« on: February 09, 2015, 06:56:43 PM »
Good point. My hope is that they have improved on the banding issues when lifting shadows over thr Mkiii. If they did that it would be enough for me to consider a pre-order. If not, then i'm going to most likely wait and see what the 5dmkiv is going to offer. I know it's wishful thinking but I hope Ken Rockwell's guess that the mkiv will be 36MP happens to come true.
Personally I'm in the market for a landscape camera. As such, I am not interested much in marginal reductions in banding or reductions in dark current. I can put more actual dynamic range (the ratio between maximum pixel signal and pixel read noise) to work in landscape photography. My 5D III does fine for birds and wildlife at high ISO, so I don't care about the high ISO performance of the 5Ds. I would use it pretty exclusively at ISO 100-200.
I am pretty sure I know what Canon is talking about with regards to lower noise floor. It sounds like Canon is still using their 500nm process, and measurements of Canon cameras over the years have shown that however they build their system, smaller pixels have lower read noise. The problem is that because they are still consuming so much die space with transistors and wiring thanks to their ancient 500nm process, they suffer significant losses in terms of photodiode area (i.e. photodiode pitch to actual full pixel pitch ratio is small). Photodiode area is what affects FWC...greater area, higher FWC...lower area, lower FWC.
Both read noise and FWC are directly linked to photodiode and pixel area...so they both grow or shrink together. THAT is why Canon can have lower read noise with the small 4.1 micron pixels of the 7D II, but not actually have higher dynamic range than the higher read noise of the 6.5 micron pixels of the 6D. Dynamic range is the ratio between FWC and read noise. If they both shrink and grow with each other, the ratio remains the same, DR remains the same. Increase FWC or reduce read noise, or both, and the ratio increases, and DR increases. You need the ratio, the range between read noise and signal saturation, to GROW in order to have increased DR. With more FWC at the same read noise, you can gather more light before the highlights clip...and not need to reduce exposure (not need to shift exposure down to make more room for the highlights). Similarly, with lower read noise and the same FWC, you can gather the same amount of light before the highlights clip, but the lower read noise means you don't lose as much in the shadows. Increase FWC and reduce read noise and you gain at both ends! Leave them the same, or reduce either or both, and you lose DR.
Other manufacturers have solved this problem a few ways, but one of the key improvements with other manufacturers is that they use 180nm, 130nm, 90nm or 65nm processes. Sony has used 180nm and 90nm processes, Samsung currently uses a 65nm process. That means Sony transistors take up 2.7x to 5.5x less space, and Samsungs transistors take up 7.7x less space, on the sensor die, than Canon transistors. It's basically taking a fat border of transitors from around each pixel, and replacing it with a thin, thinner, or very thin border. That leaves more area in the center of the pixel for photodiode (light sensitive photon-to-electron conversion surface area). Canon could very likely realize immediate gains if they would just drop their ancient 500nm process and move to 180nm process. That might gain them a stop right there. If they could figure out a way to flatten their read noise curve like most other manufacturers, that could gain them a stop or more.
So long as they do not do those things, however, I don't foresee Canon actually gaining on the read noise front. The day Canon releases a camera with either the same read noise and a meaninfully larger FWC, or meaningfully lower read noise and the same FWC, or higher FWC and lower read noise (doubtful in the foreseeable future) is the day we will finally see a real-world increase in dynamic range.