Although the spectrometer was initially designed to use a subsection
of a NICMOS3 detector,
we found later that very good performance can be obtained
on an even larger area. Using 256 pixels in the wavelength direction, we have a
spectral coverage of almost 
. This means that with a single
grating setting we can measure a complete J spectrum and have good
coverage in H and K.
The best 
subsection was selected on the basis of good
cosmetics (low percentage of bad pixels) and low dark current and readout
noise. We measured the percentage of
bad pixels, the dark current, and the readout noise
via laboratory tests based on sets of images taken
at a series of exposure times of a spatially uniformly illuminated
scene, and without any illumination
(by substituting the filter with a cold stop).
The readout noise is determined as the mean standard deviation of each pixel in the stack of short integration times where the dark current is negligible. The dark current and gain measurement are based on two linear regressions: values of dark frames as a function of exposure time in the first case, and spatial medians of the stack variance relative to the stack median in the second one. Details of these tests are presented in Vanzi et al. (1995). In Table 1 (click here), we present the results of further tests carried out in April 1995.
