The next major innovation from Sony is to combine the image processing logic with the sensor die using a "3D" layered die attachment process. That won't really revolutionize IQ ...
Maybe not revolutionize but will constitute the next major step in performance.
The big thing about the stacked/3D design is that the circuitry layer can be made ... cheaply ... on a state-of-the-art chipmaking process (say 45nm).
This will allow the integration of sophisticated on-sensor signal processing (per-pixel ADCs, dual conversion gain, etc.) - and in turn better overall sensor performance, which might prove impossible to match by a 'standard' sensor.
So, the stacked/3D design has a lot of promise, actually.
And unlike current BSI techologies, it will be much more cost-effective to scale to DSLR sensors.
Sorry, but I highly doubt stacked IC design will allow per-pixel ADC. Per-pixel ADC, just like any other per-pixel circuitry, would need to be PART OF THE PIXEL. That would have to be done on the sensor die, right on the photodiode. That would consume diode space, reducing FWC, which would have an impact on IQ. That is assuming an ADC could even FIT...they are much more complex than something like per-pixel CDS or a basic amplifier, which is a very simple thing in comparison.
I am not really sure what you would gain with per-pixel ADC over what Sony already achieved with per-column ADC anyway...Exmor's readout is about as clean as it is going to get without a more direct way of reducing dark current noise, which at this point is best done with very significant active cooling (i.e. peltier and a highly effective heat removal system, say copper heatsink and copper heat pipes, for freezing or sub-zero temperatures). Exmor already eliminated the downstream components that introduce noise, those being high frequency circuitry, such as a clock generator or PLL.
nope, you have 2 layers so the fil factor will not be hurt, there are papers describing how it can be done
I've read the papers. The layers are not linked directly to the pixels. The sensor layer still has the readout circuitry, which then pipes that information to an image processor via a bus...it is just that the bus is directly integrated into the package, rather than being off the die and requiring transfer across significantly greater distances to some other off-die processor. I've seen no patents or papers that describe directly linking pixels to the packaged processing layer.
active cooling gives in long exposures a noticeable better result and it is exposures with several sec up to minutes .
Nothing we common user have a big benefit from except astro photographer
all sensors has a cooling with heatsink today
In a sensor that introduces no downstream noise, such as Exmor, the primary source of noise is dark current. A heat sink is a PASSIVE cooling device. I am talking about an ACTIVE cooling device...such as a small peltier. Even though Exmor introduces practically zero downstream noise (the ADC introduces practically none as all the high frequency components are located away from the CP-ADC circuitry), it still has about 3e- read noise. That noise contribution comes from dark current in the sensor itself, and is the remnant that CDS is not correcting for. The only way to eliminate that is to cool the sensor. CMOS efficiency seems to reach its peak at about -80°C, at which point noise from dark current is ~200x LOWER than at room temperature. Cool the sensor, and the 13.23 stops of DR the D800 has would likely approach the theoretical maximum...say 13.999 stops of DR. That is another 2/3rds of a stop, which is still significant enough to be useful. It may be most
useful for astrophotographers, but I figure I could use 20 stops of DR at least for my landscape photography...I'd still happily take the extra 2/3rds of a stop that a supercooled sensor would give me. That would also make a higher bit depth that much more viable...a 16-bit ADC on a sensor with active cooling...I see no reason why you couldn't get 15.9 stops of DR.
I don't foresee a peltier being used in a consumer device any time soon...the power requirements are more in line with using a fuel cell than a battery. But, actively moving heat away from the sensor rather than just with a passive heatsink would still introduce efficiency gains, and reduce dark current noise that CDS cannot compensate for.