Introduction

Near infrared focal plane technology has developed rapidly during the past decade. Detectors having more than one million pixels are now in routine operation at 8 meter class telescopes. In parallel to the increased array format both the readout noise and the detector darkcurrent have been reduced by more than one order of magnitude. The combination of a powerful 8 meter telescope and state of the art detector technology largely expands the scope for astronomical observations in the infrared.

In November 1998 the Infrared Spectrometer and Array Camera ISAAC was installed at Antu, the first of the four the VLT telescopes. Initially, the short wavelength arm was equipped with a Hawaii 1024x1024 HgCdTe array and the long wavelength arm with a 256x256 InSb array. In February 2000 this array was replaced by the Aladdin 1024x1024 InSb array. The Aladdin array mounted in ISAAC was the outcome of a best effort contract with Raytheon to manufacture 6 Aladdin arrays .

At present the second generation of VLT instruments is being built with an ever increasing demand for infrared arrays. The cryogenic echelle spectrometer CRIRES will be equipped with a mosaic of three Aladdin arrays and the multiobject spectrograph NIRMOS will house four 2Kx2K HgCdTe arrays. In order to meet this demand, ESO is participating in funding the development of the LPE 2Kx2K HgCdTe arrays having a cutoff wavelength of lc=2.5 mm . These arrays are based on the 1Kx1K PACE1 technology , . Furthermore, a contract was placed for the manufacture of four MBE 2Kx2K HgCdTe arrays having a cutoff wavelength of 1.9 mm. These arrays will be mounted in the multiobject spectrograph NIRMOS.

In order to handle the preprocessing of a continuous data rate of 40 megabytes/sec generated by 32 parallel video channels of the Aladdin array, we had to develop the high speed data acquisition system called IRACE , . The flexible architecture of the IRACE system is well suited to be expanded to a 128 channel system needed to read out the four 2Kx2K NIRMOS arrays each having 32 parallel outputs. A prototype of the 128 channel system is operational.

HgCdTe arrays

For medium resolution spectroscopy in J and H, the dominant radiation source is the bright line emission of the OH air glow. In between the OH lines the remaining continuum emission can be as low as 753 phot. s-1arscsec-2m-2mm-1 generating a detector signal of 0.17 e/sec/pixel as measured with ISAAC and elsewhere . In principle the detector integration time can be increased until the photon shot noise exceeds the detector readout noise. In practice however, for integration times longer than a few hundred seconds, warm pixels, array nonuniformities and cosmic rays degrade the overall performance. For this reason the shot noise limit should be reached at the shortest possible integration time. This requires a detector read noise of well below 10 electrons rms. However, the dominant source of read noise is the source follower in the unit cell of the HAWAII multiplexer. It contributes a noise level of ~10 electrons rms with double correlated sampling which has to be applied to eliminate the KTC noise. Further improvement of the readout noise relies on multiple sampling techniques.

Sampling strategies for multiple nondestructive readouts

Two sampling methods are in use, Fowler sampling and Follow-Up-the-Ramp sampling (FUR) . With n being the number of nondestructive readouts and SNRDC being the signal to noise ratio for a simple double correlated clamp, the improvement of the signal to noise ratio for Follow-Up-the-Ramp sampling SNRFUR is described by equation 1.

Multiple Sampling with Cryogenic Operational Amplifiers

In an ongoing effort to reduce the readout noise we have tried different readout schemes for the Hawaii 1024x1024 MCT array. The readout multiplexer has an on-chip output source follower for each quadrant to which the internal video signal bus is connected. The user has also direct access to the internal signal bus. The bus can be connected to +5 volt by an external 200KW load resistor. The internal bus is directly connected to the input of an external operational amplifier, located as close as possible to the detector signal pins for maximum immunity to noise pickup. Therefore the amplifiers have to operate at cryogenic temperatures. A linear CMOS operational amplifier was selected (Texas Instruments LinCMOS TLC2274) which contributes ~ 3 electrons rms to the readout noise of a double correlated clamp. The design of differential data line drivers using these cryogenic CMOS amplifiers is shown in and has been described in more detail elsewhere . In this configuration the on-chip output source follower is not used. Hence, its glow is eliminated, when multiple sampling is applied to reduce the readout noise. This is one of the biggest advantages of off-chip amplifiers.

