That is ingenious. By having alternating small and large photosensitive areas coupled to fixed size charge storage devices, you get high and low sensitivity pixels with high commonality.
In theory you could have an exponential series to have even more dynamic range.
Or alternatively, if you use powers of two for the sizes, you could use 1-bit sensors, though you probably couldn't do a high bit depth that way. But it'd be extremely fast, as there wouldn't need to be a real A/D conversion. Maybe concentric pixels of varying sizes...
In theory you could have an exponential series to have even more dynamic range.
Or alternatively, if you use powers of two for the sizes, you could use 1-bit sensors, though you probably couldn't do a high bit depth that way. But it'd be extremely fast, as there wouldn't need to be a real A/D conversion. Maybe concentric pixels of varying sizes...