Christian Buil has an interesting CCD/CMOS comparison. Though specifically concentrating on spectroscopy,he covers camera noise, amp glow and linearity
With high resolution spectroscopy of faint objects at the limit of detectability the sky signal is insignificant so camera noise becomes the most significant source of noise. To minimise read noise contribution, exposure lengths used with CCD are typically >1200sec and it takes many hours to get a sufficiently noise free stack of darks, though cloudy nights can be used of course!
Although low, the read noise in CMOS is not insignificant and once the typical slit width relative to pixel size is considered, is not that different from that of good CCD because of the ability to bin pixels with CCD so there is not much latitude for combining shorter exposures with CMOS (though an spectrograph optimised specifically for the small pixels found in common CMOS sensors could perhaps avoid this). The thermal noise is also typically higher for CMOS cameras which need much lower temperatures to match that of a CCD so although the performance gap is not large these days, CCD still appears to have the edge in performance (Though not in cost per area) when used with currently available commercial spectrographs.