Figure shows the reduction of readout noise by Follow-up-the-ramp sampling with multiple nondestructive readouts. After the array is reset it is read out nondestructively at a frame rate of 1 Hz. The number of readouts is proportional to the integration time. Each frame samples the integration ramp and improves the accuracy of the regressional fit of the integration ramp. The measured readout noise indicated by circles in Figure was computed for each pixel from a sequence of frames taken under identical conditions. The data points show the peak of the noise histogram of the corresponding noise image. The fit is performed in real time during integration. In principle, the readout noise can be reduced to negligible amounts as described by . The expected noise reduction by multiple sampling is shown by the curve decreasing with time. The other curve, increasing with time, is due to some kind of shot noise. In an earlier publication this shot noise was erroneously attributed to the detector darkcurrent . However, careful darkcurrent measurements carried out by sampling the integration ramp every 15 minutes during a total stare time of 3 hours at a detector temperature of 65 K yielded a darkcurrent of 1.3 10-2 electrons/s as shown in Figure . The shot noise of the detector darkcurrent is too small to explain the increase of readout noise encountered for stare times longer than one minute with more than 64 nondestructive readouts as shown by the measured data points in Figure .

 

Other sources for this excess noise had to be investigated such as the multiplexer glow shown in which shows a raw image taken with 512 nondestructive readouts. In the center of the top and bottom edges and in the top and bottom right corners there are localized, bright radiation sources. The radiation extends but is attenuated towards the center of the array. Since we do not use the on-chip source follower it cannot contribute to the multiplexer glow. The center of the glow is localized at the end of the shift registers. It depends linearly on the number of nondestructive readouts as shown in Figures and . From this we conclude, that the glow is caused by the shift registers of the multiplexer.

In Figure the intensity of the glow is plotted versus the number of readouts for the area at the bottom edge of the array close to the radiation source, indicated by the small rectangle in Figure . The glow is 49 electrons per pixel per nondestructive readout of the full frame. In Figure the same curve is shown for the dark region far from the radiation source as indicated by the large rectangle in Figure . In this region the glow drops to 0.20 electrons per pixel per nondestructive readout. The four glow centers sitting on the Hawaii chip have to be masked in order to avoid ghost images reflected onto the detector by optical surfaces of the instrument.

The measured glow value of 0.2 electrons/pixel/readout for the center of the array was used to calculate the shot noise of the shift register glow. It is represented by the increasing curve in Figure . The measured readout noise fits well to the predicted noise which consists of two contributions, the readout noise of Follow-up-the-ramp sampling and the shot noise of shift register glow.

 

To further investigate the nature of shift register glow, a bare Hawaii multiplexer was imaged onto a CCD while in operation and performing multiple readouts. The magnification of the multiplexer image formed on the CCD was 1.087 and the focal ratio at the Hawaii multiplexer was f/5.37. The CCD was a 2Kx4K EEV CCD having a gain of 2.8 electrons/ADU and a pixel size of 15 mm. Figure shows an optical CCD image of the glow centers at the end of the shift registers of the Hawaii multiplexer. The separation of the glow centers is 19.565 millimeters. The points are 310 mm away from the edge of the sensitive area of the detector. The quantum efficiency of the EEV CCD drops from 0.1 at a wavelength of l=0.95 mm to 0.04 at l=1 mm. The exposure time of the CCD was 3 hours.

The Hawaii multiplexer was continuously read out during each 3-hour CCD exposure. For each exposure the readout speed was changed. In this way the number of multiplexer readouts per CCD exposure was varied from 1082 readouts to 17312 readouts per CCD exposure. In Figure the intensity of the glow centers at the end of the Hawaii shift registers is plotted versus the number of full frame readouts of the multiplexer. The signal is scaled to 2p steradians assuming an average quantum efficiency of 0.1 for the EEV CCD. The electroluminescence depends linearly on the number of multiplexer readouts. Each glow center emits 324 photons per full frame.

