Are there really 20 M AF sites? If I look at the picture at page three (daf-area.jpg) it looks much more like a few hundred AF sites instead of 20 million.
That would make sense to me, I cant's see the benefit of reading out every pixel and processing 16M (80% of 20M) AF sites as opposed to having one AF point for every 1000 pixels or so in the same area. And it is probably a lot easier to maintain good high ISO performance that way.
Henrik Herranen: So Ricoh hasn't been able to put there a proper anti-alias filter (short flange distance, angle, etc). My guess would be that there is a very good likelihood that there will be no proper infrared filter either. Perhaps it would be a good idea to put this module through the same motions as the Leica M8 eventually was. Or, as a minimum: point a generic IR remote controller at it and see whether you can see the light flashing on the screen. If yes, there may be problems especially in tungsten light.
If there's really no IR filter and no anti-alias filter, this would potentially be a very good astro camera. The IR filter in most current digital camera's take away much of the near infrared red that is very important for astro photography (Hydrogen Alpha nebulae light at 656 nanometers)