« on: February 26, 2015, 09:42:05 PM »
This is the way I see it. And of course Lee Jay and PBD can add/correct me if needed.
To increase high ISO performance to me is to increase DR at high ISO/light-limited situations. I need to increase FWC or max signal per pixel or QE, however you look at it. You need to lower read noise. HOW you do those two things I'm not really commenting about but if you can do that you can increase S/N at high ISO. You already have less read noise with smaller pixels so why can't you increase the size or efficiency of the photodiode in the pixel? I've been in discussions about smaller parts in and around the pixel to make way for larger photodiodes, for instance. I can also see where the FWC could be more important than read noise and overcome the higher read noise by adding more signal because signal is additive whereas noise is added SQRT. So in that case, larger pixels might still win. This of course is all at equal sensor size and equal technology.
Increasing FWC (probably a poor term to use, max saturation is probably better) is definitely a way that you can improve DR at higher ISO. Reducing read noise can certainly help, but at higher ISO read noise is already quite low, 3e- or less these days usually, and it's tough to complain about that. The saturation point at higher ISOs is usually only a couple thousand e-, sometimes as little as a few hundred e-, so increasing the charge capacity of each pixel is probably the better way to improve SNR at high ISO.
You aren't thinking about this correctly.
At high ISO, FWC is limited artificially by all the gain. The cells are fully capable of holding a lot more charge, and they do, but the A-D saturates because of all the analog gain.
Assuming you are at high ISO because you are light-limited (not a bad assumption, IMHO), then the ONLY way to increase DR is to reduce read noise (assuming Bayer dyes, same QE, etc.).
I understand it perfectly.
I agree that if you are light limited (a possible use case for high ISO), then you would need lower read noise to increase DR. I disagree that's the only use case for high ISO. I frequently shoot at high ISO when there is plenty of light, easily enough to saturate the entire sensor in a fraction of a second, because I need motion-stopping shutter speeds for very fast motion (just watch a Chickadee or Bushtit sometime...those things NEVER stop moving, and they make these ultra fast micro-moves that blur with shutter speeds lower than around 1/2000th or so). Increased EQE and increased FWC would result in greater IQ at the higher ISO settings I often need to use for these birds. Personally, I would rather not go with larger pixels to achieve that higher FWC though...I want my resolution.
So, I disagree that the only way to increase DR at high ISO is to reduce read noise. Just look at the A7s...that sucker has a MASSIVE FWC (true full well capacity, the base ISO maximum charge capacity of the photodiode) of 155557e-! It's saturation point at say ISO 12800 is 1298e-. Now that's thanks to having a greater fill factor...more total light sensitive surface area in the sensor, and per pixel, because the pixels are huge. The saturation point at the same ISO for the 6D is 604e-. Both cameras have similar read noise at that ISO, 1.8e- and 1.6e- respectively, but one has 9.7 stops DR and the other has 8.4 stops. Why does one have more DR than the other, if the only way to improve DR at high ISO is to reduce read noise? The camera with the higher DR actually has higher read noise! The 6D has a lower fill factor, less total light sensitive surface area.
The capacity of a photodiode is primarily limited by it's area, and the sensitivity is limited by area and EQE. Sensor sensitivity is affected by total light sensitive area in the sensor. So if Canon had made the 6D with the 65nm process Samsung is using, they could have increased the photodiode area. I don't have time at the moment to actually calculate how much area increase Canon could achieve without changing pixel size, but suffice it to say they could increase light gathering capacity and sensitivity by increasing photodiode area. There are still EQE losses due to the use of microlenses and CFA. Improve microlenses (aspheric lenses have been researched to better focus off-axis light onto photodiodes), replace the CFA with color splitters, move to BSI and basically gain nearly the entire surface area of the sensor as light sensitive area, and you increase the amount of light reaching each photodiode, which makes them saturate faster, thus utilizing that increased capacity, therefor allowing you to reduce gain further (which reduces the amplification of everything, noise included.) That improves IQ, even at high ISO.