-control
chemical processes for emulsion deposition & development makes them
very difficult to calibrate quantitatively.
- Each individual plate (or at the best, each co-processed "batch" of
plates) has different characteristics.
Conversion of data to quantitative form (e.g. with microdensitometers) is
slow and cumbersome.
In 1907, Joel Stebbins
(UIll, UWisc) began testing various types of photoelectronic
detectors. These, such as selenium phototubes, were largely
experimental and temperamental, but use of PE devices spread after the
World War II greatly accelerated these technologies, with mass
production of high quality vacuum tubes and sensitive electronics for
detection of faint signals.
The key design for astronomy was the photomultiplier tube (PMT), the first
widely used example of which was the RCA 1P21. The initial detector
in a PMT is a photo-emissive cathode surface, made from
alkali metal compounds, which ejects a single electron in response to
a photon absorption. A series of other "secondary electron emissive"
surfaces (the "dynode chain") amplifies this into a burst of
~106-7 electrons. (See illustration below.)
PMT's require cooling to suppress dark current and were typically
operated at dry-ice temperatures (-78C). The photocathode QE's of
PMT's in the optical region reached 20-40% by the 1960's, with very
wide acceptance bands.
Although PMT's were initially operated in a "direct current" mode, the
pulse-like character of PMT output led to the adoption in astronomy of
the same kinds of high speed digital electronics that had been
developed earlier for accelerator and nuclear physics. This kind of
"pulse-counting" operation offers excellent noise rejection and
permits the detection of individual photons from cosmic
For more details on PMT's and PMT-based photometers, see the articles
by Lallemand and Johnson in Astronomical Techniques, ed. W. A. Hiltner
(Stars & Stellar Systems Vol II).
PMT's became the workhorses of multicolor photometry and
spectrophotometry (e.g. the Johnson & Morgan broad-band UBV system)
after 1950. They featured excellent linearity and stability, which
implied an unprecedented capacity for accurate calibration of
photometric measurements. PMT photometry routinely reached the 1%
level for flux calibration. In turn, this was responsible for an
explosion in quantitative astrophysics.
Despite their excellent performance and their
broad wavelength response, PMT's were single element
devices. They were not easily adaptable to 1-D, let alone 2-D,
applications (such as imaging), despite heroic attempts such as the Palomar Multichannel Photoelectric Spectrometer (Oke,
Around 1960, a plethora of efforts began to develop robust 1-D and 2-D electronic devices suitable for
astronomy. A dozen or so of these produced practical systems (e.g.
the Secondary Electron Conduction Vidicon, the Intensified Dissector
Scanner, the Intensified Reticon Scanner, the Image Photon Counting
System, and the Multi-Anode Microchannel Array detector).
Most of these employ some type of image
intensifier vacuum tube as a key element. This technology was
developed for military night vision systems. Intensifiers produce
easily detectable light pulses in a 2-D image field from individual
incident photons by accelerating the photo-electrons to high energies
and/or producing a cascade of electrons, then focussing these on a
luminescent output screen. Disadvantages include the frequent use of
high voltages (e.g. 25 kV) and serious image blur which, however, can
be reduced by the use of pulse centroiding electronics. Fiber
optic input/output plates were commonly used with intensifiers after
1970. Microchannel plate (MCP) intensifiers have often been
used in space astronomy missions (including HST, GALEX, and FUSE).
SOLID STATE ARRAY DETECTORS:
But by far the most successful 2-D devices for astronomy emerging in
the last 30 years have been solid
state, semiconductor arrays. These are based on
photo-conductive materials fabricated with embedded
microelectronic integrated circuits on thin wafers by photolithography. Although lower quality devices can
be mass-produced by microchip "foundries," professional grade
detectors still need to be custom-processed.
Solid state arrays are now employed as astronomical detectors from the
X-ray to the far-infrared. The most widely used is the Charge-Coupled Device (CCD), which
operates at wavelengths from the X-ray to ~1 µ. The
basic CCD architecture was invented at Bell Labs in the late 1960's.
By the mid-70's, CCDs were being tested as astronomical detectors.
They did not come into widespread use until ca. 1980, following
extensive development efforts associated with the Wide Field & Planetary Camera on the Hubble Space
Telescope, led by J. Westphal, J. Gunn, and C. R. Lynds.
