|
|
AUSTRALIAN
NATIONAL UNIVERSITY System Design Note 8.00 Created: 5 April 2000 Last modified: 5 April 2000 |
NIFS SCIENCE DETECTOR TRADE-OFFS
Peter J. McGregor
Research School of Astronomy
and Astrophysics
Institute of Advanced
Studies
Australian National
University
Revision History
|
Revision No. |
Author & Date |
Approval & Date |
Description |
|
Revision 1 |
Peter J. McGregor 20 October 1999 |
Mark Downing 27 October 1999 |
Original document. |
|
Revision 2 |
Peter J. McGregor 05 April 2000 |
|
Reformatted for Word 2000. |
|
|
|
|
|
Contents
4 HAWAII-2
HgCdTe/Sapphire PACE Technology Arrays.
5 HAWAII-2
HgCdTe/CdZnTe MBE Technology Arrays
6 NIFS Science
Detector Recommendation
1 Purpose
The purpose of this document is to discuss trades in selecting a suitable detector for the Gemini Near-infrared Integral-Field Spectrograph (NIFS). Devices considered are the Rockwell HAWAII-2 2048´2048 detector and a 1-2.5 mm 2048´2048 CdZnTe detector emerging from the NGST detector program at Rockwell.
2 Applicable Documents
|
Document
ID |
Source |
Title |
|
RSAA |
NIFS Performance Model |
|
|
|
|
|
3 Introduction
The Gemini Near-infrared Integral-Field Spectrograph is designed to use a 2048´2048 science detector. The science requirements for the instrument dictate a wavelength response from 1.0-2.5 mm and demand low dark current and low read noise performance. The higher dark currents inherent in detectors with sensitivity to 5 mm and the difficulty associated with blocking high thermal backgrounds at wavelengths beyond 2.5 mm further motivate the selection of a 2.5 mm cutoff device for NIFS. NIFS is a fast-tracked instrument that is expected to be available on Gemini North by mid-2002. Meeting this schedule requires delivery of an engineering array around mid-2001 and delivery of the science detector by the beginning of 2002. This short timescale demands a conservative approach to emerging technologies.
Rockwell Science Center are currently developing the 2.5 mm cutoff, 2048´2048 pixel, HAWAII-2 HgCdTe array for astronomy. This device is an evolution of the successful 1024´1024 HAWAII-1 array. It uses PACE technology in which the HgCdTe detector material is deposited on a sapphire substrate. The HAWAII-2 array is the baseline detector for NIFS.
Rockwell Science Center are also developing low dark current 5 mm cutoff HgCdTe arrays for applications on NGST. These devices use HgCdTe detectors deposited on lattice matched CdZnTe substrates using molecular beam epitaxy (MBE). The adaptation of this technology to 2.5 mm cutoff detectors promises improved performance over the HAWAII-2 arrays. We discuss this possibility below, but emphasize that Rockwell has no immediate plan to develop 2.5 mm cutoff devices based on this technology.
Rockwell Science Center are also experienced in producing HgCdTe detectors on silicon substrates. Silicon substrates offer some of the advantages of the CdZnTe substrate; silicon provides a better lattice match to HgCdTe than sapphire (but not as good as CdZnTe), and silicon substrates are obviously an excellent thermal expansion match to silicon multiplexers. Astronomical detectors based on this technology may be viable, but Rockwell appear to have no plans to develop such devices in the short term. We do not discuss this option further.
