Recent papers and talks on Quanta Image Sensor (QIS)

Started Jun 24, 2013 | Discussions thread
Eric Fossum
OP Eric Fossum Senior Member • Posts: 1,189
Re: Recent papers and talks on Quanta Image Sensor (QIS)

Roland Karlsson wrote:

Eric Fossum wrote:

Roland Karlsson wrote:

Eric Fossum wrote:

I think the main problem is that a 12b counter takes a lot of circuit area and BSI is not going to solve that problem. I am not sure what the advantages are of having the counter in the pixel. We already know having an ADC per pixel is only useful for certain applications. Assuming the count rate per pixel is the same with the counter in or out of the pixel, how do you see the counter helping with the data rate?

The reason for the counter should be to avoid any AD converter at all.

But ... currently AD conversion has very little noise ... so it might be uninteresting.

We don't need an ADC for the QIS. The data is already digital if we can achieve low enough read noise. Aggregating bits to form a pixel value can be done in many ways. Just counting bits in a local neighborhood is maybe the simplest way, and the closest analogy to the way current sensors work. Counting bits in time to perform data compression, which is what I think you started talking about is also simple. I just don't see any advantage to having the counter in the pixel since it takes up a lot of circuit area.

Sorry you missed my analogy to having and ADC per pixel and your proposed counter per pixel (actually such a counter per pixel was proposed a long time ago by NHK I think). The point is that the cost in sensitive sensor area is not worth putting the circuit in the pixel, at least for most applications.

I do understand how the QIS works. This is a thought experiment.

Lets make an ordinary detector, where you have 16 bit bit depth per pixel. Thats plenty today, more common is 12.

Lets compare it to a jots detector with the same amount of bits. I.e. there are 4x4 jots per pixel.

The jots based image has four times higher resolution, but the gradiation per pixel is poor.

So - how do you plan to compete?

Sure, actually the comparison is more like 16x16 jots in space @ 200 nm pitch = 3.2 um in space, and 16 or 32 fields in time giving 16x16x16=4096.

The discussion of multibit QIS or qDIS is to go to maybe 3-4b depth on readout which would increase this storage density to 16x16x16x16 which is certainly adequate.  Reduction of the jot to 100 nm pitch further improves this by another factor of 4.

Last, but hardly least, changing the integration time per field, as discussed in the paper, can yield much higher exposure-referred dynamic range with good exposure-referred SNR.

-- hide signature --

*Have fun, make some money, and learn along the way*

 Eric Fossum's gear list:Eric Fossum's gear list
Sony RX100 II Nikon Coolpix P900
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow