Improvements in Operating the Raytheon 320 # 240 Pixel Si:As Impurity Band Conduction Mid-Infrared Array

Transcription

1 Publications of the Astronomical Society of the Pacific, 115: , 2003 December The Astronomical Society of the Pacific. All rights reserved. Printed in U.S.A. Improvements in Operating the Raytheon 320 # 240 Pixel Si:As Impurity Band Conduction Mid-Infrared Array S. Sako, 1,2 Y. K. Okamoto, 3 H. Kataza, 4 T. Miyata, 5 S. Takubo, 1,2 M. Honda, 1,2 T. Fujiyoshi, 2 T. Onaka, 1 and T. Yamashita 1,2 Received 2003 June 29; accepted 2003 September 12 ABSTRACT. We have developed a new operating method for the Raytheon 320 # 240 pixel Si:As IBC detector (hereafter 320 # 240 FPA), the largest mid-infrared array detector currently available designed for ground-based astronomy. The array shows quite a strange behavior when a bright source is observed. In images that contain a stellar object sufficiently brighter than the background flux, the signal level of the 320 # 240 FPA drops in the vicinity or along the same row or column of the stellar object. The level-drop phenomena are caused mainly by transient variations in the characteristics of the source followers in the detector readout circuit (ROIC). We successfully corrected the variations by correlated quadruple sampling, which reduces the level-drop phenomena appreciably. We evaluated the properties of the ROIC and optimized the bias voltages supplied for the array. The detective quantum efficiency of the 320 # 240 FPA was also estimated. 1. INTRODUCTION The Raytheon 320 # 240 pixel Si:As Impurity Band Conduction (IBC) high-background infrared Focal Plane Array (FPA), hereafter 320 # 240 FPA, is the largest mid-infrared detector currently available designed for ground-based astronomy and has a high quantum efficiency in the wavelength range from 2 to 28 mm (Love, Estrada, & Lum 1999). The 320 # 240 FPA employs a hybrid structure consisting of the CRC- 774 Silicon Cryo-CMOS Read-Out Integrated Circuit (ROIC) interconnected to a backside-illuminated Si:As IBC photodetector. The IBC detector is electrically connected to unit cells of the CRC-774 ROIC by an array of indium bumps. The 320 # 240 FPA is operated at a temperature between 4 and 12 K. The IBC detector consists of an infrared active layer with heavily doped impurities and a blocking layer that is lightly doped. This architecture provides a high quantum efficiency and a low dark current. Several mid-infrared instruments for large telescopes, such as COMICS for the Subaru 8.2 m (Kataza et al. 2000; Okamoto et al. 2003), Michelle for the Gemini North 8.0 m (Glasse, Atad-Ettedgui, & Harris 1997), T-ReCS for the Gemini South 8.0 m (Telesco et al. 1998), and TIMMI-2 for the ESO 3.6 m (Reimann, Weinert, & Wagner 1998), employ the 320 # 240 FPA. 1 Department of Astronomy, School of Science, University of Tokyo, Hongo, Bunkyo, Tokyo, Japan, ; sako@subaru.naoj.org. 2 Subaru Telescope, 650 North A ohoku Place, Hilo, HI; Institute of Physics, Center for Natural Science, Kitasato University, Sagamihara, Kanagawa, Japan, Institute of Space and Astronautical Science, Japan Aerospace Exploration Agency, Yoshinodai, Sagamihara, Kanagawa, Japan, Kiso Observatory Institute of Astronomy, University of Tokyo, Tarusawa, Mikake, Kiso, Nagano, Japan, In this paper, we describe difficulties we encountered during the test observations of COMICS. We also propose a method to reduce them greatly and evaluate its effectiveness. In 2, the principle of ROIC operation is summarized, while in 3, the level-drop phenomena are described in detail and a correlated quadruple sampling (CQS) method to correct them is proposed. Measurements of the source follower (SF) gains in the ROIC are given in 4. In 5, we describe a method to measure the detective quantum efficiency (DQE) of the array and present the results. 2. OPERATIONS OF ROIC The CRC-774 ROIC consists of 320 # 240 unit cells, which correspond to the detector pixels, and the readout circuits of MOS FETs. The ROIC has 16 identical and independent circuits (channels), each of which reads out the detector signals from 20 # 240 unit cells through four source followers (SF1 SF4), two multiplexer circuits that select rows (MUX1s) or columns (MUX2s), and a sample-hold circuit (SH) (see Figs. 1 and 2). The photoelectrons generated by the IBC detector are directly injected to the integration capacitance (C INT )inthe unit cells. At the beginning of the readout process, one of the 240 rows is selected by 20 MUX1s. These 20 signals are stored in the SH capacitances (C SH ) through SF1 and SF2. Then their signals are selected by MUX2s and FET switches sequentially and are read through SF3 and SF4. Figure 3 illustrates the readout sequence on the floor plan. The detector signals of individual channels are sent to 16 V OUT lines simultaneously. We employ a sampling rate of 3 or 5 ms per sample. Common external 28 biases and 16 clocks are supplied to the 16 channels to operate the ROIC. The SF currents for SF1, SF2, and SF3 1407

