Every so often there needs to be a "reset"And finally, I think it will be a dual processor.... but I am wondering if the time has come for one processor to be optimized for stills and the other processor optimized for video.
That's an interesting thought. I guess it depends on the frame rate. If they only bump up to 10fps, I think a single DIGIC7 could handle it (easily...with room to spare for a whole ton of other stuff). That would be ESPECIALLY if they move the ADC onto the sensor die and make it column parallel. They do have a patent for CP-ADC with a Dual-Scale Ramp ADC (the dual-scale just allows the ADC to operate at different rates based on some trigger factor...say sensor heat...more heat, higher noise, slower readout, less noise...switch from high speed readout to low speed readout when possible under higher heat, and you could counteract the increase in dark current noise...I have no idea what the trigger factor would be to switch from the higher speed to the lower speed or vice versa in an actual product, though.) Parallelizing ADC and putting that logic on the die also reduces load in the DIGIC itself...it would then solely be responsible for digital pixel processing, in which case a DIGIC 7 that has no ADC units could theoretically have even more processing power than a DIGIC 7 that did include the ADC units. So instead of being 7x faster than a DIGIC 6, it might end up being 12x or 14x faster.
If you had one of those dedicated to stills/af/metering, and one dedicated to video, you could really do a hell of a lot with the video. Canon should be able to surpass what the Bionz X in the A7s does easily, achieving ultra low noise ISO 400k, maybe even 800k.
It happened in lenses when AF started to come out. Canon took the brave step of completely redesigning the interface to handle the requirements of digital communication between lens and body and we jumped from FD to EOS... this might be the time for a similar jump on the inside of the camera.
15 years ago, the innards of digital cameras were a collection of ic's with specific functions and the communication between them was both complex and at the same time, fairly limited. At the moment, Canon is doing a major redesign to fit DPAF onto it's sensors and presumably, onto a smaller fabrication technique. This gives us four big possibilities on the sensors. First, with less wasted surface area, the amount of captured light goes up and with that, so does the performance. The second one is with smaller fabrication comes smaller transistors and that means shorter electrical paths and that gives us higher speed. Third is that with smaller transistors and lower voltages, we get less heat, and that means lower noise and longer battery life. Fourth is that with smaller transistors the a/d can be integrated into the sensor and that gives us less noise and a cleaner design.
With all the analog confined to the sensor, one can now communicate digitally between the sensor, processor(s), display devices, and storage medium(s). Gigabyte per second transfer rates are easy to achieve.
15 years ago, there was no video on DSLRs.... now it is a standard feature.... yet we have the same general purpose DIGIC doing both..... it might be time for a split into a dedicated stills processor and a dedicated video processor.
This could be the time for a "reset" of the internal architecture of the Canon DSLRs.....
Regarding the shorter electrical paths bit...I think you may be conflating CPUs with sensors. In a CPU, you can shrink the whole package...and yes, that results in shorter distances for electrons to travel. In a sensor, it isn't really the same. There is a very minimal amount of logic that occurs at the pixel...basically amplify. The distance that amplified charge travels is the same for any given sensor size thought...the top row of pixels is going to have to travel the full 24mm height of the light-sensitive area, plus the additional millimeter or so of masked and calibration pixels around the border, plus the extra millimeter to the CDS unit, plus another short distance to some voltage output (or in the case of CP-ADC, to the ADC units). Shrinking the transistor size doesn't really change charge travel distance in CIS devices, since the whole die generally remains the same. Even if you shortened the distances in the logic around the pixel itself, that is such a trivial distance compared to the total readout distances, I don't think it would be meaningful.
As for lower voltages, for cameras used for normal photography, again I don't think the difference in noise from using a lower voltage on the sensor itself is going to matter much. In Canon's current setup, the sensor itself is actually very low noise. Based on Roger Clark's work, some of Canon's more recent sensors have as little as 1.5e- worth of electronic noise introduced by the sensor itself (or maybe it was 1.2e-). When your FWC is around 30ke- for APS-C and around 67ke- to 92ke- for FF, that is effectively meaningless. The primary source of read noise comes from the downstream (and off the sensor die) electronics, the off-sensor bus, secondary downstream amp and the high frequency ADC units. Those suckers are adding up to 35e- or so worth of noise.
I do believe moving the ADC onto the sensor die is the biggest move that could reduce noise. Eliminating the bus, secondary amp, and massively increasing the parallelism of the on-die ADC units would allow them to operate at a significantly lower frequency. The reduction in operating frequency would have the most significant noise reduction effect. If Canon follows Sony's design, moving the clock and other high frequency components off to a remote corner of the die, away from the ADCs, would practically eliminate the noise they cause.
I also agree that once the analog signal is converted to a digital signal, then you have the freedom to move it around at high speed. You can use error-corrected transfer, and sure, ship the information around at gigahertz error-corrected speeds.
I think that DIGIC is still handling both still and video because up to date, the DIGIC houses the ADC units. Once the ADC units are on the sensor, then sure, I think it would be a lot easier to have some kind of switching unit on the digital bus between the sensor and the processors, and you could dedicate one to stills processing and another to video processing. However, if you have a sufficiently powerful DSP with enough horsepower, it would probably be more energy efficient to hose both of those dedicated processors in a single unit. Especially of the ENTIRE unit can be dedicated to image processing, since the ADC units wouldn't be taking up die space or power.