Section 2: CCDS & APS/CMOS Flashcards
types of pixelated semiconductor devices used for digital imaging
CCDs
APS
CMOS (examples of APS)
CCDs readout image by
moving the stored charge across the image
both CCDs and APS are subject to
uneven response, thermal noise, hot pixels, electronic noise, cosmic ray hits…
CCDs/CMOS work via the
photoelectric effect in a semiconductor - electron energised by photon
semiconductors: in an isolated atom,the atomic energy levels are
well spaced out
semiconductors: in solids, atomic levels form
‘blended’ bands - the low energy bands are filled by electrons, up to fermi level
semiconductors: the last filled level is the
valence band
band gap
between valence and conduction bands
typically a few eV
for electrical conduction, electrons must be able to
move between energy levels (so the full bands cannot participate)
in conductors, the valence band
is not full and overlaps with conduction bands
electron conduction arises when
an electron moves from the valence band into a higher energy state
get an electron in conduction band and hole in valence band
how can a valence electron gain energy and jump the band gap?
- thermally (don’t like)
- photon transferring energy (ie photoelectric effect)
semiconductors: photons will produce
electron-hole pairs
doped semiconductors
adding a small amount of a different atom with more/less valence electrons
doped semiconductors are used to
improve condition and help store the charges
p type doping
eg boron into silicon
fewer valence electrons so extra hole above valence band
n type doping
eg arsenic into silicon
more valence electrons so extra electrons below conduction bands
can use p or n-type
individually or combined in a p-n junction to store the charge produced by the photons
p-type silicon held under a small bias voltage - setup
a thin insulator layer of SiO2 is placed onto the p-type Si and on top of this is a metal electrode (known as surface channel device)
p-type silicon held under a small bias voltage - depletion region
the positive holes are driven away from the small positive bias voltage near the electrode
electrons migrate to near electrode
electron hole pairs created by photons will have the electron stored
in the potential well near the electrode, at the top of p-type Si, below the SiO2
stored charge is directly proportional to the
number of photons falling on it (in IR/optical regime)
full well capacity
max no of electrons that can be held before pixel saturates
CCD readout by
applying sequences of voltages along the columns and down the rows of the CCD, transferring charge from one pixel to the next
charge transfer efficiency describes
the fraction of charge transferred per pixel with the semiconductor, typically >0.999
very little lost - very efficient
for speed, each column in the CCD is
shifted at the same time
(must be precisely timed)
analogue to digital converter
converts voltage to data numbers (DN)
CCD readout process
- move to next column
- last column shifts down
- readout pixel goes through amplifier, ADC and then into memory
some devices read out all rows simultaneously, in this case…
each row has its own amplifier and ADC
3 phase readout scheme
each pixel has 3 electrodes, connected in parallel at voltages Φ1, Φ2, Φ3
voltage is varied, allowing charge to migrate but also be kept separate
3-phasesteps of the 3-phase process
- charge stored in middle of each pixel below Φ2 (Φ1=0,Φ2=+V, Φ3=0)
- charge moves to right to below Φ3 (Φ1=0,Φ2=+Vdown, Φ3=+V up)
- charge moves to right to below Φ1 i next pixel (Φ1=+V up,Φ2=0, Φ3=+Vdown)
- read out rightmost column
contributions to the CCD DNs that must be dealt with in processing
Pij, Tij, Eij and Cij
Pij
contribution from pixel charge due to photons from the astronomical source
Tij
contribution from pixel charge due to thermal effects (also called dark current)
Eij
contribution from readout process
Cij
contribution from pixel charge produced by cosmic ray hits
Xij=
Pij+Tij+Eij+Cij
thermal noise Tij arises from
thermal energy in the CCD material, leading to lattice vibrations, called phonons.
Energy of these vibrations can create electron-hole pairs in the absence of illumination.
what are dark frames
exposures with no illumination falling on the CCD
dark frames are exposed for long enough to capture thermal patterns
electronic noise can arise in
-Transfer of charge from pixel to pixel
– Amplification of readout voltage - need low-noise amplifier
– Measurement of amplified voltage
an average readout noise is often quoted as
σ0=the standard deviation of Eij
eg: σ0= 10 electrons RMS
(refers to the RMS readout noise in units of electrons per pixel)
quantisation noise
Conversion of the analogue voltage into a digital signal (DN) also
introduces an additional error term: “quantisation noise”
bias frames
exposures of zero duration, without light falling on the CCD, which capture the various sources of 𝑬𝒊j
bias frames also provide
the pixel to pixel structure in the electronic noise
Typically, a series of bias frames is acquired and averaged to reduce
the SNR on the bias frame values.
