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 represent the third of the three most important technical innovations for observational UVOIR astronomy in the second half of the 20^th century (the others being space telescopes and interactive computing). They are now nearly ubiquitous in observations made at wavelengths between the near-IR (~1 µ) and the X-ray.

Sensitive 2-D area detectors like CCD's are critical not just for astronomical imaging and photometry but for many other state-of-the-art applications, including high precision astrometry with Gaia and large-scale, multi-object spectroscopic surveys (such as the Sloan Digital Sky Survey). They have transformed the way astronomy is done.

I. REFERENCES FOR CCD'S

General

• Buil, CCD Astronomy
  
• Howell, Handbook of CCD Astronomy
  
• Kitchin, Astrophysical Techniques
  
• Rieke, Detection of Light : general introduction to detector physics
  
• LLM Sec. 7.3.3.
  
• MacKay, Ann Rev Astr Ap, 24, 255, 1986 "CCDs in Astronomy" (review)
  
• Timothy, PASP, 95, 810, 1983: discussion of many detector types

CCD Data Reduction and Applications

• Abbott, "In Situ CCD Testing"
  
• Massey, A User's Guide to CCD Reductions with IRAF (1997)
  
• Craig & Chambers (Astropy), CCD Data Reduction Guide (Python)
  
• Tyson, JOSAA, 3, 2131, 1986: high precision technique
  
• Hook & Fruchter, "Dithering, Sampling and Image Reconstruction"; also see the HST DrizzlePak Handbook.
  
• Stetson, PASP, 106, 250, 1994 and references therein. Widely-used ALLFRAME and DAOPHOT point-source CCD photometry programs.
  
• Franx et al. AJ, 98, 538, 1989: sample application to surface photometry
  
• Bertin & Arnouts, A&AS, 317, 393, 1996 SExtractor---source ID and extraction software for digital images. Obtain software here.

II. GENERAL DETECTOR CHARACTERIZATION

QUANTUM EFFICIENCY QE = percentage of photons incident on detector that produce measurable signals Strong wavelength dependence (e.g. threshold cutoffs set by work function/band gap) Typical peak values: Eye: 1-2% Photographic plate (Pg): 1-2% Photomultiplier tube (PMT): 20-30% CCD: 70-90% IR array [HgCdTe]: 30-50%

Schematic QE curves for various classes of detector

• "Detective quantum efficiency" is defined to be [(SNR_out)/(SNR_in)]², where SNR is the signal-to-noise ratio and "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 used.

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 the 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 increase with the signal

NOISE CHARACTERISTICS

See discussion of various types of detector noise in Lecture 7 notes

RESOLUTION

• Temporal
  • 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 an image, we need at least 2 pixels per resolution cell, but in practice there may be many more. A lower pixel density implies "undersampled" imaging. A significantly higher pixel density does not provide more information and is a waste of pixels.
    For best pixel matching in 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.

• Spectral
  • Most UVOIR detectors are broad-band and have inherently poor spectral resolution. One must use external elements (filters, gratings, photonics) to provide spectral resolution

  • 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 ASTRONOMY:

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 most applications in astronomy from around 1900 to 1950 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:
1. Very long exposure times (up to a week in early applications), meaning limiting fluxes thousands of times fainter
   
2. A permanent, objective record of astronomical images and spectra
   
3. Geometric stability -- important for astrometry and morphology
   
4. Large formats for wide field surveys
   
5. Extension of observations to EM bands beyond the "visual" band. The classic "photographic" band is centered around 4500 Å. Astronomical photography ultimately covered the range 3200-10000 Å.

However, the photographic emulsion had serious limitations with which astronomers had long struggled:
1. It is relatively insensitive, with QE ~ 1%.
   
2. Pg materials have non-negligible thresholds (a minimum of ~4 photoelectrons per grain). "Reciprocity failure" refers to the fact that the threshold exposure must be delivered within a certain critical time period or the photoelectrons on a grain will be lost.
   
3. Pg materials have a strongly non-linear response function, and limited dynamic range.
   
4. These properties plus the use of hard-to-control chemical processes for emulsion deposition & development makes them very difficult to calibrate quantitatively.
   
5. Each individual plate (or at the best, each co-processed "batch" of plates) has different response characteristics.
   
6. Conversion of data to quantitative form (e.g. with microdensitometers) is slow and cumbersome.

For details on astronomical photography at low light levels, see Smith & Hoag 1979, ARAA, 17, 43.

