"The novel image sensors of the invention, and methods of using them, are based on various stacked photodiode pixel designs. In one aspect, the stacked photodiode pixels of the invention provide a means of extending dynamic range. The designs comprise, in an exemplary embodiment, an upper and a lower photodiode. In the event of a high photon flux wherein the upper photodiode becomes saturated, the attenuated light received by the lower photodiode may be used to determine the photon flux incident upon the pixel, extending the dynamic range of the system beyond what is capable with a single photodiode.
In another aspect, the stacked image sensors and associated methods of the invention provide a system for color discrimination in low light conditions. In combination with the signal received by the upper photodiode, the attenuated light signal received by the lower photodiode may be used to determine color, based on the predicted attenuation profile of the device, which is wavelength dependent."
Strongly wavelength dependent, how can the final image be reconstructed if the "color" of the photons in each pixel is unknown?
ReplyDeleteBy using a color filter (120 in the image)?!
DeleteColor filters are not monochromatic, there is already a significant difference in absorption depth between one sigma left of the filter's peak and one sigma right of it.
DeleteMaybe foveon style.
DeleteWhat is not absorbed in the top layer is absorbed in the 2nd layer. So, for sure you can get two-color images.
DeleteIn our forthcoming 2015 IISW paper, Leo Anzagira will discuss the use of a DBR inserted between the layers, two different DBRs on different pixels, to generate 4 signals and after processing, 3 colors. It works pretty well - not as good as RGB, but you don't waste many photons either. That is the key point.
Yes for color but here the point is to increase DR by having one node not saturating when the other saturates. The problem is that you do not know the QE ratio between the two notes to match the two layers. It may cause high PRNU in reconstructed images.
DeleteSorry Arnaud, I do not understand your point. Well, I understand that if one signal out of 4, say, saturates, color reproduction becomes difficult - but that is true for all color sensors. In the Forza patent app they are saying if the top detector saturates, you can still get some unsaturated signal in the lower detector. Of course the color is sort of ruined but spatial data might still be obtained. This is also obvious. Lastly, it sounds like you are worried about possible PRNU in that case. For security for example, this is of lesser importance. Still, I imagine you are saying something more important than these things so please explain more.
DeleteThe question is whether a larger full-well by stacking is justified, compared to larger form factor/ larger pixels in terms of cost, yield, noise performance and production challenges.
ReplyDeleteGood question for now larger form factor is cheaper but once stacking is mainstream and teething issues have been sorted I expect it'll end up being viable option. However I expect the gains from additional processing in the lower layer will outweigh larger full well especially if we get to the stage that it can be read out at high speed or non-destructively a large number of times per exposure.
DeleteIndeed. I also think the cost is not favourable for such application.
DeleteI'd rather use stacking for placing ADCs underneath a small group of pixels for much higher parallelism, or for voltage mode global shutter. Place the pixel storage capacitors on the second silicon and you have an unbeatable shutter efficiency..