2 1408 SAKO ET AL. Fig. 1. Schematic diagram of the CRC-774 ROIC (Love et al. 1999). The bias voltages that COMICS adopts are presented in italics. The voltages of the dark and saturation levels are shown with dotted arrows. are supplied through the current mirror circuits controlled by the external biases. 3. LEVEL-DROP PROBLEMS AND CQS 3.1. Level-Drop Phenomena Figure 4 shows sample images of a bright star. Positive and negative stellar images due to chopping within the frame are seen. One would notice several spurious structures along the same row and column of the source in addition to the positive and negative stellar images. These structures are frequently seen for objects sufficiently brighter than the background flux. It is observed not only for pointlike sources but also for bright Fig. 2. Simplified diagram of the CRC-774 ROIC. Fig. 3. Floor plan of the CRC-774 ROIC. The ROIC consists of 16 identical and independent circuits (channels). Each channel has 20 # 240 unit cells. The detector signals are read from the bottom of each channel.

3 IMPROVEMENTS IN OPERATING THE RAYTHEON ARRAY 1409 Fig. 4. Images of bright stars obtained with the normal operation. These images were acquired by COMICS with the secondary-mirror chopping, and the off-beam images are subtracted from the on-beam image. The stellar brightness and the background flux in the left image differ from those of the right image. Four level-drop phenomena are seen in these images; a gradation pattern isolated within a channel that starts from the strong signal to the upper side (pattern 1), a level drop in the same column and row of the strong signal (patterns 2 and 3), and a level drop in the same positions as the strong signal in every channel (pattern 4). The intensities of level-drop phenomena depend on the stellar brightness and the background flux. Fig. 5. Dark image of the 320 # 240 FPA obtained with the normal operation. White dots are hot pixels. The signal level drops slightly at the two diffuse sources as well as in spectroscopic observations. These phenomena can be classified into four patterns: a gradation pattern isolated within a channel that starts from the strong signal to the upper side (pattern 1), a level drop in the same column and row of the bright source (patterns 2 and 3), and a level drop in the same positions as the bright source in every other channel (pattern 4). Patterns 3 and 4 are seen only when the object is brighter than the background. The same phenomena are also seen for a hot pixel in raw images (see Fig. 5). Since the IBC detector is a single-layer photoconductor without an array structure, the phenomenon along the pixels of the detector array should be attributable to the CRC-774 ROIC. This interpretation is supported by the fact that the level-drop phenomena are found for a hot pixel in an image of the CRC- 774 ROICs that is not interconnected to the IBC detector (see Fig. 6). Figure 7 shows an output waveform of the 320 # 240 FPA in dark circumstances. The pixel sampling rate is 5 ms, or 200 khz. The voltage level of the signals should be at a constant dark level. However, Figure 7 shows that the signal level changes after reading out a hot pixel. The signal recovers to the normal level in a few milliseconds, or a few hundred pixels, producing a level-drop image of pattern 1. In Figure 7, one also notices other level-drop phenomena in the next row of the same column of the hot pixel. The signal recovers in a few tens of milliseconds. This produces a level-drop image of pattern 2. A strong signal affects other pixels and causes the leveldrop phenomenon. In other words, an incident photon affects signals over many pixels, producing an electron point-spread function (e-psf). An image is supposed to be the superposition of each e-psf. If we can determine the e-psf with a sufficient accuracy, we can reconstruct the image by deconvolution. We columns on the border between the channels; however, these patterns can be corrected by chopping subtraction. The 320 # 240 FPA is operated at 8 K. attempt to estimate the e-psf by using the signals around hot pixels in the dark image and assuming that the e-psf does not vary with pixels. Hot pixels are considered to have effects similar to strong signals generated by incident photons because they originate from the unit cells. The relationships between the intensity of hot pixels relative to the dark level and the amount of the signal drops are shown in Figure 8. The amount of level drop is approximately proportional to the intensity of the hot pixel, but the plot shows large scatter, indicating that the e-psf is not Fig. 6. Image of the CRC-774 ROIC that is not interconnected to the IBC detector. White dots are hot pixels. Horizontal lines in the top of the image are caused by bad rows. The CRC-774 ROIC is operated at 8 K.

4 1410 SAKO ET AL. Fig. 7. (a) Level-drop phenomena in the output waveform of the 320 # 240 FPA under dark conditions. (b) The magnified figure around a particular leveldrop phenomenon. The pixel data are read every 5 ms. The voltage level of the signals changes after reading out a hot pixel. The resultant signal produces the level-drop image of pattern 1. In this figure, one can also notice other level-drop phenomena in the next row of the same column of the hot pixel. This corresponds to the level-drop image of pattern 2.

5 IMPROVEMENTS IN OPERATING THE RAYTHEON ARRAY 1411 of the hot pixel drops. Therefore, the level-drop phenomenon of pattern 2 must be related to the transient variation in the characteristics of SF2 and/or SF3, when the input signals to SF2 and SF3 are largely changed. The characteristics of SF1 may be affected by the drastic variation in the input signal as SF2, SF3, and SF4. However, the characteristic variations should have no effects unless observing sources fluctuate at a high frequency with a large amplitude because the input line of each SF1 is connected to a single detector pixel. Fig. 8. Plot between the intensity of hot pixels relative to the dark level and the amount of signal drop. stable. The deconvolution with the average e-psf will introduce a significant uncertainty. Therefore, to resolve this problem, it is required to investigate the mechanism of the leveldrop phenomenon and improve the operation of the 320 # 240 FPA Mechanism of Level-Drop Phenomena The level-drop phenomena of patterns 3 and 4 occur in all the channels simultaneously. The signal line of a channel is not connected to other channels, but the bias lines are shared with all the channels. Thus, we considered patterns 3 and 4 to be caused by variations in the bias voltages when the input signals are drastically changed. The level-drop phenomenon of pattern 1 continuously occurs even when the column multiplexer (MUX2) is switched to another line. Thus, it is inferred that this phenomenon originates from SF4. Figure 9 shows the output waveform in the case that the pixel signals are read out alternately with the bias voltage of V CLOUT. The bias voltage is read out by turning on pclout and connecting the bias line to the gate of SF4. After a hot pixel, the readout level of V CLOUT drops. Since a constant voltage is supplied to the bias line, the level-drop phenomenon of pattern 1 can be attributed to the transient variation in the characteristics (such as the gain or the offset) of SF4 by electric charging or thermal heating, when the input signal to the SF4 is drastically changed. The level-drop phenomenon of pattern 2 is considered to arise from SF2 and/or SF3 because this phenomenon occurs only in a certain column. Figure 10 shows the output waveform when 20 pixel signals in a row are read out and then the voltage on V CLCOL bias is read out through the same SFs for the 20 signals. The bias voltage is read out by turning on pclcol and connecting the bias line to the gate of SF2. When a hot pixel is read out, the voltage level of V CLCOL in the same column 3.3. Correlated Quadruple Sampling (CQS) Here we propose a new readout method for the 320 # 240 FPA to reduce the level-drop phenomena. As mentioned in 3.2, the level-drop phenomena of patterns 1 and 2 are due to the transient variations in either the gain or the offset of the SFs. The characteristic variations can be monitored by referring to constant voltages in the ROIC through the SFs and corrected. In order to correct the level-drop phenomenon of pattern 1, we devised a scheme to read out the signal referring to the constant voltage of the V CLOUT bias (Ref1). Figure 11 shows a