Wafer containing a number of co-fabricated CCDs
and other devices.
Semiconductors are crystalline materials which are not
normally good conductors of electricity but which can be made to
conduct under certain circumstances. The central useful property of
semiconductors employed in astronomy is that their electrical
properties change significantly after absorption of photons.
Absorption of a photon can push a valence electron into the conduction
band and produce a potential electrical signal. The photon energy
must exceed Eg, which implies that there is a maximum
wavelength for excitation given by:
Lammax = 12,400 Å/Eg(eV)
Obviously, materials with band gaps in the few eV range are of interest
as potential UVOIR detectors. Band gaps and max wavelengths for some
important materials are given in the following table:
BAND GAPS: The properties of semiconductors are manipulated by
changing the structure of their internal energy levels, which are
spread out into "bands" by the proximity of the component atoms of the
solid. The "valence" band, corresponding to the outermost electrons in
a normal, unexcited atom, is the lowest energy band where electrons
are able to move between ions. However, no net conduction occurs as
long as the band is full. Above this lie the "conduction" bands,
which are not filled, and where electrons will move freely in response
to external EM fields. The separation between the valence and first conduction band
is called the "band gap energy", Eg.
Different materials have a wide range of band gaps. "Conductors" have
a zero gap, meaning that electrons are always available to transfer
charge. "Insulators" have very large gaps, implying zero conduction
except under extreme circumstances. "Semiconductors" have
Photons are primarily absorbed by electrons in the valence band.
For photon energies above Eg, the electron is boosted
to the conduction band, leaving a hole behind. If a positive
voltage is applied at one side of the semiconductor, the electron
will be attracted toward it while the hole will be repelled.
DOPING: The "elemental" semiconductors are those elements in group
IVa of the Periodic Table (including Si and Ge). These have four
electrons in their valence shells, half the maximum allowed. These
are shared among the ions in "co-valent bonds." There are many other
types of "compound" semiconductors, however, composed for instance of
atoms from group IIIa and Va of the Table; two of
these are listed in the table above.
The electrical properties of pure semiconductors can be dramatically
altered by adding ("doping with") small amounts (~1 part in
106) of an impurity. The result is called an "extrinsic"
By adjusting the amount of doping, the band gap of the semiconductor
(donor/acceptor levels in diagram at right) can be customized. By
layering n/p regions, the response to applied potentials can be adjusted
to create a large variety of electronic devices.
- n-type: a material with more than 4 valence electrons is added
(As, from group Va, in the illustration). The extra electrons
cannot be accommodated in the valence band and so occupy the conduction
band. They represent a persistent set of negative carriers
- p-type: a material with fewer than 4 valence electrons is
added (e.g. B, from group IIIa). This has one fewer electron than normal
and creates a small "vacuum" in the electron sea of the valence band.
This is called a "hole." As valence electrons shift to fill it, the
hole propagates like a positive charge in the opposite direction.
The holes represent a persistent set of positive carriers.
- In pure semiconductors, conduction electrons and holes can also
exist, if electrons are excited by thermal effects, for instance. But
they always occur in pairs. Electrons and holes in n- and p-type
materials, respectively, have no counterparts. Extrinsic material is
electrically neutral but is more responsive than pure materials to a
difference in electrical potential.
Doping affects energy levels
V. BASIC CCD DESIGN
Apart from sensitivity, the key design issues for solid state arrays
are to localize photon-produced charges on their surfaces and then
arrange to amplify and read them out without distorting the image or
introducing unacceptable amounts of noise.
A CCD is a charge-transfer device. Its storage, transfer, and
amplification electronics are all fabricated on a single piece of
silicon (unlike most IR arrays). During an exposure, it traps
released photoelectrons in small voltage wells. After the exposure,
the electrons are shifted in a series of "charge-coupled" steps across
the CCD surface, amplified, read out of the CCD, and stored in a
computer memory. This is "destructive readout"---i.e. one cannot
read the same signal again (unlike other array architectures, where
each pixel is coupled to a separate amplifier).
