The MIPS-24 Si:As detector array is read non-destructively every ~0.5 seconds up to total integration time of 3, 4, 10, or 30 seconds, depending on the observing mode chosen. It became clear before launch that dowloading every 0.5 sec sample of the array would make it impossible to store that information on the spacecraft and download it in a timely way. Therefore, an on-board slope-fitting algorithm was developed in order to preserve the dynamic range of the integration while drastically reducing the data storage requirements. This mode is called ''SUR'' or ''sample up the ramp'' (where the ramp samples are fit, and not saved), as opposed to the ''RAW'' mode which retains all of the samples but which is not available in any of the observing AOTs. For SUR mode, a 2-dimensional (128x128x2) FITS data cube is downlinked; the first plane being the the slope image (the fit to the full exposure using the samples) and the second plane being the ''difference'' image, which corresponds to the difference between the first 2 reads. The difference image is non-zero for those pixels that exceed a threshold count rate. The BCD pipeline merges these two images into one FITS file, by replacing the pixels that are ''soft'' saturated in the slope image with unsaturated pixels from the difference image. Calibrated slope and difference images are also separately produced by the BCD pipeline. The final BCD product therefore has replaced pixels in it where appropriate. We note that the uncertainty of the replaced pixels are higher than in the rest of the image due to the short exposure time.
The silicon array is annealed at the beginning of every campaign in order to restore the responsivity of pixels affected by latents. This was done for the entire mission with the exception of the MIPSGAL I/II Legacy programs, when the Silicon array was annealed every 12 hours; these programs often observed saturated sources in the Galactic Plane, and thus the recovery from latent images was important. The responsivity of the array (overall flat fielding) drifts for ~1 hour after the anneal, so science observations were typically taken after that time has passed.
We note that, unlike many other instruments, the A/D converter for the MIPS-24 array reaches saturation much before the detector full well is reached. The A/D saturates at about 1/3 of full well. This means that the array is largely still in the linear regime when the A/D converter saturates. Table 2.4 summarizes the performance of the silicon arrays as known from both lab and on-orbit tests.
Table 2.5 summarizes the performance of the Ge:Ga arrays as known from both lab and on-orbit tests. The response of the 70 µm detectors is linear to about 10%. This high degree of linearity is one of the primary benefits of the capacitive transimpedence amplifier (CTIA) readout design of these arrays, as well as the use of the scan mirror to modulate the signals within the 'fast' response of the detectors. For the 160 µm array the 'fast' and 'slow' response times are not as widely separated, and the linearity of the response is correspondingly worse than for the 70 µm array. The pipeline corrects for all well-characterized nonlinearities.
Saturation for any particular pixel occurs at 4x105 electrons, and is very abrupt due to the use of the CTIA readouts. The saturation value given in Table 2.5 is intentionally smaller than this value to allow for measured pixel-to-pixel variations in responsivity of the arrays, such that an exposure designed using the nominal responsivity values will not saturate any of the pixels in the arrays. Intra-pixel variations in responsivity are uncharacterized, but are expected to be relatively unimportant because the PSF is well sampled. Such variations may be a greater concern for the wide FOV mode of the 70 µm array, where the pixels are relatively large. However, in all cases, the high sampling redundancy in the MIPS data will tend to make the effects of these variations average out. Dead space between pixels is very near zero.
** Read noises on orbit are 1000-1500 electrons due to cosmic ray effects.
Cross-talk between pixels that receive a radiation hit during an exposure and neighboring pixels is less than 1%. The rate of cosmic ray hits on the large Ge:Ga pixels is high, about once per 12 seconds. The impact of cosmic ray strikes on data quality is minimized by eliminating cross-talk and by employing a sophisticated scheme for detecting and removing cosmic ray hits in pipeline-processing of the data.