dark image produced by the difference between the signal and Ref1 (signal Ref1), which are read out alternately as described in 3.2. The figure clearly shows that the level-drop phenomenon of pattern 1 completely disappears. These results indicate that the prime cause of the level-drop phenomenon is likely to be the offset variation and that the characteristic timescale of the transient variation is much longer than the pixel readout rate. 6 To correct the level-drop phenomenon of pattern 2, we read out the signal referring to the constant voltage of the V CLCOL bias (Ref2). Figure 12 shows a dark image produced by the difference between the signal and Ref2 (signal Ref2). Again, it can be seen that the level-drop phenomenon of pattern 2 also completely disappears and that this method is very effective in the correction for pattern 2. Combining both methods, one can correct the level-drop phenomena of patterns 1 and 2 at once. Figure 13 shows the output waveform when these two methods are simultaneously applied. Four kinds of signal the pixel signal (signal), the V CLCOL reference (Ref2), the V CLOUT reference for the pixel signal (Ref1a), and the V CLOUT reference for the V CLCOL reference (Ref1b) are needed to correct both level-drop phenomena. We call this scheme correlated quadruple sampling (CQS). A dark image obtained by (signal Ref1a) (Ref2 Ref1b) is shown in Figure 14. The CQS is fairly successful in correcting both patterns 1 and 2. Figure 15 shows the image of a pointlike source taken with the CQS, in which the level-drop phenomena of patterns 3 and 4 are not seen because of the faintness of the object. Comparison with Figure 4 clearly demonstrates the effectiveness of the CQS. 6 The offset variation of an SF is generally smaller than its gain variation.

6 1412 SAKO ET AL. Fig. 9. (a) Output waveform of the case that the pixel signals are read out alternately with the bias voltage of V CLOUT.(b) Magnified figure around level-drop phenomenon. The bias voltage is read out by turning on pclout and connecting the bias line to the gate of SF4. After a hot pixel, the readout level of V CLOUT drops.

7 IMPROVEMENTS IN OPERATING THE RAYTHEON ARRAY 1413 Fig. 10. Output waveform of the case that 20 pixel signals in a row are read out and then the voltage on V CLCOL bias is read out through the same SFs for the 20 signals. The bias voltage is read out by turning on pclcol and connecting the bias line to the gate of SF2. When a hot pixel in the 13th column is read out, the voltage level of V CLCOL on the same column of the hot pixel, 13th column, drops Restriction of Operation The level-drop phenomena of patterns 3 and 4 cannot be corrected only by the CQS (see Fig. 16, top panel) because they are related to variations in the bias voltages. The intensities of patterns 3 and 4 are a few tenths of a percent of that of the stellar object. When the object is significantly brighter than the background flux, the level-drop phenomena become recognizable. In order to reduce patterns 3 and 4, analytical corrections are required in addition to the CQS. The signal levels of patterns 3 and 4 drop by an equal amount in every channel; i.e., the pattern is common to all the channels. They can be corrected Fig. 12. Dark image obtained with the correlated double sampling (signal Ref2). The level-drop phenomenon of pattern 2 completely disappears. for by subtracting an image of a certain channel without any astronomical sources. In an actual application to astronomical data, we obtain improved images by subtracting the median channel image of those without noticeable astronomical sources (see Fig. 16, bottom panel). This result indicates that the prime cause of patterns 3 and 4 is also the offset variation in the ROIC. One should be aware that it is not easy to correct patterns 3 and 4 in images of extended bright sources and spectroscopic data of bright sources. To make a correction for the level-drop phenomena in these observations, at least a single-channel image without sources is needed. The CQS greatly reduces the level-drop phenomena, but it introduces a new limitation in the operation. In the CQS, the amount of data produced becomes 4 times larger and so does Fig. 11. Dark image obtained with the correlated double sampling (signal Ref1). The level-drop phenomenon of pattern 1 completely disappears. Fig. 13. Output waveform of monitoring both references (CQS). After the sequence (signal Ref1) for a certain row is executed, (Ref2 Ref1) is performed for the same row.

