The way described in a Canon patent maybe 18 months ago is:an ultrafast pair of exposures with a few stops difference could make shots possible that aren't, with current technology
Every receptor cell has TWO charge buckets and the sensor can instantly change between charge being distributed in one bucket or the other.
This can be alternated EXTREMELY rapidly, for instance 1 nanosec in bucket 1, 999 in bucket 2, repeating over and over.
So say you take a 1/1000 sec photo, and there's even motion blur in the photo. You'd get 1000 slices of exposure in each bucket, with bucket 1 being 10 stops underexposed. As long as subject movement was slow enough, the result would look continuous in both exposures. Now, if the object moved well over 1000 pixels in the resulting picture, then the underexpose would show the motion as a little disconnected. But that's a fast object to move so far in 1/1000 sec! In the normal case of something moving only a few pixels or a few hundred, and further not so focused that it only illuminates a single pixel while not illuminating neighbors at all, it'd look very smooth.
Besides this capability of nearly doubling dynamic range, you also can use the sensor for global shutter: when finished with a (normal DR) shot in bucket 1, just switch to bucket 2 which you never bother to read. Then take your time to read out bucket 1 and you get the effect of global shutter.
Under normal shooting, of course, you'd think each charge bucket was about half the normal size and thus limited in headroom for bright objects. But in this case it alternates 50/50 between bucket 1 and 2, so can store about the same charge across two buckets the the standard design could in 1.
There's one more trick in the patent I forget.
Anyway, I'm about half-convinced we'll see this special new sensor in a camera slotted above the R5 within a year or two, and if so it will be actually quite revolutionary.