Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
• In photon detectors, the light interacts with the detector material to produce free charge carriers photon-by-photon. The resulting miniscule electrical currents are amplified to yield a usable electronic signal. The detector material is typically some form of semiconductor, in which energies around an electron volt suffice to free charge carriers; this energy threshold can be adjusted from ~ 0.01 eV to 10eV by appropriate choice of the detector material.
• In thermal detectors, the light is absorbed in the detector material to produce a minute increase in its temperature. Exquisitely sensitive electronic thermometers react to this change to produce the electronic signal. Thermal detectors are in principle sensitive to photons of any energy, so long as they can be absorbed in a way that the resulting heat can be sensed by their thermometers. Thermal detectors are important for the X-ray and the far-infrared through mm-wave regimes.
• In coherent detectors, the electrical field of the photon interacts with a locally generated signal that downconverts its frequency to a range that is compatible with further electronic processing and amplification. Downconversion refers to a multi-step process in which the incoming photon electrical field is mixed with a local electrical field of slightly different frequency. The amplitude of the mixed signal increases and decreases at the difference frequency, termed the intermediate frequency (IF). This signal encodes the frequency of the input photon and its phase.
Three Ways to Detect Light
Following: Lord Rosse image of M33 vs. Hubble image demonstrate how critical detector technology is.
A modern electronic detector (in this case, the 2k X 2K photodiode arrays used in NIRCam for JWST)
Photon detector and semiconductor terminology:
Photon absorption and creation of free charge carriers in semiconductors is the basic process behind photo-detection. The electrical behavior of a semiconductor is usually indicated with an energy level (or band gap) diagram: the low energy levels indicate all crystal bonds in place, higher levels for free electrons, and the energy to break the bonds as the energy band gap. Compare the diagram of crystal structure (above) with the band gap diagrams (below). To free an electron in intrinsic material (1) requires a certain energy indicated by the band gap. It takes less energy to free charge carriers from impurities (2) and (3).
Terminology to describe detector behavior:
• quantum efficiency
• indirect vs. direct absorption
• linearity
• dynamic range
• resolution
• time response
• spectral response
• responsivity
• noise
• shot noise
• Johnson/kTC noise
• excess noise (e.g., electronic)
The net absorption is characterized by the absorption coefficient, a. Note the difference between direct and indirect absorption (e.g., silicon vs. GaAs). Quantum mechanical selection rules do not permit transfers at the band gap energy for indirect absorbers.
)3(,1)(
1
0
1)(
00da
eS
daeSS
ab
The quantum efficiency is:
A decent detector will have close to linear response over some range of signal, and will completely saturate at some high level. In between, it is possible to recover information – the range of signal where useful information can be obtained is the dynamic range.
We characterize the resolution in a number of ways, including the MTF. Although the time response can have complex behavior, we will deal mostly with simple resistance-capacitance (RC) behavior:
)9(0,/
0
tRCt
e
RC
v
outv
The spectral response drops abruptly to zero at the band gap or excitation energy. The responsivity is the amps out per watt of signal in. It rises linearly to the cutoff wavelength for an ideal detector (assuming one charge carrier per absorbed photon).
Johnson and kTC Noise
kTVC N2
1
2
1 2 R
kTdfI J
42
• The circuit below has both potential energy (charge on the capacitor) and kinetic energy
(Brownian motion of electrons in the resistor)
• From thermodynamics, we have kT/2 of energy with each degree of freedom, leading to:
• To control this type of noise, we need to operate
the detector cold and build it to have a very large
resistance.
• Must also operate cold to control thermally
generated currents
Here is what the very simplest
detector might look like
C R
In a simple photoconductor, the photon is absorbed between the
contacts. The performance is limited because this detector must have
an extremely long RC time constant (since R must be very large to
minimize thermal noise), and the geometry cannot be adjusted to
separate the R from the C.
