An InSb charge amplifier for use in a spectrometer array
Don Hartill, Franco Lisi, and Andrea Bettarini
We present a cryogenic charge amplifier system developed for an InSb photodiode array for use in high-
resolution infrared spectrometers. After a general introduction, the detailed circuit diagram of the chargeintegrator is presented. The integrator features a new optoelectronic method for resetting the integratingcapacitor; the whole charge amplifier is placed in the cold Dewar to reduce microphonics and to increase noise
immunity. The analysis of the expected performance is given along with the results of some preliminary tests
performed on a two-channel version of the system. A single-channel version under current use in the TIRGO
photometer shows noise figures in good agreement with our prediction; the measured noise is -210 electrons
in 1-s integration time, which corresponds to a NEP of 2.8 X 10-17 W/V, i in the K band (2.2 /m).
1. Introduction
A charge amplifier system has been developed foruse with a seven-element InSb linear photodiode ar-ray' to be placed in the focal plane of a high-resolutioncooled grating spectrometer. This spectrometer willbe used on the IR telescope of Gornergrat (TIRGO)operated by the Arcetri Observatory for the CNR ofItaly. 2
The system we present here is an improved versionof the charge amplifier developed for the TIRGO InSbphotometer.3 The InSb photometer has now beenretrofitted with a single-channel version of the systemwe describe.
The use of a charge amplifier avoids the inherentnoise of the large feedback resistor present in the clas-sic transimpedance amplifier 4 and provides large dy-namic range (106-107) by simply changing the mea-surement time. There are other examples of chargeamplifiers used in InSb infrared instrumentations; ourapproach presents two advantages. The first comesfrom placing most of the total system gain into thecharge amplifier operating at liquid nitrogen tempera-ture in close proximity to the InSb array. This sub-stantially reduces microphonics and noise pickupproblems because of the small physical size of thefeedback loop and short rigid connections betweendetectors and preamplifiers. The second advantage
D. Hartill is with Cornell University, Physics Department, Ithaca,New York 14853; the other authors are with Osservatorio di Arcetri,Largo E. Fermi 5, 50125 Firenze, Italy.
comes from our reset scheme6 which allows the use ofInSb photodiode arrays with either common cathodesor common anodes. This opportunity removes theneed for custom InSb arrays which arises with otheroptoelectronic reset schemes.5
II. System Description
A block diagram of the full charge integrating sys-tem is presented in Fig. 1 which for simplicity showsonly one channel of the eight in detail. There are threemain parts: (a) the eight FET-input integrating pre-amplifiers inside the cold Dewar (seven for the InSbarray and one for control purposes); (b) the set ofcorrelated double samplers; and (c) the sequence con-troller. Each correlated double sampler includes twogain stages and a pair of sample-and-hold circuits withselectable rise time. A multiplexer scans the eightchannels and the analog signals are converted by a 12-bit ADC. The sequence controller generates the set oflogic signals needed for proper system timing of thecharge integrators, and it provides the data and controlinterface to the on-line computer (PDP 11/34).
Each preamplifier operates as an integrator with thedetector current accumulated as a voltage ramp on afeedback capacitor during the integration time T. Atthe end of the measurement cycle the integrator israpidly reset by illuminating the two series FET junc-tions, in parallel with the capacitor, with an LED.3
The LED light output is controlled by driving the LEDwith a current determined by the integrator output.As can be seen in Fig. 2, the integrator output is fed to adifferential amplifier with a gain of '40. The ampli-fied signal drives the reset control circuit and a pair ofsample-and-hold (S/H) circuits. The integration cy-cle begins with a reset-enable pulse (Fig. 3), discharg-ing the feedback capacitance. At the end of the reset
Fig. 1. Block diagram of theInSb charge amplifier system.Only one channel of eight is
shown.
Fig. 2. Detailed schematic of the charge amplifier, reset driver, and sample-and-hold circuit. Vbi1s is a variable voltage which provides aknown bias at the charge amplifier input. R and C8are selectable to change the low-pass filter time constant.
the integrator output begins to decrease linearly; thetrailing edge of the reset pulse triggers both the startand the stop pulses, closing the two acquisition switch-es of the S/H circuits. The time constant r of the RXCcells in Fig. 2 sets the typical delay time that each S/Hneeds to follow its own input, so that the circuit be-haves like a single-pole low-pass filter. The width ofthe start pulse is a compromise between good low-passfilter settling with accurate ramp tracking and thepractical need of having the measurement cycle rea-sonably efficient. When the start S/H opens, the inte-gration period begins and continues until the stop S/Hswitch opens after a time T. At this moment thesecond differential amplifier output is sampled andthe signal is converted by the ADC. Shortly after, the
reset-enable pulse is triggered, and a new measure-ment cycle begins. The sequence controller generatesthe timing signal with a 12-bit X 1024 ring memorywhich is scanned by a 10-bit counter. Each block ofdata is stored in another 12-bit X 1024 RAM, which isread by the PDP 11/34 at the end of the programmedintegration cycles.
