The SPAD image sensor developed by Canon overcomes the longstanding difficulties of achieving high SPAD pixel counts. By adopting a new circuit technology, Canon was able to realize a digital image resolution of 1MP. Exposure time can be shortened to as little as 3.8ns. In addition, the sensor is capable of up to 24,000 fps with 1 bit output, thus enabling slow-motion capture of fast movement within an extremely short time frame.
The sensor also features a high time resolution as precise as 100 ps. With a high resolution of 1MP and high-speed image capture, it is also able to accurately perform 3D distance measurements.
The camera was jointly developed with scientists at the Swiss Federal Institute of Technology in Lausanne and published in OSA Optica.
First ~MP should be from Panasonic ISSCC 2016 (albeit with some rounding)
ReplyDeleteHello Anonymous,
DeletePersonally I have always wondered about this so a while back I got in touch with the author's of the Canon/EPFL and Panasonic works and they both kindly offered their point of view.
What I understood is that although the device uses avalanche multiplication there was no quantification of its quench behavior. Also, while the device did show a high avalanche gain of 10^4 you could say it is relatively lower than a typical SPAD device gain of 10^6 or so. So strictly speaking it is an APD sensor. Panasonic certainly made a point of their 'capacitive quenching' technique in later publications and only then do they explicitly refer to their VAPD as a SPAD.
Nevertheless the 2016 VAPD Panasonic sensor is certainly worth a mention as it achieved avalanche based single photon counting at such a small pitch and high resolution early on compared to typical SPAD ones.
The attractiveness of SPADs remains in their time-resolved capability which subsequent Panasonic sensors and this Canon/EPFL sensor offer. The continuous drive towards ever smaller and high resolution SPAD sensors by Canon/EPFL and Panasonic in form of the VAPD are undoubtedly tremendous achievements!
Tarek
Any figure on pitch, size or power?
ReplyDeleteThere is an open access paper on this work: https://www.osapublishing.org/optica/abstract.cfm?uri=optica-7-4-346
ReplyDeletethere are 2 flavors of pixels - both 9.4 um. One pixel uses per SPAD 1bit memory and consumes 284 mW @ 24kfps. The other device uses transistor sharing for the read path and hence achieves better FF/PDE but consumes much more power 535 mW. Unclear why that is the case. Both SPADs have impressively small median DCR of ~<2 cps at room temperature. Power should largely depend the light intensity (amount of avalanche bursts) - unclear at which light level this was evaluated - couldn't find anything in the paper