PHOTOELECTRIC ASTRONOMY TO 1980:

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 1920's. 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 ~10^6-7 electrons and deposits this on an anode. (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, where the average induced current was measured, 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 sources.
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 after 1950, e.g. the Johnson & Morgan broad-band UBV system (see Lecture 14 for details). They featured excellent linearity and stability, which implied an unprecedented capacity for accurate calibration of photometric measurements. PMT filter 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 efforts such as the 33-channel Palomar Multichannel Photoelectric Spectrometer (Oke, 1969).
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 electronographic film camera, 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 50 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.
The image at the right shows a wafer containing a number of co-fabricated CCDs and other devices.
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. CCD's were invented at Bell Labs in 1969 by G. Smith and W. Boyle (Nobel Prize, 2009). 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.

IV. SEMICONDUCTORS

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.

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", E_g.

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 intermediate gaps.

Absorption of a photon can push a valence electron into the conduction band and produce a potential electrical signal. The photon energy must exceed E_g, which implies that there is a maximum wavelength for excitation given by:

Lam_max = 12,400 Å/E_g(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:

Material	Eg(eV)	Lammax (Å)
Pure Si	1.1	11,300
GaAs	1.43	8,670
InSb	0.36	34,400
Hg1-xCdxTe	1.55x	8,000/x

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" semiconductor. 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. 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 doping p-type doping Doping affects energy levels

Photons are primarily absorbed by electrons in the valence band. For photon energies above E_g, 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.

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. 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 layer of silicon-oxide has been grown. P-silicon can be made to have very few free electrons ("high resistivity") before exposure to light; this is important for best performance. A set of thin conducting electrodes of semitransparent "polysilicon" are applied.

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 10 µ thick.

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 the 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 of the CCD is a column of "serial shift" electronics and an output amplifier. There is only one amplifier in the design shown. 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 along the rows toward the serial shift column. 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 electrons. Each parallel transfer places the contents of one pixel from 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.

"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

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 professional-grade chips.

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 "Frontside-Illuminated" chips.
  Mitigation:
  
  • Use special thin, polysilicon material for electrodes; but these 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 a ~10 µ deep unit and illuminate from backside; greatly improves blue response. The technique was pioneered by Mike Lesser at the University of Arizona.

  Difficulties with thinned chips:
  • Fragile
    
  • 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). Produce strong spatial modulation of response or "Fringing". Especially serious for terrestrial night sky emission lines. Can be well calibrated for narrow-band filters or for broad-band filters. Harder for intermediate band filters.
    
  • Thinning reduces red response. For good response at 5000-10000 Å, thick (~500 µ) front-illuminated chips are preferred.

• CHARGE TRANSFER EFFICIENCY (CTE)
  • CTE can be better than 99.999% per transfer from one pixel to the next, but it has to be, since the throughput of a chip with 2048 required shifts = CTE²⁰⁴⁸.

  • It is possible to mitigate the effects of CTE reduction by revising the technical design of a CCD (changing the number of phases or clocking cycles; adding "minichannels") or operationally by adding a pedestal background signal (see below).

VII. ADVANTAGES OF CCD DETECTORS

• HIGH QUANTUM EFFICIENCY: Reaching peak values of 80-90% in the optical band. Much effort was expended to reach these high levels. Sample response curves for three CCD's from ESO are shown below right.
  
  • 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 detections requires larger telescopes rather than better detectors. This was an important factor in the post-1980 rush to build ever larger telescopes.
    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.

• LINEAR RESPONSE across the whole range of operation until an exposure approaches full-well capacity (typically 200,000 e/pixel). This allows much improved flux calibrations and much higher precision for flux measurements at all levels than other 2-D systems.

• EXCELLENT DYNAMIC RANGE: Typically 10⁴.

• WIDE WAVELENGTH RESPONSE: Intrinsically sensitive from X-ray to ~1 µ. Arrays based on other materials (e.g. InSb₂, HgCdTe) with similar architectures are usable in the IR.

• GEOMETRICALLY STABLE: excellent for astrometry

• ONLY LOW VOLTAGES REQUIRED (~5-15 v)

• RELATIVELY CHEAP, AVAILABLE, SIMPLE compared to other digital 2-D systems. Standard chips cost ~$2-200 K.

