Imaging Developer Claims Massive Low-Light Improvements
Are these gains theoretical only at this stage, is there a working prototype?
Why wouldn't the additional noise generated by the extra readouts of the charges outweigh the benefits of the selective binning?
If there are any on-die changes to the substrate, wouldn't these possibly reduce the fill factor as well, offsetting gains?
If Rick is monitoring, I'm curious if he has seen this patent and can comment on the similarities and differences between his own...
We think that given market direction, the biggest impact will be in CMOS sensors. No one builds something like this right now, although charge domain binning in CMOS has been done before, and the technology is proven. We have performed the multiple binning in CCDs in our lab, and the actual gains are exactly what you'd predict theoretically. Not surprising. This is not rocket science and astronomers (like me) have been doing sensor binning for decades to increase low-light sensitivity. The new idea here is to do multiple, different binnings in a single exposure without performing a distractive read between the different binnings, and then to use the best parts of the multiply-binned exposures to form the very best image available.
Thanks Rick.
Can you clarify these statements?
Pixon developed additional sensor design components that allow the same electrical charges to be read multiple times, once without binning, and then 1 or more additional times as an image binned different ways (2x2, 3x3, etc.) as necessary
and
The new idea here is to do multiple, different binnings in a single exposure without performing a distractive read between the different binnings
Are you saying there are multiple reads between binnings or not?
Hi. I'll try to clarify. Regarding the "additional sensor design components", there are lots of possibilities. And frankly, what will be implemented would be up to a licensee of our technology. Concerning the binning and readout scheme, again there are multiple possibilities. But illustrate with one of the simplest. Here the sensor would be exposed and readout as usual. Then, without reseting the collected charges, charges in a group of pixels would be binned on the gate of a single output amplifier and the sensor would be readout again. The sensor is then reset and is ready for another exposure. So in this example there are two binnings per exposure, one readout for each binning, followed by sensor charge reset.
Here the sensor would be exposed and readout as usual. Then, without reseting the collected charges, charges in a group of pixels would be binned on the gate of a single output amplifier and the sensor would be readout again.
Thanks. So what I was trying to ask originally, in imprecise language no doubt, was whether the noise generated from the first read degrades the charge read out in the second read.
Ah. Thanks for the clarification. No, the noise of the first read should not contaminate. It is a voltage noise in the readout amplifier, but it should not generate any signal charge that is collected on the gate of the amplifier(s). So the charge signal is uncorrupted. When the charge is binned and regathered on another amplifier gate the noise voltage associated with the new readout is different, independent, and the level of noise generated is that associated with, and typical of, a single amplifier read.
No, the noise of the first read should not contaminate.
Generally speaking, we hear that "the act of measuring alters that which it measures", imperceptibly in everyday experience, e.g. a tire pressure gauge must let some air out of the tire to get a reading, but in a quantum system this can be quite significant.
Does this not apply to multiple readouts (measurements), or is it an insignificant effect?
No. I am a physicist and understand what you are saying. But this acts as a semi-classical system. The charge is preserved. If the transfer is efficient (in CCDs the transfer is 99.999%, or so, efficient), meaning that charge is not lost, we are okay. Now CMOS devices are not that efficient since the voltages (and wells) are not that large (deep). Still, charge is nearly 100% preserved, and signal is not lost. FET gates in CMOS devices are classical devices, and have very high resistance, meaning that there is no leakage or loss (or addition) of charge. And the noise associated in reading the unit cell is transitory and doesn't affect the charge on the amplifier gate. All in all, this doesn't act like a "quantum system" in the sense of measurement affecting the system, or change in signal charge. And measurement uncertainties don't enter here, with the exception that there are read noises when the FET is read due to various noise processes. And depending on the sophistication of the readout circuitry, and the unit cell architecture, some of this noise can be measured and cancelled (e.g., CKT noise, etc.)
Thanks, that makes sense.
One thing I'm curious about, what about inpixel FPN fixed pattern noise readouts, (for noise reduction) which can't be measured with out charge destruction? Do you do without them in the non-binned readouts or somehow capture a single FPN reading (per frame) applicable to all the readouts?
Hope I'm making sense. :)
Fixed pattern noise is always a problem. But this can be corrected if one takes care. And no, we want to measure this if possible, and perform all the corrections that we can. Also, the fixed pattern noise will be reduced in the binned images, as binning is sort of a averaging procedure. Still this is a concern and needs to be respected.
Doesn't Sony have this in there ultra low light cameras?
Not sure, but I just saw this video using a new Starvis Sensor.
Looks like a nice sensor, but what kind of product name begins "Starv"?
No, Sony doesn't have this in their low-light sensors. They do perform charge-domain charge transfer, however, the technology needed for ExDRA. However Sony uses this to perform global shuttering on their devices. They transfer charge, but do not bin them.
Any progress getting this adopted in the industry?
Hi Luis. No, we have not yet been successful in having this adopted. We have presented it to multiple significant players. Some were quite interested, but were not capable of integrating this in their product. Others show adequate capability to adopt it, showed modest interest, but were too busy bringing out new features of their own to which they were already committed. But we haven't given up, and are looking into ways to ease the barrier to adoption.
What’s your take on this imager/camera?
Canons New 4 Million ISO 2MP Camera
Do you think it’s all they say? Is the price likely to come down significantly?
Ask questions and get answers to your physical security questions from IPVM team members and fellow subscribers.