Digitimes published first stamp-size picture from the new OV8810 1.4um BSI sensor. With such a size I can say nothing about its image quality. Click on picture to make it a little bit larger:
Yesterday I was in the luxury position to have a side-by-side demo of the 8M, BSI sensor, 1.45 um of Omni with their 5M, FSI sensor, 1.75 um. Sensitivity of the 8M is a factor 2 less than the 5 M due to the smaller pixel and despite of the BSI. Overall impression : there is still room to improve as far as colors are concerned, but that was also true for the 5 M. The demo was done with a lot of light. When I asked to lower the light level, it was said this was not possible .... Albert T.
Well, it may be hard to comment on specific performance issues but there are some important points (a) it exists as a mass production part (b) at the scale shown, it takes a decent picture (c) a factor of 2 less sensitivity considering the scaling overhead with pixel size is not too shabby. (i.e. for same design rule, fill factor probably shrinks more than 2/3 when the pixel area is shrunk by by 1.5x)
I am hardly a Omnivision booster but for a first generation high volume BSI sensor I think they appear to be doing alright, at least based on this limited information.
(a) 1.75um pixel uses 0.13um process, while 1.4um BSI uses 0.11um process, so a relative photodiode size should not be much less. (b) BSI fill factor is close to 100%, while in small FSI pixels the effective fill factor is often limited by metal's aperture. If somebody from Chipworks reading this, can you confirm? You have Omnivision's 1.75um reverse engineering, right? (c) The stack hight is much smaller in BSI. I would expect to see some sensitivity improvement because of this fact.
But it's quite impressive to see BSI in consumer product, close to mass production!
Yesterday I was in the luxury position to have a side-by-side demo of the 8M, BSI sensor, 1.45 um of Omni with their 5M, FSI sensor, 1.75 um. Sensitivity of the 8M is a factor 2 less than the 5 M due to the smaller pixel and despite of the BSI. Overall impression : there is still room to improve as far as colors are concerned, but that was also true for the 5 M. The demo was done with a lot of light. When I asked to lower the light level, it was said this was not possible ....
ReplyDeleteAlbert T.
Albert,
ReplyDeleteVery interesting. Thank you for sharing your impressions.
Well, it may be hard to comment on specific performance issues but there are some important points
ReplyDelete(a) it exists as a mass production part
(b) at the scale shown, it takes a decent picture
(c) a factor of 2 less sensitivity considering the scaling overhead with pixel size is not too shabby. (i.e. for same design rule, fill factor probably shrinks more than 2/3 when the pixel area is shrunk by by 1.5x)
I am hardly a Omnivision booster but for a first generation high volume BSI sensor I think they appear to be doing alright, at least based on this limited information.
-EF
EF,
ReplyDeleteI tend to agree with you, with few reservations.
(a) 1.75um pixel uses 0.13um process, while 1.4um BSI uses 0.11um process, so a relative photodiode size should not be much less.
(b) BSI fill factor is close to 100%, while in small FSI pixels the effective fill factor is often limited by metal's aperture. If somebody from Chipworks reading this, can you confirm? You have Omnivision's 1.75um reverse engineering, right?
(c) The stack hight is much smaller in BSI. I would expect to see some sensitivity improvement because of this fact.
But it's quite impressive to see BSI in consumer product, close to mass production!
Oops, my bad: Both Omnivision 1.75um and 1.4um BSI use same 0.11um process.
ReplyDelete