• RELATIVELY LOW NOISE compared to many other classes of astronomical detectors, e.g. Pg plates, Reticons, SEC vidicons, etc. Dark current is largely suppressed by cooling. But noise is not negligible. Typical equivalent read noise now is 2-10 e/pixel,

• SMALL PIXELS: typically 10-30 µ. Usually an advantage in devising instrument/detector combinations, but designers should match pixel size to 1/2 of smallest resolvable picture element in the optical system.

• SPECIAL FORMAT/READOUT DESIGNS: By changing electrode structure and clocking cycles, one can arrange for many different integrate/readout modes.
  Rapid clocking/video: inherent in CCDs intended for TV applications and useful in some astronomical applications (e.g. speckle interferometry). For bright sources, readout rates of 100 MHz are possible.
  Drift Scanning:
  The basic drift-scanning technique (Schneider et al. 1994) is to clock the chip slowly along its parallel registers while moving the telescope to keep a target centered on the moving electron cloud as it builds up. The chip is continuously read out 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 parallel registers 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 it out to obtain much better sky subtraction (e.g. in the near-IR where atmospheric OH emission is very bright and variable).
  On-chip binning is possible by changing clocking to combine electrons from several pixels together before reading out. E.g. combine a 2x2 pixel region on the chip into a single output pixel. This suppresses the effect of amplifier readout noise on each pixel in the final data and also reduces memory and storage requirements. It adds considerable practical flexibility to CCD systems. Obviously, however, binning reduces the resolution of the output image. Binning 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.

• UBIQUITY: CCDs and other solid-state array devices are now almost universally used in astronomy (amateur & pro). Photographic materials and older electronic detectors have been phased out, except for specialized applications.

• 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 that 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 near-real-time assessment of data and hence 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 are less than 7 cm square. They cover only a small fraction of the high quality imaging field on typical modern telescopes. Device size is limited by the small fabrication yield of large, defect-free chips. The largest routinely available chips are 4096x2048.
  • However, mosaicking technology is now well developed. Typical mosaics are now constructed from 4096x2048 chips.

  • The largest CCD mosaic camera, containing 201 CCD's, is being built for the Large Synoptic Survey Telescope

  • Other 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 is necessary. Thermoelectric coolers are usually not adequate, so cryogens (e.g. liquid N₂) are required. Introduces many practical complications. Chips with excellent low-T performance differ from commercial types used at room temperature in digital cameras, cellphones, etc.

• READOUT NOISE: Produced by variations in amplifier gain. There has been much effort to reduce readout noise, which is now typically equivalent to 2-10 e/pix.
  Even at these very low levels, RON can compromise some types of observations (spectroscopy, surface photometry), see Lecture 7. 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.
  RON effects in some applications can be reduced by 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, rather than CCD's.

• FULL WELL CONSTRAINTS: Bright sources which over-fill pixels can produce "blooming" or "bleeding" along columns, making parts of the chip useless even far from the source. The best solution is operational: e.g. by simply excluding overbright sources or by breaking a long exposure into a series of shorter ones such that saturation is avoided -- and then combining/stacking the result.

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 special exposures ("dome flats" or "twilight flats"). Special observing procedures to suppress flat field effects include: 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 (see below). These methods result in reduced field of view because not all parts of the original field will have uniform exposures. FRINGING: see above. Caused by interference effects within chip. SENSITIVITY TO COSMIC RAYS: High energy particles produce strong, localized signals in CCDs, especially in thick chips. Their effects must be removed during processing. CR's are a major problem for spacecraft detectors (e.g. on HST). Mitigation requires that exposures be broken into multiple parts (called "CRSPLITs" on HST) so that CR events can be identified. CR hits can be removed from processed 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 X-rays from some materials in the vicinity of CCD's, e.g. special glasses used for windows, can produce a diffuse background which adds noise to observations.

CCD Flatfield Frame (AAO)       CR's on HST/WFPC2 2400 sec exposure

• CHARGE TRANSFER EFFICIENCY: Because of "traps" within the CCD's, good CTE is often possible only for signal levels above a threshold of ~10-50 e/pix. For good efficiency at low light levels, this requires adding a "fat zero" (typically 1000 e/pix) electronically or by "preflash" from a diffuse light source. Unfortunately, this also creates added noise (10-30 e/pix). Long-term exposure of CCD's to radiation in space on missions like HST systematically increases the trap density, meaning that CTE reduction is a serious problem there
  .
• AVAILABILITY HOSTAGE TO COMMERCIAL MARKET: CCD availability has always been driven more by commercial and military applications than science. Scientific CCD manufacture represents only a few percent of the overall $18B array detector market. A serious issue, since "active pixel sensors" (e.g. CMOS devices), a different technology with amplifiers incorporated in each pixel, now dominate the larger imaging array industry. Requirements for good performance at low (astronomical) light levels are considerably different than for room-temperature, short-exposure, mass-produced equipment.

