The IRS uses a 256
256 InSb array produced by SBRC. We have recently
(Dec 96) replaced the original engineering grade array with the array
which had been used in COB. The new array has significantly better
cosmetic quality and lower dark current. The well capacity is
considerably lower but satisfactory.
| The IRS Detector | |
| Type | InSb |
| Pixel | 30 micron |
| Operating Temperature |  35 K |
| Gain | 8 e-/ADU |
| Bias | 0.55 V |
| Read Noise | 32 e- |
| Dark current | 0.5 e-/sec |
| 2% nonlinear | 3500 ADU |
| 4% nonlinear | 5000 ADU |
We normally operate the array with a bias of 550 mv. This provides a gain of 8 e-/ADU and a read noise of about 32 e- for double-correlated sampling. With a 550 mv bias the array has a useful well depth of about 5000 ADU. To avoid large nonlinearity affects we would recommend that you operate the array at less than 3000 ADU. This is particularly critical for short exposures since there is a time delay of about 0.3 seconds between the array reset and the first read of the double-correlated sampling. Thus, for a minimum exposure time of 0.30 seconds the array has already collected signal for 0.3 seconds before the first read and this charge is not indicated in the final difference image. With the 550mv bias the total dark current is less than 0.5 e-/sec for a single pixels.
The IRAF task ctio.irlincor can be used to correct IRS data for nonlinearity. With a 550mv bias irlincor should be run with coeff1=1.0009, coeff2=0.0441, coeff3=1.416.
For programs in high background, the bias may be increased to 0.70-0.75 v with a significant increase in well depth, but with an increase in high-dark-current array artifacts (which are less important under those conditions). For example, increasing the bias to 600mv increases both the well depth and the dark current by 50%. It also increases the number of hot pixels. Thus, for normal operation we suggest the low bias value of 550 mv. For very high background observations (L and M) where the detector dark current is negligible compared to the natural background increasing the bias to 700mv will yield a gain of 7.41 e-/ADU with a useful well depth of about 9000 ADU. Once again to avoid large nonlinear effects we recommend that only about half of full well be used (i.e. 4,500 ADU).
Even with a fairly long fdly (the time between the array reset and the first read) this array appears to be rather sensitive to how it was operated in the previous exposure. Thus, if you take a short exposure (during set up) and then a long exposure (an observation) the first long exposure will be strange. Thus, it is best to take a ``junk'' exposure with the same integration time, low noise reads and coadds as your first science exposure before you take data.
Since the MUX of the SBRC array allows any pixel to be addressed and read nondestructively you can reduce the noise of the array by reading each pixel several times. The number of low noise reads is set by the parameter lnrs. But the MUX also produces some photons each time it is read and these photons contribute to the dark current and thus the noise of the array. The minimum effective read noise is achieved with 8 low noise reads. Since the low noise reads take extra time you can use 8 reads only on exposures of 2.4 seconds or longer. For very high signal levels the ADC hits its rail at about 15000 ADU. At or above this level there is no way to recover the true sky intensity from the difference image, since the second read is held at a constant level. This creates the odd phenomenon that brighter sources actually cause lower signals since only the signal level at the first read increases and the thus the difference becomes smaller.