Record sensor temp alongside raw, apply different noise compensation depending on temp in post.
A Peltier can help build the profiles - by heating and cooling to a range of temperatures as needed.
Fixed Pattern Noise
General explanation: http://en.wikipedia.org/wiki/Fixed-pattern_noise
AXIOM Alpha measurements:
C0 + C1 * exposuretime
C0: Constant component of dark frame energy. The non-uniformity across columns is probably due to each ADC response.
C1: Exposure dependant component of the dark energy. If all pixels are read in parallel, the non-uniformity across rows might come from temporal variation of thermal noise. But why is it exposure dependant ?