Or, more technically, why can’t the values of photosites be read while exposing? As that would allow a camera to store versions of a RAW with varying exposure times (“subexposures” I’ll call them for lack of a better name).
Maybe this could even be made possible in existing consumer cameras with a firmware update? Unless photosites get reset in the process of having their values read, all it’d have to do is repeatedly read photosites values and write the subexposure’s RAW array to file every n millis until it reaches the max exposure time.
Benefits I can think of:
- Exposure time could be “changed” in post as you would have your choice at which RAW you want to use. This would be useful e.g. to maximize the amount of light while minimizing motion blur in an action shot.
- It could possibly be useful in reducing digital noise by merging RAWs? Although this would mess with how motion blur looks as it would be merging in subexposures with less motion blur.
Limitations I can think of (what else am I missing that renders this idea impossible?):
- Obviously you would have to store an entire extra RAW file for every subexposure. This would be VERY expensive storage/memory-wise.
- It may increase digital noise as electronics close to the sensor have to read the values of the photosites more than just once.
- If photosites values are read row-by-row, the last-read rows will be more exposed while the first-read rows will be less exposed as reading takes time, and the shutter is still open. But this seems correctable. Also, this delay seems pretty negligible as the process of reading happens practically instantaneously.
- (see @lijat’s comment) There may be overheating issues with constantly rereading photosites and because of the overhead of copying many RAW arrays from memory to storage quickly. Excessive sensor heat also introduces noise to my knowledge. Although, an operation like this would be similar to using a very fast continuous shooting mode. I believe a camera would handle it reasonably well.
The benefits seem very worth it despite the limitations. Why don’t camera companies do this right now? Is it impossible for some reason I don’t see? Or is it simply not economical/marketable?
It may be that I just don’t know the right terminology, but I can’t find a single piece of writing about this idea anywhere on the internet.