detector ii: ir arrays, bolometers - caltech astronomydmawet/teaching/ay... · bib devices in large...

Detector II: IR arrays, bolometers

Ay122a: Astronomical Measurements and Instrumentation, fall term 2015-2016D. Mawet, Week 6, November 13, 2015

Discovery of infrared radiation

The existence of infrared was discovered by William Herschel in 1800. Herschel measured the temperature of sunlight of different colors passing through a prism and found that the temperature increased just outside of the red end of the visible range. He later showed that these “calorific rays” could be reflected, refracted and absorbed just like visible light.

Discovery of Infrared Radiation

The existence of infrared was discovered by William Herschel in 1800. Herschel measured the temperature of sunlight of different colors passing through a prism and found that the temperature increased just outside of the red end of the visible range. He later showed that these “calorific rays” could be reflected, refracted and absorbed just like visible light.

Summary of digital imaging steps

Copyright 2008 Society of Photo-Optical Instrumentation Engineers

This paper was published in the Proceedings of the SPIE Conference on Astronomical Instrumentation (2008, Marseille, France) and is made available

as an electronic reprint with permission of SPIE. One print or electronic copy may be made for personal use only. Systematic or multiple reproduction,

distribution to multiple locations via electronic or other means, duplication of any material in this paper for a fee or for commercial purposes, or

modification of the content of the paper are prohibited.

page 3 of 14

2. HYBRID CMOS IMAGE SENSOR ARCHITECTURE

The most advanced IR arrays are made with a hybrid CMOS architecture, which combines the best qualities of infrared

detector material with the performance achieved by advanced CMOS integrated circuits. A hybrid CMOS image array,

shown in Fig. 2, uses the infrared detector layer for the first steps of digital imaging – detection of light and collection of

photocharge into pixels. The CMOS circuit is fabricated in the same silicon foundries that produce computer chips, but

special amplifiers are required to sense the very small packets of photocharge produced by faint astronomical sources.

The silicon readout integrated circuit (ROIC) converts the charge to voltage with an amplifier in each pixel, and transfers

the signal to the edge of the array. The analog-to-digital conversion can be done on the imaging array, or in the

associated focal plane electronics. The pixels of the detector array are attached to the pixels of the ROIC via indium

interconnects, one indium bump per pixel. Since the ROIC multiplexes the signals from each pixel to the off-chip

electronics, the ROIC is sometimes referred to as a “multiplexer”, although multiplexing (signal transfer) is just one of

the functions provided by the ROIC.

Fig. 2: Hybrid CMOS image array architecture (left) and the 6 steps of digital imaging. Drawing on left is courtesy of Laser Focus World. CMOS denotes complimentary metal-oxide-semiconductor.

The fabrication of large, high performance IR arrays involves three key technologies:

1. Growth and processing of the HgCdTe detector layer

2. Design and fabrication of the CMOS ROIC

3. Hybridization of the detector layer to the CMOS ROIC

The next two sections will discuss the HgCdTe detectors and the CMOS ROICs. The technology for hybridization

requires precision and strength at the same time. The indium bumps must be precisely aligned to within 1 µm across the

entire focal plane array. This alignment must be maintained while the detector and ROIC are “cold welded” to each

other by pressing the detector and ROIC detector together, with a pressure up to several hundred kilograms. Teledyne

now routinely hybridizes arrays up to 4×4 cm (such as the H2RG-18) and is developing 6×6 cm hybridization capability

in 2008 for a program that is producing 2K×2K, 30 µm pixels. This development enables the next generation IR array

for astronomy, the H4RG-15, a 4K×4K, 15 µm pixel array. A subsequent expansion to 8×8 cm hybrids is foreseen for

2009-10, which will enable an 8K×8K, 10 µm pixel array.

3. HgCdTe – HIGH PERFORMANCE INFRARED DETECTOR MATERIAL

The best detectors of infrared light are crystal lattices that provide well-defined separation between the valence band,

within which electrons are captive, and the conduction band, where electrons are free to move through the crystal lattice.

The energy separation between the valence and conduction bands is called the bandgap of the material. If a photon with

Anti-reflective coatings

R1=(n0-n1)2/(n0+n1)2

R2=(n1-ns)2/(n1+ns)2

Conditions for destructive interference Phase: Φ1-Φ2 = π Amplitude: R1 = R2

=> e = λ/4n => n1 = sqrt(n0 . ns)

Rayleigh condition

Charge generation via photoelectric effectAn incoming photon excites an electron from the the valence band to the conduction band: hν > Eg

Eg = energy gap of material Critical wavelength: λc (μm) = 1.238 / Eg (eV)

conduction band

valence band

E

Ege-

hν

Material name Symbol Eg (eV) λc (μm) Op. Temp. (*)

Silicon Si 1.12 1.1 163-300

Indium Gallium Arsenide InGaAs 0.7 1.7 77-200

Mer-Cad-Tel HgCdTe 1.00-0.09 1.24-14 20-80

Indium Antimonide InSb 0.23 5.5 30

Arsenic doped Silicon Si:As 0.05 25 4

(*) to keep dark current low (thermal electrons)

Introductory notes on photovoltaic IR Detectors

• Single pixel IR detectors have long used the photovoltaic effect

• Diode is formed at the junction between a p- and n- doped semiconductor

• This pn junction generates an internal electric field to separate the photon generated electron-hole pairs

• Migration of holes and electrons changes the electric field, hence there is a voltage change across the junction which can be measured

Intrinsic vs extrinsic IR detector material

• Intrinsic photoconductivity = natural photoconductivity of pure semiconductor material

• Extrinsic photoconductivity = synthesized photoconductivity of doped semiconductor material

• The gap energy Eg is replace by the impurity energy Ei :

• e- gets elevated from the valence band to an impurity level to create a hole

• Alternatively, e- may be freed by elevating it from an impurity level to the conduction band

• Either the hole or the electron can then move through the material in response to the electric field in the detector.

