Previous Topic: CCDs: Detection Limits and Array Equation Lecture Index Next Topic: Reduction of CCD Data

ASTR 5110, Majewski [FALL 2017]. Lecture Notes

ASTR 5110 (Majewski) Lecture Notes



  • Howell, Handbook of CCD Astronomy, Chapter 3 (and other places!).

  • Rieke, Detection of Light, Chapter 6.


Despite the huge advantage that CCDs have over photographic plate detectors, astronomers continue to drive the creation of devices that are even closer to the ideal of a detector with 100% QE.

  • The demands on large telescopes makes this an important research investment:

    Every fractional increase in QE means an equivalent reduction in either light gathering power or integration time (at same light gathering power) needed to do same S/N work.

  • QUESTION: Why do you think astronomers are driving this kind of work and not the television/broadcast industry?

Understanding QE in detail:

  • Peak Q.E.'s for CCDs are from ~40-85% (compared to a few percent for photographic plates).
  • The Q.E.'s of CCDs vary with wavelength.
  • An important source of this variation is the ability of photons of different wavelengths to penetrate the silicon.
    • Imagine an incoming photon stream (flux) at the front surface of a Si CCD chip as
      F(φ) in photons / sec. The flux at some depth z in the silicon is given by:

      where is the Coefficient of Intrinsic Absorption, which describes how easily the photon is absorbed and is a function of temperature, T, and wavelength, λ.
    •   λ (1/μ)
          T=300 K T=77 K
      blue 4000 Å 5.0 4.0
      red 6000 Å 0.5 0.25
      infrared 8000 Å 0.1 0.005
      infrared 10000 Å 0.01 0.002

  • Note:

    • Photons are pretty well "stopped" by about four "scaleheights" (where an exponential scaleheight is given by 1/, also called an "optical depth").

      This means that blue photons are pretty much totally absorbed by ~ 4/ ~ 4/(4μ-1) ~ 1μ
    • On the other hand, a significant number of photons make it past only one scaleheight.

      This means that infrared photons can travel a significant distance, to > 1/, or > 200 μ
    • Note, λ > 11000 Å electrons do not have enough energy to make electron - hole pairs in silicon and pass completely through the device.
    • The plot below shows the overall wavelength dependence of absorption of photons as a function of energy, where in this case the absorption length is defined as the distance for which 1/e of the incoming photons (63%) are absorbed.

    From Howell, Handbook of CCD Astronomy.
  • Thus, based on the above wavelength dependence considerations of the intrinsic absorption, , of silicon, we see how a CCD's blue and red limits of sensitivity are actually defined, apart from the bandgap energy:
    • Sensitivity in the blue is limited by weak penetration of photons (on order of only microns). Many photons are absorbed before reaching the depletion zone. Need to be worried about thinness of SiO2 and other layers to cross. To increase the efficiency in blue, decrease thickness of Si to be crossed.
      • THIN CCDs, "backside illuminated" desired.
    • Sensitivity in red requires substrate thick enough to have enough opportunity to absorb weakly interacting red photons.
      • THICK CCDs best, typically "frontside illuminated".


  • Easier to make (cheap)
  • Thick substrate and surface layers OK for red photons to ~ 11,000 Å
    • low, can travel 500 μ or more -- the thicker the CCD, the more sensitive -- i.e. more electrons intercepted by Si to make electron-hole pairs -- higher Q.E.

    • But note:
      1. Loss of resolution with increasing thickness -- electrons can "get lost" on way to depletion zone.
      2. More substrate = more chances for dark current.
  • Crossing gate and SiO2 layer limits best Q.E. to about 50%.
  • Ways to make a thick CCD sensitive in the blue, and even the UV, involve the applications of substances to the CCD that act as a wavelength converter.
  • Most commonly used substances are fluorescent.
  • How does wavelength conversion take place in fluorescence?

    • Fluorescent materials contain molecules that have a variety of vibrational and rotational modes:

      More about vibrational modes...if you're interested.
    • These modes have associated energy states that are quantized.