As the pixels are readout, the stored charged is converted to a
voltage, producing a time varying analogue signal often measured relative to an arbitrary reset voltage
why is the analogue voltage is often measured relative to an arbitrary reset voltage?
to help with noise reduction
“correlated double sampling”
the arbitrary reset voltage can result in negatives values due to noise in the reset level (electronics) and pixel signal (i.e. noisy in low light). So…
make sure the signal is always >0 (which the ADC needs),add a fixed offset voltage, the “bias level”, to the pixel signal.
the gain characterises
how many photoelectrons there are per integer value in the readout, i.e. the number of e- per DN
(normally chosen so max no of electrons stored = max DN value the electronics can handle
because semiconductor material may vary…
each CCD pixel responds differently to illumination.
Some give a higher number of photo-electrons per photon, some lower.
CCD can also have surface spots or flaws
flat field
By exposing the CCD to a uniform light source we can see CCD response to illumination
this exposure is the flat field
the flat field corrects for
non-uniform CCD response, giving the image that would be produced if every CCD pixel were identical
what needs to be done to flat field to give Fij
normalise (divide by average)
pixel variation corrected photo-electron is
Nij=Pij/Fij
Thermal and electronic noise often cannot be distinguished in the
dark image so
combine into instrumental background Bij where
Bij=Tij+Eij
to calibrate the CCD, we need to first
subtract the background from
the DN, and then adjust for the non-uniform response
Nij
the DN per pixel due to astronomical dources
Nij=(Xij-Bij)/Fij
cosmic rays are
energetic subatomic particles from space
originate from the sun and galactic/extragalactic sources
if a CCD is exposed for a long time or a CCD is in space then
cosmic rays impact it and cause pixels or groups of pixels to saturate
how to deal with cosmic rays - median or mean
median is more meaningful as CR values»_space; normal so mean is misleading
when is combining exposures not useful?
when the source is changing (often in solar astronomy)
CCD quantum efficiency peaks in
the optical but the wavelength response can be broadened into the UV by coatings
when source is changing - what are the solutions
can replace ‘bad’ pixels by spatial filtering
common methods are mean or median filtering
each pixel in an image is replaced by the mean/median value of the immediate surrounding pixels
CCD material (semiconductor and the SiO2 insulating layer is very
reflective shortwards of 400nm
anti-reflection coatings improve quantum efficiency down to
about 350nm
below about 350nm, reponse is further improved by
a coating called Lumogen which absorbs UV and re-emits as optical
Short wavelength photons (UV/EUV: 10-400nm) have a 1/e absorption
depth of
10-100nm
so absorbed before they reach the depletion region
CCD materials can now be made very thin (<10nm) and illuminated from the ‘rear’ side to give
increased sensitivity at wavelengths <400nm
at wavelengths <250nm, one photon can generate
more than one electron hole pair
non-linear response
the limited full well of a CCD pixel limits the
CCD dynamic range and can lead to blooming
dynamic range
ratio between the brightest and faintest sources that can be recorded
blooming
photoelectrons overflow from one potential well to the next along conduction paths (rows)
leads to bright streaks which cannot be corrected other than cosmetically.
An “APS detector” is a detector in which
individual pixels contain the
photosensitive material and an amplifier
each pixel can be read out individually
no moving charge about like in a CCD
most common APS
CMOS
CCD noise
Single amplifier deals with
output from all pixels; high
bandwidth needed, leading
to high noise. But many
years of development of
materials so dark and
readout noise small
CMOS noise
Each pixel has own
amplifier; can be low
bandwidth, leading to
low noise. But suffer
from larger dark and
readout noise (less
heritage than CCD)
CCD vs CMOS nosie
Overall, CCD
historically better for
low noise and hence
preferred for low light,
but CMOS are
catching up
CCD QE
> 90% with appropriate
adaptations, and > 60% over
large range in wavelength
CMOS QE
Best ~60% (due to
additional circuitry on
pixel), lower in the red
CCD vs CMOS qe
CCD preferred for
operation at low light
levels, but CMOS
catching up fast
CCD readout rate
Slower (10ms) - charge must
be stepped across in
sequence; shutter closed
during readout
CMOS readout rate
Fast (ms) - all charge
read out (near)
simultaneously; shutter
can be open during
readout or no shutter
CCD vs CMOS readout rate
Unimportant for most
astronomy (but fast is
good for AO systems
and solar work)
CMOS better
CCD uniformity of response
Can be made large and very
uniform. Single substrate,
single readout process.
