To be technically accurate, the noise is higher on APS-C because the gain is higher (must be in order to produce the same ADU values after ADC). Think of an APS-C pixel like a FF pixel used at a higher ISO. On average, ISO 100 on APS-C is about the same as ISO 250-400 on FF from a noise standpoint. This is because the smaller pixel area means the photodiode area is smaller, and charge capacity in a photodiode is primarily based on area.
Jon, I always appreciate your answers and your patience with those of us less technologically inclined. It's frankly a sharp contrast to some other people here who prefer to serve up every answer with sarcasm.
I do wonder though, based on your example of ISO 100 noise on APS-C being comparable to ISO 250-400 on full frame, what do you think is a reasonable high-end for a 7DII? ISO 6400 is pretty darn impressive on the 5DIII and I would be very pleased if ISO 1600 on the 7D could match that, which is seems like it should be possible based on your comments.
What are you looking for in ISO performance from a 7DII?
I guess there are two factors to ISO settings, for different ranges of the ISO "spectrum", so to say. At low ISO, read noise is the critical factor. The lower your read noise, the greater your dynamic range. By reducing read noise from levels that used to be pretty normal to the industry (20-40 e-, depending on pixel size), Sony Exmor (which has a relatively constant 3- read noise) was able to properly utilize the dynamic range allowed by 14-bit ADC. This affects ISO settings 100-400.
For the other end of the range, high ISO, read noise is a factor, however more important than read noise is pixel quantum efficiency. Thanks to very efficient CDS, or correlated double sampling, Canon already has very low read noise at high ISO (from around 3.5e- to less than 1.7e- at the highest native settings), so their sensor performance is largely physics bound. Increasing quantum efficiency is the only real way to reduce noise at high ISO. The 7D has a Q.E. of 41%. Assuming we want a "true"
one full stop improvement in high ISO performance (i.e. a reduction in apparent noise by one full stop) without increasing pixel size, then quantum efficiency would need to be doubled (twice the real sensitivity, twice the rate of conversion of photons to charge). That means a Q.E. of 82%. In Canon's best sensors recently, like the 6D, they have achieved Q.E. around 50-51%. The best Q.E. for room temperature CIS these days is around 60-65%.
If we figure Canon makes some amazing strides in their sensor fabrication technology, and are able to achieve 65% Q.E., that is about a half stop improvement in high ISO performance. I don't believe Canon can reach 82% Q.E. without taking a more radical approach. The only time I've read of such a real sensitivity being achieved is with extreme cooling, usually a dual-stage TEC (peltier) cooling system with a passive or passive/active cooling system for that (i.e. a heat pipe setup to a heatsink which is further cooled by a fan.) A lot of astrophotography CCD cameras use dual-stage TEC cooling with a fan (and the good ones, the FF sensor ones, cost about $4000-6000!)
There was mention, a while ago back near the beginning of the year, that Canon might try to employ some kind of active cooling technology. A simple fan probably wouldn't do much...all it would really serve to do is cycle the air locked inside the camera body, so eventually the ambient temperature is going to increase and the benefit of having a fan would be largely negated. Some kind of peltier, however, along with proper heat venting or other form of expelling heat to the exterior of the camera body, could reduce sensor temperature by a lot, thereby reducing dark current and increasing Q.E. I don't know how much thermoelectric cooling would be practical. You have a delicate balance of power usage (peltier's suck power like it was candy) and cooling capacity. Canon would need a battery capable of holding a much greater charge, and one capable of providing a higher continuous voltage. Practically speaking, I am not sure the digital photography world is ready for thermoelectric cooling yet.
So, at best, absolute best, I suspect we will see a 1/3 to 1/2 stop improvement in high ISO performance in the 7D II, assuming the pixel count (and pixel size) stay the same.
If pixel count increases, pixel size must decrease, so I suspect we will see a 1/3 stop improvement at most, if that (assuming Canon actually achieves 65% Q.E.) If Canon increases the pixel size, then that will implicitly result in larger area. A megapixel reduction along with an increase in Q.E. could result in better high ISO performance. I don't really expect that to occur...the trend is, has always been, and will likely always be towards higher and higher megapixel count.
So, assuming Canon makes some modest gains in Q.E., increases megapixel count to around 22 megapixels (give or take 2mp), does NOT use any kind of thermoelectric cooling...I don't foresee any real improvement in high ISO at all. I see it staying roughly the same, which is saying something at the very least if megapixel count does indeed increase to 24mp.
As a side note, since it would take an increase to 82% Q.E. for the 7D II to gain a true ONE stop improvement in high ISO performance, we can never hope to see a true two stop improvement. The 7D II, nor any successor, nor any new pro-grade APS-C line of cameras from Canon or anyone else, will ever perform as well as a FF sensor that has larger pixels. So long as the average pixel size for FF sensors remains larger than the average pixel size for APS-C sensors, FF sensors will always perform better at high ISO. Nothing we can do about that...its just physics.
Additionally, on a composition and size-normal basis (i.e. when scaling the output images of FF and APS-C sensors to the same size...equivalence), FF sensors will always perform better than APS-C sensors, no matter what the pixel size. Assuming you frame your scene identically with a FF camera and an APS-C camera, the FF camera is going to gather more total light, period. Since you can usually pack more larger pixels into the area of a FF frame than an APS-C frame, the FF image will always be sharper and have less noise than the APS-C. Even if the FF sensor had pixels the same size as the APS-C, or even smaller than the APS-C, when normalizing the results the FF sensor will always
do better. (One possible case where APS-C might achieve parity with FF is if, at ISO 100, the APS-C sensor had a stop or two better dynamic range...then, you might get similar results, but I doubt APS-C would ever produce a better result than FF.)