 

 

 

Filter for Video Signal by Subpixel Sampling

The shift register glow limits the possibility to reduce the readout noise of the Hawaii detector by increasing the number of nondestructive readouts. As can be seen in Figure the readout noise has its minimum at 6 electrons rms with 64 nondestructive readouts. Even in the center of the array far from the glow centers the shot noise of shift register glow becomes the dominant noise source, if more nondestructive readouts are used. An alternative method for improving the noise statistics is to use more than one ADC conversion to sample the video signal of each individual pixel in the time interval during which this pixel is addressed. In this case the time interval spent to read out one pixel will be much longer than the time constant of the analog low pass filter. In the most simple case the pixel value is the digital average of all ADC conversions. In effect, subpixel sampling constitutes a digital low pass filter of the video signal. Since off-chip amplifiers are used, no additional glow is generated by slowing down the frame rate and increasing the pixel time to apply the subpixel sampling. The noise statistics is improved without additional clocking of the shift registers and without additional multiplexer glow.

For low flux spectroscopic applications the detector integration times will be 600 seconds or longer. Since the number of readouts has to be limited to values n ~ 64 and a nondestructive readout takes ~ 1 second, the condition
Tint >>nTread holds. In this case Fowler sampling is better than Follow-up-the-ramp sampling by a factor of as discussed in chapter 2.1. The best performance with Hawaii arrays can be achieved by combining the subpixel sampling and Fowler sampling. For full frames the readout noise can be as low as 3.6 electrons rms.

Reset Anomaly

After the integrating node capacity of the detector is reset, the detector signal is a strongly nonlinear function of time. The data in Figure show the detector signal taken in the double correlated sampling mode as function of integration time observed for two different flux levels.

The triangles represent dark frames and the diamonds represent H-band images. The integration ramp is a nonlinear function of time. Fortunately, the difference between the two flux levels represented by squares, is linear.For this reason it is mandatory to subtract a dark frame from the science exposure and both exposures must have exactly the same integration time. After reset the integration ramp needs more than 30 seconds to stabilize and become linear. Because the initial nonlinear part of the integration ramp is unstable, it is the origin of several image blemishes like row crosstalk, the dc-gradient at the beginning of the readout being seen as a bright bar in the center of the array, and extra noise.

If flux levels are as low as a few photons per second, the detector can be operated for hours without resetting the array. The array can be operated as usual while the reset clock is switched off. In this way the reset anomaly is eliminated and extremely uniform and deep images can be obtained with raw images. Of course the array has to be reset, before it saturates. Depending on the application part of the array may be allowed to saturate. For instance, sky lines may saturate without affecting the clear spectral regions on the array.

In the pixel to pixel noise is compared for different readout modes. In the left column of Table 1 the rms noise of the pixel to pixel variation of raw frames is calculated for pixels within the rectangle displayed in the images of . The right column shows the pixel to pixel variation calculated from the difference between two successive frames normalized to the rms noise of a single readout by multiplying the noise with a factor of . It is remarkable, that resetless operation of the Hawaii array yields the same pixel to pixel noise for the raw image and for the difference between two images. Furthermore, the raw image is blemish-free as can be seen in the right image of . The two images in have been taken by imaging a grid of pinholes onto the detector with a focus wedge inserted into the collimated beam. The focus wedge produces 5 images for each pinhole.

For comparison, the left image shown in was obtained by Follow-up-the-ramp sampling with 32 nondestructive readouts and 1 sample / pixel. In the center of this image a bright horizontal stripe can be seen at the quadrant borders. It is caused by the readout topology of the array. Two quadrants have their first rows being read out in the center of the array. The stripe of increased intensity at the beginning of the readout is unstable and a region of increased noise. We tried different clocking schemes to reduce the effect. Best results were obtained by clocking the fast shift register without application of the frame start pulse. The number of dummy clock cycles is equivalent to the readout of 64 detector rows.