8 1414 SAKO ET AL. Fig. 14. Dark image obtained with the CQS (signal Ref1a) (Ref2 Ref1b). The level-drop phenomena of both patterns 1 and 2 completely disappear. the data-taking time. If raw data are co-added in the memory on the array controller, the increased data size is not a serious problem (Sako et al. 2003). The period for a single readout depends on the response speed of the ROIC. When the pixel signals are read out at a rate faster than 3 ms, the readout noise increases significantly. Thus, the shortest period for a single readout should be 3 ms, which results in a frame rate of greater than 60 ms with the CQS. The frame rate is equivalent to the minimum integration time in the integrate-while-read mode 7 that we use. In the Q-band (17 24 mm) imaging observations, the integration time should be less than a few tens of milliseconds to prevent detectors from saturation because of the high background radiation. To achieve an integration time 7 The integrate-while-read mode is one of the operation methods of the 320 # 240 FPA. In this mode, the readout of a pixel is carried out while integrating other pixels. Fig. 16. Top panel: Level-drop phenomenon of pattern 3 in the image of a bright pointlike source taken with the CQS and the partial readout operation of 120 rows. The intensity of the level-drop phenomenon is 0.3% of that of the stellar object. Bottom panel: Improved image by subtracting the median channel image of those without noticeable astronomical sources from the upper panel image. shorter than the frame rate, the partial readout operation 8 or the electric ND operation may be required. 9 In addition, the CQS enhances the readout noise because the final signal is derived from four sets of data. In our case, the CQS-enhanced readout noise is 1.7 times, but not 2 times as expected because the offset variations in the bias voltages may be corrected by the CQS. The enhancement of noise does not 8 We read out a part of array by skipping readout sequences of certain rows to shorten the integration time. 9 The electric ND operation shortens the integration time by increasing the reset time of the C INT. Fig. 15. Image of a bright pointlike source taken with the CQS. Comparison with Fig. 4 clearly demonstrates the effectiveness of the CQS. Patterns 3 and 4 are not seen in this image because of the faintness of the object. Fig. 17. Relationship between V CLOUT and V OUT measured by turning on pclout-switch. These data are best fitted by a linear function V OUT (V) p 0.902V CLOUT 1.27.