    • When a photon is initially absorbed,

      1. can put the molecule into an excited state of higher vibrational energy levels.
      2. Collisions of the molecule with its environment allows some shedding of kinetic energy with no photon emission.
      3. The molecule can ladder down vibrational modes until, if the process takes long enough, can at some point instead undergo a spontaneous emission of a (lower energy) photon.
    • Note a "fluorescent" process involves very quick timescales -- it is "instantaneous" so that the effect stops if electron source removed (e.g., the cool black-lite poster in your friend's dorm room).

        Compare to "phosphorescence", wherein the molecules can absorb a photon and find a metastable intermediate, semi-forbidden state, that only slowly leaks off energy (e.g., "glow in the dark" materials).

  • Polycyclic Aromatic Hydrocarbons (PAHs)
    • The most common materials used to make CCDs more sensitive is a class of molecules containing carbon rings called Polycyclic Aromatic Hydrocarbons (PAHs).

    • PAHS have a number of possible vibrational modes involving the ring nodes.

      • A vibrational mode of a dehydrogenated Pyrene molecule
    • For related reasons to their effectiveness as wavelength translators, PAHs also represent an important molecule found in interstellar dust, where they are commonly seen emitters of infrared radiation.

    • Space telescope CCDs (early) used Coronene - absorbs UV (1000 - 3500 Å ), re-emits at 5500 Å (chips good > 4000 Å).
      • Vibrations of a dehydrated Coronene molecule:
    • Another good substance is Lumigen Phosphorous (Highlighter ink!) -- Absorbs < 4800 Å, re-emits at 5250 Å.
    • "Laser dyes" are most efficient, when different ones are chosen carefully to convert photons to progressively lower energies (longer wavelength). These cascades can reach 20% efficiency, but tricky to apply and some very toxic.
  • Note that these PAH substances are generally very carcinogenic, which makes them unpleasant to use in the lab.


The heart of the CCD camera we use for ASTR 3130 is a thick CCD chip made by Kodak.

The heart of the new ST-8E camera is an enhanced ("E") version of the KAF-1600 detector from Kodak

A few years ago we sent our ST-8 cameras to the camera manufacturer for the chips to be coated for increased blue/UV sensitivity.

Note the improvement in the QE response.

What does this mean for the gain in blue, red, green sensitivity for the same integration time?


(26817 bytes) 2ABG-G.TIF
(26817 bytes) 2ABG-R.TIF
(26817 bytes)




ENH-BL2.tif (26817 bytes) 2ENH-GL.TIF (26817 bytes) 2ENH-RL.TIF (26817 bytes)




(22456 bytes)



Below is the information from the ASTR 313 manual explaining the characteristics of the ST-8 CCD. You should now be able to understand what most of this table means!

Model ST-8 CCD Specifications
CCD Kodak KAF-1600 + TI TC-211
Pixel Array 1530x1020 pixels, 13.8x9.2 mm
Total Pixels 1,500,000
Full Well Capacity 40,000 e-
Dark Current 1 e-/pixel/sec at 0o C
Antiblooming Standard (non ABG as option) -- Limits "bleeding" of bright source photoelectrons into neighboring pixels

Readout Specifications
Shutter Electromechanical
Exposure 0.11 to 3600 seconds, 10 ms resolution
Correlated Double Sampling Yes
A/D Converter 16 bits
A/D Gain 2.3 e-/ADU
Read Noise 15 e- RMS
Binning Modes 1x1, 2x2, 3x3
Pixel Digitization Rate 30 kHz
Full Frame Acquisition under 60 seconds

Optical Specifications (for an 8-inch telescope at F/10, e.g., as in "Night Lab")
Field of View 24x16 arcminutes
Pixel Size 0.9x0.9 arcseconds
Limiting Magnitude Magnitude 14 in 1 second
(for 3 arcsec FWHM stars) Magnitude 18 in 1 minute


  • Because blue photons are absorbed within a few microns it is desirable to avoid the gate / SiO2 MOS layers.
  • Illuminate CCD from behind.
  • But, in order for photoelectrons to be collected efficiently and without resolution losses, best to form in or very near depletion zone -- have to thin CCD.
    • Can get 80% Q.E. if substrate 10 μ thick CCD.
    • Obviously, lose some red sensitivity.
    • Can be hard to thin evenly - need 1 μ accuracy or get large Q.E. variations.
    • Use strong acids to dissolve substrate.
    • Thin CCD fragile - subject to bending as support structure cools and shrinks -- Glue to glass on backside.
    • Hard to make = expensive, but nice Q.E.!
  • After thinning, unavoidable oxidation of Si on backside -- 20 Å wide SiO2 layer that can trap shortest λ UV photons.
    • UV Flood - Use UV lamp to ionize O2 in air, settles negative ions on layer which then drive photoelectrons into substrate. Lasts several months in vacuum.
    • Flash Gate - Attach a fine metal layer with negative bias.