4 HAWAII-2 HgCdTe/Sapphire PACE Technology Arrays
The 2048 ´ 2048 HAWAII-2, 18 mm pixel arrays use similar technology to the 1024 ´ 1024 HAWAII-1 arrays. Both devices have a 2.5 mm wavelength cutoff. They are expected to have similar performance with a single double-correlated sample read noise of ~ 9 e. Read noises of ~ 4 e are likely using eight double-correlated (i.e., Fowler) samples. The dark current performance of the HAWAII-1 array is not well documented, due partly to the difficulty of measuring extremely low dark currents. Finger et al. (1998) report a mean dark current of < 30 e/hr (< 0.0083 e/s) for a HAWAII-1 array operated at 78 K. This very low measurement is limited by electrical drifts in the data system. The Rockwell Science Center WWW pages show a dark current distribution with a mode of ~ 0.01 e/s/pixel for a HAWAII-1 array at an operating temperature of 78 K (Figure 1). Kozlowski et al. (1998) plot a different dark current distribution for a HAWAII-1 array operated at 78 K. This has a mode of ~ 0.026 e/s/pixel and a high dark current tail extending to ~ 0.15 e/s/pixel. Bailey et al. (1998) quote a mean dark current of 0.05 e/s/pixel with > 99.66% of pixels having < 1 e/s dark current for a HAWAII-1 array operated at 77 K and 0.5 V reverse bias. Mackay et al. (1998) report a mean dark current at 90-110 K for three of their HAWAII-1 arrays of ~ 0.1 e/s/pixel and ~ 2 e/s/pixel for an earlier fourth array. They note that for their devices ~ 10% of all pixels have dark currents > 5 times the mean, ~ 4% have dark currents > 10 times the mean, and ~ 1% have dark currents > 20 times the quoted mean value. These hot pixels behave in a predictable and repeatable way. The latter two high dark current measurements are partly due to the higher detector reverse bias voltage or higher operating temperature used. We conclude that HAWAII-1 arrays are capable of achieving modal dark currents as low as ~ 0.01 e/s/pixel when operated below 70 K and with reverse bias voltages < 200 mV, but that the dark current distribution has a tail extending to > 0.1 e/s/pixel. The node capacitance is ~ 40 fF at this reverse bias (Hodapp et al. 1996), so the well depth is ~ 50,000 e. PACE technology devices have fast output amplifiers permitting sample times of ~ 5 ms/pixel, but they suffer from declining quantum efficiency at wavelengths shortward of ~ 1.3 mm (QE ~ 60% declining to < 50%) and significant persistence effects due to lattice mismatch between the HgCdTe detector and its sapphire substrate. Problems with amplifier glow in HAWAII-1 devices are expected to be solved in the HAWAII-2 devices, but this is yet to be verified.
Figure 1: Dark current distribution for HAWAII-1 array (RSC WWW pages).
A typical quantum efficiency curve for a HAWAII-1 array is shown in Figure 2 (Rockwell Science Center WWW pages). Representative read noise values as a function of the number of correlated double samples for a HAWAII-1 device are shown in Figure 3 (Hodapp et al. 1996).
Figure 2: Quantum efficiency function for HAWAII-1 detector (RSC WWW pages).
Figure 3: Read noise for a HAWAII-1 array measured at 65 K with 250 mV bias as a function of number of correlated double samples (Hodapp et al. 1996).
The HAWAII-1 PACE array is known to exhibit fringing in spectroscopic applications (Figure 4; Hodapp et al. 1996). This is caused by interference effects in the sapphire substrate on which the HgCdTe detector material is grown. It is likely that HAWAII-2 PACE arrays will show a similar effect.
Figure 4: Interference effects in a dispersed image recorded with a HAWAII-1 array (Hodapp et al. 1996). The spectrum ranges from the K band in the lower order to ~ 0.85 mm in the upper order.
The rate of ionizing events per pixel for the HAWAII-2 array with 18 mm pixels will be more than a factor of two lower than for the 27 mm pixel ALLADIN 1024´1024 InSb arrays used in NIRI and GNIRS. Mackay et al. (1998) quote cosmic ray detection rates with their HAWAII-1 arrays of ~ 1 event per square centimeter per minute, corresponding to ~ 815 events detected by a HAWAII-2 array in a 3600 s integration. Higher event rates may be experienced on Mauna Kea.
The first HAWAII-2 bare multiplexer was delivered to Klaus Hodapp in October 1999. Science detectors are expected to be delivered to the original HAWAII-2 consortium members by mid- to late-2000. Rockwell then intends to market devices to non-consortium customers. This allows ~ 12-18 months for delivery of the NIFS science detector in late-2001.
5 HAWAII-2 HgCdTe/CdZnTe MBE Technology Arrays
The HgCdTe/CdZnTe MBE double layer planar heterostructure (DLPH) technology being developed by Rockwell for NGST solves many of the shortcomings of PACE technology devices by improving the lattice match between the HgCdTe detector and the CdZnTe substrate. This reduces the number of lattice defects and so produces lower dark currents, higher more uniform quantum efficiency, and eliminates the persistence problem. MBE devices with 5 mm cutoff typically have dark currents of ~ 0.01 e/s/pix, as low as or lower than 2.5 mm cutoff HAWAII-1 arrays and near the theoretical diffusion limit. If the same near-theoretical performance is achieved in 2.5 mm cutoff devices, these arrays should have dark currents approaching 0.001 e/s/pixel. MBE devices with 5 mm cutoff and no AR coating have measured quantum efficiencies that are flat at ~ 72% longward of 1.2 mm with no discernible degradation when cooled from 85 K to 60 K. The quantum efficiency of these devices is limited by surface reflection losses; Rockwell expects to achieve quantum efficiencies of ~ 85% over the 1.0-2.5 mm range using suitable anti-reflection coatings. This is made possible by the better refractive index match of CdZnTe. MBE devices with 5 mm cutoff show a dramatically reduced residual image of only 0.3-0.5% on the read after heavy saturation, with no detectable persistence on subsequent reads. The same should be true of 2.5 mm cutoff devices.
