Omnivision patent application US20140374862 "CMOS image sensor with integrated silicon color filters" by Chiaying Liu, Keh-chiang Ku, Dyson Hsinchih Tai, and Wuzhang Yang presents a silicon color filter 170, 172 combined with SiO2 microlens 155:
what are the main advantages of such silicon based color filters please ? Since the silicon can not cover 100% the pixel area, so the color selectivity could be low.
So this is like foveon but in the same plane. Same colour overlap problems, worse resolution but probably easier to manufacture and a little less noisy (bigger full well perhaps?)
Actually, the color overlap problem would be even worse here than in Foveon sensors because it is proposed that the photodiodes all be the same depth. This means that both the blue and green channels have more red response than in the Foveon devices. Thus, the off-axis terms in any color transformation matrix would be very large.
I was curious how this worked out quantitatively. I could not get anything that looked promising, FSI or BSI. BSI would be difficult to fab anyway, I think. Note that Fig 4 is "backwards" with red on the left instead of the right. Either I am missing something, or Omnivision decided to patent it anyway, just in case some application comes up later where it is useful. Better safe than sorry I suppose.
Maybe they made the wavelength axis in the "backwards" direction to match the sequence of the diodes in Fig 3. What is really amusing is that the red channel has the lowest red response.
what are the main advantages of such silicon based color filters please ? Since the silicon can not cover 100% the pixel area, so the color selectivity could be low.
ReplyDeleteSo this is like foveon but in the same plane. Same colour overlap problems, worse resolution but probably easier to manufacture and a little less noisy (bigger full well perhaps?)
ReplyDeleteActually, the color overlap problem would be even worse here than in Foveon sensors because it is proposed that the photodiodes all be the same depth. This means that both the blue and green channels have more red response than in the Foveon devices. Thus, the off-axis terms in any color transformation matrix would be very large.
DeleteI was curious how this worked out quantitatively. I could not get anything that looked promising, FSI or BSI. BSI would be difficult to fab anyway, I think. Note that Fig 4 is "backwards" with red on the left instead of the right. Either I am missing something, or Omnivision decided to patent it anyway, just in case some application comes up later where it is useful. Better safe than sorry I suppose.
ReplyDeleteEric,
DeleteMaybe they made the wavelength axis in the "backwards" direction to match the sequence of the diodes in Fig 3. What is really amusing is that the red channel has the lowest red response.
Dave