Examples of extrinsic silicon detector spectral response

low-noise contacts. Although the potential of largeextrinsic silicon FPAs for terrestrial applicationshas been examined, interest has declined in favourof HgCdTe and InSb with their more convenientoperating temperatures. Strong interest in dopedsilicon continues for space applications, particu-larly in low-background flux and for wavelengthsfrom 13 to 20 lm, where compositional control isdifficult for HgCdTe. The shallower impurity en-ergies in germanium allow detectors with spectralresponse up to beyond 100 lm wavelength andmajor interest still exists in extrinsic germaniumfor wavelengths beyond about 20 lm.

To maximize the quantum efficiency and de-tectivity of extrinsic photoconductors, the dopinglevel should be as high as possible. This idea isrealized in blocked impurity band (BIB) devices.The longer spectral response of the BIB Si:As de-vice compared with the bulk Si:As device (see Fig.12) is due to the higher doping level in the formerthat reduces the binding energy of an electron. Fora detailed analysis of the BIB detector see Szmu-lowicz and Madarsz [35].

BIB devices made from either doped silicon ordoped germanium are sensitive in the IR wave-length range of 2 and 220 lm. BIB devices in largestaring array formats are now becoming com-mercially available. The best results have beenachieved to date for Si:As BIB hybrid FPAs pro-duced by Hughes Technology Center in Carlsbad[36,37] and Rockwell International Science Center

in Anaheim [38]. Hybrid FPAs with Si:As BIBdetectors operating in 4–10 K temperature rangehave been optimised for low, moderate, and highIR backgrounds. The 256! 256 format with 30lm pixels and 240! 320 format with 50 lm pixelsare available for low- and high-background ap-plications, respectively. Antimony-doped silicon(Si:Sb) arrays and 128! 128 pixel Si:Sb hybridFPAs having response to wavelengths >40 lmhave been also demonstrated, primarily for use atlow and moderate backgrounds. Germanium BIBdevices have been developed on an experimentalbasis, but they have not been reported in large 2Darray formats yet.

4.6. GaAs/AlGaAs QWIPs

Among the different types of quantum well IRphotodetectors (QWIPs), technology of the GaAs/AlGaAs multiple quantum well detectors is themost mature [39,40].

QWIP technology is based on the well-devel-oped A3B5 material system, which has a largeindustrial base with a number of military andcommercial applications. QWIP cannot competewith HgCdTe photodiode as the single device es-pecially at higher temperature operation (>70 K)due to fundamental limitations associated withintersubband transitions [17]. However, the ad-vantage of HgCdTe is less distinct in tempera-ture range below 50 K due to problems involvedin a HgCdTe material (p-type doping, Shockley–Read recombination, trap-assisted tunnelling,surface and interface instabilities). Even thoughthat QWIP is a photoconductor, several its prop-erties such as high impedance, fast responsetime, long integration time, and low power con-sumption, well comply requirements of large FPAsfabrication. Due to the high material quality atlow temperature, QWIP has potential advantagesover HgCdTe for VLWIR FPA applications interms of the array size, uniformity, yield and costof the systems.

Fig. 13 shows two detector configurations usedin fabrication of QWIP FPAs. In the bound-to-continuum QWIP the photoelectron can escapefrom the quantum well to the continuum transportstates without being required to tunnel through the

Fig. 12. Examples of extrinsic silicon detector spectral re-sponse. Shown are Si:In, Si:Ga, and Si:As bulk detectors and aSi:As BIB (after Ref. [34]).

A. Rogalski / Infrared Physics & Technology 43 (2002) 187–210 199

Pros and cons of extrinsic detectors

• Pros:

• Can operate at much longer wavelength, because it takes less energy to free a charge carrier from an impurity atom than from an atom of the semiconductor crystal material

• Cons:

• Extrinsic photoconductivity is far less efficient because of limits in the amount of impurity that can be introduced into the semiconductor without altering the nature of the impurity states.

Tuning the bandgap/cutoff of Hg1-xCdxTe detectors






page 4 of 14

energy greater than the bandgap is absorbed, an electron will be excited from the valence band and placed into the

conduction band. The photocharge in the conduction band can be collected into pixels and measured.

A unique type of infrared detector material is Mercury-Cadmium-Telluride (HgCdTe, referred to as “Mer-Cad-Tel”, or

“MCT”). HgCdTe is special since its bandgap depends on the mixture of Mercury and Cadmium, and the bandgap can

be tuned by more than an order of magnitude, from less than 0.1 eV to greater than 1.5 eV. A more exact expression of

this ternary compound is Hg1-x

CdxTe, for which 50% of the atoms are Tellurium, and the remaining 50% of atoms are

composed of a mixture of Mercury and Cadmium (when x=1, HgCdTe reduces to CdTe). The longest wavelength of

light that can be sensed by a detector is inversely proportional to the bandgap, and Teledyne uses growth by molecular

beam epitaxy (MBE) to produce HgCdTe arrays with wavelength sensitivity for near-IR (1.7, 2.5 μm), mid-wave IR (5

μm), long-wave IR (8-10 μm) and very long-wave IR (up to 18 μm). The hybrid CMOS architecture enables pixels with

100% fill factor and high QE. The quality of the HgCdTe material continues to improve and the JWST specification of

dark current of less than 0.01 electrons per pixel per second (at 37 K operating temperature) is being achieved for 2.5

and 5 μm cutoff H2RG arrays.

Fig. 3 presents the bandgap and cutoff wavelength of Hg1-x

CdxTe as a function of the cadmium fraction, x. This plot is

derived from the equation presented by Hansen et al4

, where x is the cadmium fraction and T is the temperature in

degrees Kelvin.

Fig. 3: Bandgap and cutoff wavelength of Hg1-x

CdxTe as a function of the cadmium fraction, x.

Due to the temperature dependence of the bandgap, it is important to define the temperature of operation as well as

wavelength range of operation when fabricating HgCdTe material. This is especially important for the smaller bandgaps

that corresponds to cutoff wavelengths in the 12 to 18 µm range.

