Check out the work of Yadid-Pecht (conditional reset) and Hornsey (light to frequency modulation), they did exactly the same thing as what is shown here. Please note that it is a 3T pixel with dark issues we have seen two decades ago ....
With stacked architectures, one can think about so many circuits per pixel not being instant product death. Still, there is a power issue in the continuous time preamp, and in the continuous time comparators for the pulse train generator (multiply by 12 million pixels), and it still a lot of circuitry for a 1.5um pixel or smaller. It must be nice to be a computer scientist and not have to worry about pesky practical product hardware issues...although MIT LL ought to know better. On the other hand, why not just increase the counter bit depth to 20 or 30 bits, or make the converter non-linear, or a hundred other things for improving DR, if we are not worried about circuit area or power per pixel?
This approach has also been investigated by thermal imager community for years now. The power consumption and the residual charge measurement remains issues.
Indeed, some of the cited papers are from the innovative work by Brian Tyrell and others at MIT/LL. But, that is for expensive, cooled, DoD application with array sizes of 0.065 Mpixels (as of 2009).
It sounds very similar to the Pixim technology. Does any one know the differences?
ReplyDeleteLooks it won't save time, maybe when ISP is not strong, this can help reduce data to process.
ReplyDeleteIt seems similar to Rambus's proposal.
ReplyDeleteCheck out the work of Yadid-Pecht (conditional reset) and Hornsey (light to frequency modulation), they did exactly the same thing as what is shown here. Please note that it is a 3T pixel with dark issues we have seen two decades ago ....
ReplyDeleteWith stacked architectures, one can think about so many circuits per pixel not being instant product death. Still, there is a power issue in the continuous time preamp, and in the continuous time comparators for the pulse train generator (multiply by 12 million pixels), and it still a lot of circuitry for a 1.5um pixel or smaller. It must be nice to be a computer scientist and not have to worry about pesky practical product hardware issues...although MIT LL ought to know better.
ReplyDeleteOn the other hand, why not just increase the counter bit depth to 20 or 30 bits, or make the converter non-linear, or a hundred other things for improving DR, if we are not worried about circuit area or power per pixel?
There's no need to worry about power consumption to publish a paper :)
DeleteThis approach has also been investigated by thermal imager community for years now. The power consumption and the residual charge measurement remains issues.
ReplyDeleteIndeed, some of the cited papers are from the innovative work by Brian Tyrell and others at MIT/LL. But, that is for expensive, cooled, DoD application with array sizes of 0.065 Mpixels (as of 2009).
Delete