BASIC STRUCTURAL ELEMENT: The basic element in a CCD design is a
"Metal-Oxide-Semiconductor" capacitor. See the illustration
above. This serves both to store photoelectrons and to shift them
wholesale. The bulk material is p-silicon on which an insulating
later of silicon-oxide has been grown. P-silicon can be made to have
very small numbers of free electrons ("high resistivity") before
exposure to light; this is important for best performance. A set of
thin conducting electrodes of semitransparent "polysilicon" are
Before exposure, the central electrode is set to a positive bias while
the two flanking electrodes are set negative. This creates a
"depletion" region under the central electrode containing no
holes but a deep potential well to trap electrons. The region shown
is about 10µ thick.
BUCKET BRIGADE: The resulting transfer and readout process is illustrated
in the animation below:
ADU CONVERSION: For storage in memory, the electrical signal generated
by the amplifier must be digitized. This is done by an
"analog-to-digital converter". This is normally adjusted
such that one digital unit corresponds to more than one
photo-electron. Typical values of this conversion are 2 to 8
electrons per stored digital unit.
The stored values are called "ADU's", for analog-to-digital-unit. The
corresponding constant of transformation, normally quoted in units of
"electrons per ADU", is often called the "Gain" (although this is confusing
nomenclature because a larger Gain results in reduced ADU values).
Note that the use of such a conversion importantly affects the
statistical properties of the recorded signal. If x is the
recorded signal in ADU's, y is the original signal in photo-electrons,
and G is the gain, then from
Lecture 8 we see that:
Var(x) = Var(y)/G2
OPERATION SEQUENCE: During exposure (controlled by a separate
shutter), light enters through the "front-side" electrodes.
Photoelectrons generated under the central electrode will be attracted
toward the electrode and held below it. The corresponding holes will be
swept away into the bulk silicon. Good performance requires little
diffusion of electrons away from the potential well.
The surface of the CCD is covered with MOS capactitors in a pattern
like that at right. In this particular design, there are three
electrodes per pixel. A single pixel is shown shaded in the diagram.
Typical pixel sizes are 10-40 µ. The "parallel
shift" registers are shown as rows running across the whole face
of the CCD. These are separated by insulating "channel stops."
At one end is a column of "serial shift" electronics and an
output amplifier. Note that there is only one amplifier in this
design. Contemporary large chip designs involve several amplifiers
(but always many fewer than the number of pixels!).
At the end of the exposure, readout of the collected electrons is
accomplished by cycling ("clocking") the voltages on the electrodes
such that the charge is shifted bodily to the right along the
rows. The figure at the right shows how this is done. Good
performance depends on near-100% transfer of the electrons
to/from each electrode with no smearing or generation of spurious
Each parallel transfer places the contents of one pixel in each row
into the serial register column. This column is then shifted out
vertically through the output amplifier and into computer memory
before the next parallel transfer occurs.
VI. CCD DESIGN ISSUES
CCDs have undergone a long optimization process since 1980.
Contemporary designs have excellent performance but still require
careful calibration in order to overcome inherent limitations. There
is also only a handful of reliable manufacturers of
Here are some of the issues affecting electronic design and
manufacture of CCDs:
- INTRINSICALLY LOW BLUE RESPONSE (< 4500 Å):
Caused by absorption in bulk Si and by electrode structures in
- Use special thin, polysilicon material for electrodes. But cannot
be completely transparent.
- Special Coatings: "Anti-reflection" coatings trap
photons, causing multiple reflections as in Fabry-Perot etalons, and therefore
enhance absorption. "Downconverter" coatings are phosphors which
absorb blue photons but emit green photons at wavelengths where the
CCD QE is higher (e.g. "mouse milk," coronene, lumogen).
- "Thinned, Backside-Illuminated" chips: shave off silicon
subtrate, leaving only ~10µ deep unit; illuminate from
backside; greatly improves blue response. For techniques, see Mike
Lesser's UofA website.
Difficulties with thinned chips:
- Non-uniform thinning
- Surface trapping by SiO2 layer of photo-electrons produced
nearby (shorter wavelengths)
- Interference effects if wavelength ~ chip thickness (i.e. in
IR). Strong spatial modulation of response = "Fringing".
Especially serious for night sky emission lines. Can be well
calibrated for narrow-band filters or for broad-band filters. Hard
for intermediate band filters.
- Thinning reduces red response. For good response
5000--10000 Å, prefer thick (~500µ)
- CHARGE TRANSFER EFFICIENCY (CTE)
- CTE can be better than 99.999% per transfer, but has to be, since
throughput of chip with 2048 required shifts = CTE2048.