Note how the QE of this chip has been enhanced by thinning and backside illuminating.

Spec Value
Dewar IR Labs
Liquid nitrogen cooled, 1-liter capacity
Heating resistor
Hold time is > 36 hours
Chip SiTE 2048-ENG
Back-illuminated, thinned to enhance blue response
Operating Temp. Unstable about -100o C
Optimal operating temperature is ~ 110o C
Lowest achievable temperature is -134o C
Format 2048 x 2049; 24 μm square pixels
Overscan Default is columns 2049-2100
High Gain Inverse Gain: 1.5 e-/ADU; Read noise: 11.5 e-
Low Gain Inverse Gain: 2.8 e-/ADU; Read noise: 11.5 e-
Field of View 12.5' x 12.5'
1 pixel = 0.365"
CTE 0.99998 - 0.99999
Dark Current A good exercise for the student
Full Well > 150,000 electrons / pixel
Readout 150 s (2.5 min) for entire chip
maximum ADU = 65536
No on-chip binning
Random horizontal shifts occur following shifting by N lines
4 available readouts (A-D), but only one amplifier (A)


  • The Bias Level
  • Most CCDs today have 16 bit amplifiers, so that they can create 216=65536 different possible integer numbers, or greyscale levels.

    To use 16 bits to describe an integer, one can either either use them as:

    • 16-bit signed numbers, meaning one bit is used to describe the sign of the integer, yielding numbers from -32767 to +32768.

      In such a system, once a pixel exceeds +32768, the value recorded for the pixel "wraps around" to -32767 (and then increases).

    • 16-bit unsigned, meaning one uses the extra bit to give another power of 2, at the loss of being able to describe negative levels. This means that they cover the range 0 to 65355.

    Since in a normal CCD picture, one expects any normal pixel to have a zero or positive number of photoelectrons, it doesn't make sense to use a whole bit to allow description of negative integers, and take advantage of the extra bit to increase the recordable dynamic range in the output image.

    • However, due to the readnoise problem, it is possible for a normal CCD pixel to have a (slightly) negative value.

    • Other effects (such as an improperly ground-referenced amplifier due to drifts in the CCD electronics or the actual ground level) can also yield negative output values for normal pixels.

    To properly account for the above problem, but get the substantial dynamic range benefit of 16 bit unsigned output, CCD electronics will add a pedestal level, called the bias level, to shift all pixel levels up into the positive range.

    • This "zero" level is typically a few hundred to ~1000 ADU.

    • The plot below shows the distribution of pixel values in a "bias frame" (a CCD image taken after "wiping" all of the pixels clean of dark and photoelectrons and then instantaneously reading out the pixels after a 0 second exposure.

      From Howell, Handbook of CCD Astronomy.
    • Note that in a well-behaved CCD where all pixels are about equal in their electronic set-up and dark current, one will see a Gaussian distribution of ADU values centered around the "mean bias level" or "mean zero level".

    • Since in principle the pixels are at the nominal "zero" level with no dark or photoelectrons, the width of the Gaussian distribution should reflect the read noise of the CCD amplifier in ADU:

      σADU = RN / G

      In reality, no array is perfect, and the above should be considered as giving an upper limit to the read noise.

  • Drifts in the Readout Amplifier and the Overscan
    • During readout, reference voltage can drift, changes "bias" level -- results in varying "0-point".
    • Solution: Overscanning.

      • If chip has N columns, clock the MUX register N+M "pixels" through amplifier. The M extra reads of the amplifier correspond to no actual pixels on CCD -- i.e. represents pixels with 0 charge.

      • Thus, the "charge level" one reads from these non-existent charge packets tells you what the system is currently putting out as ADU signal for a 0 voltage.

      • After the fact, measure this 0-point level for each CCD row and subtract off the ADU level for each pixel in the row. Do for each row in the image to rectify the wandering 0-point level.