Ill. Charge Amplifier
The unique feature of the charge amplifier is thereset switch. It consists of two reverse biased gate-channel JFET junctions that are illuminated by anLED. The LED light output is controlled by a gatedsense amplifier connected to the integrator output.The dual JFET (with the top of the case cut away) and
Fig. 3. Waveform diagram for one integration cycle. T is theintegration time between two successive samples of the integrator
output.
the red LED are placed at opposite sides of a shortblack PVC pipe; the whole assembly is made light-proof. In normal operation the switch-off resistancedoes not present any detectable leakage current. Abias voltage can be applied to the photodiode for mea-surement by means of a resistor in the source circuit ofthe input FET (one-half of a National 2N6484 dualJFET). By using the dynamic load configuration, theinput 2N6484 has a voltage gain of'-100. An addition-al gain stage ('100) is provided by a TI TL074 opera-tional amplifier (op amp). The TL074 is a quad-integrated BIFET op amp which operates well at liq-uid nitrogen temperature. Compared with room tem-perature its gain-bandwidth product at low tempera-ture is reduced by a factor of 2 and the power supplycurrents are also lower by a factor of 2. Its equivalentseries noise voltage is similar to its room temperaturevalue and is a good choice for this application. Arelatively large 10-pF feedback capacitor is used tooffset any possible nonlinear effects due to the voltage-dependent capacitance of the gate-channel junction ofthe reset FET connected to the input node. The 1-MQresistor effectively shorts to ground any capacitiveeffects of the gate-channel junction connected to theintegrator input and limits the leakage currentthrough that junction. To restore the system gainreduced by the large feedback capacitor, a gain boostscheme is used to connect this capacitor to the integra-tor output, resulting in an equivalent feedback capaci-tance of -1 pF. The cost of this method is the reduc-tion of the open loop gain by the same factor, whichincreases the system sensitivity to transient effects dueto the relatively large input capacitance of the amplifi-er. With the gain boost the total open loop gain is'1000. As mentioned earlier it is very easy to adapt acommon-anode array to the integrator by merely re-versing the polarity of the two gate-channel FET junc-tions which reset the integrating capacitor.
The charge amplifiers are mounted on a G-10 print-ed circuit board which is fastened to the Dewar cold
in -nR~ vi +in O D Filters
stop
Fig.4. Equivalent circuit of detector and amplifier for noise analy-sis. Cin is the equivalent input capacitance. The amplifier is as-
sumed to be noiseless.
plate so that the heat load is transferred directly to thecold bath. The high impedance node in the inputstage has been connected in air to get rid of leakagecurrents caused by the poor insulation properties ofthe board. The power dissipation of each integratoramounts to -90 mW.
IV. Noise Analysis
The InSb photovoltaic detector can be modeled as acurrent generator with an equivalent parallel imped-ance given by a dynamic resistance RD and an intrinsiccapacitance CD. By operating the detector at zerobias, 1/f noise is not a problem: the largest noisesources are the parallel resistance RD and the shotnoise of the input currents. The detector Johnsonnoise is represented by an equivalent current genera-tor in = 4kT RD. The input transistor in the pream-plifier is the other main contributor to the systemnoise. The JFETs present two major noise sources inthe frequency range of interest: the shot noise due tothe leakage current of the gate-channel junction, andthe thermal noise generated by the carriers in thechannel. A flat spectral density is assumed for noise inthe following computation. The first noise source isnot always negligible in standard operating conditionsand must be modeled as a parallel noise source, whilethe other noise source is currently modeled by anequivalent series voltage source en = VFdepending onJFET type and operating temperature.
Figure 4 illustrates the charge amplifier replaced byan ideal noiseless amplifier with the appropriate noisegenerators. Cin is the equivalent input capacitance ofthe preamplifier and RF is the equivalent resistance ofthe optoelectronic reset. Cin is dominated by the gate-drain capacitance reflected to the input by the Millereffect, with the high voltage gain of the first stagemaking Ci. relatively large. The total output noise isgiven by the weighted sum of the two noise generatorsin Fig. 4; the weighting functions are the impulse re-sponse h(t) of the circuit and its time derivative h'(t).To perform this sum, Campbell's theorem may beused7 ; the noise voltages are
Vbas (V)Fig. 5. I-V characteristics for the input resistors at liquid nitrogentemperature (triangles = channel 1, circles = channel 2). To findthe effective voltage across the input the voltage must be multipliedby 2.7 X 10-3 (see Fig. 2). The two channels show unequal inputoffset voltages which are due to the mismatch of the input transistor
pairs.
the input resistances are 0.42 TO (channel 1) and 0.41TO (channel 2). The measured values of the inputresistors agree well with their nominal values at roomtemperature when a temperature correction is applied.