- Radiation damage to CCD's in space seriously decreases CTE over several
Operational: add (electronically or with diffuse light source) a
pedestal background signal (a "fat zero") over whole chip to increase
mean electron density per pixel. However, adds additional noise and
not suitable for very faint source applications.
Technical design: change number phases, clocking cycles; add
VII. ADVANTAGES OF CCD DETECTORS
- HIGH QUANTUM EFFICIENCY: To 80-90% at peak in optical. Much
effort was expended to reach these high levels.
- This had tremendous impact on astronomical imaging &
spectroscopy. It meant the detection threshold with any instrument
was extended 4-5 magnitudes over film and that therefore a 1-m
telescope could now pursue the kind of science previously possible
only with 4-m class telescopes.
- A key corollary: since we are already near 100% QE, at least in the
optical region, achieving significantly lower threshold fluxes
requires larger telescopes rather than better detectors.
NB: "Quantum yield" can be over 100% for far-UV and X-ray photons
(i.e. more than one photoelectron can be generated but
fewer than 100% of photons produce responses).
Sample CCD QE curves (ESO)
- LINEAR RESPONSE: Until approach full-well capacity (typically 200,000
e/pixel). This implied much improved flux calibrations and much higher
precision for flux measurements at all levels.
- EXCELLENT DYNAMIC RANGE: Typically 104.
- WIDE WAVELENGTH RESPONSE : Intrinsically sensitive from X-ray to
~1µ. Other materials (e.g. InSb2, HgCdTe)
with similar architectures are usable in the IR.
- GEOMETRICALLY STABLE: good for astrometry
- ONLY LOW VOLTAGES REQUIRED (~5-15 v)
- RELATIVELY CHEAP, AVAILABLE, SIMPLE: Compared to other
digital 2-D systems. Standard chips cost
- RELATIVELY LOW NOISE : Compared to many other classes of
astronomical detectors, e.g. Pg plates, Reticons, SEC vidicons, etc.
But noise is not negligible. Typical read noise now 2-10 e/pixel, and dark
current is largely suppressed by cooling.
- SMALL PIXELS : Typically 10-30 µ. Usually an advantage,
but should match pixel size to 1/2 of smallest resolvable picture element
in optical system.
- SPECIAL FORMAT/READOUT DESIGNS: By changing electrode structure
and clocking cycles, one can arrange for many different integrate/readout
Rapid clocking/video: inherent in CCDs intended for TV application. For
bright sources, readout rates of 100 MHz are possible.
Basic technique is to clock the chip slowly along columns while moving
the telescope to keep a target centered on the moving electron cloud.
Continuously read out the chip to produce a strip-image of the sky.
The integration time is set by the drift time across the chip.
The simplest approach is to align the CCD columns east-west, keep the
telescope fixed on the sky, and clock the chip westward at the sidereal
rate. A sample application of drift scanning is described here.
Drift-scanning is now a standard method for wide-field CCD sky
surveys, most notably the Sloan Digital Sky Survey
"Nod and Shuffle": this technique takes advantage of the
capability to shift an image wholesale on the CCD without reading out to obtain
much better sky subtraction (e.g. in near-IR where atmospheric
OH emission is very bright and variable).
On-chip binning: change clocking to combine electrons from several
pixels together before reading out through amplifier. E.g. combine a
2x2 pixel region on the chip into a single output pixel. This reduces
the effect of amplifier readout noise on each pixel in the final
data. Also reduces memory and storage requirements. Adds
considerable practical flexibility to CCD systems. Obviously,
however, reduces the resolution of the output image. Is useful
for applications such as imagery of very faint, extended sources (e.g. galaxy
halos), low spatial resolution spectroscopy, photometry of point sources
under poor seeing conditions, etc.
- UBIQUITOUS : Now almost universally used in astronomy
(amateur & pro). Photographic materials and older electronic
detectors being phased out.
- IMMEDIATE DIGITAL CONVERSION OF DATA: The other advantages of
array detectors notwithstanding, it is the immediate conversion of
astronomical signals into a form capable of computer
storage/analysis which has so dramatically transformed UVOIR
astronomy since 1980. The practice, pace, & scope of UVOIR astronomy
are entirely different now than in the "photographic" era that
preceded widespread use of CCDs and other array devices. Digital
conversion of images and spectra has enabled quantitative analysis of
observations on a scale not possible before.