The growth method that produces the highest performance HgCdTe material is MBE. A MBE machine, Fig. 4, grows

the HgCdTe layer in an ultra-high vacuum, starting with a CdZnTe (cadmium-zinc-tellurium) substrate that has a lattice

spacing that is nearly identical to the lattice spacing of HgCdTe (6.4 Ǻ). The MBE machine evaporates and deposits Hg,

Cd and Te onto the CdZnTe substrate. The HgCdTe detector layer is slowly and precisely grown, one atomic layer at a

time in a very pure environment. An atomic layer is deposited every 1-2 seconds, and it takes 4-6 hours to grow an

HgCdTe detector layer. With MBE, the mixtures of Hg, Cd and Te, and additional doping materials, can be precisely

controlled. The feedback on material growth is provided by “spectroscopic reflection ellipsometry”, a technique of

measuring light reflected from the HgCdTe surface during material growth. The intensity and polarization of light

reflected from the HgCdTe surface provides a highly accurate measurement of the HgCdTe composition. This “bandgap

engineering” enables MBE to grow complicated structures in the HgCdTe layer that provide the highest level of

performance.

( )xTxxxEg 211035.5832.081.093.1302.0

432 −×++−+−= −

Most IR detectors are hybrid CMOS






page 3 of 14

2. HYBRID CMOS IMAGE SENSOR ARCHITECTURE

The most advanced IR arrays are made with a hybrid CMOS architecture, which combines the best qualities of infrared

detector material with the performance achieved by advanced CMOS integrated circuits. A hybrid CMOS image array,

shown in Fig. 2, uses the infrared detector layer for the first steps of digital imaging – detection of light and collection of

photocharge into pixels. The CMOS circuit is fabricated in the same silicon foundries that produce computer chips, but

special amplifiers are required to sense the very small packets of photocharge produced by faint astronomical sources.

The silicon readout integrated circuit (ROIC) converts the charge to voltage with an amplifier in each pixel, and transfers

the signal to the edge of the array. The analog-to-digital conversion can be done on the imaging array, or in the

associated focal plane electronics. The pixels of the detector array are attached to the pixels of the ROIC via indium

interconnects, one indium bump per pixel. Since the ROIC multiplexes the signals from each pixel to the off-chip

electronics, the ROIC is sometimes referred to as a “multiplexer”, although multiplexing (signal transfer) is just one of

the functions provided by the ROIC.

Fig. 2: Hybrid CMOS image array architecture (left) and the 6 steps of digital imaging. Drawing on left is courtesy of Laser Focus World. CMOS denotes complimentary metal-oxide-semiconductor.


1. Growth and processing of the HgCdTe detector layer



The next two sections will discuss the HgCdTe detectors and the CMOS ROICs. The technology for hybridization

requires precision and strength at the same time. The indium bumps must be precisely aligned to within 1 µm across the

entire focal plane array. This alignment must be maintained while the detector and ROIC are “cold welded” to each

other by pressing the detector and ROIC detector together, with a pressure up to several hundred kilograms. Teledyne

now routinely hybridizes arrays up to 4×4 cm (such as the H2RG-18) and is developing 6×6 cm hybridization capability

in 2008 for a program that is producing 2K×2K, 30 µm pixels. This development enables the next generation IR array

for astronomy, the H4RG-15, a 4K×4K, 15 µm pixel array. A subsequent expansion to 8×8 cm hybrids is foreseen for

2009-10, which will enable an 8K×8K, 10 µm pixel array.

3. HgCdTe – HIGH PERFORMANCE INFRARED DETECTOR MATERIAL

The best detectors of infrared light are crystal lattices that provide well-defined separation between the valence band,

within which electrons are captive, and the conduction band, where electrons are free to move through the crystal lattice.

The energy separation between the valence and conduction bands is called the bandgap of the material. If a photon with

ROIC = readout integrated circuit, converts the charge to voltage with an amplifier in each pixel, and transfers the signal to

the edge of the array

Note about CMOS

• Complementary metal–oxide–semiconductor (CMOS)

• Refers to a manufacturing technique

• “metal–oxide–semiconductor” is a reference to the physical structure of certain field-effect transistors, having a metal gate electrode placed on top of an oxide insulator, which in turn is on top of a semiconductor material.

CMOS Field Effect Transistor

FET

Indium bump interconnects

3

The readout circuit (to be described below) requires just three CMOS transistors per pixel (and associated electrical traces), which occupy only a fraction of the area of a typical 18µm wide pixel. This leaves space for the relatively large electrical contact pad that provides the interconnect path to the diode array, lying in the plane above it.

Figure 2

The technology for making the vertical connection between the dissimilar materials in the light sensing and signal processing layers is the key to IR detector manufacture. The contact is constructed by depositing a thick layer of Indium on each pad, one per pixel, of the readout IC (through an etched photo-resistive mask). Matching “Indium bumps” are deposited on the underside of the photodiode array. The tops of the Indium bumps must be accurately coplanar and very clean so that when the bumps on the detector layer and the silicon layer are precisely aligned then squeezed together, a cold weld is formed making a permanent electrical and mechanical connection – one per pixel. Currently it is possible to connect 4 million pixels with only a few hundred failures. A low viscosity epoxy is then wicked into the <10um wide spaces between the Indium columns and the detector layer is then polished and etched until it is only ~10um thick. This is a very complex and delicate process with yield problems at every step, so the top quality devices carry price tags in the $250-500K range, making IR detectors five to ten times as expensive as CCDs. Operation: See Figure 3. Before the exposure, the reset switch is closed, so that the photodiodes are reverse biased by a hundreds of millivolts. The CMOS transistor, which buffers the diode voltage has essentially zero gate leakage at the low temperature required for optimal photodiode performance, so the change in voltage on the photodiode is dominated by the electron-hole pair generation by photons. During the exposure the diode junction acts as a capacitor, which stores this photogenerated charge. The electric field induced by the reverse bias has driven all mobile charges (carriers) from the P-N junction (depletion region), which then acts as an insulator (dielectric) between the capacitor “plates”, which are formed by the conductive outer regions of the diode where

The technology for making the vertical connection between the dissimilar materials in the light sensing and signal processing layers is the key to IR detector manufacture. The contact is constructed by depositing a thick layer of Indium on each pad, one per pixel, of the readout IC (through an etched photo-resistive mask). Matching “Indium bumps” are deposited on the underside of the photodiode array. The tops of the Indium bumps must be accurately coplanar and very clean so that when the bumps on the detector layer and the silicon layer are precisely aligned then squeezed together, a cold weld is formed making a permanent electrical and mechanical connection – one per pixel. Currently it is possible to connect 4 million pixels with only a few hundred failures. A low viscosity epoxy is then wicked into the <10um wide spaces between the Indium columns and the detector layer is then polished and etched until it is only ~10um thick.