9 IMPROVEMENTS IN OPERATING THE RAYTHEON ARRAY 1415 Fig. 18. Relationship between V SHPCHG and V OUT measured with psh switched off and pshpchg on; V CLOUT and V OFFSET biases were set to 4.6 and 2.8 V, respectively, at the measurement. These data are best fitted by a linear function V OUT (V) p 0.688V SHPCHG affect imaging or low-dispersion spectroscopic observations, in which noises are dominated by the background radiation. In high-dispersion spectroscopic observations, the enhanced readout noise slightly reduces the sensitivity. 4. ANALYSIS OF ROIC INTERNAL CIRCUITS The transimpedance 10 is one of the most fundamental properties for an infrared detector and is deeply related to the detectivity. It is generally derived from the ratio of the signal to the square of the shot noise. However, the IBC detector has an excess noise due to the gain dispersion as described in 5. Because one cannot distinguish between shot and excess noises, the standard method gives us an incorrect value for the transimpedance. Here we employ an alternative method to measure the SF characteristics with the assumed capacitance of C INT Deriving Transimpedance of the CRC-774 ROIC Important characteristics of the SF for the transimpedance measurement are the gain and the threshold voltage, which determines the offset level of the output signal. These characteristics are determined from the response of the output signal to the input signal of the SF. 11 For SF4, we measure the response of the output voltage of the ROIC (V OUT )tov CLOUT bias by turning on pclout-switch (cf. Figs. 1 and 2). Figure 17 shows the relationship between V CLOUT and V OUT. These data are best fitted by a linear function V OUT (V) p 0.902V CLOUT 1.27, and thus the gain (A SF4 ) and the threshold voltage (V th4 ) of SF4 are derived as and 1.27 V, respectively. 10 The transimpedance corresponds to the conversion factor between output voltage and generated electrons. 11 Our experiments are carried out at a temperature of 9 K. Fig. 19. Relationship between V OFFSET and V OUT measured with the psh switched off. Although V OUT should be inversely proportional to V OFFSET, the relationship breaks down at V OFFSET 1 V th 2.5 V. In addition, the threshold voltage depends on the V PW bias. These data are best fitted by a linear function V OUT (V) p 0.728V OFFSET 3.98 for V OFFSET! 2.5 V. In a similar way, we apply the V SHPCHG bias as an input signal for SF3 with psh switched off and pshpchg on in order to obtain the characteristics of SF3. The relationship between V SHPCHG and V OUT is shown in Figure 18, and the data are best fitted by a linear function V OUT (V) p 0.688V SHPCHG 0.125; V CLOUT and V OFFSET biases were set to 4.6 and 2.8 V, respectively, at the measurement. As described in 4.2.2, the relationship between the gain (A SF3 ), the threshold voltage (V th3 ) of SF3, and V OUT is expressed as V p A {A [(A V V ) V ] OUT SF4 LS SF3 SHPCHG th3 OFFSET V CLOUT } V th4 p 0.728(ASF3VSHPCHG V th3) 0.841, (1) where A LS is the gain of the level shifter of the ROIC. Therefore, A SF3 and V th3 are derived as and 1.33 V, respectively, from the fitting result. To obtain the characteristics of SF2, one should apply the V CLCOL bias as an input signal for SF2 with pclcol switched on. However, we could not control the V CLCOL bias independently because our system operates both the V CLCOL bias and the fixed V SS bias by a common bias line. We assume that the threshold voltage of SF2 (V th2 ) is 1.30 V, a mean value between V th3 and V th4. The relationship between the gain of SF2 (A SF2 ), V th2, and V OUT is described as V p A (A {[A (A V V ) V ] OUT SF4 LS SF3 SF2 CLCOL th2 th3 V } V ) V OFFSET CLOUT th4 p 0.688(A V V ) (2) SF2 CLCOL th2