Among other things, rapid digital processing allows
much improved use of observing time---notably in surveys (e.g. for
variable sources in MACHO and supernova searches; moving targets such
as asteroids/Kuiper Belt objects; or combined imaging/spectroscopic
surveys such as SDSS, 2dF).
VIII. LIMITATIONS/DISADVANTAGES OF CCD'S
- SMALL SIZE : Individual chips typically less than 7 cm square.
Cover only ~20% of high quality imaging field on typical modern
telescope. Size limited by small fabrication yield of large,
defect-free chips. Largest routinely available chips are 4096x2048.
- However, MOSAICKING technology now well developed. Typical
mosaic now uses 4096x2048 chips.
- Examples: (click on images below for enlargements):
- Sloan Digital Sky Survey: 54 chip mosaic for drift scanning in 5 bands simultaneously
- Canada-France-Hawaii-Telescope MegaPrime mosaic: 40 4096x2048 chips covering
1 degree FOV
Sloan DSS Camera
Sloan DSS Camera
CFHT MegaPrime Mosaic
- CRYOGENIC COOLING REQUIRED: To reduce dark noise, cooling to below
-100 C necessary. Thermoelectric coolers usually not adequate, so
cryogens (e.g. liquid N2) required. Introduces many practical
complications. Optimal low-T chips differ from commercial types used
at room temperature in digital cameras, etc.
- READOUT NOISE: Produced by variations in amplifier gain. Much
effort to reduce. Now typically 2-10 e/pix.
What matters is not the noise per pixel but rather the total noise per
image area, which can extend over many pixels, depending
on the application.
Even at these very low levels, RON can
compromise some types of observations (spectroscopy, surface
photometry), see Lecture 7.
Can reduce RON effects in some applications using on-chip binning (see
above). In some cases, it may be more advantageous to use "pulse
counting" detectors, which can unambiguously detect individual photons,
- FULL WELL CONSTRAINTS: Bright sources which over-fill pixels can
produce "blooming" or "bleeding" along columns, making parts of
the chip other than the immediate vicinity of the source useless.
Best solution is operational.
- RESPONSE NONUNIFORMITIES ("FLAT FIELD" EFFECTS): Caused by small
variations in masks used to manufacture chips, deposition
irregularities, thinning variations, etc. Typically 5%
pixel-to-pixel. Color dependent. Requires extensive calibration,
with color-matching to targets. Use "dome flats" or "twilight
flats." Special observing procedures to reduce flat field effects
- Drift Scanning: see above.
- Multiple Offset: Break exposure into 4-5 parts, offset ~50
pixels between exposures. Combine exposures during data reduction.
"Dithering" is a related technique involving smaller offsets to
achieve sub-pixel spatial resolution. These methods result in reduced
field of view because not all parts of the original field will have uniform
- SENSITIVITY TO COSMIC RAYS: Especially thick chips. Must remove
effects during processing. CR's are a major problem for
spacecraft detectors (e.g. HST/WFPC2). Requires that exposures be
broken into multiple parts (called "CRSPLITs" on HST) so that CR
events can be detected. CR hits can be removed from data frames, but
this always leaves "holes" which have less exposure than other
parts of the image.
- SENSITIVITY TO X-RAYS: An advantage for X-ray astronomy, but some
materials in vicinity of CCD's, e.g. special glasses used for windows,
can produce a diffuse background of X-rays which add noise to
CCD Flatfield Frame (AAO)
CR's on HST/WFPC2 2400 sec image (extract)
- CHARGE TRANSFER EFFICIENCY: Good CTE is often possible
only for signal levels above threshold ~10-50 e/pix. For low light
levels, require adding "fat zero" (typ. 1000 e/pix)
electronically or by preflash. Creates added noise (10-30 e/pix).