InSb arrays

In 1996 ESO placed a contract with SBRC to produce six 1024x1024 InSb arrays on a best effort basis , . All six arrays have been delivered and the performance of all arrays has been evaluated.The delivery of the sensor chip assemblies took longer than anticipated due to problems with hybridized arrays. Some of the arrays display Photon Emitting Defects (PED's). These areas of excessive glow are caused by a current flowing between the detector and the multiplexer. They only show up after hybridization and are not intrinsic to the multiplexer. A technique has been developed at Raytheon to remove these PED's by a photolithographic etching technique minimizing the loss of pixels to 2x2. With the same etching technique cracks in the InSb material can also be terminated. The InSb diode arrays are thinned to the required thickness. Since the InSb is thinned, it behaves like a rubber band and compensates the thermal mismatch between InSb and silicon during cooldown.

Two Aladdin arrays out of the lot are science grade and have been mounted in the VLT instruments ISAAC and CONICA. shows the Aladdin array installed in the ISAAC detector mount with flexible copper boards for connections to biases, clocks and video outputs. The detector is thermally insulated by four flexible manganin boards inside the aluminum ring. The detector operates at 30 K, the detector board at 70 K. The temperature gradient is maintained by the manganin boards due to their high thermal impedance.

Dark current

One of the arrays has cosmetic defects. Several cracks in the InSb material show up as dead lines in the array. However, this array has the lowest darkcurrent ever reported for InSb. The darkcurrent measured with a newly developed technique is 4 10-3 e/s at an operating temperature of 25 K.

At these darkcurrent levels the detector signal is dominated by changes of the dc level of the video signal caused by temperature fluctuations. Dead pixels in the corner of the array can be used to monitor this drift. These pixels consist of multiplexer unit cells which are not connected to an infrared diode.

 

Their signal represents an ideal monitor for the temperature dependence of the complete acquisition chain including the Si multiplexer. The temperature dependence of reference pixels and pixels in the central part of the array used to measure the darkcurrent may be different as shown by the plots in . The intensity of uncorrelated samples is plotted versus temperature for two different areas on an InSb array. The temperature dependence of dead reference pixels in upper right corner of array is br=1.7 103 e /K as shown in the right plot of , whereas the temperature dependence of pixels in the central area of the array, which are used for the darkcurrent measurement, is bv=2.4 103 e /K. If the signal of the reference area is subtracted from the video signal by taking into account the factor of bv/br, which is a smoothly varying function of position within the array, the remaining signal of the infrared diode is decontaminated from all effects masking the true signal of the infrared sensor.

shows the integration ramp of a dark exposure. The raw signal is represented by triangles and the corrected signal is represented by squares. The corrected signal exhibits a linear integration ramp corresponding to a darkcurrent of 4 10-3 e/s/pixel. If this monitoring technique is not applied, darkcurrent measurements are masked by thermal drifts. The deviation of the measured integration ramp (triangles) from the corrected integration ramp (squares) was caused by a temperature drift of 8 10-3 K/hour. Since the temperature sensor used for stabilizing the detector temperature is not on the sensor chip assembly but mounted off-chip close to the detector, drifts could not be eliminated completely, even though an active control loop was stabilizing the detector temperature.

 

 

 

Readout Noise and quantum efficiency

The Aladdin array used for this measurement was hybridized to an Aladdin 2 multiplexer. The dependence of readout noise on the number of nondestructive readouts is shown in . The integration ramp was measured by continuously applying multiple nondestructive readouts. The frame time for a nondestructive readout is 42.6 ms. The integration time is proportional to the number of nondestructive readouts. The noise was calculated from a series of dark exposures as explained above for the Hawaii array. The expected noise reduction by multiple sampling is shown by the curve decreasing with stare time. The curve increasing with time is due to shot noise of multiplexer glow and darkcurrent. The measured data points well match the combined noise (read & shot noise). With 128 nondestructive readouts the readout noise can be reduced from 70 erms to 9.7 erms.

Meanwhile, the multiplexer design has been revised and the Aladdin 3 multiplexer is reported to have a noise performance which is improved by a factor of 1.5.

We also tried to use cryogenic operational amplifiers next to the detector. Initial tests were carried out with a 256x256 InSb array. In our setup the power dissipation of the cryogenic amplifiers raised their temperature to 114 K. As a consequence, the detector signal generated by the thermal radiation of the amplifiers was ~200 e/sec. The cooling of the amplifiers and the optical screening of the detector from thermal radiation of the cryogenic amplifiers must be improved, if cryogenic amplifiers are to be used with InSb detectors.