This is a very complex and delicate process with yield problems at every step, so the top quality devices carry price tags in the $250-500K range, making IR detectors five to ten times as expensive as CCDs.

Note about multiplexer (MUX)

Since the ROIC multiplexes the signals from each pixel to the off-chip electronics, the ROIC is sometimes referred to as a “multiplexer”, although multiplexing (signal transfer) is just one of the functions provided by the ROIC (see below).

Monolithic vs hybrid focal plane arrays (FPA)

In the monolithic approach, some of the mul-tiplexing is done in the detector material itself thanin an external readout circuit. The basic element ofa monolithic array is a metal–insulator–semicon-ductor (MIS) structure as shown in Fig. 2(c). AMIS capacitor detects and integrates the IR-gen-erated photocurrent. Although efforts have beenmade to develop monolithic FPAs using narrow-gap semiconductors, silicon-based FPA technol-

ogy with Schottky-barrier detectors is the onlytechnology, which has matured to a level ofpractical use.

Hybrid FPAs detectors and multiplexers arefabricated on different substrates and mated witheach other by the flip-chip bonding (Fig. 3) orloophole interconnection. In this case we can op-timise the detector material and multiplexer inde-pendently. Other advantages of the hybrid FPAs

Fig. 2. Monolithic IR FPAs: (a) all-silicon; (b) heteroepitaxy-on-silicon; (c) non-silicon (e.g., HgCdTe CCD) (after Ref. [14]).

Fig. 3. Hybrid IR FPA with independently optimised signal detection and readout: (a) indium bump technique, (b) loophole tech-nique.

190 A. Rogalski / Infrared Physics & Technology 43 (2002) 187–210

Monolithic FPA

Monolithic vs hybrid focal plane arrays (FPA)

In the monolithic approach, some of the mul-tiplexing is done in the detector material itself thanin an external readout circuit. The basic element ofa monolithic array is a metal–insulator–semicon-ductor (MIS) structure as shown in Fig. 2(c). AMIS capacitor detects and integrates the IR-gen-erated photocurrent. Although efforts have beenmade to develop monolithic FPAs using narrow-gap semiconductors, silicon-based FPA technol-

ogy with Schottky-barrier detectors is the onlytechnology, which has matured to a level ofpractical use.

Hybrid FPAs detectors and multiplexers arefabricated on different substrates and mated witheach other by the flip-chip bonding (Fig. 3) orloophole interconnection. In this case we can op-timise the detector material and multiplexer inde-pendently. Other advantages of the hybrid FPAs

Fig. 2. Monolithic IR FPAs: (a) all-silicon; (b) heteroepitaxy-on-silicon; (c) non-silicon (e.g., HgCdTe CCD) (after Ref. [14]).

Fig. 3. Hybrid IR FPA with independently optimised signal detection and readout: (a) indium bump technique, (b) loophole tech-nique.

190 A. Rogalski / Infrared Physics & Technology 43 (2002) 187–210

Hybrid FPA

Fabrication of IR arrays


1. Growth and processing of the detector layer (e.g. HgCdTe)



Molecular beam epitaxy (MBE)






page 5 of 14

Fig. 4: Two of the molecular beam epitaxy (MBE) machines that Teledyne uses for growing high quality HgCdTe detectors. The

machine on the left can hold 3-inch wafers, the machine on the right can hold 10-inch wafers.

Since HgCdTe is a direct bandgap semiconductor, it is a very efficient absorber of light. The absorption depth of the

photons in HgCdTe, i.e. the distance over which 1-e-1

(63%) of the light is absorbed, is shown in Fig. 5. For high

quantum efficiency, the thickness of the HgCdTe detector layer should be at least 3 absorption depths, so that at least

95% (1-e-3

) of the light can be detected. A rule of thumb is that the thickness of the HgCdTe layer should be at least

equal to the cutoff wavelength.

Fig. 5: Absorption depth of photons in HgCdTe as a function of cutoff wavelength. Values are shown for 77K temperature. For high

quantum efficiency, the thickness of the HgCdTe detector layer should be at least equal the cutoff wavelength of light.

Since the absorption of light near the bandgap is less efficient, the quantum efficiency of a detector layer decreases in the

region of the cutoff wavelength as shown in Fig. 6. The industry standard is to define the cutoff wavelength (λco) as the

wavelength where the QE is 50% of peak. Technically, this λco does not exactly correspond to the bandgap of HgCdTe

mixture, but this definition of λco is a useful parameter for specification of an IR focal plane array. Since the QE starts

decreasing at 90-95% of λco, the λco for an astronomy array is usually set at 5-10% beyond the longest wavelength

required for science, so that high QE is obtained for all wavelengths of interest. For ground-based telescopes, the

standard λco for J & H band is 1.8 µm, the standard for J, H, K is 2.5 µm, and for J through M bands, the standard λco is

5.3 µm.

3” MBE at Teledyne 10” MBE at Teledyne

Unit Cell (UC) for Direct Readout (DRO)

CHAPTER 1. INTRODUCTION 14

during integration. In addition, this multiplexer can be programmed to address a single

pixel or small subsection of pixels for continuous readout, enabling very short read times.

ResetVoltage

DetectorSubstrateVoltage

Unit�CellSource�Follower�FET

Drain

UNITCELL

Read�EnableClock

Reset�EnableClock

To�Output�FET andCurrent�Supply

IntegratingNode

Unit�CellSource�Follower�FET

with�InverterDetectorDiode

Figure 1.8. Unit cell schematic for Direct Readout (DRO).