Detector type #1, Si:X IBC
• Idea is to create a large resistance for all but the photo-electrons
•Physical structure to left, band diagram to right; structure is a thin intrinsic
layer, then to right of it a heavily doped absorbing layer, then to right, a contact
• An absorbed photon elevates an electron to the conduction band, from which
it can migrate to the contact unimpeded. Thermal charges in the impurity band
are blocked at the impurity layer, so dark current is low.
• Detector type of choice for 5 – 35mm
Use of these detectors in
an array requires some architecture
changes, to allow attaching the
readout (to the left in these drawings).
Also, very high purity must be
achieved in the silicon to allow for
complete depletion of the IR-active
layer, or the quantum efficiency will
suffer (or the bias will have to be set
too high, increasing the noise –
w is the width of the depleted region,
tB the blocking layer thickness, NA the
minority impurity concentration, and
Vb the bias).
Here is a state-of-the-art Si:As IBC
array (made by Raytheon for MIRI on
JWST).
It is 1024 X 1024 pixels and at ~
6.7K delivers dark current < 0.1 e/s,
read noise of ~ 15 e rms, and
quantum efficiency > 60% from 8 to
26mm (~ 5% of maximum still at
28.3mm and > 30% at 5mm). The
pixels are 25mm on a side. Impurities
are at the 1012 cm-3 level. That is,
within a pixel there are 1015 silicon
atoms and 2 X 104 impurity atoms (2
X 10-9 %)
A special process is used for the
readout circuit so it works well at
such low temperatures.
Similar devices but poorer readout
performance were made by DRS
Technologies for WISE.
Detector Type #2, Photodiodes
• A depletion region is created by doping to form a junction between n-type
and p-type material
An electron has a probability of 0.5 of lying at the Fermi energy level. Dopants shift the
Fermi level; n-type up and p-type down. When different doping is applied to adjacent
semiconductor volumes, electrons flow until the Fermi level is level, bending the bands.
When all the free charge carriers
have been captured or driven out
of a volume of semiconductor, it is
said to be depleted. When
backbiased (so external voltage
tries to drive electrons up the
contact potential), the resistance
is large.
When a photon is absorbed,
the freed charge carriers
diffuse through the material
until one of them
encounters the junction,
which it is driven across by
the internal field to create
the photocurrent. The
diffusion coefficient and
length are:
)17(,q
TkD
m
)18(.DL
where m is the mobility
(characterizes ability of charge
carrier to migrate) and is the
recombination time (goes as
T1/2).
L goes as T3/4
Diodes attached tostrong substrate (readout)and then thinned:SBRC InSb arrays, usedfor SIRTF and largeformat ‘Aladdin’ device
Diodes on “back” of transparent substrate:Rockwell HgCdTearrays, used forNICMOS and for largeformat ‘Hawaii’ devices
• For operation with low dark current, low temperatures are needed
• Absorbing layer may need to be thinned to 10 - 20mm to collect charge
carriers at these temperatures
• Two approaches are shown below
• The yields in doing this can be low, partly explaining the high prices of these
arrays.
Material Cutoff wavelength mm)
Si 1.1 (indirect)
Ge 1.8 (indirect)
InAs 3.4 (direct)
InSb 6.8 (direct)
HgCdTe ~1.2 - ~ 15 (direct)
GaInAs 1.65 (direct)
AlGaAsSb 0.75 – 1.7 (direct)
• Here are some photodiode materials and their cutoff wavelengths.
HgCdTe has a variable bandgap set by the relative amounts of Hg and Te in
the crystal. AlGaAsSb behaves similarly.
• Indirect absorbers will have poor QE just short of the cutoff
An example: Teledyne HgCdTe
for all the JWST instruments
except MIRI. Here is the
architecture of a pixel. Photons
come in from the right and the
contact to the readout is to the
left. The cap layer has the
Hg/Te ratio adjusted to
increase the bandgap, so free
charge carriers are repelled
without actually having a
discontinuity in the crystal.
These arrays come with either
2.5 or 5mm cutoff. A competing
technology is diodes in InSb,
with a cutoff just beyond 5mm.
An example: the HgCdTe
arrays for NIRCam and
other JWST instruments.
