The FET switches in the CRC463 multiplexer are not perfect switches as
assumed above, but instead have finite gate capacitances that act as
sinks of charge that would otherwise remain on the detector node
capacitance. The reset pedestal (i.e., the amount the detector node
voltage jumps by when the reset is taken off) is due to a
redistribution of charge from the
detector node capacitance to the reset FET gate capacitance that
occurs when the the reset FET gate voltage (i.e., the reset clock
voltage) moves positive to switch the reset FET off. This amounts to
100 mV of lost detector reverse bias (or well depth).
Similarly, a further 400 mV of detector bias is lost when the
row and column select FETs are switched off to deselect the pixel.
This constitutes a movement of 500 mV of charge off the
detector node compared to the normal operating detector reverse bias
that remains of only 200 mV. The true pedestal is
therefore significantly larger than the reset pedestal seen if each
pixel is reset and read on one pass through the array.
Fowler & Gatley (1990, ApJ, 353, L33) show that the read noise can be reduced by performing multiple non-destructive passes through the array at the beginning and end of the integration ramp. By resetting each pixel on one pass through the array, and sampling the detector node voltage on subsequent passes through the array, the true pedestal is removed from the data. Each time a pixel is selected charge is redistributed from the row and column select FETs back onto the detector node capacitance. Fowler claims that the read noise is predominantly due to the kTC noise associated with this charge redistibution. By performing multiple non-destructive passes through the array at the beginning and end of the integration, the read noise is reduced by the square root of the number of passes.
Readout Method 5 implements Fowler sampling in this way. The number
of reads at each end of the integration is set by the FNDR parameter
(CASPIR/FNDR=...). For applications where low read noise is
required, at the expense of increased frame readout time, method 5 is
the preferred readout method. This is likely to be the case when
using the grisms.