Figure 1.8 illustrates the unit cell (UC) operation for DRO. The UC may be indi-

vidually selected to be read or reset or both simultaneously. When the UC is selected

for reset, the Reset Enable Clock connects the Reset Voltage to the Integrating Node.

This resets the actual bias across the Detector Diode to the amount determined by

the difference between the Reset Voltage and the Detector Substrate Voltage. Once

the Reset Enable Clock disconnects the Reset Voltage from the Integrating Node,6 the

Detector Diode may debias due to photo-current or dark current and the voltage at the

Integrating Node will change accordingly.

When the voltage at the Integrating Node is selected for reading, the Read Enable

Clock connects the Unit Cell Source Follower (UCSF) Current Supply to the source

of the UCSF FET, which in turn enables the current to flow through the UCSF FET.

When this happens, the voltage on the integrating node is amplified, first by the UCSF

FET, and subsequently by the Output FET, before being read at the output. The

gain of the source follower amplifiers (UCSF FET and Output FET) is designed to be

6The voltage on the Integrating Node is actually modified when the Reset Voltage is disconnected

from it by charge redistribution in the reset line (see Section 3.2.4). The actual bias across the Detector

Diode after this redistribution is what debiases by photo-current or dark current, thus changing the

voltage at the Integrating Node.

FET = field effect transistor

Detailed operation

• Reset switch is closed to set the photodiode in reverse biased mode by Vr ~ -100 mV

• Reverse bias => depletion region widens => diode is high-resistance insulator

• Thermal leakage is negligible at low temperature for both the photodiode and MOSFET

• Change in voltage dominated by electron-pair generation, which drops the voltage by ΔV

Correlated Double Sampling (CDS)

5

Figure 4: Sample timing for the last pixel in a CDS frame.

It is reasonable to ask why two samples are necessary. The most obvious effect of the first sample is to remove the DC offsets which are intrinsic to the readout circuit and which vary from pixel to pixel. One might consider calibrating these offsets less than once per exposure, however our inability to correct perfectly for changes in these offsets is one of the primary sources of error. There are uncertainties in the charge left on the diode after reset, both due to thermal noise in the reset switch resistance and the voltage to which it is connected. The buffer transistor offset voltage varies rapidly with temperature changes, and is sensitive to supply and detector substrate bias voltage changes. Even if external noise sources could be reduced, the initial sample would still be required to subtract offset drift due to low frequency noise (drift) in the buffer transistors. When the diode voltage is sampled only once at the start, and again at the end of the exposure, we call this “Correlated Double Sampling” (CDS). This is similar to the CDS performed when reading a CCD, except that the signal samples span the exposure time rather than a pixel time. Clearly the 105 to108 greater time between samples makes infrared detectors very much more susceptible to sources of zero-point drift than CCDs. MOSFETs generate a drain-source current in response to a voltage on the gate. The gate is insulated from the FET channel so leakage at cryogenic temperatures is negligible and independent of drain-source current. Thus the accumulated charge is measured non-destructively. We will take advantage of this later to allow a more precise estimate of the incoming flux in the presence of electronic noise (referred to as “read noise”). Thus far we have considered how charge is integrated on the sensing node, and how the exposure time is defined, but we have not considered how one accesses the outputs of millions of pixels packed into a small area.

The exposure time is the time between samples and not the time since reset. Signal arriving prior to the first sample is ignored (subtracted from the final sample), so the exposure duty cycle is not 100% but approaches it when the exposure times are long.

Fowler N sampling (Al Fowler is a NOAO engineer)

15

Fowler sampling

Figure 9: Comparison of CDS timing and Fowler Sampling,

Classical “Correlated Double Sampling” involves nothing more than sampling the signal after reset and subtracting this from the final value. Any signal accumulated outside these two samples is invisible, except for its effect on linearity and dynamic range. Fowler sampling is a simple variant for improving the read noise. Since the signal is read non-destructively, multiple samples at the beginning and end of the exposure can be averaged to reduce the effective read noise. (The detector readout software will do this calculation in real time). An alternative way to think of this is as a set of partially overlapping CDS pairs, which are averaged. The effective exposure time is the difference between the Nth sample in each group and not time between first and last sample. See Figure 9. In the limiting case where the Fowler samples are tightly packed at either end of the exposure, the exposure duty cycle is high, and the signal and photon shot noise is nearly identical in each “fowler pair”. When the fowler sampling time becomes a larger fraction of the total integration time then the shot noise in each pair is only partially correlated between pairs.

• Can you see why it is better to scan the whole array N times rather than reading each pixel N times, averaging, then moving to the next pixel?

1. Measure and plot spatial Read Noise vs #samples for constant exposure time.

<…/Ay105/fowler/1_dark60s/>

Take darks with constant effective exposure time = 60s

Fowler sampling is a simple variant for improving the read noise. Since the signal is read non-destructively, multiple samples at the beginning and end of the exposure can be averaged to reduce the effective read noise. (The detector readout software will do this calculation in real time). An alternative way to think of this is as a set of partially overlapping CDS pairs, which are averaged. The effective exposure time is the difference between the Nth sample in each group and not time between first and last sample.

2D Multiplexor

6

Figure 5: Schematic layout of a readout multiplexor for an infrared detector array.

The pixels are accessed via a 2D multiplexor. Each pixel output has a single MOSFET which is driven hard on or off, to act as a switch. Outputs of all switches in the same column are connected to a bus. The control lines for the pixel-select switches in each row are ganged together. Only one row-enable line is active at a time: every pixel in the selected row is connected to a different column bus. At the edge of the array, each column bus is connected via a switch to the output buffer. To raster through the pixels one enables a row then sequentially selects columns, then repeats for the next row. To operate a 2048x2048 array one does not need to generate 2048 row and column select signals. These are the outputs of serial registers running along two edges of the multiplexor. A single bit is loaded into the row/column-select shift register at frame/line-start then shifted by the row/column clock. In some devices the row clock is derived from the output of the column shift register so only a frame start pulse, one clock and a reset control line are required. From 3 to 10µs are typically needed to access each pixel. To reduce the time needed to raster through all pixels the columns are subdivided into groups each served by a separate output buffer. The 2048x2049 pixel array you will be using has 32 outputs each serving 64 consecutive columns.

