IEEE Solid-State Circuits Society publishes a short tutorial "Circuit Configurations on stacked CMOS Image Sensor" presented at 2019 VLSI Conference by Yusuke Oike, Deputy Senior General Manager and Distinguished Engineer at Sony:
Although BSI is certainly exciting, especially when applied to global shutter, we have seen BSI implemented in many rolling shutter devices in the past. Integrating the ADC into every pixel however seems very new. Aside for the impressive number of ADCs, Sony indicates this architecture would reduce power consumption related to pixel amplification. Wonder what that means in practical terms versus similar sensor with column ADCs.
Integrating an ADC into every pixel is not new. This was already published in 2001 bij S. Kleinfelder of Stanford Univ. Having an ADC per pixel, and having the ADC located in the second layer of a stacked device is new.
The idea of digitizing at the pixel is even older, by far.
Ando @NHK - 1984 US Patent 4,710,817 Fowler et al @Stanford - 1994 ISSCC Fossum, Pain @JPL - 1995 US Patent 5,665,959 (photon counting!)
and probably others. Once you can stack devices, moving the ADC under the pixel itself is obvious, and even mentioned in my 1995 patent.
Making the technology feasible, and performing well, is full credit to Sony. But as usual, Sony marketing always overreaches in communicating their achievements.
I guess he meant that even though the pixel level ADC has a very high current during conversion, if consider the in one frame time, the pixel AD time duration is much short than column ADC. In the end, the pixel AD wins in power in one frame time is considered
Trying here to understand the benefits of pixel ADC. Except for the questionable power saving of comparators, there seem to be many issues: cost (pixel ADCs consume more overall area than column ones), noise (large delta between signal conversion and reset convertion=high 1/f and RTS noise), large leakage (signal stored on FD node for ms time), PLS,...So, what is the killer benefit?
The ADCs themselves are not more efficient. They are too small to make them efficient. In fact, there will be a lot of integrity issues with all these small ADCs operating at the same time. That said, I think you can make a true global shutter. ADCs convert the signal immediately, there are no leaky, sensitive analog memory cells. Then, readout can be faster and lower power. You don't have to use energy and take the time to charge and settle these long and heavy analog column lines. That's all, folks.
Thanks for the feedback. You are right about heavy analog lines to be driven, but here you have to drive lots of digital lines. Also not completely sure with your definition of "there are no leaky, sensitive analog memory cells". In a global shutter sensor you store the signal on relatively large capacitors or in charge domain for a few ms time. With pixel ADCs you store the signal on the very leaky FD node for the same time. Not sure it is a better global shutter approach.
Although BSI is certainly exciting, especially when applied to global shutter, we have seen BSI implemented in many rolling shutter devices in the past. Integrating the ADC into every pixel however seems very new. Aside for the impressive number of ADCs, Sony indicates this architecture would reduce power consumption related to pixel amplification. Wonder what that means in practical terms versus similar sensor with column ADCs.
ReplyDeleteIntegrating an ADC into every pixel is not new. This was already published in 2001 bij S. Kleinfelder of Stanford Univ. Having an ADC per pixel, and having the ADC located in the second layer of a stacked device is new.
ReplyDeletethe latter one makes pixel level AD really meaningful
DeleteThe idea of digitizing at the pixel is even older, by far.
DeleteAndo @NHK - 1984 US Patent 4,710,817
Fowler et al @Stanford - 1994 ISSCC
Fossum, Pain @JPL - 1995 US Patent 5,665,959 (photon counting!)
and probably others. Once you can stack devices, moving the ADC under the pixel itself is obvious, and even mentioned in my 1995 patent.
Making the technology feasible, and performing well, is full credit to Sony. But as usual, Sony marketing always overreaches in communicating their achievements.
This comment has been removed by the author.
DeleteI guess he meant that even though the pixel level ADC has a very high current during conversion, if consider the in one frame time, the pixel AD time duration is much short than column ADC. In the end, the pixel AD wins in power in one frame time is considered
ReplyDeleteTrying here to understand the benefits of pixel ADC. Except for the questionable power saving of comparators, there seem to be many issues: cost (pixel ADCs consume more overall area than column ones), noise (large delta between signal conversion and reset convertion=high 1/f and RTS noise), large leakage (signal stored on FD node for ms time), PLS,...So, what is the killer benefit?
ReplyDeleteThe ADCs themselves are not more efficient. They are too small to make them efficient. In fact, there will be a lot of integrity issues with all these small ADCs operating at the same time.
ReplyDeleteThat said, I think you can make a true global shutter. ADCs convert the signal immediately, there are no leaky, sensitive analog memory cells.
Then, readout can be faster and lower power. You don't have to use energy and take the time to charge and settle these long and heavy analog column lines.
That's all, folks.
Thanks for the feedback. You are right about heavy analog lines to be driven, but here you have to drive lots of digital lines. Also not completely sure with your definition of "there are no leaky, sensitive analog memory cells". In a global shutter sensor you store the signal on relatively large capacitors or in charge domain for a few ms time. With pixel ADCs you store the signal on the very leaky FD node for the same time. Not sure it is a better global shutter approach.
Delete