I respectfully disagree on all your points
- Small pixels reduce sensitivity, not dynamic rage, but the whole point with the isocell sensor is to increase sensitivity in a small pixel. Because photons are discrete, your dynamic range can be no better than 10*log(photon count/pixel). To get 10 bit dynamic range you need 10e3 photons/pixel.
- The megapixel game is not meaningless. I use a large printer, and with a 25Mpix sensor, the result is a lot better than with a 10Mpix sensor. The print actually has a resolution of 12 000 Mpix!
The quantum efficiency, QE, of most backlit sensors ranging from the best DSLR to the Samsung is all around 10%. (Human eye and astronomical cameras can be up to 100% i.e. detect single photon.)
10% QE is about 5 picoLumens per pixel sensitivity, and here is where the sensitivity comes in. 1 lux= 1 EV = 1 lumens/m2 = a bit more than bright moonlight. Assume you have an f1 lens. now you will need 5 nanoLumens/pix for 10bit DR.
A 7mm lens will give you 3.8E15 photons/s, so each of the 16Mpix will get 2.38E8 photons, or 2.38E7 LSB. This should equal 24bit dynamic range. This is with a lot of generous assumptions like an f1 lens, no statistical noise, no thermal noise etc, but still enough photons to give good dynamic range in the darker parts of a photo.
This should give some insight into some of the fundamental limits.