CCD-World mailing list has a nice post by Jim Janesick:
"PTC funnies surface when...
1). When CTE [Charge Transfer Efficiency - ISW] isn't behaving which smooth's out the shot noise measurement. The CTE problem has to be pretty bad though for it to influence e-/DN. Why use the sensor in the first place ?
2). When V/e- nonlinearity is present (e.g., CMOS). However, one additional step to the standard PTC approach makes the technique viable. Here you will interestingly find that PTC will need to keep track of two gain terms. i.e., SIGNAL (e-/DN) and NOISE (e-/DN). For they are different as signal varies.
3). When the incoming wavelength is < 4000 A. Short wavelengths produce multiple electron hole pairs (i.e., quantum yield). PTC reacts with a e-/DN gain shift (decreases with decreasing wavelength).
4). Software engineers. :-)