Now I have to admit I didn't understand one word of this...so Brian, I'll depend on you going forward to explain what this means in language used by us mere mortals...I'd love to know the signal processing algorithm used for CMOS sensors, how much is hardware- how much is firmware/software.
The noise at base ISO is driven by "dark current", the value that a pixel gives without any light hitting it. Saturation count is the maximum value that a pixel can deliver, ie more light hitting it makes no difference. The M9 sensor is somewhere around "2" for dark current and "50,000" for saturated count. The data sheet is not available for the KAF-18500, and I am estimating that based on other datasheets published with devices in the same 6.8um detector family and comparing performance with the M8. Kodak published the KAF-10500 long sheet. So it's easy to determine the dynamic range, the ratio of dark-current to saturation. The M8 had twice the dark-current of the M9, and performance of the M9 at 2500 is almost as good as the M8 at ISO 640. Other factors such as "Quantum Efficiency" also get improved with sensor design.
I'd love to read the data sheets on the newer CMOS sensors, but they are often not published.