10 1416 SAKO ET AL. Fig. 20. Simple diagram of the level shifter: Cgs and Cgd represent the potential capacitances between the gate and the source and one between the gate and the drain, respectively. When pclcol-switch was turned on and V CLCOL was set to 5.50 V, V OUT indicated 2.51 V. Using equation (2), A SF2 is calculated to be The derived gain is nearly equal to the gain of SF3 or SF4 and is consistent with the SF gain value shown in the documentation of the manufacturers (Love et al. 1999). The ROIC has clamp capacitors between SF1 and SF2 for correlated double sampling (CDS). Since the clamp circuit transfers the voltage variation of C INT charged by photoelectrons to SF2, one cannot use V RSTUC 12 bias as an input signal of SF1 to measure the characteristics. Thus, we assume that the gain of SF1 (A SF1 ) and the threshold voltage of SF1 (V th1 ) are and 1.30 V, respectively, which are mean values of those of SF2, SF3, and SF4, because of difficulties in measuring the SF characteristics. The combined gain of four SFs (A SF )is expressed as A p A A A A A (3) SF SF1 SF2 SF3 LS SF4 p Assuming that the capacitance of C INT is 1.85 pf, 13 the transimpedance of the detector (Z det ) is given by where q is the electron charge. q Z p A (4) det SF C INT 8 p 5.15 # 10 V/e, 12 V RSTUC corresponds to the reset voltage of the integration capacitor (C INT ). 13 We operate the detector with V cap p 0.0 V, so the capacitance of C INT should be set to 1.85 pf according to the document provided by the manufacturer (Love et al. 1999). We have not confirm this capacitance experimentally. Fig. 21. Plot between the square of the noise and the signal that was obtained by measuring a radiation source. The slope of solid line corresponds to Z det bg. The dashed line indicates the relationship in the case of bg p Operational Limitation of Level Shifter V OFFSET FET Switch The level-shifter circuit in the ROIC between SF3 and SF4 shifts the voltage level of the input signal by the voltage difference between V CLOUT and V OFFSET. We recognize that the level shifter does not work in a certain setting. When we operate the level shifter with the psh switched off, we can measure the response of V OUT to V OFFSET bias. Figure 19 shows the relationship between V OUT and V OFFSET in such an operation. Although V OUT should be inversely proportional to V OFFSET, the relationship breaks down at V OFFSET 1 V th 2.5 V. In addition, the threshold voltage depends on the V PW bias. The pcloutswitch cannot be turned off completely for V OFFSET 1 V th. Therefore, V OFFSET must be set to less than 2.5VatV PW p 6.0 V. We operate the detector with V OFFSET p 2.8 V Gain of Level Shifter An ideal level-shifter circuit has a gain of unity, whereas that of the ROIC does not. A MOS FET generally has potential capacitances, one between the gate and the source ( C gs ) and one between the gate and the drain ( C gd ) (see Fig. 20). The gain generated by the level shifter with C and C is expressed as C LS gs A LS p, (5) C C iss where C LS is the capacitance of the level shifter capacitor and C iss { Cgs Cgd. The relationship between the input voltage of LS LS the level shifter ( VIN ) and the output voltage ( VOUT) corresponding to the input voltage of SF4 is LS LS LS VOUT p A LS(VIN V OFFSET) V CLOUT. (6) gd

11 IMPROVEMENTS IN OPERATING THE RAYTHEON ARRAY 1417 TABLE 1 Characteristics of the CRC-774 ROIC Name Value Description A SF SF1 gain a A SF SF2 gain A SF SF3 gain A LS Level-shifter gain A SF SF4 gain A SF Total gain p A SF1 A SF2 A SF3 A LS A SF4 V th1 (V) SF1 threshold voltage b V th2 (V) SF2 threshold voltage b V th3 (V) SF3 threshold voltage V th4 (V) SF4 threshold voltage Z det (V/e ) # 10 8 Transimpedance c Note. The detector was kept at a temperature of 9 K in the measurement. a Mean value between A SF2, A SF3, and A SF4. b Mean value between V th3 and V th4. c We assumed that the capacitance of C INT is 1.85 pf. Fig. 23. Clock patterns for the partial readout method with CQS. The data in Figure 19 for V OFFSET! 