      Example of the overscan variation in a WIYN Mini-Mo CCD image, and a fit to remove the general trend. Courtesy Ricardo Munoz.

    • Overscanning IS done on the Fan Mtn. CCD but is NOT done on the ST-8 CCD at McCormick Observatory.

  • Linearity:

    A great advantage often quoted for CCDs over, e.g., photography, is their linearity. But, strictly speaking, CCDs do exhibit some nonlinearities, though usually (but not always) small.
    • For a given pixel, let:

        F be the incident flux (in photons per second)

        S be the recorded signal level (in ADU)

        Q be the quantum efficiency

        G be the gain

        t be the integration time (in seconds)

      Then, for a strictly linear system:

      In reality, Q is a function of the accumulated charge:

      Where the expression in parentheses is the accumulated charge in time t.

      Thus, we are left with a sensitivity curve much like that we saw earlier for photographic plates, except that the strictly linear part is a larger fraction of the unsaturated regime:

    • In older CCDs one often saw a non-linear "toe" at low count levels due to the first few electrons getting "trapped" in crystal impurities.

      To overcome this, and get into the linear regime, some CCDs were equipped with "pre-flash" mechanisms to "fill" these traps (just as we discussed for photographic work, which has exactly the same problem).

    • In general, one does not need to worry about linearity with good quality CCD cameras, except in the case when the pixels become very full...

      From Howell, Handbook of CCD Astronomy.

      Deviations from non-linear in Sloan Digital Sky Survey CCDs.

  • Blooming:

    For sources that overfill their respective pixels with photoelectrons that overfill the well capacity, the electrons need to go somewhere, and this is usually in adjacent, charge-coupled pixels. The result is called blooming.

    • Example of blooming:

      Blooming visible in the bright stars around M42. The black stripe is unrelated.

    • Additional problems arise with overfull wells having degraded CTE, so that the blooming affects those pixels whose electron packets must pass through the affected pixel to get read out.

    • Some CCDs (like the ST-8), have either or both an anti-blooming barrier and/or a drain between the photosites on the CCD.

      Upper figure from Majewski notes. Lower figure from Howell, Handbook of CCD Astronomy.

    Note that a related problem is that one sees with exposing the entire CCD to too much light so all pixels are saturated with photoelectrons.

    • This used to be a real problem, and could even cause permanent damage to a CCD.

      In the least, it used to be the case that recovery from this situation would take hours, requiring many "wipes" of the charge to remove it all.

    • More modern CCDs are more immune from this problem, though it is still not good to do this.

  • Dynamic Range of a CCD
  • The dynamic range of a CCD is limited by its maximum useful level. There are several considerations for defining this:

    • Obviously, the full well capacity is an ultimate limit for a pixel.

      If the full-well capacity of the chip is FWC electrons, then the largest value that can be recorded is, in principle, FWC / G, unless ...

    • ... this is higher than 2b, where b is the number of bits in an unsigned b-bit amplifier (or 2b-1 if using signed output).

      Here the dynamic range is limited by the amplifier.

    • A more practical limit than the FWC is probably where the chip becomes non-linear in ADU, if this is less than 2b.

    • Note that if there is a bias level, this shift in ADO must be taken into account.

    Adopting the definition used in acoustics, the dynamic range, D, of a CCD is given in decibels by:

    D(db) = 20 log10 ( maximum level / RN )

    • Each 3dB is a doubling in range.

    • A CCD with 100,000 e-=FWC and RN = 10 e- has an 80 dB dynamic range.

  • Cosmic Rays
  • Perhaps the most insidious problem one deals with when using CCDs are cosmic rays.

    • High energy particles of unknown origin (probably supernovae and solar flares) that interact with silicon and produce many electron-hole pairs -- sharp, star-like points at random places in image. See this site if interested in knowing more.
    • Set of zero second exposures (bias frames) showing different cosmic ray patterns. Note that the sizes of the blemishes depend on the strength of the CR, its angle of approach, and whether it hits near the edges or centers of pixels.

    • If CR comes in at a small angle to CCD, i.e. nearly parallel to surface, can leave a streak of affected pixels.
    • Close up of a Hubble Space Telescope CCD image showing a large number of cosmic ray streaks. Photo courtesy of Dominion Astrophysical Observatory, HIA.

