Micron filed a patent application US20080291310 on 4T pixel with no global RST transistor wiring. The principle is quite self-explanatory from the picture below:
Getting rid of the reset pin is a big step foward. I hear that the omnipixel 3-HS uses this kind of thing to improve light sensitivity, but I've never actually seen anything with those kind of OV pixels. Don't know of anybody else who has found a way around the reset problem.
Maybe this is not only for the improvment of the light sensitivity. Eliminate RST routing provides a routing for GND line to connect substrate in each pixel for example ...
There is a payment for simplicity. Now that RST comes vertically they need to reset the entire pixel array to read a particular row. This looks like unnecessary power consumption. Performing RST once a frame may cause issues capturing an overexposed scene (assume A-nodes leak to gnd). One may also expect lower voltage range and higher PRNU through source follower capacitance reattached from RST to pix_out.
Perfoming RST once a frame is probably not an option in shared pixels. Power penalty might be not that big - even assuming 10nF array reseted every 10us to 3V we get just 3mA current.
Yes, it was presented on the conferences already. However, the patent application is broader than the conference papers, as it talks about few different implementations of this idea.
Getting rid of the reset pin is a big step foward. I hear that the omnipixel 3-HS uses this kind of thing to improve light sensitivity, but I've never actually seen anything with those kind of OV pixels. Don't know of anybody else who has found a way around the reset problem.
ReplyDeleteYes, RST routing elimination is a big deal. It's great to hear that Omnivision thinks in the same direction.
ReplyDeleteMaybe this is not only for the improvment of the light sensitivity. Eliminate RST routing provides a routing for GND line to connect substrate in each pixel for example ...
ReplyDeleteThere is a payment for simplicity. Now that RST comes vertically they need to reset the entire pixel array to read a particular row. This looks like unnecessary power consumption. Performing RST once a frame may cause issues capturing an overexposed scene (assume A-nodes leak to gnd). One may also expect lower voltage range and higher PRNU through source follower capacitance reattached from RST to pix_out.
ReplyDeletePerfoming RST once a frame is probably not an option in shared pixels. Power penalty might be not that big - even assuming 10nF array reseted every 10us to 3V we get just 3mA current.
ReplyDeleteThis idea was already presented at the International Solid-State Circuits Conference. Does anyone know whether the idea is already used in products ?
ReplyDeleteYes, it was presented on the conferences already. However, the patent application is broader than the conference papers, as it talks about few different implementations of this idea.
ReplyDelete