These remarkable improvements are partially offset by two difficulties with current CdZnTe MBE technology. The first is that CdZnTe and silicon are a poorer thermal expansion match than sapphire and silicon. This causes stresses between the detector substrate and the multiplexer during thermal cycling which makes early devices of this type prone to debonding between the detector and multiplexer. Rockwell have schemes for addressing this problem, but these are still developmental. The second difficulty is that CdZnTe can currently be obtained only in small wafers that accommodate just one 2048´2048 device. This means that devices must be produced individually on production lines that are in high demand. This is inefficient and the yield will be relatively low, so early devices are expected to be expensive.
We do not know whether HgCdTe/CdZnTe MBE arrays will exhibit fringing as HAWAII-1 PACE arrays do. It is possible that this effect will be reduced due to the better refractive index match between HgCdTe and CdZnTe.
HgCdTe/CdZnTe MBE technology has so far only been applied to 5 mm cutoff devices for the NGST development, and to date only for 1024´1024 pixel arrays. The NGST development is likely to be the main focus for Rockwell over the next two years. It is therefore questionable whether they will have the resources to develop 2.5 mm cutoff devices on a timescale suitable for NIFS.
6 NIFS Science Detector Recommendation
It is clear that, when fully developed, HgCdTe/CdZnTe MBE devices will be superior to the current HAWAII-2 arrays. The higher and more uniform quantum efficiency will boost signals at all wavelengths, but especially in the J band. Most NIFS measurements are likely to be performed in the K band where ALTAIR will perform best. However, NIFS will be used in the J band to measure Pb 1.282 mm and [Fe II] 1.257 mm emission-lines in low redshift objects and Ha in high redshift galaxies where the improved sensitivity will be most beneficial. NIFS is expected to be detector noise limited for most observations in the J and H bands, so the signal-to-noise ratio will improve linearly with the quantum efficiency gain. Image persistence in a HAWAII-2 PACE array would be problematic in at least two ways; bright OH airglow lines will leave remnant images when switching between gratings, and bright standard stars and point spread function reference stars will leave remnant images in science object data. Both of these complications will be greatly reduced, or eliminated, with a HgCdTe/CdZnTe MBE array.
Sensitivity improvements will also flow from the lower dark current of a HgCdTe/CdZnTe MBE device. We consider the likely noise sources in limiting NIFS observations in order to quantify these gains. The NIFS performance model (NIFS Performance Model, SDN0004.01) has been used to estimate the contributions due to dark current, read noise, and background emission detected between OH airglow lines to NIFS observations at the central wavelength of each of the four baseline gratings. The mean of the dark current distribution is assumed to be 0.01 e/s/pix, but we noted above that actual dark currents may be up to 10 times larger than this in a typical HAWAII-2 PACE array. The read noise is assumed to be 5 e. The background flux that NIFS will detect between OH airglow lines is quite uncertain. The NIFS performance model includes contributions from thermal emission from the telescope and sky (which dominates in the K band), a fixed sky continuum based on measurements in the H band (but assumed to be present at all wavelengths), and a flux contribution from scattered OH airglow line emission. The scattered contribution is calculated assuming that 10% of the total OH emission within the grating bandpass of interest is scattered uniformly across the array. Typical values for a 3600 s integration using each of the gratings, with and without the scattered term, are listed in Table 1. The statistical noise values listed in Table 1 are optimistic in the sense that they assume that the fixed dark current pattern has been accurately removed by subtracting a noiseless stable dark image. This dark image will probably be the mean of many 3600 s dark exposures (subtracting a single 3600 s dark exposure adds one further read noise and one further dark current noise term in quadrature). Table 1 shows that limiting NIFS observations in J and H will have comparable noise contributions from the three noise sources considered and K band observations will be background-limited, if the dark current is as low as 0.01 e/s/pixel. Dark current will be a significant noise source at all wavelengths if dark currents are as high as 0.1 e/s/pixel with a HAWAII-2 PACE array. A dark current reduction to below 0.001 e/s/pixel, feasible with a HgCdTe/CdZnTe MBE device, will realize a reduction in random noise at J and H over that shown in Table 1 of > Ö1.5 if scattered OH emission is significant, and > Ö2 if scattered OH emission can be ignored. The improvement at K will be small. If we compare to a HAWAII-2 array with 0.1 e/s/pixel dark current, the sensitivity improvement is large at all wavelengths.
