Sunday, January 16, 2022

Quanta Image Sensor Presentation

Eric Fossum presented a keynote at Phototics Spectra Conference last week (free registration) "Quanta Image Sensors: Every Photon Counts, Even in a Smartphone." Few slides:


  1. So the quanta pixel is a classic 4T pixel by reducing the capacitance of the floading diffusion to minimum and the dynamic range is recovered by digital post processing? am I right?

    1. In essence yes. But moreover the idea I guess it to minimize pixel pitch as much as possible beyond what one could accept for regular CIS due to full-well constraints etc. And they you want to do massive spatial and temporal oversampling. Challenge here in the long run is to reduce readout power, although it should help that the ADC resolution can be drastically reduced. E.g. Eric's students have put some research on 1b-ADCs. Still quite some data to read. And then you have to work on the post processing as you said to deal with motion vs. noise. So there are some challenges beyond what is typically observed in CIS. Plenty of room for research for future students...

  2. How does the current denoising algorithm utilize the single-photon resolving capability of QIS? I guess you can estimate an average photon flux based on Poisson distribution and the photon-counting histogram, but I am curious if it has advantages in terms of the photon budget compared to the simple averaging.

  3. (continued from the previous comment) I think the computational photography using 1-b bit-plane is very interesting, and would like to know if there will be something else than the motion alignment with higher timing resolution.


All comments are moderated to avoid spam and personal attacks.