2.5 V can be fitted by a linear function V OUT (V) p 0.728V OFFSET Then A LS is derived to be from the fit since its slope corresponds to A LS A SF4. The fact that A LS is not unity suggests that C iss is not sufficiently smaller than C LS. The clamp circuit between SF1 and SF2 is also subject to a similar gain mechanism, but we did not evaluate it experimentally because one cannot set the voltage of the clamp circuit capacitor to an arbitrary value. The characteristics of each SF and the optimized biases are summarized in Tables 1 and 2, respectively. Fig. 22. Clock patterns for CQS. 5. MEASUREMENTS OF DQE The IBC detector has a photoconductive gain G larger than unity. The gain process is noisy because the statistics of the interactions determines the gain. Therefore, the detector noise exceeds the shot noise estimated from the photoelectron number. We should evaluate effects of the excess noise on the sensitivity to estimate the detection limit. Using the gain dispersion 2 2, the ratio of signal to noise in observations where b p AG S/AGS the background limits the detection is expressed as signal NthGZ s t h p p N s, (7) noise N thbgz N b bk where N s and N bk are the incident photon numbers per second from the source and the background radiation, respectively, t is the integration time, h is the quantum efficiency, and Z is the system transimpedance. We define the detective quantum TABLE 2 Optimized Biases Name Value Unit V SSCLK V V SS /V CLCOL V V DETGRV V V NEG V V DETGRDRG V V GATE V V OFFSET V V CASUC V V RSTUC V V PW /V SSD V V SHPCHG V V CLOUT V V CAP V I MUX ma I GG ma I GG ma I GG ma Note. The detector was kept at a temperature of 9 K in the measurement. bk

12 1418 SAKO ET AL. efficiency (DQE) as h DQE p. (8) b The DQE represents the effective quantum efficiency in observations. We can experimentally obtain bg from 2 noise p ZbG. (9) signal Figure 21 shows the relationships between the square of the noise and the signal obtained by measuring a radiation source with Ns p Nbk. The inclination of the function corresponds to ZbG. Thus, using Z det of equation (4) we derive bg p (10) The conversion factor between the generated electron and the incident photon is given by hg. From the measurement of a radiation source of the known flux, we obtained the conversion factor as hg (11) Therefore, the DQE of the IBC detector is estimated from equations (10) and (11) as DQE (12) 6. SUMMARY We proposed the CQS method to reduce the level-drop phenomena of the 320 # 240 FPA. When a strong signal is read out, the offset characteristics of SFs vary transiently. In the CQS, the offset variations are monitored and corrected by referring to the fixed bias voltages. The CQS provides significant improvements in the actual observations. We also estimated the characteristics of SFs and optimized the biases. The DQE was derived using the SF characteristics. This work was supported by the Subaru project. We would like to thank all the members of the Subaru telescope and the staff at the Advanced Technology Center of the National Astronomical Observatory of Japan. We also thank K. Ando and the staff at Raytheon Systems for useful discussions. APPENDIX The clock patterns for the CQS and for partial readout method are shown in Figures 22 and 23, respectively. The sampling rate of presented clock patterns is 5 ms per sample, or 200 khz. The partial readout method is provided by skipping readout sequences for certain rows. REFERENCES Glasse, A. C., Atad-Ettedgui, E. I., & Harris, J. W. 1997, Proc. SPIE, 2871, 1197 Kataza, H., Okamoto, Y., Takubo, S., Onaka, T., Sako, S., Nakamura, K., Miyata, T., & Yamashita, T. 2000, Proc. SPIE, 4008, 1144 Love, P., Estrada, A., & Lum, N. 1999, 320#240 Si:As IBC High- Background Astronomy IRFPA: User s Guide and Operating Manual (Lexington: Raytheon) Okamoto, Y. K., Kataza, H., Yamashita, T., Miyata, T., Sako, S., Takubo, S., Honda, M., & Onaka, T. 2003, Proc. SPIE, 4841, 169 Reimann, H., Weinert, U., & Wagner, S. 1998, Proc. SPIE, 3354, 865 Sako, S., Kataza, H., Miyata, T., Okamoto, Y. K., Takubo, S., Honda, M., Onaka, T., & Yamashita, T. 2003, Proc. SPIE, 4841, 1211 Telesco, C. M., Pina, R. K., Hanna, K. T., Julian, J. A., Hon, D. B., & Kisko, T. 1998, Proc. SPIE, 3354, 534