They are 2048 X 2048
pixels in size, with 18mm
pixels. Dark currents at ~
37K are 0.004 e/s for the
short cutoff and 0.01 e/s for
the long. The QE is > 90%
from 1mm to close to the
cutoff and > 70% between
0.5 and 1mm. The read
noise is ~ 6-7e rms.
Similar devices but
somewhat lower
performance have been
made by Raytheon for
VISTA.
Avalanche Photodiode (APD)
The APD allows gain and
even single photon counting
with a solid-state device.
Avalanche multiplication
occurs when the charge
carriers are accelerated to
high enough energy to free
additional charge carriers and
so one can produce an
avalanche of many. This is a
stochastic process and hence
brings extra noise, described
by the gain dispersion:
or by the excess noise factor,
where G is the gain and G is the standard deviation of the gain.
• To minimize G want the gain to occur in a very thin region of the detector
• Use bias to deplete absorption region
• Large field that produces avalanche is confined to the junction
• For many materials, avalanching involves both holes and electrons
• Silicon avalanche photodiodes have F 2
• For HgCdTe, the holes can hardly move. Technically we say the hole
mobility is much less than the electron mobility
• Therefore, can have gain in HgCdTe avalanche photodiodes of up to 30
with F < 1.1 – this makes possible detectors with impressive capabilities for
very rapid readout with low noise
• Another approach is to increase the gain to yield a pulse for a single absorbed
photon
• There is a dead time of hundreds of nsec to quench the avalanche
• Thus, this only works for very low photon rates
• Good when very accurate timing is needed on low signals
Avalanche Photodiode Design
The Array Revolution in Infrared Astronomy
1968, 5-m telescope,
3 nights, single
detector
~ 2000, 1.3-m
telescope, 8
seconds, 256 X
256
~ 2006, 6.5-m
telescope, 1 hour,
1024 X 1024
(4 minutes with the
2x2 2048 X 2048
NIRCam mosaic)
Detector Type #3: Image Intensifier
• Vacuum photodiode - similar in physics to semiconductor photodiode
• Lower quantum efficiency because photoelectron has to escape from the
photocathode in to the vacuum space (photoelectric effect)
• Operates well at room temperature because the vacuum has high resistance (the
limiting issue for its electrical performance is the thermal release of charge carriers
from the photocathode)
Band gap diagram for a photocathode
The work function, , is the energy required for the electron to escape
Some old-time image intensifiers
All suffer from signal-induced backgrounds. Their outputs are an amplified
version of the inputs and any leakage back to the input contaminates the signal.
Modern use is in the ultraviolet and with a microchannel array as the amplifier,
providing an electronic output with no feedback of light.
The microchannel is coated with a material that releases a number of electrons
when hit by an energetic electron. A high voltage is maintained from left to right
in this diagram to accelerate the electrons and create more at each impact.
However, this process is noisy because of the small number of electrons freed
on the first impact.
The GALEX image intensifiers use a microchannel plate for amplification and its
output goes to a grid of electrodes that encode the position where it hit the
photocathode with delay lines. That is, by measuring the relative output times of the
left and right ends of the x delay line, one determines the position.
Photomultipliers are also based on the vacuum photodiode with an
amplifier that works by multiple stages of avalanching. The first stage can
be manufactured to release many electrons (perhaps 20) and reduce the
contribution to the noise.
Infrared Detector Arrays
• Best performance with silicon integrated circuit readout • Cannot manufacture high quality electronics in other semiconductors
• CCD-type readout has charge transfer problems at cold temperatures
• Direct hybrid construction • Fields of indium bumps evaporated on detector array, readout amplifiers
• Aligned and squeezed together - very carefully
The readout: a source follower
simple integrating amplifier
• As charge accumulates on the gate
capacitance of the MOSFET, it
modulates the current in the channel
• To keep from saturating, the charge
is reset from time to time
How should we sample the output?
kTC or Reset Noise
kTVC N2
1
2
1 2 R
kTdfI J
42
• The circuit below has both potential energy (charge on the capacitor) and kinetic energy
(Brownian motion of electrons in the resistor)
• From thermodynamics, we have kT/2 of energy with each degree of freedom, leading to:
)24(.2
2
q
SCkT
NQ From the left expression we get:
For C = 10-13 F and T = 40K, the noise is about 45 electrons rms. A
NIRCam array has a read noise of 6 - 7 electrons. How is this done?