CMOS uniformity of response
Can be very nonuniform! Each pixel and its amplifier can be
different.
CCD vs CMOS uniformity of response
CCD preferred when
need larger and more
uniform coverage (i.e.
synoptic surveys).
CMOS need more flatfielding.
CCD dynamic range
Larger full well capacity
(bigger pixels and lower
noise)
CMOS dynamic range
Often smaller pixels,
and higher noise, so
lower full well capacity
CCD vs CMOS dynamic range
CCD preferred for higher
dynamic range, but is
blooming an issue?
CCD blooming
No insulation along rows,
all pixels exposed for
same duration ->
blooming
CMOS blooming
Individual pixels can be
read out, draining
charge before
blooming
CCD vs CMOS blooming
CMOS preferred but
“anti-blooming”
techniques help in CCDs
CCD spectral coverage
CCDs adapted to function
across the spectrum
CMOS spectral coverage
CMOS have poor IR
response and are
harder to thin for UV/Xray (but possible)
CCD vs CMOS spectral coverage
CCD better outside the
optical range. But
APS/CMOS can count
individual high energy
photons
CCD flexibility
On-chip binning and
region-of-interest
selection possible.
CMOS flexibility
Arbitrary groups of
pixels can be read out.
CCD vs CMOS flexibility
CCD readout needs
circuits; CMOS readout
needs software &
computing power
CCD power
High power, to drive
readout voltages and
maintain potential wells.
CMOS power
Low power - amplifiers
only turned on at
readout
CCD vs CMOS power
If power an issue (i.e.
small satellite) then
CMOS preferred.
steps to change the DN into physical units steps
- convert DN to photo-electrons per pixel
- covert photo-electrons per pixel to photons per pixel an CCD illumination
- convert illumination to flux arriving at telescope
to Convert DN to photo-electrons per pixel, need to know
gain (e per DN)
N=n/g +/- sigma0
N=readout nosie, n=no of e-, sgima0=readout noise
total noise (sigma (N)) is a combination of
readout noise (sigma0) and photo-electron counting noise (sigman = sqrt(n)) -poisson
photon-counting noise in electrons
sqrt(Ng)
photon-counting noise in ND
sqrt (Ng)/g = sqrt(N/g)
adding readout and photon-counting in quadrature gives
sigma^2(N) = sigma0^2 +N/g
form of y=mx+c with m=1/g
can calculate the read noise and gain from CCD output by
- Take several flat fields images at different illumination levels
(giving different N) - Divide each flat field image into different sub-areas, and
measure ⟨𝑵⟩, 𝛔(𝑵) (the stdev) in each sub-area - Plot 𝝈^𝟐 (𝑵) vs ⟨𝑵⟩
- Fit with an equation of the form 𝝈^𝟐 (𝑵) = b +k⟨𝑵⟩
- So 𝒃 is gives the readout noise, and 𝒌=𝟏/𝒈
how to estimate the gain and readout noise without having to fit a line to the data
If the 𝝈^𝟐 (𝑵) vs <𝑵> is given as a
log-log plot</𝑵>
The number of photo-electrons produced by a pixel 𝒏𝒊𝒋 is
𝒏𝒊𝒋 = Nij g
quantum efficiency
Q: number of photo-electrons generated per incident photon.
Q=1 = 100% efficiency
number of photons incident at that pixel is
nph = nij/Q
The illumination, 𝑱, of the CCD in the telescope focal plane is
the energy s-1 m-2 arriving at the CCD.
J can be calculated from nph if
the pixel size is known, duration of
exposure, and if we define a mean energy of detected photons.
CCD and instrument will respond differently to
photons of different
wavelengths, described by passband 𝑷(lambda)
passband
probability of detecting a photon of wavelength lambda
mean wavelength of the photons detected
integral of P(λ) s(λ) λ dλ / integral of P(λ) s(λ) dλ
where s(λ) is a normalised source spectrum
For an extended source, covering many pixels, 𝑱 will typically
change from pixel to pixel
equation relating illumination to integrated flux assumes
no losses in the telescope
how to find surface brightness B from equation relating J to F
Dividing by the solid angle 𝛀 subtended by the pixel, and the width ∆𝝂 of the spectral passband
