Because of the fixed read-out patterns available for use in MULTIACCUM mode (the SAMP-SEQs), in order to make an exposure of total integration time a minute or two, it is necessary in most modes to make a significant number of readouts. This leads to a significant volume of data to process. Additionally, the readouts are initially stored in a buffer in the NICMOS flight computer. A maximum of 100 readouts can be stored in this buffer, after which the contents of the buffer must be dumped to the Solid State Recorder. A full dump of 100 reads takes almost three minutes. At the time of writing, this process cannot occur in parallel with other NICMOS observations, which means that for every 100 images obtained, nearly three minutes of overhead time are incurred. Obviously, for programs requiring many exposures of only a few minutes' duration, if MULTIACCUM readout mode is used this can lead to very large overheads. ACCUM mode, on the other hand, yields only a single image in memory for each exposure, and so is much less expensive in overheads. Efforts are underway to reduce this overhead time, by making it possible to schedule buffer dumps in parallel with exposing the detectors in some circumstances, but this effort is not yet complete. ACCUM mode may thus appear attractive for programs requiring many exposures. Finally, because almost all MULTIACCUM exposures will consist of many readouts (typically ten or more), the amplifier glow signal, and the resulting photon noise, will be quite significant. In ACCUM mode on the other hand, it is possible to adopt just two reads (one initial and one final), greatly reducing the amplifier glow signal. This may be advantageous in cases where the background and target are very faint (e.g., narrowband line images).
There are a variety of disadvantages to ACCUM mode. First, the ability which is present in a MULTIACCUM exposure to filter out CR hits which occur during the exposure is lost. We find for NICMOS that typically between 2 and 4 pixels are hit per second per camera by CRs: most of these are low energy and so can be filtered out of a MULTIACCUM exposure by the calnica software. In ACCUM mode the process of CR removal requires separate exposures, and is a time consuming piece of post-processing as it is for WFPC2. Second, the ability to detect pixel saturation, which again is done automatically for MULTIACCUM observations by calnica, can in some circumstances be lost in ACCUM mode. Th ereason for this is that the time elapsed between the first read for each pixel and the reset immediately prior to the read is approximately 0.2 seconds. During this time, pixels exposed to a bright target will accumulate significant signal, which is then present in the first read. When this is subtracted on-board in ACCUM mode, all the charge accumulated in the time between reset and read will be subtracted. If the pixel has saturated during the exposure, the difference between initial and final reads will be less than the expected saturation value for the pixel, and thus it may be impossible to recognise that the pixel as saturated. Thus in the case of bright targets, erroneous signal levels may be recorded in ACCUM mode. Third, in ACCUM mode, even if pixel saturation is detected, it is not possible to repair the data obtained in the saturated pixel. In MULTIACCUM mode, pixels which have saturated can be repaired by using the results of previous, unsaturated reads during the same exposure. Finally, because of the many advantages of MULTIACCUM over ACCUM, during Cycle 7 rather few observers are using ACCUM mode. As a result, the quality of the calibration of dark current for ACCUM mode is likely to be significantly lower than that for MULTIACCUM.
In conclusion, in cases where a multitude of short duration exposures must be made per orbit, or where there is very low background and a desire to minimise the amplifier glow contribution to the noise, ACCUM may possibly (but not necessarily) be a good choice. In all other cases it is likely that MULTIACCUM will yield the best results, and we recommend that all observers attempt to use MULTIACCUM.