Only one row-enable line is active at a time: every pixel in the selected row is connected to a different column bus. At the edge of the array, each column bus is connected via a switch to the output buffer. To raster through the pixels one enables a row then sequentially selects columns, then repeats for the next row.

Readout time

From 3 to 10μs are typically needed to access each pixel. To reduce the time needed to raster through all pixels the columns are subdivided into groups each served by a separate output buffer/channel. A typical 2048x2048 H2RG has 32 outputs each serving 64 consecutive columns.

Note that since the exposure time is defined by the times at which the pixels are read, then the exposure for the last pixel is displaced from the first by the time it takes to scan through whole array. (~10μs*2048*2048/32 = 1.3s)

Pros of on-pixel integration

• Electronic shuttering: Exposure time is the time between initial and final reads, not time since reset. Charge accumulated between reset and first sample has no effect on noise but can consume dynamic range and affect linearity.

• Since readout is non-destructive, noise reduction is possible by combining multiple samples. (Fowler Sampling, Sample Up the Ramp). However the improvement isn’t as good as √N due to temporal correlations: there is significant noise power or systematic drift on frame-to-frame timescales.

Cons of on-pixel integration

• CDS occurs across the exposure time. Consequently IR detectors must be DC coupled and are at least 1000 times more sensitive than CCDs to electronic drifts and temperature changes.

• Because each pixel has a different signal path, there is no “overscan” (as in a CCD) to provide an accurate zero-point reference. The next best thing is to use dark pixels in the image area, or unconnected pixels around the edges to mimic the zero- point drift of the image pixels. These “reference pixels” don’t tell us where zero is but do tell us how much it has changed.

Cons of on-pixel integration cont’d

• Read noise increases with exposure time. With good electronics, the read noise is reduced to the 1/f noise in the pixel buffer transistor and detector material.

• The charge to voltage conversion is non-linear. The signal is accumulated on the detector diode capacitance. The reverse bias applied by the reset is discharged by the photocurrent causing the width of the depletion region to be reduced, so the diode capacitance increases: the voltage change of for a given charge increment drops.

• When observing bright sources or in high background, substantial charge can be accumulated between the reset and first sample, eating into the apparent dynamic range.

Notes on IR arrays (vs CCDs)

• Many parameters vary considerably from pixel to pixel. (e.g. dark current, QE, noise, temperature sensitivity).

• Dark current is higher and is more steeply dependent on temperature. In the best IR detectors this is just due to the lower bandgap, but it is often the case that imperfect surface passivation during manufacture degrades the dark current and causes large dark current variations from pixel to pixel.

• Dark current takes several hours to fully stabilize (!!! major problem at observatories) after a perturbation such as a temperature or bias voltage change (eg cycling power).

Example of IR arraysIR Arrays

IR arrays from Teledyne






page 2 of 14

� OCO – Orbiting Carbon Observatory (spectrographs)

− 1024×1024 pixel arrays (N = 3), HgCdTe (λco

= 2.5 µm), Visible silicon PIN (λ = 0.4-1.1 µm)

� Hubble Space Telescope (HST), Wide Field Camera 3 – WFC3 (imager)

− 1024×1024 pixel array (N = 1), HgCdTe (substrate-removed, λ = 0.6-1.7 µm)

� James Webb Space Telescope (JWST), Near Infrared Spectrograph (NIRSpec), Near Infrared Camera

(NIRCam) and Fine Guidance Sensor (FGS) (imagers and spectrograph)

− 2048×2048 pixel arrays (N = 15), HgCdTe (substrate-removed, λ = 0.6-2.5 µm and 0.6-5.0 µm)

− Engineering model deliveries have been made, with flight units to be completed during 2008

� Carnegie Observatory, Four Star (imager)

− 2048×2048 pixel arrays (N = 4), HgCdTe (substrate-removed, λ = 0.6-2.5 µm)

� European Southern Observatory, X-Shooter (spectrograph)

− 2048×2048 pixel arrays (N = 1), HgCdTe (substrate-removed, λ = 0.6-2.5 µm)

Additionally, during the past 2 years, TIS has made several significant technological advancements

1,2

, including:

� Substrate-removal of HgCdTe became a routine process for IR arrays. Substrate-removed HgCdTe (see

section 3) increases quantum efficiency (QE) in the near IR, provides visible response, eliminates fringing in

the substrate, and removes fluorescence from cosmic rays that are absorbed in the substrate.

� The H2RG image array and the SIDECAR ASIC reached TRL-6 and are fully qualified for use in space

astronomy instrumentation.

� The SIDECAR ASIC was selected for the repair of the Advanced Camera for Surveys (ACS) instrument of

the Hubble Space Telescope. Installation will be made during the October 2008 servicing mission.

� Spaceflight packages were developed for the SIDECAR ASIC: JWST cryogenic package, and HST

hermetically sealed package.

� The first SIDECAR ASICs were delivered to ground-based astronomy and the SIDECAR ASIC cryogenic

board was developed for placing the SIDECAR within ground-based astronomy cryostats.

� A 4096×4096, 10 µm pixel ROIC, named the H4RG-10, was designed, fabricated and successfully tested.

Developed primarily for use with a visible silicon PIN, this ROIC has also been used to fabricate an IR array.

The H4RG-10 development continues in 2008.

� A high speed, low noise readout integrated circuit, entitled Speedster128, was designed for IR adaptive optics

wavefront sensing and silicon-based x-ray detection. This array has 128×128 pixels, and readout noise of <5

electrons CDS at 900 Hz frame rate. Fabrication of the Speedster128 has commenced.

� A 4K×4K, 15 µm (H4RG-15) IR array has been designed and if funded, will commence fabrication in 2008.

