i'm not sure why people keep saying super high ISOs like 200K that will will look good won't be possible. I'm sure there are advances in science and technology that none of us have any idea about. they just made a camera that can capture light moving, and i bet everyone would have said that no way will that ever happen like a year ago.
i seriously doubt that anyone here is an optical engineering expert with a multi-billion dollar firm to back our research, so its hard to take any argument (including my own) seriously.
I know its fun to guess, and i understand that people are making valid points, but the honest-to-goodness truth is that no one here knows nuthin'.
One need not be a certified optical engineer with a Ph.D to understand the concepts. Its pretty basic, and boils down to total light per time interval. At ISO 204800, the time interval is generally going to be VERY short (there are exceptions, such as say night sky or astrophotography). The shorter the time, the more impact the random nature of light is going to have on the final image. People talk a lot about read and electronic noise...however in the grand scheme of things, it is photon noise, or photon "shot" noise as I prefer the term, that completely dominates. The reason photos appear noisy at high ISO has nothing to do with the amount of electronic noise generated in the sensor...by the time you pass ISO 400 electronic noise results in 2-3 electrons per pixel, vs. a maximum saturation level of thousands to tens of thousands of electrons. Its a minuscule percent.
If you think of light passing through a lens and onto a sensor as a "rain" of photons, then we can use real rain as a corollary. If you watch a flat surface, such as a dry, flat concrete slab made up of small one foot squares (pixels), during a rain storm. Assuming a constant and moderate amount of rain, under a short observation, say 5 seconds, you will have a widely dispersed pattern of infrequent drops visible on the slab, and maybe a few strikes per square. Under a longer observation of say 5 minutes, you will have a much denser dispersal pattern, with the infrequent dry spot but mostly a wet slab, and just about every square will have at least one drop. As another corollary, quantum efficiency would be the rate of photons that strike an actual concrete square in our slab, and not one of the gaps between concrete squares. If our concrete slab has 3" gaps between each square, were going to lose a fair number of rain drops. Some drops may strike a concrete square, then drain off into one of the gaps. The "quantum efficiency" of our concrete rain catching slab is not ideal. We could improve it by reducing the size of the gap between each square, to say 1/2". We can also improve it by say creating a small ridge around the edge of each square to hold more of the raindrops that strike it. Our quantum efficiency is a lot higher now. To throw in an extra factor, a heavy rain would be like having a wide aperture, where as a light rain would be like having a narrow aperture.
A high ISO photo is like the concrete slab for 5 seconds. From a "total light captured" perspective, it doesn't matter how large the pixel size is when talking about very high ISO...the random nature of light and discrete nature of photons will mean that few, rather than many, pixels will actually encounter a photon at all, and for those that do, the total amount of photons will be low. A wider aperture increases the rate of photon "rain", so you can improve the number of pixels that encounter photons by using a faster lens. But even doing that, the grand total amount of light is still going to be considerably lower than a lower ISO setting in more light...because were working with a "light photon rain", rather than a "heavy photon rain".
The only way you can really make a high ISO setting produce the same kind of image quality as a much lower ISO setting is to increase the amount of available light. This can either be done by increasing the illumination of the scene, or using a faster lens. In the case of ISO 204800...a MUCH faster lens with near-perfect qualities at apertures we have yet to hear of. We can also improve quantum efficiency, however the thing about Q.E. is that it will only matter if we are actually "losing" photon strikes to heat or reflection. If we had a latticework of wood laid over the top of our concrete slab to represent readout wiring, any rain drops that strike the latticework can't be counted as a captured rain drop on our slab. Same deal with a sensor...there is a lot of electronic wiring for each pixel that can get in the way, converting a photon to heat. Photon strikes at the right angle of incidence can even reflect off of the surface of the sensor that is not actually part of a photodiode, and even with a microlens, just as with any lens, the right angle of incidence can cause reflection of a photon rather than capture. With backilluminated sensors with multiple layers of microlenses, and say potentially even nanocoated microlenses to avoid reflection, along with all of the advanced electronic noise reduction present in Sony Exmor sensors, we might be able to push 60-70% Q.E. To get much farther than that, we would need to start applying active cooling to reduce the temperature of our sensors to well below zero, improving electronic efficiency and making photon capture and conversion into electrons more effective. We might be able to push 80% or more at that point (much like scientific-grade CCD sensors.)
If we can surpass 80% Q.E. in a commercial-grade image sensor, we might see usable ISO 204800 with acceptable IQ for regular use. It will never offer the same kind of IQ as much lower ISO settings, such as 1600 and less, but it could theoretically be useful. It will still be noisy, particularly in the kinds of situations where one would actually need it....say photographing the auroras without such long exposures that the fine, helical filamentary nature of them become blurred into nothing (and even then, we might actually need ISO 819200 or higher to REALLY capture an Auroral discharge in full detail.)