• To avoid reset or kTC noise, we have to use readout strategy (b) or ( c)
• Then, if R = 1019 (say), RC 106 sec
• ( c) lets us take out some forms of slow drift by sampling with the reset
switch closed. However it adds root2 more amplifier noise
• Modern arrays can support strategy (b) even for integrations of 1000 -
2000 seconds
• In general, the reset noise is
• The kTC noise is then (45e)*(1 - exp(-t/RC) ) = 0.1e for 2000 seconds
RCteq
kTCnoiseread /
21
with additional components from amplifier noise and other sources.
Vout
time
More readout options: “Fowler sampling” to the left and “multi-accum”
to the right. Both address averaging out the high frequency electronic
noise (e.g., from the output amplifier). Fowler sampling has an
advantage in principle for lower net read noise, multi-accum is more
robust against upsets, e.g., cosmic ray hits.
Now consider the operation of array of readout amplifiers:
This approach has
random access and
reads nondestructively.
That is, we can read
out any pixel we want,
and we can read it and
then leave it exactly as
it was with the same
signal. This allows
interesting read out
patterns, like Fowler
sampling (multiple
samples to drive down
amplifier noise) or
sampling up the
integration ramp.
Charge Coupled Devices
(CCDs)
CCDs are actually more
complex than infrared arrays,
so we have saved them for last.
The top shows the physical
arrangement of a pixel, and (b)
and ( c) the situation before and
after exposure to light. The
electrodes deplete the silicon
near them and create potential
wells where free charge carriers
collect.
To keep the photoelectrons
from getting trapped at the back
surface, special coatings are
applied that bend the bands to
repel them.
Charge Coupled Devices
(CCDs)
Note the resemblance to the
simple photoconductor.
Reading out a CCD
When it is time to read out the signal, it is transferred along the array by
manipulating the voltages on the electrodes. This illustration is for a 3-phase
CCD, but similar strategies work for 4-phase and, with a bit of a trick on the
electrode design, on 2-phase. The isolation of the charge packets provides the
high resistance necessary to minimize thermal noise.
The charges are brought to the output amplifier in (a) in-line transfer, (b )
interline transfer, or (c ) frame transfer architectures. Astronomers
generally prefer (a), in which case the chip continues to collect signal as it
is read out. However, when better time definition is needed, interline
transfer is used (for example, for X-ray detection).
)25(.2/1)0
2( NnCTE
N
The charge transfer efficiency must be extremely high. If the proportion of
charge lost in n transfers is , then the noise in transferring N0 charges is
Consider a high performance 2048 X 2048 CCD with 2 electrons read noise.
If it is 3-phase, it takes about 12000 transfers to transfer the most distant
charge package out. Consider a signal of 100 electrons. Then, if the CTE =
0.999999, the CTE noise is 1.5 electrons! Incredibly high charge transfer
efficiency is the key to the performance we need.
However, as we have drawn the CCD, the photo-electrons collect at the
interface between the electrode and the silicon crystal, where there are lots
of open crystal bonds that trap electrons and degrade the CTE. The noise
achievable with such “surface channel” CCDs is many hundreds of
electrons.
Another source of traps is damage by cosmic rays. The CCDs in the ACS
(Advanced Camera for Surveys) on HST are getting seriously damaged in
this way and their charge transfer efficiency is dropping.
The trapping problem is fixed
with a buried channel. A thin
layer of silicon (about a
micron) is doped oppositely
to the rest of the detector
wafer and the resulting
potential minimum causes the
electrons to collect in the bulk
crystal, not at the electrodes.
This does not work at low
temperatures (< 70K), and if
the wells are over-filled then
the CCD has some electrons
that revert to surface channel,
i.e., serious latent images
and bad CTE.