Additionally, the CRISM spectrometer in the Mars Reconnaissance Orbiter has gone into operation, and is performing

very well. These arrays are 640×480 pixels, and both IR (λco

= 2.5 µm) and visible silicon PIN arrays are operating in

the CRISM instrument. Also, the 1024×1024 pixel (λco

= 4.8 µm) IR array in the Deep Impact mission continues to

perform well since launch in 2005, and the spacecraft added another comet rendezvous in 2010 as the EPOXI mission.

In this paper, we provide a summary of the technologies, products and ongoing developments for IR sensors at TIS. A

companion paper, entitled “Teledyne Imaging Sensors: advanced silicon CMOS sensors for x-ray to near-IR”, presented

in the visible sensors section of this conference, reports on silicon-based imaging sensors of TIS

3

.

Fig. 1: Examples of Teledyne infrared imaging sensors (left to right): WFC3 1K×1K (HST),

CRISM 640×480 (Mars Reconnaissance Orbiter), H2RG-18 2K×2K (JWST and ground-based astronomy)






page 11 of 14

5. SIDECARTM ASIC – FOCAL PLANE ELECTRONICS IN A SINGLE INTEGRATED CIRCUIT

In parallel with the development of the H2RG, Teledyne developed an application specific integrated circuit (ASIC) that

interfaces directly with the H1RG and H2RG and provides all of the functionality required from focal plane electronics

(FPE). The SIDECAR (System for Image Digitization, Enhancement, Control and Retrieval) ASIC, shown in Fig. 11,

provides significant reduction in the size, weight and power of the FPE9.

The SIDECAR contains a programmable microprocessor, bias generators, clock generators, amplifiers and analog-to-

digital converters (ADCs). Up to 36 analog inputs can be accommodated in parallel, with choice of 500 kHz, 16-bit

ADC or 10 MHz, 12-bit ADC (36 ADCs operate in parallel). The SIDECAR presents a digital interface to instrument

electronics, and with LVDS (low voltage differential signal) communication, the SIDECAR can be placed several meters

from the instrument electronics. All operation of the SIDECAR is fully programmable via LVDS communication lines.

The SIDECAR ASIC has been selected for use in 3 of the 4 instruments of the JWST. Two features of the SIDECAR

were important factors in its selection for JWST:

1. Low power operation: For JWST operation - 4 ports continuously read at 100 kHz pixel rate and 16 bit

digitization - the SIDECAR uses 11 mW power at 37K. The low power operation enables JWST to place the

SIDECAR within the very cold (37K) instrument module which is located 4 meters cable length from the

electronics located in the warm section of the observatory.

2. Low noise performance: The SIDECAR noise is negligible when compared to the H2RG readout amplifier, so

that the total noise of the H2RG-SIDECAR system is set by the low noise H2RG operation.

The SIDECAR ASIC was also selected for the repair of the Advanced Camera for Surveys (ACS) instrument in the

Hubble Space Telescope. In this system, the SIDECAR will be used to operate two 4K×2K CCDs. The ACS Repair

will take place during HST Servicing Mission 4, which is scheduled for October 2008. For the ACS Repair, a new

hermetically sealed spaceflight package was developed for the SIDECAR ASIC. The SIDECAR ASIC packaging is

shown in Fig. 11.

Fig. 11: SIDECAR ASIC and its packages. The ground-based astronomy package (left), the Hubble Space Telescope hermetically

sealed package (center), and the James Webb Space Telescope cryogenic package (right).

Both the H2RG and the SIDECAR have undergone environmental testing in spaceflight packages and have been

demonstrated to NASA’s Technology Readiness Level 6 – an important milestone for acceptance in space missions.

Both the H2RG and the SIDECAR have shown a high level of radiation hardness, and are now being considered for

other space missions. The SIDECAR has been operating the 4096×4096 pixel infrared camera (mosaic of four H2RGs)

at the University of Hawaii 88-inch telescope since early 2007, and the SIDECAR is being incorporated into new

instruments for several ground-based observatories, including, but not limited to: Carnegie Observatory (Four Star)

wide field imager, W.M. Keck Observatory (MOSFIRE multi-object spectrograph being developed by UCLA), Carnegie

Observatory (FIRE spectrograph being developed by MIT), Subaru Telescope (HiCIAO imager being developed with U.

Hawaii), Gemini Observatory (Gemini Planet Imager instrument being developed by UCLA).

Typical raw image

19 July 2010 NOAO Gemini Data Workshop 14

So, here’s what we have to deal with..

•  Raw K-band image of field shows stars, but also substantial sky signal

•  Sky signal intensity varies over field

–  Large-scale variations •  Illumination

•  Quantum efficiency variations

–  Small-scale variations •  Pixel-to-pixel variations

•  Array defects

•  High dark current pixels (mavericks)

•  These can be corrected by appropriate calibration images

–  Dark frames (bias)

–  Flatfield images

Raw K-band image of field shows stars, but also substantial sky signal

Sky signal intensity varies over field: – Large-scale variations:

Illumination Quantum efficiency variations

– Small-scale variations: Pixel-to-pixel variations

Array defects

High dark current pixels

These can be corrected by appropriate calibration images:

Dark frames (bias) Flatfield images

Let’s try this CCD-like recipe

Obtain science images

Obtain calibration images

Dark frames at same integration time

Flatfield images of uniform target

Subtract dark frame from science images

Divide dark-subtracted images by flatfield

=> Image of science field with uniform sky level

Subtract (constant) sky level from image

But, here is what we get.....

– Better, but still see substantial sky variations


When we try this (CCD style)…

•  Obtain science images

•  Obtain calibration images –  Dark frames at same integration time

–  Flatfield images of uniform target

•  Subtract dark frame from science images

•  Divide dark-subtracted images by flatfield

•  Image of science field with uniform sky level

•  Subtract (constant) sky level from image

•  But, here is what we get…..