• DATA GLUT! CCDs typically produce 2 x N_pix bytes of data per readout, where N_pix 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 solid-state storage are now much improved, but long-term stability of the massive amounts of data now being produced by array detectors is a non-trivial issue. Cf. the LSST.

IX. EXAMPLE CCD SYSTEMS

• HST WIDE FIELD PLANETARY CAMERA 2 (1992)
  • 4 CCD's, optically mosaicked with beam-splitter
    
  • Loral 800x800, 15 µ pixel chips
    
  • Front-illuminated, lumogen phosphor coating for UV response
    
  • RON 5 e/px; QE (6000 Å) 35%

• SLOAN DIGITAL SKY SURVEY (1997)
  • Drift scanning mosaic (not contiguous) containing 54 chips
    
  • Tektronix/SITe 2048x2048, 24 µ pixel chips in 5x6 array; for simultaneous broad-band photometry. Both frontside and thinned backside with AR coating. QE (6000 Å) ~60%; RON < 5 e/px
    
  • Tektronix/SITe 2048x400, 24 µ pixel chips; for astrometry & focus check. Frontside illuminated, RON < 15 e/px.

• HST WIDE FIELD CAMERA 3 (2001)
  • 2 CCD's, physically mosaicked, contiguous
    
  • Marconi 4096x2048, 15 µ pixel chips
    
  • Thinned, back-illuminated; anti-reflection coating for enhanced blue/UV response
    
  • Minichannel for improved CTE
    
  • RON 3 e/px; QE (6000 Å) 70%

• FAN MOUNTAIN CCD SYSTEM (Gen II)

X. CCD HIGH PRECISION CALIBRATION PROCEDURE

A. DATA REQUIRED

• Bias frames
  
• Dark frames (optional)
  
• Flat field frames
  
• Sky flat/fringe frames (optional)
  
• Flux calibrator fields
  
• Target frames

B. REDUCTION AND DATA-TAKING PROCEDURES

• SUBTRACT BIAS FRAME: Bias frames (zero exposure time) are taken with the chip unexposed to light. These measure the electronic pedestal of the chip. For high precision, average many bias frames, before and after observations. Check for bias drift during the night. With some chips, you can determine the electronic bias level from the "overscan" region on the chip.
  • 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: This removes the residual pixel-to-pixel variations (typical 5%). Make exposures of the 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 because of color sensitive effects.
  [Note: photon statistics must yield a SNR significantly in excess of the final desired SNR -- i.e. if you want SNR = 100, the net flat field image must contain well over 10,000 photons per pixel.]

  • Optional: SUBTRACT SKY FLAT/FRINGE FRAME: Remove night sky emission line fringing effects (worse in near-IR) by observing an uncrowded field in the night sky. Take several exposures, moving telescope by, say, 25 arc-sec between them. Use the "adaptive modal filter" technique to zap star images and create a mean sky frame. Scale to target frames and subtract. Resulting flat field as good as 0.1% can result. A sky flat determined this way is needed for each filter.

• TARGET FRAMES: USE MULTIPLE OFFSETS, DITHERING OR DRIFT SCANNING: For faintest possible photometry, use the multiple-offset exposure technique -- e.g. 500 sec exposures shifted by about 20 arc-sec each -- to reduce flat field errors and identify cosmic rays. You always want more than one exposure anyway for "reality checks" and empirical determination of photometric errors and transient effects.

• REGISTER AND COMBINE TARGET FRAMES: Re-register the offset frames to sub-pixel accuracy (using, e.g., "Drizzle" softare), creating an image "stack."

• REMOVE COSMIC RAY EVENTS: Median filter the signals in each pixel in the stack. Outlying signals caused by cosmic rays and other transients are thereby rejected from the combined image. You must track the remaining "good" exposure time for each pixel.

• COSMETIC TREATMENT: trim off the "lost" parts of the image. Depending on the application, you can interpolate over or mask out bad pixels.