In the following it is useful to refer to the integratoroutput signals measured in counts at the ADC output.The calibration of the system gives 1 ADC count for acharge of -100 electrons into the integrator input.
To verify the noise analysis of Sec. IV, two tests werecarried out. In the first test the time constant -r wasfixed at 1 ms so that the series noise in Eq. (5) was keptconstant, and the squared output noise ENC' wasmeasured against the integration time Ti. In the sec-ond test the integration time was fixed at 0.1 s so thatthe parallel noise in Eq. (4) was kept constant and thesquared output noise ENC2 was measured against thetime constant rs. In each test the shot noise due to theinput current was subtracted in quadrature.
Both channels gave very similar results in the firsttest. A typical set of results and the best-fit line,
ENC2 = 29.5Ti + 3.0 (measured),
are shown in Fig. 6, where ENCa is in (counts) 2 and Tiin seconds. The squared noise charge computed byEqs. (4) and (5), assuming T = 100 K for RD and e = 4nV/VW, is
ENC2 = 26Ti + 1.6 (estimated).
This measurement confirms the theoretical linear de-pendence of ENC2 on the integration time, and thenoise excess is partly ascribed to the input resistorbeing at a temperature somewhat higher than 100 K.
Figure 7 shows the results of the second test: ENCbis plotted vs T-' (including both the internal timeconstant of charge amplifier and the S/H time con-stant). The best fit to the data is given by the line
ENC = .002TS1 + 2.7 (measured),
(f A)
60
40
20
V2(series) = 1/2e2 | I h(t)l dt + /2e2/Rp
X | h h(t)l 2dt, (2)
where CT = C + CD + CF and Rp = RDRF/(RD + RF).From Fig. 4 h(t) is easily found by standard Laplace
transform techniques. We assume that the amplifierhas an open-loop midband gain AO with a dominantpole at so giving v = -viAoso/(s + so). Because of thegain boost the standard formulas must be modified byo = R2/(R1 + R2), so that CF = aCp and RF = Ri/a.
With the above assumption and using the typicalamplifier listed in Table I, we obtain
h(t) = -1/CF[exp(s+t) - exp(st)], (3)
where s = -sOAOCF/CT and s = -1/(RFCF). Thefunction h(t) looks like a rough step with a rise time ofseveral microseconds and a very long fall time of sever-al seconds.
The complete system response to the input noisemust include the low-pass filter and the S/H section inthe correlated double sampler. The series noise volt-age can be substantially reduced by choosing suitabletime constants for the S/H, while the parallel noisecontribution is determined by integration time. If rsis the equivalent time constant resulting from the in-teraction of the amplifier and the S/H time constants,the impulse response in Eq. (3) must be modified ac-cordingly. Taking into account the temporal limits ofeach integration and substituting Eq. (3) into Eqs. (1)and (2), the equivalent noise charges ENC (parallel)and ENC (series) at the system input are, to first order,
ENC2(parallel) = 2kT/RDTi, (4)
ENC (series) = /2enCT/r 0 , (5)
again using the values of Table I.
V. Test Results
Preliminary tests were carried out with two-channelversion of the charge integrating system mounted in aDewar with the cold plate at solid nitrogen tempera-ture (50 K). In our test setup the photodiodes weresimulated with 0.465 T resistors8 in parallel with 20-pF capacitors.
By changing the bias voltage it is possible to inject aknown current into the integrator, and in Fig. 5 wepresent the V-I characteristics of the input resistors.From the slopes of the V-I curves we determine that
Fig. 6. Squared equivalent input noise charge vs integration timeTi. The continuous line is a best-fit approximation to the measuredvalues (circles) and the dashed line is the prediction of the theoreti-
cal analysis. The time constant is 1 ms.
ENC'
counts2)
40
30
20
10
1.0 4.9 90 15 185 20
rS1 s1)
Fig. 7. Squared equivalent input noise charge vs the inverse of thetotal time constant Tsa. The continuous line is a best-fit approxima-tion for the measured values (circles) and the dashed line is theprediction of the theoretical analysis. Integration time is 100 ms.
with r in seconds. The typical values of CT and en givean estimated squared noise contribution at Ti = 0.1 s of
ENC = 0.OO16T-1 + 2.6 (estimated).
The linear dependence of ENCb on r1 is confirmed.The small noise excess is partly due to an underesti-mate of the parallel noise as in the first test and partlyto the systematic effects arising from the turn-on tran-sient at the beginning of a series of integrations; in thefinal version the system is cycled continuously.