- AVAILABILITY HOSTAGE TO COMMERCIAL MARKET: CCD availability has
always been driven more by commercial and military applications than
science. Scientific CCD manufacture represents only about 10% of the
overall $1B CCD market. A serious near-term issue, since industry is
moving to "active pixel sensors," a different technology with
amplifiers incorporated in each pixel. Requirements for good
performance at low (astronomical) light levels are considerably
different than for room-temperature, short-exposure, mass-produced
- DATA GLUT! CCDs typically produce 2 x Npix bytes of
data per readout, where Npix is the number of pixels. For
a 8096x8096 mosaic, this is 130 MB. Data storage/manipulation was a
serious problem when CCD's were first introduced, and this influenced
the style of data processing systems such as IRAF. Disk and tape
storage are now much improved, but long-term stability of massive amounts
of data now being produced is a non-trivial issue. [All that data is
really important and worth saving...isn't it?]
IX. EXAMPLE CCD SYSTEMS
- HST: WIDE FIELD PLANETARY CAMERA 2 (1992)
- 4 CCD's, optically mosaicked with beam-splitter
- Loral 800x800, 15µ pixel chips
- Front-illuminated, lumogen phosphor coating for UV response
- RON 5 e/px; QE (6000 Å) 35%
- HST: WIDE FIELD CAMERA 3 (2001)
- 2 CCD's, physically mosaicked, contiguous
- Marconi 4096x2048, 15µ pixel chips
- Thinned, back-illuminated; anti-reflection coating for enhanced blue/UV
- Minichannel for improved CTE
- RON 3 e/px; QE (6000 Å) 70%
- SLOAN DIGITAL SKY SURVEY (1997)
- Drift scanning mosaic (not contiguous) containing 54 chips
- Tektronix/SITe 2048x2048, 24µ pixel chips in 5x6
array; for simultaneous broad-band photometry. Both frontside and
thinned backside with AR coating. QE (6000 Å) ~60%; RON < 5
- Tektronix/SITe 2048x400, 24µ pixel chips; for
astrometry & focus check. Frontside illuminated, RON < 15 e/px.
- FAN MOUNTAIN CCD SYSTEM (Gen II)
X. CCD HIGH PRECISION CALIBRATION PROCEDURE
A. DATA REQUIRED
B. REDUCTION PROCEDURE
- Bias frames
- Dark frames (optional)
- Flat field frames
- Sky flat/fringe frames (optional)
- Flux calibrator fields
- Target frames
C. EXAMPLE SCIENTIFIC APPLICATIONS
- SUBTRACT BIAS FRAME : Bias frames (zero exposure time) taken with
chip unexposed to light from telescope. These measure electronic
pedestal of chip. For high precision, average many bias frames,
before and after observations. Check for bias drift during night.
With some chips, can determine electronic bias level from
overscan region on chip. For optical "fat zero" subtract average of
- Optional: SUBTRACT DARK FRAME: Median filter many long (~30
min) dark exposures; note possible LED activity of CCD electrodes.
Scale result to integration time of each data frame before subtracting.
- DIVIDE BY TWILIGHT OR DOME FLAT FIELD: To remove residual
pixel-to-pixel variations (typical 5%). Make exposures of twilight
sky (good diffuse source, but tricky to get exposure level right; only
2 chances per night). Or make many exposures of diffusely illuminated
screen in dome (disadvantage: often not uniformly illuminated). Must
be done for each filter used (color sensitive effects).
[Note: photon statistics must yield S/N significantly in excess of
final desired S/N---i.e. if desire S/N = 100, require over 10,000
photons per pixel in net flat field image.]
- Optional: SUBTRACT SKY FLAT/FRINGE frame: Remove night sky
emission line fringing effects (worse in near-IR) by observing
uncrowded field in night sky. Take several exposures, moving
telescope by, say, 25 arc-sec between them. Use "adaptive modal
filter" technique to zap star images and create mean sky frame. Scale
to target frames and subtract. Resulting flat field as good as 0.1%
can result. Require a sky flat determined this way for each
- USE MULTIPLE OFFSETS, DITHERING OR DRIFT SCANNING FOR TARGET FRAMES:
For faintest possible photometry, use multiple-offset exposure
technique---e.g. 500 sec exposures shifted by about 20 arc-sec each---
to reduce flat field errors. Always want more than one
exposure anyway for "reality check," empirical determination of
photometric errors, cosmic ray removal, etc.
- INTERPOLATE OVER OR MASK OUT BAD PIXELS/REMOVE COSMIC RAY
- REGISTER AND COMBINE TARGET FRAMES: Re-register offset frames to
sub-pixel accuracy (e.g. "Drizzle" softare). Median filter. Trim off
"lost" parts of image.