Table 1: Predicted NIFS Noise Contributions with a HAWAII-2 PACE Array
|
|
With Scattered OH |
Without Scattered OH |
||||||
|
Grating |
(Read Noise)2 |
Dark Signal |
Sky Signal |
Noise |
(Read Noise)2 |
Dark Signal |
Sky Signal |
Noise |
|
J1 |
25 |
36 |
31 |
9.6 |
25 |
36 |
8 |
8.3 |
|
J2 |
25 |
36 |
35 |
9.8 |
25 |
36 |
9 |
8.4 |
|
H |
25 |
36 |
41 |
10.1 |
25 |
36 |
14 |
8.7 |
|
K |
25 |
36 |
251 |
17.7 |
25 |
36 |
223 |
16.9 |
In practice, HgCdTe/CdZnTe MBE arrays will deliver even greater sensitivity gains because removal of the fixed dark current pattern during data reduction will be less crucial for data obtained with these arrays. The dark current pattern will dominate raw 3600 s integrations at J and H with a HAWAII-2 PACE array. The NIFS performance model currently uses a Gaussian dark current distribution with s = 0.0065 e/s/pix, based on Figure 1. The RMS noise in a raw 3600 s dark integration is ~ 23.4 e due to this dark current pattern, larger by at least a factor of two than the J and H random noise values in Table 1 which are based on data in which the dark current pattern has been subtracted perfectly. The dark current pattern of a HAWAII-2 PACE array will need to be stable and to be measured accurately if this systematic effect is to be reduced to below the level of the shot noise due to the dark current itself (~ 6 e or ~ 17% of the mean dark current). In contrast, it will only be necessary for the dark current pattern of a HgCdTe/CdZnTe MBE array to be stable at the level of the read noise to achieve the same result (~ 5 e or ~ 140% of the assumed mean dark current).
The large sensitivity and stability gains possible with a HgCdTe/CdZnTe array due to both its higher quantum efficiency and lower dark current must be balanced against the likely availability of these devices on timescales appropriate to NIFS. Rockwell are not yet developing 2.5 mm cutoff HgCdTe/CdZnTe MBE detectors. If schedule is the prime consideration, the best course for NIFS is to order a HAWAII-2 PACE array. This will ensure delivery of a suitable detector array. We should also encourage Rockwell to pursue the development of HgCdTe/CdZnTe MBE arrays, and hope to convert the order to the more advanced technology if science grade arrays are likely to be available by late-2001. This approach will lead to significantly inferior performance from NIFS at J and H, compared to using a HgCdTe/CdZnTe MBE array, and may exclude observations of Ha in high redshift galaxies, for example.
On the other hand, holding out for a HgCdTe/CdZnTe MBE array is likely to delay delivery of the instrument.
It will be possible to retro-fit a 2048´2048 pixel HgCdTe/CdZnTe MBE array to NIFS if these become available after NIFS is commissioned. However, we note that HAWAII-2 PACE devices and HgCdTe/CdZnTe MBE devices have opposite polarity; the HAWAII-2 PACE detectors are n-on-p while the MBE detectors are p-on-n (Kozlowski et al. 1998). This should be considered in the design of the detector controller.
7 References
Bailey, R., Arias, J., McLevige, W., Pasko, J., Chen, A., Cabelli, C., Kozlowski, L., Vural, K., Wu, J., Forrest, W., & Pipher, J. 1998, SPIE, 3354, 77
Finger, G., Biereichel, P., Mehrgan, H., Meyer, M., Moorwood, A. F. M., Nicolini, G., & Stegmeier, J. 1998, SPIE, 3354, 87
Hodapp, K.-W., Hora, J. L., Hall, D. N. B., Cowie, L. L., Metzger, M., Irwin, E., Vural, K., Kozlowski, L. J., Cabelli, S. A., Chen, C. Y., Cooper, D. E., Bostrup, G. L., Bailey, R. B., & Kleinhans, W. E. 1996, New Astr., 1, 177
Kozlowski, L. J., Vural, K., Cabelli, S. A., Chen, A., Cooper, D. E., Bostrup, G., Cabelli, C., Hodapp, K.-W., Hall, D., & Kleinhans, W. E. 1998, SPIE, 3354, 66
Mackay, C. D., Beckett, M. G., McMahon, R. G., Parry, I. R., Piche, F., Ennico, K. A., Kenworthy, M., Ellis, R. S., & Aragon-Salamanca, A. 1998, SPIE, 3354, 14
Appendix A: List of Figures
|
Figure 1 |
rsc_idc.gif |
|
Figure 2 |
rsc_dqe.gif |
|
Figure 3 |
read_noise.gif |
|
Figure 4 |
fringe.jpg |
|
|
|