Ever since Nikon announced their patent on a post-Bayer sensor using dichroic mirrors, everyone has been trying to figure out how they could do that technically. One problem, for example, is that at each pixel you need silicon space to hold all those electrons you're storing, you need room to move the electrons off the sensor or integrate them on the sensor into a pixel value, and you need room for the mirrors themselves.
The answer, it appears, is that a post Bayer sensor is still arrayed as individual colored "photosites" at the deepest level, but you put something on top that "groups" the light for a pixel. The currently circulating documents show a large 12 micron microlense arrangement feeding smaller 6 micron single color "photosites" that live in a pattern arrayed around the blue photosite. The arrangement at first looks a bit like a Fujifilm sensor honeycomb, but that's misleading. The end result is a 7.2mp FX sensor that has better than D3 levels of sensitivity (!) and slightly better than D3 levels of resolution. Seems like a logical candidate for a D4, doesn't it?
You might have balked at the "slightly better than D3 levels of resolution" statement. After all, 7mp is less than 12mp. But that 7mp is RGB pixels, not individual R, G, or B photosites. Just as the Foveon sensors provide more resolution and acuity than you'd expect from their final pixel count, so would Nikon's design. Edge acuity would be excellent in such a system, though the low pixel count could produce visible stairsteps on some diagonal lines.
All this looks to solve most of the problems that were discussed after the original Nikon idea was first announced. The one remaining hurdle appears to be how do you manufacturer the mirrors in the silicon?