The sensor is 12bit A to D - can't change that until we work on the next camera, but it should be more than adequate for our initial needs, especially with the constraints of running at 60fps.
[a href=\"index.php?act=findpost&pid=67022\"][{POST_SNAPBACK}][/a]
I am replying from a context of CMOS still cameras, so the application may be a bit off.
It seemed to me at one time that 12-bit digitization was the main bottleneck to DR in Canon CMOS DSLRs, as ISO 1600 collects much more real-world, absolute range at the darker end, which is not present in the shadows of ISO 100. I thought that the reason was only 4096 (or less) levels, but recent experiments indicate that ISO 100 simply has about 7x as much noise, relative to absolute sensor exposure. ISO 100 is the noisiest ISO, in an absolute sense (it is quieter in practice, only because the metering calls for more absolute exposure at ISO 100). Posterizing the ISO 1600 shadows by robbing them of the 4 least significant bits does not detract from them in any significant way, and they still look much cleaner than ISO 100 with the same exposure index.
Conclusion - readout noise is currently the most significant limiting factor in dynamic range. A clean 12-bit sensor readout (readout noise at half the single bit level) would give spectacular DR at the pixel level, and even more at the image level (especially video, which "averages" pixel values in our perception, over time, while stills only average over space). IMO, Digitization of more than 12 bits of depth is of no particular value until readout noise is conquered.