The measured quantum efficiency of the Aladdin arrays is high for the complete spectral range and summarized in table 2

 

.

Quantum efficiency of 1024x1024 InSb array

band	J	H	K	L	M
quantum efficiency	0.89	0.73	0.88	0.74	0.81

Crosstalk and Readout Speed

To cope with the high thermal background emitted by the telescope and the atmosphere for broad band imaging in the L and M band, the Aladdin array has to be read out at the maximum speed. The array has 32 parallel video outputs. The ESO data acquisition system IRACE is using 16 bit 2 MHz ADC's for each of the 32 video outputs and can process in real time 30 1Kx1K frames/sec , . However, the readout speed is limited by the analog bandwidth of the Aladdin 2 multiplexer. To boost the analog bandwidth, the current in the output stage was raised by increasing Vload from 0 V to 5 V and by reducing the load resistor from 100 KW to 20 KW. Vref was decreased from -2 V to -4.5V in order to increase the current reference for the Islew current mirror on the column bus. By implementing these modifications we could reduce the rise time of the video signal to 556 ns. With this rise time a complete frame of can be read out in 36 ms. Yet, there is an additional overhead of 11.5 ms needed at the beginning of the row readout to let the internal bus stabilize before the readout of pixels starts. Resetting a the columns contributes another 4.6 ms resulting in a total minimum frame time of 52 ms for a pixel time of 1.1 ms. The Aladdin readout is organized in four independent quadrants. In each quadrant 8 adjacent pixels in a row are read out simultaneously. Due to this special readout topology a brightly illuminated pixel has a shadow image 8 pixels further down the row as shown in the image inserted in . For fast readout this shadow image is generated by the limitations in rise and fall times of the detector signal. In the intensity of the shadow image normalized by the intensity of the real image is plotted as function of readout time for Vload= 0 V and Vload = 5 V. The crosstalk can be reduced from 26% to 7% for frame times of 52 ms and from 0.66% to 0.38% for frame times of 98 ms if Vload is increased from 0 V to 5 V. With these Aladdin settings broadband L images can be obtained in ISAAC covering a field of view of 72x72 arcsec at the 8 meter Antu telescope of the VLT. This provides a unique observing facility.

.

Conclusions

Second order effects of the Hawaii 1 detector have been examined. Even though cryogenic off-chip operational amplifiers have been employed, the reduction of readout noise by multiple sampling techniques becomes limited by the electroluminescence of the shift registers. To compensate for the limited number of nondestructive readouts because of multiplexer glow a digital filter can be applied to the video signal by subpixel sampling. To avoid the reset anomaly, the Hawaii array can be operated by temporarily switching off the reset clock when operating under low flux conditions. On raw unsubtracted images a pixel to pixel rms noise of 3.6 electrons rms has been demonstrated.

A darkcurrent of 0.004 e/sec has been measured with an Aladdin array by monitoring the thermal drift of the multiplexer with unhybridized pixels in the corner of the array and subtracting the drift from the video signal. This is the lowest darkcurrent reported for InSb. With multiple sampling the readout noise of the Aladdin array could be reduced to 9.7 erms. The frame rate of the Aladdin 2 array could be boosted to 20 frames/sec. The crosstalk is 0.5% if the frame rate is reduced to 10 Hz, which is adequate for broad band imaging in L on an 8 m telescope for a field of view of 72x72 arcsec.

Future multiplexer designs should carefully address the problem of multiplexer glow and screen the IR diode array in the best possible from multiplexer glow to allow unrestricted application of multiple sampling techniques. A reference row should be provided to mimic the video signal and provide a reference signal for a true differential data acquisition chain employing off-chip cryogenic amplifiers.

Acknowledgments

The authors wish to thank U. Weilenmann, J. Rouchet and L. Vanzi for performing some of the measurements with the Hawaii 1 array mounted in SOFI at the NTT telescope in LaSilla and J. Cuby and C. Lidman for helpful discussions.