To confirm the dependence of the output seriesnoise on the input capacitance Cin, the gain of the firststage was reduced by a factor of 4 and the gain in thesecond stage was raised accordingly, so that the equiv-alent input capacitance was reduced by the same fac-tor. The measured series noise was reduced to -% ofthe series of the unmodified integrator.
The results of the tests show substantial improve-ments over the charge amplifier used in the InSb pho-tometer.3 With an integration time of 1 s, RD = 400
f A)
-25
-30
-20 -10 0 V (V)
Fig. 8. I-V characteristics of the 0.63-mm diam InSb photodiode ata temperature of 50 K. The dashed line is the tangent at zero-biaspoint. Iand Vaxes are linear. The equivalent resistance is given bythe slope of the dashed line; the result is -4.5 TQ. The InSb
detector was not flashed.
GQ, and a filter time constant of 3.5 ms, the totalequivalent noise charge is -600 electrons, which agreeswith Eqs. (4) and (5). To estimate the series noisecontribution, the input of the charge amplifier waskept open. The thermal contact between the inputFET and the cold bath was improved, so that the inputleakage current was reduced to -50 electrons/s. FromEq. (4), which assumes flat noise density, and with afilter time constant of 1 ms, series noise should be <100electrons; we measured a total noise of -150 electrons.
Another set of tests was carried out with a single-channel integrator mounted in the TIRGO photom-eter. The InSb photodiode, built by Cincinnati Elec-tronics, has a diameter of 0.63 mm. From its biascurve at a temperature of -50 K (Fig. 8), a zero-biasdynamic resistance of 4.5 TO is determined. For thisvalue of RD, Eq. (4) gives 100 electrons of parallel noiseand Eq. (5) gives -150 electrons of series noise, with atotal noise charge of -180 electrons. Operating thephotometer at low background, with a total input cur-rent of -5 X 104 electrons/s and when the effects ofshot noise of input currents are taken into account andproperly subtracted, we obtain a noise level of 210electrons in 1-s integration time. This is in goodagreement with what we predicted.
The present noise level corresponds to a noise equiv-alent power (NEP) in the K band (2.2 um),
NEP = hc 2s = 2.8 X 10-'7 W//
for a quantum efficiency of = 0.7 and an integrationtime of 1 s.
VI. Conclusions
An eight-channel low-noise charge amplifier systemfor InSb arrays has been described and its performancereported.
A single-channel version of the system was used inthe InSb photometer at the TIRGO telescope. When
the InSb detector was blinded, a noise figure of 210electrons in 1-s integration time was attained. Theperformance of our system can be compared with oneof the best IR photometers currently in use in astrono-my: the IRPS of the Anglo-Australian Observatory.The published intrinsic noise figure for that system9 is170 electrons. On the 1.5-m TIRGO telescope thephotometer sensitivity is limited by photon shot noisefrom sky and telescope at all wavelengths (1.1-5.0 gm),even with the smallest diaphragm (7 sec of arc). For a3 a measurement in 30-min integration time, the limit-ing magnitude is -19 in the J band (1.2 ,um).
At present, the opportunity of further improve-ments is being investigated. In particular, it is possi-ble to lower both the input leakage current and theseries noise of electronics. Input leakage current isdue mainly to the self-heating of the input FET; theproblem will be reduced by lowering the temperatureof the FET. This will allow low-noise operation withthe very low background levels of the spectrometer. Itis also possible to reduce the series noise from theelectronics to much lower values by simply controllingthe time constant r, and the first stage gain. The priceis a small reduction in the efficiency of the measure-ment cycle when r, is relatively long.
References
1. Cincinnati Electronics Corp., Cincinnati, OH.2. P. Salinari, "The TIRGO Observatory," in Proceedings, Second
European Southern Observatory Infrared Workshop, Garching(1982), p. 45.
3. D. Hartill, "A New InSb Charge Amplifier for Application to aSpectrometer Array," in Proceedings, Second European South-ern Observatory Infrared Workshop, Garching (1982), p. 215.
4. D. N. B. Hall, R. S. Aikens, R. Joyce, and T. W. McCurnin,"Johnson Noise Limited Operation of Photovoltaic InSb Detec-tors," Appl. Opt. 14, 450 (1975).
5. J. R. Barton and D. A. Allen, "An Integrating Preamplifier forIndium Antimonide Infrared Detectors," Publ. Astron. Soc. Pac.92, 368 (1980).
6. F. S. Goulding et al., "An Opto-Electronic Feedback Preamplifierfor High-Resolution Nuclear Spectroscopy," Nucl. Instrum.Methods 71, 273 (1969).
7. V. Radeka, "Signal, Noise and Resolution in Position-SensitiveDetectors," IEEE Trans. Nucl. Sci. NS-21, 51 (1974).
