Canon patent application for improving dual ramp ADC sensor design
There's more that goes into getting high DR and low noise out of today's sensors than simply slapping an ADC onto the sensor and calling it a day.
Many small improvements such as this patent have to be done to improve the response of the sensor and make us all happy.
Canon utilizes what is called a dual ramp ADC for its column parallel ADC sensors that are now in just about every single camera that canon makes sans 7D Mark II .. maybe.
ADC's on the sensor itself aren't infallible. In the case of a ramp ADC, a comparator compares the voltage of the ramp against that of the pixel. If the voltage is the same or higher on the ramp, then comparator turns ON. the time it took to turn on determines the digital value. In the case of a ramp ADC, the time determines the digital value, not the pixel voltage itself.
It's very precise and pretty easy to do on the chip.
However, all sorts of problems can come into play if the ramp voltage isn't exactly precise, or it drifts. This is where Sony EXMOR has added in what they call a post ADC CDS (correlated double sampling) that attempts to remove any ADC error. Given how good the EXMOR Sony sensors are, it does a pretty bang up job of it.
Canon has to keep track of all possible error (aka noise floor) errors can occur and eliminate them on the sensor itself.
On top of that, a dual ramp has it's own issues. A dual ramp is used to speed the process up. As you can imagine, the higher the value stored in the pixel, the longer it will take the comparator to turn on. now times that by 4000 values per row and Houston, we have a problem here. We can't get the data off fast enough!
The second ramp is a sharper slope than the normal ramp and depending on the value threshold stored in the pixel, it will use the normal ramp or the faster ramp to get the value, so the whole sensor read doesn't slow down to a crawl of you are shooting a scene with alot of data in the highlights. However, that usually comes at a cost. The faster the ramp, the less accuracy because of the less time the comparator has to "flip the switch" when the comparison turns true.
This is what canon explains here in this patent:
A solid-state imaging device can be provided with the picture element part by which a plurality of pixels were arranged, and the converter which carries out analog to digital conversion (AD translation) of the pixel signal from each pixel, for example. It may succeed in an AD translation by measuring time until the size relation of the ramp signal and pixel signal changes for example, using the ramp signal with which a signal value changes to linearity with a lapse of time.
By the way, the technology chosen from the a plurality of ramp signals based on a photographing condition, operational mode, etc. may be used in the ramp signal used for preparing and performing [ in which the rate of change (inclination) of the signal value per unit time generates or generates several each other different ramp signals] an AD translation. For example, if time which an AD translation takes can be shortened if inclination uses a large ramp signal, and inclination uses a small ramp signal, accuracy of an AD translation can be made high. As shown in a Patent document 1, it is also possible to determine any of a plurality of ramp signals are used based on the signal value of a pixel signal.
This patent assists in making the comparison a high precision comparison regardless of the ramp used.
Good stuff if this comes in subsequent generations of our sensors. Each little step along the way will improve the sensors. Maybe not dramatically, but they aren't far off what the best sensors can do now as shown by the 5D Mark IV sensor.
This is also why the 6D Mark II may very well have older (and cheaper) generation of the sensor with ADC onboard, with less advanced sampling and error handling as the more complex (and expensive) full frame sensors do in the 5D Mark IV and 1DX Mark II. It also may not have ADC's, it's a guess on our part - Chipworks would be able to tell us if they published it .
Canon does need to do an updated CMOS sensor whitepaper and fill us in with the technologies they are using in today's sensors.
Japan Patent Application 2017-175565