References

 

1. A. Moorwood, J.G. Cuby, P.Ballester, P. Biereichel, J.Brynnel, R. Conzelmann, B. Delabre, N. Devillard, A. VanDijsseldonk, G. Finger, H. Gemperlein, C. Lidman, T.Herlin, G.Huster, J. Knudstrup, J.L. Lizon, H. Mehrgan, M. Meyer, G. Nicolini, A. Silber, J. Spiromilio and J. Stegmeier, "ISAAC at the VLT", The Messenger 95,pp. 1-5,1999.
2. A.W Hoffman, K.J. Ando, A.D.Estrada, J.D.Garnett, N.A.Lum, P.J.Love, J.P. Rosbeck, K.P.Sparkman, A.M. Fowler, J. Pipher and W.Forrest "Near IR arrays for ground-based and space based astronomy", Proceedings SPIE 3354, pp. 24-29, 1998.
3. A. M. Fowler, I. Gatley, F.J. Vrba, H.D.Ables, A. Hoffman and J. Woolaway, "Aladdin, the 1024x1024 InSb Array: Test Results", Proceedings SPIE 2475, pp. 27-33, 1995.
4. CRIRES - VLT High-Resolution IR Echelle Spectrometer, http://www.eso.org/instruments/crires.
5. NIRMOS Near InfraRed MultiObject Spectrograph, http://www.eso.org/instruments/nirmos.
6. VISTA Visible & Infrared Survey Telescope for Astronomy, http://www-star.qmw.ac.uk/~jpe/vista.
7. A.K.Haas, C. Cabelli, J.L. Kozlowski, W.E. Tennant, J.T. Montroy, Y. Bai, D.E. Cooper, C.A. Chen, G. Bostrup, J.M. Arias,J. Bajaj, J.Chow, S.Bhargava, K. Spariosu, D. Edwall, J.D.Blackwell, T.Pepper, K.Vural, K.W.Hodapp, D.N.B. Hall, G.G. Price, and J.A. Pinter, "2048x2048 Visible/Infrared HgCdTe Focal Plane Arrays (FPAs)", August 1999.
8. L.J. Kozlowski, K. Vural, S.A.Cabelli, C.Y. Chen, D.E. Cooper, G.L Bostrup, D.M. Stephenson, W.L. McLevige, R.B.Bailey, K. Hodapp, D.Hall and W.E. Kleinhans, "2.5 mm PACE-I HgCdTe 1024x1024 FPA for Infrared Astronomy", Proceedings SPIE 2268, pp. 353, 1994.
9. K.W.Hodapp, J.L.Hora, L.L.Cowie, M.Metzger, E.Irwin, D.Hall, K. Vural, L.J. Kozlowski, S.A.Cabelli, C.Y. Chen, D.E. Cooper, G.L Bostrup, R.B. Bailey, W.E. Kleinhans, "The Hawaii Infrared Arrays: Testing and Astronomical Characterization of Prototype and Science-Grade Devices", to be published in New Astronomy.
10. G. Finger, P. Biereichel, H. Mehrgan, M. Meyer, A.F.M. Moorwood, G. Nicolini and J. Stegmeier "Infrared Detector Programs for the VLT Instruments at the European Southern Observatory", Proceedings SPIE 3354, pp. 87-98, 1998.
11. . M. Meyer, G. Finger, H. Mehrgan, G.Nicolini and J Stegmeier, "The ESO Infrared Detector High Speed Array Control and Processing Electronic IRACE", Proceedings SPIE 3354, 1998.
    
12. H.Mehrgan and M. Meyer,"Infrared Array Control Electronics IRACE Design Description", Internal Report, http://www.eso.org/projects/iridt/irace/documents.html.
13. T. Maihara et al, "Observations of the OH Airglow Emission", PASP 105, pp 940-944, 1993.
14. J.D. Garnett and W.J. Forrest,"Multiply sampled read limited and background limited noise performance", Proceedings SPIE 1993, pp. 395-402, 1993.
15. C.J. Skinner, "On- orbit properties of the NICMOS detectors on HST", Proceedings SPIE 3354, pp. 2-13, 1998.
16. G. Finger and G. Nicolini, "Interquadrant row crosstalk", http://www.eso.org/~gfinger/hawaii_1Kx1K/crosstalk_rock/crosstalk.pdf, Internal report, 1998.