- CALIBRATE FLUX SCALE: Observe "standard stars" in recommended CCD
calibrator fields (e.g. star clusters) to determine nightly
atmospheric extinction and telescope throughput. No need to take on
non-photometric nights (variable clouds), but no flux calibration
either. Important that calibrator stars overlap targets of interest
in color. These frames are reduced exactly like target frames,
apparent fluxes extracted, and correction factors to determine true
fluxes of targets are obtained.
- APPLY GOOD PHOTOMETRIC REDUCTION SOFTWARE: for source IDs,
flux measurements (point or extended source).
=====> RESULT : PHOTOMETRIC CALIBRATION
GOOD TO 0.005 MAG
- Subpixel registration of dithered HST imaging by A. Fruchter's
"drizzling" technique. On left is one original frame (I-band, 2400
sec exposure). On right is result of drizzled processing of 12 such
frames. Combined image has limiting magnitude of I ~ 28. It
would be impossible to reach such levels with photographic detectors.
- Example of improvement in color-magnitude diagram for star
cluster 47 Tuc. On left is 1977 state-of-the-art CMD based on
widefield photographic images with photoelectric calibrations
(Hesser & Hartwick, 1977). On right is a 1987 CMD derived from
CCD photometry (Hesser et al. 1987). The greatly improved photometric
precision reveals new features of astrophysical interest: e.g.
the thinness of the main sequence near turnoff, which places
strong limits on the range of ages present in the cluster.
XI. NON-UVOIR USE OF CCD & RELATED DEVICES
A. X-RAY: CHANDRA/ACIS DETECTOR
ACIS is a 10-CCD (1024x1024 chips) focal plane array used on Chandra for both imaging
and spectroscopy. It uses both back-illuminated and front-illuminated
It is operated in a pulse-detection mode, unlike the standard
procedure at UVOIR wavelengths.
Each X-ray photon releases more than one electron in the
CCD, in fact, the mean number released is ~ EXR/3.7 eV.
Since Chandra operates at ~ 5 keV, the average electron cloud
corresponding to one photon has ~ 1000 electrons.
The standard operation sequence is to expose for 3.2 seconds, then
rapidly read out the array in 40 ms. The resulting image is
analyzed by on-board software to catalogue the x,y position and
the pulse amplitude of each valid photon pulse.
Because the pulse amplitude is proportional to the photon energy,
ACIS achieves a spectral resolution of R ~ 10-50.
A difficulty with the ACIS design is that if more than one photon
strikes the same pixel during the exposure time, the counting
analysis becomes distorted, and the photon flux is underestimated.
This is called "pileup." Fortunately, most X-ray sources are faint
enough that this is not a problem.
B. INFRARED ARRAYS
The table in Sec. IV above shows that pure silicon photoconductor
arrays will not work at IR wavelengths, but there are a number of
other materials that will.
There are many varieties of IR detectors in use today. Some of
these are monolithic, i.e. fabricated on single subtrates like
CCD's, and some are
hybrids in which the readout electronics
are fabricated separately from the photon-detection devices.
Hybrids typically use silicon wafers for the readout electronics.
Some actually use CCD-type architecture. The readout is connected to
the photon-sensing material using conducting "bump-bonded" indium
studs. If the wafers ultimately produce readout though a small number
of output amplifiers, they are called "multiplexers" or "MUX's".
a cross-section of the HST/NICMOS infrared detector, which uses
a HgCdTe ("mercad-telluride") photon detector.
IR arrays will be discussed further in the guest lecture by
Mike Skrutskie and in ASTR 512 (Spring 2004).
ADDITIONAL WEB LINKS
CCD-World (forum for
discussions about CCDs and their use)
Example Systems & Development Efforts:
November 2009 by rwo
Bandgap images from Britney's Guide to Superconductors.
MOS capacitor illustration from Molecular Expressions. Bucket brigade
animation and front/back illumination drawing by C. Tremonti (JHU).
CCD design drawings from C. MacKay, Annual Reviews (1986). Most other
images from public observatory sites. Text copyright © 1989-2009
Robert W. O'Connell. All rights reserved. These notes are intended
for the private, noncommercial use of students enrolled in Astronomy
511 at the University of Virginia.