It is currently believed that the bands are caused by a pull-down of the amplifier bias when the amplifiers of another NICMOS camera are being used at the same time. Since the readouts of parallel NICMOS exposures are staggered by fractions of readout time (~1/3 of a readout time), then bands will appear where the readout of another camera stops or starts. The bands appear in approximately the same place on the detector, but are not 100% spatially repeatable due to timing pattern differences. Also, the polarity of the bias jumps is not always in the same sense or of the same amplitude, making it somewhat non-repeatable.
Since this feature is often (but not always) found in the final read of a multiaccum sequence, it can usually be minimized by recalibrating the image, eliminating the last read. If the loss in exposure time is important, then running CALNICA up to the flatfielding step and subtracting the column (or row, whichever is perpendicular to the fast read direction) average of the image (after expanding this average to a 256x256 image) and finishing the CALNICA processing can largely eliminate this feature. Note that the noise caused by the increased bias will remain, however. In this NIC3 example, the point source is (fortuitously) in a low bias region, and peaks at over 28,000 cts/sec while the bias jumps between -1, 11, and 23 cts/sec. Correction may not be necessary in this case.
Since Cycle 11, we have implemented a modified readout sequence for the three NICMOS cameras which reduces the probability that a detector is being reset while another is read. This procedure will be completely transparent to users and should significantly reduce the electronic bands problem.