@jirista: Thank you so much for taking your time to elaborate your explanation for me. Now I got a faint idea of what it is all about. I highly appreciate that! Has there leaked any info (patents) that Canon are changing to the CP-ADC approach sometime soon? or let's say within the 5DIIIs or 6Ds body cycle? Could the rumored 5DX contain at least some first components towards this system? Looking forward to read about CP-ADC approaches related to future canon sensor designs....Cheers, Pedro
Glad I could help.
I think the reason Exmor is so good is rather misunderstood...I even misunderstood it for a long time. Then I read a PDF article on it, and it all made sense. On-die parallel digital processing is really the way of the future, I think.
Regarding what Canon has, I really can't say. A lot of their patents are a bit over my head. I have a decent understanding of image sensor technology and electronics, so I can generally understand the bulk of their CIS patents, but when it comes right down to the exact mechanisms, I am not an expert, and I cannot say exactly 100% what Canon has up their sleeve. Also, I've only read the patents that have been granded and are publicly available...I'm sure Canon has patents they have filed over the last year or two that are pending grant, and not yet available to the public. What they may have pending could be anything...some form of CP-ADC is a possibility, CIS active thermal cooling is definitely another...but I have not seen explicit patents about them so far.
What I do know Canon has are regarding on-die hyperparallel readout logic. This does NOT include ADC, it is just column-wise parallel read. One of the readout patents described per-column (rather than per-pixel) amplifiers, which I thought was interesting. Canon also has some interesting noise reduction patents. I don't understand them all (still need to finish reading and researching most of them). One that I did understand involved a power source disconnect feature in readout logic. The concept seemed to boil down to the idea that noise-generating dark current exists only when an active power source is flowing current through a circuit. Disconnecting the active power source, and using capacitance and existing charge to actually read out the pixel value, eliminated dark current as a source of noise.
The majority of the patents I have read seem to involve on-die readout logic. I am not sure that the stuff that occurs on-die (at least in Canon designs) is really the primary source of noise. Canon uses off-die ADCs housed in their DIGIC chips, across a high speed bus. I think the source of the most hated noise in Canon sensors...horizontal and vertical banding noise (HBVN, as I term it), is largely caused by the fact that read pixel data must then travel along a bus, after which they must be converted by high frequency ADCs. Assuming Canon has no on-die digital readout technology, I think their best option to reducing noise would be to eliminate that bus and move the ADCs on-die, in a much more parallel fashion than they have now. Even if it is not fully column-parallel, greatly increasing the parallelism beyond 16 and putting ADC on-die should help to reduce noise considerably.
There is also that ever-present die shrink. As pixel sizes get smaller, Canon is losing out in terms of Q.E. more and more. A 500nm process was quite good for APS-C and FF until a few years ago, however as pixel densities continue to increase, and as more and more logic is put on the same die as the pixels themselves, a 500nm process is going to really start hurting Canon's ability to compete on the sensor front. We've already seen it in the difference in IQ between the 5D III and D800, by an order of two magnitudes. A 180nm process is a three-fold shrink in transistor size. That means a few things. One, they can put more circuitry on-die without needing to dramatically increase the amount of power used (which, being the prime contributor to dark current noise, would have a negative impact on IQ).
Second, the usable area per pixel would increase at all sensor sizes and pixel densities. A 500nm transistor consumes a lot of space, which reduces the total photo-diode area that can actually receive photons. Microlensing helps, but it is far from perfect, and stronger, more precise microlenses are required when larger readout wiring and transistors are used. A 180nm process would allow more photodiode area to be dedicated to "pixel", and less to "transistor and wiring". Combined with better microlens technology, and possibly even better color filtering technology¥
, Canon could bring their Q.E. up to competitive levels (55-65%), which would intrinsically reduce the impact of read noise. Add in their active thermal cooling, and Canon may even be able to turn the tables on their competition, surpassing D800 IQ (even without digital readout technology).
That is a LOT of technology Canon would need to bring to bear in the next cycle. They may have it, but perfecting such technology and making it viable for mass production takes time. If they have only developed some of that technology in the last year or so, it may still be another year or two before it is all actually ready for mass commercial consumption. I think the gambit for Canon is to figure out a way to "go digital". Analog signal processing has its advantages, is an extremely well known and well understood concept, but clearly digital signal processing, in the context of CIS, is and has surpassed it. And as time goes on, hardware DSP will only get better, and I suspect more and more DSP logic will shift on-die. If Canon cannot figure out a way to patent their own on-die CIS DSP technology, they will never really catch up to their primary competition (Sony). I can't say about other competition from other CIS manufacturers...not sure if any of them are actually doing the kind of digital readout that Sony CP-ADC does.
¥ Someone posted a very interesting link to a Panasonic patent that describes a color splitting replacement for color filters, which would allow all of the light incident on the sensor to be utilized, simply by redirecting light of certain colors to pixels of the matching color...i.e. Blue, White-Red, and White+Red. Right now, around 35-40% of the light incident on any given pixel is used, with the rest being filtered out. With color splitting in place of color filtering, the light that is not "appropriate" for any given pixel is simply redirected to the neighboring pixels where it is appropriate. Per-pixel quantum efficiency could technically double, from around 30-40% to 60-80%, if such technology could be idealistically employed. Note, per-pixel Q.E. is usually different than whole-sensor Q.E...currently, several Nikon cameras have greater than 50% Q.E., with one over 60%. That is for the entire sensor. With color splitting technology, theoretically I think Q.E. could reach 90% or more at room temperature. Combine that with some active cooling technology to eliminate thermal sources of noise, and the IQ of a future-generation image sensor with color splitting could blow even the D800 away, especially in extremely low light situations. Imagine taking a photo of an Aurora at a native ISO 204800 with IQ as good as or better than ISO 51200 today.