• 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 (e.g. variable clouds), but no flux calibration for those, either. Calibrator stars should overlap targets of interest in color. These frames are reduced exactly like target frames, apparent fluxes extracted, and the multiplicative factors needed to convert to 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

XI. EXAMPLE SCIENTIFIC APPLICATIONS OF HIGH PERFORMANCE CCDs

• Deep imaging using subpixel registration. This example employs dithered HST imaging and A. Fruchter's "drizzling" technique. See the image pair below. On the left is one original frame (I-band, 2400 sec exposure). Most of the "features" in this image are cosmic ray tracks, not cosmic objects. On the right is the result of drizzled processing of 12 such frames. The combined image has a limiting magnitude of I ~ 28. It would be impossible to reach such levels with photographic plates or other less-stable and calibrateable systems..

• Color-magnitude diagrams. An example of the improvement offered by CCD's over photographic photometry in a CMD for the globular star cluster 47 Tuc is shown below. On the left is a 1977 state-of-the-art CMD based on widefield photographic images with photoelectric calibrations (Hesser & Hartwick, 1977). On the right is a 1987 CMD derived from 4-m 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, and the shape of the subgiant and asymptotic giant branches. (Later, yet higher precision, CCD imaging with HST in the near-ultraviolet bands revealed the existence of multiple populations with similar ages but different metal and He abundances in many globular clusters.)

  Photographic CMD

  CCD CMD

• High precision stellar photometry with Kepler. The Kepler spacecraft searched for exoplanet transits of 10-16 magnitude stars using a 42-chip CCD mosaic camera. To reach exoplanets of Earth size, the transit method requires extremely high precision photometry, which is possible in space given the absence of atmospheric interference. Kepler reached a photometric precision of ~30 parts per million at 12th magnitude --- the best ever achieved in astronomy to that time. A sample light curve is shown below (note the flux scale):
  
• Multi-object spectroscopy. CCD's are widely used as detectors in spectrographs. They were the basis for the flourishing of multi-object spectroscopy, which now extends to the simultaneous observation of hundreds of sources in a given field. Examples of these powerful systems are described in Lecture 10. A sample stack of 46,000 QSO spectra from the first Sloan Digital Sky Survey is shown below. It would be wholly infeasible to collect such large samples of astronomical spectra without the powerful capabilities of area detectors like CCDs.
  
• High precision astrometry with Gaia. The Gaia space astrometry mission employs a focal plane with 106 specially-designed CCDs fed by two telescopes with different lines of sight and operated in drift-scan mode. To reach its intended limit of ~20 mag, it uses the full 3300-10500 Å band available to the CCDs. It reads data only for selected targets in each field, and each of these is identified by on-board software from a special set of pre-scan mapping CCDs. Gaia achieves an unprecedented precision of about 30 micro-arc-seconds for parallaxes of stars of 15th magnitude.

XII. 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 versions.
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 ~ E_XR/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 offers an intrinsic 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 infrared wavelengths because of their large band-gaps, 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".
Here is a cross-section of the HST/NICMOS infrared detector, which uses a HgCdTe ("mercad-telluride") photon detector.

---

---

ADDITIONAL WEB LINKS

CCD-World (forum for discussions about CCDs and their use in astronomy)

Basics:

Digital Imaging/CCD Tutorial at Molecular Expressions
        (thorough, elaborate illustrations, but aimed at microscopy)
Britney's Guide to Semiconductor Physics (decorative tutorial)
Semiconductor physics at HyperPhysics (succinct, with many illustrations)
Simon Tulloch Primers on Astronomical Detectors (aimed at advanced amateurs)

Example Systems & Development Efforts:

Sloan Digital Sky Survey CCD Camera
NOIRLab CCD Mosaic Camera Manual
NOIRLab Dark Energy Camera User Guide
ESO Optical Detector Homepage
Status of CCD Development (Steve Holland, LBNL, 2012)
JPL CCD ``Delta-Doping'' Project (for high UV QE)
Scientific CMOS Sensors in Astronomy (Alarcon et al. 2023)

---

Previous Lecture

Lecture Index

Next Lecture

---

Last modified April 2023 by rwo

---

Bandgap images from Britney's Guide to Semiconductors. MOS capacitor illustration from Molecular Expressions. Bucket brigade animation and front/back illumination drawing by C. Tremonti (UWisc). CCD design drawings from C. MacKay, Annual Reviews (1986). Most other images from public observatory sites. Text copyright © 1989-2023 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.