    • Higher CR incidence at higher altitudes (less atmospheric attenuation).
      • Space telescope - severe (see image above).
      • Take "CR-splits" = multiple images of same scene, different CR pattern, "median stack" the frames (see next page).
      • QUESTION: Why might we expect more CR "streaks" in HST pictures and more compact, point-like CR events on ground-based CCD images?

    • Thicker CCDs have higher probability of intercepting CRs.
    • "CR-like" events can also come from surroundings, like radioactive glass and coatings in telescope/camera system. At McDonald Observatory, there is reputed to be a large uranium deposit under the mountain that contributes a high radioactive decay event rate.

  • Cosmetic Defects

    Take various forms:
    • Dead pixel - Pixel unresponsive to light due to defective gate, depletion zone, substrate, insulator, etc.
    • Hot pixel - Pixel with much larger dark current than neighbors.
      • Note that hot pixels can look like cosmic rays, but, unlike cosmic rays, hot pixels always produce blemishes in the same place from frame to frame.

      • If the hot pixels have a stable dark current rate you can subtract out their effects by taking a dark current frame of equal integration time:

        Raw frame

        Dark subtracted frame

    • Bad column - A defective pixel where the defect affects CTE and all charge packets that pass through the defective pixel will be destroyed (e.g., fall into a trap) resulting in a bad column in the final image.
    • CCD image of galaxy NGC4565 taken with UVa's Fan Mtn. Observatory SITE 2048 CCD showing several bad columns.

    • Note, one of the limitations on making large CCDs is the frequency with which defective pixels occur -- both a technical and economics problem.

      • If bad pixel occurs in MUX - disaster.
      • So, if chance of getting a bad pixel is 1/10000, say, then one of every five CCDs of size 2048x2048 will have a bad pixel in the MUX and is garbage.

      • Extremely hard to get perfect chip, and cost of chip goes something like (1 / # defects)

  • Electroluminescence
  • In some cases, diodes in the output amplifier can actually act as Light Emitting Diodes (LEDs) and can cause serious problems of excess light near the amplifier.

    • Solutions:

      • Turn off power to amp until needed for readout -- LED only on during readout.

        Example of electroluminescence caused by the output amplifier.
        Electroluminescence caused by the individual output amplifiers across an 8x8 Orthogonal Transfer Array, installed in the WIYN One Degree Imager (ODI). From the WIYN One Degree Imager User Manual, version 4.1.

      • Treat as "dark current" and subtract with "dark" exposure (if not strong LED that saturates pixels or that adds LOTS of Poisson noise even after mean level subtracted off).
    • Generally a defective design, and generally rarely seen now.

  • Other "Defects" in CCD Images Not Related to Chip Itself
    • Dust on nearby optical surfaces, including the CCD chip, the dewar window, and the filters will create diffraction "donuts".

      • The closer the dust to the CCD, the smaller the donut (thus allowing diagnosis of where the dust is).

        From Howell, Handbook of CCD Astronomy.

      • Generally we don't worry too much about this, because the effects can generally be removed through flat-fielding.

    • Interference from surrounding electronics - most commonly 60 Hz noise - get diagonal stripes across image (from readout + 60 Hz combo).
    • 60 Hz noise in a bias image due to interference from improperly shielded electronics cables near the CCD.
      (Image from Apache Point Observatory.)

    • Sky pollution
      • Satellites, airplanes

      A good example of sky pollution. During the observation of the galaxy pair NGC 7443-4, an airliner left traces of its signal lights on the CCD even though the field covered by this image is only about 13 arcmin. Observation was made with a 280 mm telescope and a TH7863 CCD. The integration time is 600 seconds, but the duration of the passage of the airliner was only a fraction of a second. (If the CCD image is in the red part of the spectrum, can you tell which way the airplane was moving??)

    • "Cosmic pollution"
      • Meteors, asteroids, scattered moonlight/starlight on chip dewar parts

      Scattered light streaking across a CCD image. In this case a bright star landed on the CCD chip carrier and is reflecting onto the silicon.

