It seems from responses to my post earlier this week that there's a lack of awareness about a number of things involving sensors. So here are some things to think about:
- Large sensors have been on a long march very similar to Moore's Law. In other words, every so many months there's a predictable improvement achieved. I've been trying to track this now for about 15 years, and I'd generalize it this way: every 18 months or so we see enough efficiency increase or noise reduction to create a tangible visible difference, all else equal.
- The 18-month rule is not quite as perfect in predicting as Moore's Law has been. I see variability across makers. For a long time, Canon was making the largest, most regular gains. Today that baton has passed to Sony. My point about the m4/3 sensors was that I'm not seeing any tangible gain over what amounts now to almost three 18-month periods. So some sensor makers move above the predicted gain line for awhile, some fall below it. Over time I suspect they all revert back close to the line, though we don't have enough data periods yet to fully back that prediction.
- We sometimes get sensor gains that aren't associated with still image quality. Besides all the other things that are happening, power usage, signal speed and integrity, plus better ways of addressing data for video are all going on. The m4/3 sensors seem to be progressing more on those last two areas (120 fps offload, changes in rolling shutter, etc.) than in image quality changes.
- BSI (back side integration) works better for small sensors than big. The reason is simple: in traditional sensor designs, the power and data lines are on the same surface as the light-collecting photo diodes. The process used to produce the sensor determines how small those lines can be, and sensors tend to be made on older equipment that have large process sizes (though this is changing). Even with small process dies, a cellphone sensor that isn't BSI is going to have large amounts of the surface area devoted to power and data. But the power and data lines don't get bigger for larger sensors, so the photo diodes do. The larger the sensor, the less gain you get from BSI.
- BSI has made it to the compact camera sensors. It could make it to Nikon's 2.7x sensor and make a difference. But each leap upwards in sensor size brings less gain, and it's not a trivial process without cost, so it gets less likely the larger the sensor size.