–  Better, but still see substantial sky variations

Small flatfield errors on sky still larger than faint science targets Small flatfield errors on sky still larger than faint science targets


Since the sky is the problem…

•  Subtract out the sky ( or as much

as possible) before the flatfield

correction

•  Obtain two images of field, move

telescope between

•  Subtract two images

–  Eliminate almost all sky signal

–  Subtracts out dark current, maverick

pixels

•  Divide by flatfield image

•  Result has almost no sky structure

Subtracting sky minimizes effects of flatfield errors (but noise increased by 1.4)

The sky background is the problem

Subtract out the sky (or as much as possible) before the flatfield correction

Obtain two images of field, move telescope between

Subtract two images

– Eliminate almost all sky signal

– Subtracts out dark current, maverick pixels

Divide by flatfield image

Result has almost no sky structure

Subtracting sky minimizes effects of flatfield errors (but noise increased by 1.4)

Typical sequence for IR imaging• Multiple observations of science field with small telescope motions in between (dithering)

• Sky background limits integration time • Moving sources samples sky on all pixels • Moving sources avoids effects of bad/noisy pixels

• Combine observations using median filtering algorithm • Effectively removes stars from result => sky image • Averaging reduces noise in sky image

• Subtract sky frame from each science frame => sky subtracted images • Divide sky subtracted images by flatfield image

• Dome flat using lights on – lights off to subtract background • Sky flat using sky image – dark image using same integration time • Twilight flats – short time interval in IR

• Shift and combine flatfielded images • Rejection algorithm (or median) can be used to eliminate bad pixels from final image

Illustration


Here’s what it looks like…. Sky frame

Median

Subtract sky,

divide each by

Flatfield

Shift and add


Shift and combine images •  NGC 7790, Ks filter

•  3 x 3 grid

•  50 arcsec dither offset

Bad pixels eliminated

From combined image

Higher noise in corners

than in center (fewer

combined images)

NGC 7790, Ks filter 3 x 3 grid50 arcsec dither offset

Bad pixels eliminated From combined image

Higher noise in corners than in center (fewer combined images)

Summary: data reduction strategy for IR detectors


SOURCE OBSERVATIONS

(DITHERED)

SKY

SKY OBSERVATIONS

(DITHERED)

SKY SUBTRACTED

IMAGES DARKS

DOME FLATS ON

DOME FLATS OFF

FLAT

SKY SUBTRACTED,

FLATTENED IMAGES SHIFT, ALIGN SOURCES

AVERAGED IMAGE

[MEDIAN] [MEDIAN]

[―]

[―] [―]

[ / ]

[MEDIAN]

Mid-infrared strategy (nodding & chopping)

• Sky background at 10 μm is 1,000-10,000 greater than in K band • Detector wells saturate in very short time (< 50 ms) • Very small temporal variations in sky >> astronomical source intensities

• Read array out very rapidly (20 ms), coadd images • Sample sky at high rate (~ 3 Hz) by chopping secondary mirror (15 arcsec)

• Synchronize with detector readout, build up “target” and “sky” images • But tilting of secondary mirror introduces its own offset signal

• Remove offset by nodding telescope (30 s) by amplitude of chop motion • Relative phase of target changed by 180° with respect to chop cycle • Relative phase of offset signal unchanged • Subtraction adds signal from target, subtracts offset


Mid-infrared strategy •  Sky background at 10 µm is 103 – 104 greater than in K band

–  Detector wells saturate in very short time (< 50 ms)

–  Very small temporal variations in sky >> astronomical source intensities

•  Read array out very rapidly (20 ms), coadd images

•  Sample sky at high rate (~ 3 Hz) by chopping secondary mirror (15 arcsec)

–  Synchronize with detector readout, build up “target” and “sky” images

–  But tilting of secondary mirror introduces its own offset signal

•  Remove offset by nodding telescope (30 s) by amplitude of chop motion

–  Relative phase of target changed by 180° with respect to chop cycle

–  Relative phase of offset signal unchanged

–  Subtraction adds signal from target, subtracts offset

•  http://www.gemini.edu/sciops/instruments/t-recs/imaging

chop nod

……

Bolometers

• Measure the energy from a radiation field, usually by measuring a change in resistance of some device as it is heated by the radiation

• Mainly used in FIR/sub-mm/microwave regime

• Sensitivity is measured through the Noise Equivalent Power (NEP): the power absorbed which produces S/N=1 at the output (units W/Hz0.5)

• Typically use a semiconductor resistance thermometer, and a metal coated dielectric as the absorber

Bolometers •  Measure the energy from a radiation field, usually by measuring a

change in resistance of some device as it is heated by the radiation •  Mainly used in FIR/sub-mm/microwave regime •  Sensitivity is measured through the Noise Equivalent Power

(NEP): the power absorbed which produces S/N=1 at the output (units W/Hz0.5)

•  Typically use a semiconductor resistance thermometer, and a metal coated dielectric as the absorber

Examples of bolometers

“Spiderweb” bolometer

Semiconductor bolometers from SCUBA

Components of a bolometer

• Absorber with heat capacity C

• Heat sink held at fixed temperature T0

• Small thermal conductance G between absorber and heat sink

• Load resistor RL

• Thermometer w. resistance R

• Constant current supply generating bias current I

• Device to measure voltage changes

Components of a Bolometer

•  Absorber with heat capacity C •  Heat sink held at fixed temperature

T0

•  Small thermal conductance G between absorber and heat sink

•  Load resistor RL •  Thermometer w. resistance R •  Constant current supply generating

bias current I •  Device to measure voltage changes

Schematic of a bolometer

Sources• http://spiff.rit.edu/classes/phys445/lectures/ccd1/ccd1.html

• Observational astrophysics, 2nd edition, P. Lena

• S. G. Djorgovski (Caltech, Ay122a, 2012)

• R. Smith (Caltech Ay 105 notes)

• J.W. Beletic notes (optics in astrophysics, R. & F.C. Foy editor, NATO Science Series)

• George Rieke 2007, Ann. Rev. Astr. Ap. 45, 77.

• An introduction to IR detectors (D. Joyce, NOAO)

detector ii: ir arrays, bolometers - caltech astronomydmawet/teaching/ay... · bib devices in large...

Documents