Also I do think that Foveon-like multi-layer sensors are the future. However:
1. Now, they behave poorly at high ISOs (layers response etc.). In how much amount of time do you think that they will be able to compete with Bayer on 800-6400 ISO?
Implementing the kinds of advancements Sony has built into their Exmor sensors, implementing a BSI design (which it seems the Canon patent below DOES do), using gapless microlenses and large photosites to improve QE, etc. should all help address the green and red layer response. Intriguingly, the blue layer in Foveon is at the top, so it actually has the least noise. I think this helps lead to the perception that blues are better in Foveon-based cameras...they would certainly make for much nicer skies. Red is actually the noisiest channel in a layered design, as like red and blue in a bayer design, it needs to be amplified more due to receiving less light (and being smaller in area). Lower electronic noise would help that, though.
2. I don't know if you're aware but it seems that Canon has a better patent than Foveon & Exmor for a multi-layer sensor:
I do remember reading about that. I am actually glad they have a patent, and I REALLY hope they put it into use soon. That would be awesome. That patent is interesting in that it is not a gapless microlens design. Hopefully that is something they are improving, as a higher QE would help with overall IQ in a layered design. I also think that you could keep pixel sizes larger with a layered design, since you could use the full area of a bayer-style RGBG 2x2 quad for a single stack of BGR layered pixels...that would also help with QE. The BSI design of this patent is encouraging as well, as it puts the photodiode as close to the surface of the die as possible.
I am not sure how the readout wiring works. Relatively recent patents from Sigma showed some solid improvements in their column/row activate and read wiring, making it smaller, eliminating some wires allowing greater pixel area, and improved efficiency allowing faster readout. While I don't care much for Sigma's 46mp marketing (as when putting them head-to-head against bayer designs, they only offer 15mp in terms of image size), layered sensor designs do indeed have a lot more pixels for any given output size, so readout rate will suffer unless some significant improvements are made in that area. Canon did demonstrate a high-speed 120mp APS-H camera a number of years ago...if they could fold that kind of high-rate hyper-parallel readout technology into a layered CMOS design, then they could probably push total pixel counts closer to the 100mp range (which would produce about a 33mp output image.)
In fact, in order to stay on topic , this was the sensor which I've expected in 5D3/1DX. However it seems that it isn't.
I really hope Canon does something innovative with their sensor design soon. I'm pretty much in holdout mode right now. I own a 7D, and I'm happy with everything about it except IQ in some cases. Every so often I get a shot that, for an unknown reason, shows FPN right up into the midtones. Drives me crazy, although I've found ways around that in many cases due to the kind of photography I do (birds, BIF, wildlife mostly). But it just feels so WRONG that I'm encountering such a degrading artifact in my images. And for all the money I've given Canon over the last few years (to the tune of nearly $12,000), I really EXPECT them to do better if they want to keep getting my money.
3. OTOH, related to this, there are reports that 5D3 works without the AA (OLPF) filter. And I mean it works well. (No moire etc.).
Do you think that this thing really stands? An 5D3/E? An 5D3 without AA filter?
The only cases where the AA filter was removed that I know of only removed the first half. Optical low-pass filters usually have a horizontal layer and a vertical layer, with an IR filter sandwiched between them. In the 5D III, it seems the second layer is glued to the CMOS sensor itself, so only half the OLPF was removed. That effectively reduces the anti-aliasing effect by 50%. Additionally, the cases where people described "zero moire" all seemed to be video cases. The lack of moire is more likely due to the fact that the 5D III uses 3x3 pixel binning when shooting video than the removal of the OLPF.