If the CCD is clocked out too fast, the electrons do not have time to
migrate completely to the next gate and the CTE is reduced, also causing
excess noise. One charge transfer mechanism is the thermal motion of
the electrons, which results in diffusion. The diffusion length is
)18(.DL
where D is the diffusion coefficient and is the recombination time.
)17(,q
TkD
m
and m is the mobility, a standard semiconductor parameter that is readily
available. We can re-interpret this as L is the distance between gates and
is the transfer time, and then calculate how long it takes to get all the
charge from one gate to another. A typical transfer time is 0.05ms, and we
might want to wait something like 8 transfer times to be sure all the
electrons got across, so it takes about 0.4ms per full transfer, or 1.2 ms per
pixel for a 3-phase device. If we have a 2k X 2k CCD, then it will require
about 5 seconds to read it out.
Other mechanisms assist the transfer, but this estimate is more or less
correct.
Another issue with over-
filled wells is that charge
can leak from one well to
its neighbor along the
charge transfer direction,
leading to “blooming.”
Some CCDs have extra
thick channel stops to
resist blooming, but this
reduces the quantum
efficiency and is not used
for astronomical detectors.
The buried channel CCD approach is very similar to the one using a
composition change in the HdCdTe cap layer in a Teledyne array to keep
the photo-electrons from getting trapped away from the junction.
Reading the signal out
Finally, the signal gets to the output amplifier. By using a floating gate to put
the matching charge (by capacitive coupling) on the MOSFET gate, we can
read it out while avoiding kTC noise. Since we retain the charge after
readout, we can take it to another amplifier and read it again to reduce the
net noise.
Some other aspects of CCD performance:
Pixel binning - does
not degrade the noise,
so is a painless way to
reduce the number of
effective pixels in the
array. This can reduce
data rates and also
effective read noise.
The CCD charge transfer process lends itself naturally to clocking charge
in one direction at a set rate. This capability can be useful in applications
where images drift across the detector array at a constant (relatively slow)
rate – the charge generated by a source can be moved across the CCD to
match the motion of the source. As a result, the CCD can integrate
efficiently on the moving scene of sources without physically moving
anything to track their motion.
Time-Delay-Integration (TDI)
Orthogonal charge transfer:
It is possible to arrange 4-
phase electrodes to allow
transfer of charge in either
direction and backwards
and forwards. As shown
here, transfer downward
goes 3-1-2-3, to the right
goes 4-2-1-4, and so forth.
Deep Depletion
Because silicon is an indirect absorber, CCDs tend to have poor QE near 1 micron.
The curves, in order of increasing absorption, are for 5, 10, 15, 30, and 50mm.
Just making the CCD thick reduces the
free electron collection efficiency
(particularly in the blue where the
electrons are freed in the first 10 – 20
microns). A deep depletion CCD solves
this problem by putting a transparent
contact on the back surface and placing a
bias voltage on it so drive the photo-
electrons into the wells.
CMOS detectors
are like taking
just the readout
for an infrared
array and placing
photodiodes on
the inputs of the
amplifiers.
CMOS (complementary metal oxide semiconductor) detectors
Because they can be made in
standard integrated circuit
foundries, they are relatively
cheap. They can also be made
with a lot of the support electronics
on the same silicon chip. To the
right is Sony’s diagram of how one
works.
CMOS detectors can be really big!
To the left, an X-ray detector; to the right a garden variety 18 Mpix camera detector.
However……..
For very low light levels (e.g., use in astronomy):
• CMOS detectors have poor fill factors (amplifiers compete for real estate on
the chip with the detectors)
• Questions about cryogenic performance: the fill factor issue can be fixed by
putting a grid of microlenses over the array (Canon does this for some of its
cameras), but it is not clear that such a device can be cooled. Thanks to Sony,
camera arrays are now available that are back illuminated:
An example: Fairchild CIS2521F
• 2560 X 2160 pixels
• 6.5 mm X 6.5 mm pixel pitch
• Readout speed maximum: 100
frames per second
• Read noise < 1.5 electrons rms
• Peak quantum efficiency > 52%