    • Fringing
    • (see Rieke Section 6.6.2)

      • Recall thinned CCDs:

        • Created to increase blue sensitivity and backside illuminated. e.g. RCA, TI CCDs.
        • To reduce potato chip ripple, often bonded to a glass plate with 1 μ of glue
        • This glass, if illuminated by monochromatic light, creates fringe patterns due to interference of in-phase and out of phase reflections in glass.
        • Same thing can happen off the back and front of the detector material.

        • Small variations (e.g., 0.1 microns) in the thickness from place to place horizontally changes the interference from constructive to destructive.

          Thus, large scale variations in thickness will yield "banding" patterns across the chip.

        • QUESTION: Why doesn't this happen for non-monochromatic light?

      • In what astronomical applications is this a problem?

        • Imaging with narrow wavelength filters -- both sky and sources appear as "monochromatic" to CCD.
        • Using a CCD for recording spectroscopy -- different parts of the CCD sees different wavelengths of "monochromatic" light.

          Can seriously complicate the analysis of spectroscopic data.
        • Imaging with broad band filters that include wavelengths where the night sky emits auroral emission lines. This is particularly problematical because the auroral emission varies in intensity over the timescale of minutes!
        • Important optical wavelengths where the atmosphere emits strong auroral lines and the Johnson-Cousin passbands in which they occur:

          5577 Å [OI] V band airglow
          5890,5896 Å NaI V band airglow
          6300 Å
          6364 Å
          [OI] R band airglow
          >5900 Å O,OH lines R,I bands airglow

        • Emission lines in source object (e.g., planetary nebula).
      • Removal

        • Fringing is an additive effect of sky -- tricky to remove.

          • Will not appear in domeflats typically.

          • Will not appear strongly in twilight flats.

          • Can only really see in night time images of "blank" sky.

          • Have to generate a "fringe frame" and then subtract scaled versions of this (note, adds noise).

        • "Prevention is the best cure"
        • Nowadays AR coat glass -- fringes minimized
        • Not generally a problem in thick chips
      • Example of fringing pattern seen in a thinned RCA CCD:

  • Radiation Damage in Space CCDs
  • (See Howell, Section 7.2.)

    The harsh radiation environment in space can temporarily or permanently degrade the performance (e.g., the CTE) of a CCD.

    Was known in military applications, but was found in first space-used CCDs (HST WF/PC and Galileo spacecraft to Jupiter).

    Sources of radiation damage include:

    • Solar wind, solar flares and the general background of cosmic high energy particles in unprotected environment away from Earth.

    • The South Atlantic Anomaly (SAA) is the point on the Earth's surface where its inner van Allen belt comes closest.

      • Caused by fact that the center of Earth's magnetic field is offset by 450 km from its geographic center.

      • Thus, at any altitude, the radiation intensity is higher here.

      • SAA yields 2,000 protons of 50-100 MeV per cm2 per second.

    • Radioisotope thermal electric generators that power spacecraft.

    • Planetary radiation belts (e.g., Jupiter's belts hitting Galileo).

    Two major concerns:

    • High energy photon (gamma ray) interactions creating fast electrons which in turn cause simple, localized damage and generation of numerous electron-hole pairs.

      • Causes a charge build-up in the CCD gate structures that increase dark current (by creating new levels in the band gap that can be filled by thermal electrons).

    • Massive particle nuclear reactions from neutrons or high energy photons, which tend to create large area defects by displacing silicon atoms from their lattice positions.

      • The latter effect creates trapping locations, which create (generally permanent) CTE problems.

    CTE degradation is a major issue now affecting Hubble Space Telescope cameras, with consequences for:

    • Source astrometry (with CTE varying with time, the shapes of sources -- especially stellar -- are varying with time and magnitude). Creates false proper motion signatures, for example.

    • Stellar photometry, when that photometry is done with PSF-fitting.

    Previous Topic: CCDs: Detection Limits and Array Equation Lecture Index Next Topic: Reduction of CCD Data

    All animated images from Images of KAF-1600 detector, QE curve for KAF detectors, and spectral response through different filters from Sloan linearity, hot pixel and M42 blooming images from 60 Hz noise image from All other material copyright © 2002,2009,2015,2017 Steven R. Majewski. All rights reserved. These notes are intended for the private, noncommercial use of students enrolled in Astronomy 5110 at the University of Virginia.