"The sensor can recognize light at one-tenth the brightness conventional sensors need, lending itself to a range of industrial applications such as driving and surveillance.
The number of pixels in a sensor is key to image clarity, and Canon's SPAD sensor boasts 3.2 million, making it the world's densest, more than three times the number of the company's conventional sensor. It can recognize the time of light returning in units of less than a nanosecond, allowing for more accurate object recognition.
Canon's factory in Kawasaki will begin mass-producing the new sensor in the second half of 2022. The company plans to install the sensors in its own security cameras that it is to introduce by the end of next year. It will also market the sensors to other companies.
In 2023, Canon will invest more than 21 billion yen ($185 million) to build a new image sensor plant in the Kanagawa Prefecture city of Hiratsuka, where it plans to gradually increase production of the sensor. The sensor can be made with the production technology used for CMOS sensors and at around the same cost."
Update: Few slides from the company's 2020 presentation:
Digital night vision is really coming !!!
ReplyDeleteAre there any SPAD sensors commercially available pls?
ReplyDeleteSee here, for example:
Deletehttps://www.onsemi.com/products/sensors/photodetectors-sipm-spad
https://www.sony-semicon.co.jp/e/news/2021/2021090601.html
Thanks.
DeleteAnything with at least 0.5Mpx by any chance?
Very interesting! I have a fundamental question though...
ReplyDeleteIs the input-referred noise (read noise + dark current) of SOTA SPAD image sensors lower than PPD-based CIS? Although there is no read noise in SPAD readout (if digital), SPAD dark count rate per unit area seemed to be pretty much higher than the total noise of SOTA PPD-based CIS, at RT for example. I estimated this from some of the recent publications (e.g. Optica 2020 for SPAD, and some JSSC for low-noise CIS) and I have normalized the exposure time as well. Although SPAD can be much more 'sensitive', I think input-referred noise should be smaller to claim to achieve better low-light imaging... Is there something that I am missing?
How about quantum efficiency?
I reviewed my estimation and want to revise what I said. At room temperature, (DCR x exposure time) seemed to be comparable (or in some assumption better) than CIS read noise (e.g. 0.27e-rms), but I expect DCR will rise at 60C, for example 10 times higher and will be greater than CIS read noise. In low-light imaging applications, do you control the temperature somehow?
DeleteCanon claims 60% PDP in visible band and 0.16e- equivalent noise at video frame rate and room temp. See Canon presentation here:
Deletehttps://www.imagesensors.org/Past%20Workshops/2020%20ISSW/KazuhiroMorimoto.pdf
Great, thank you for the link!
DeleteYOu can control the doping profile so that only bulk generated electrons can reach the multiplication zone. In this case, the surface generation will not be multiplied and the dark count rate can be similar to dark current in a normal CIS.
DeleteThere is already a CIS-QIS technology on that delivers deep sub-electron read noise and high dynamic range at high spatial resolution with very low dark current and very high QE. See IEEE EDL doi 10.1109/LED.2021.3072842
ReplyDeleteSo, welcome large pixel SPAD-QIS to the photon-counting image sensor games!
Ok, I guess this comment is deliberately misleading and provocative. Firstly, SPADs greatly precede QIS in both research as well as production. Secondly, QIS is far from mass production even if there is one start-up who announced 2 products. How many of those actually sell per year and how does that compare to SPADs? So welcome to QIS to the photon counting image sensor games!
DeleteBy startup, do you mean Gigajot or Hamamatsu?
DeleteHi Eric, just to avoid any misunderstanding, the above comment from Anonymous is definitely not from Canon team. I totally agree that CIS-QIS is way ahead of SPAD imagers in terms of the pixel miniaturization. Instead, SPAD has some unique features in addition to zero read noise operation, and I see a bright future for both technologies.
DeleteExcited to contribute to the photon-counting imaging world!
Sorry Anonymous, but my comment was not intended to be misleading but obviously it provoked you. I agree SPADs, in general, precede QIS in research and production, and it took a long time to get to a 1Mpixel SPAD array. SPAD-QIS refers to an image sensor, not a range sensor, and those are NOT in mass production, to the best of my knowledge, but there have been a few research devices that have been demonstrated at pixel counts less than a 1Mpixel prior to the Canon work (MIT-LL, Edinburgh/ST, EPFL, etc.). I also think Canon has only announced a future product and it is not in mass production. I wonder why you say CIS-QIS is far from mass production? Given it is a niche product, I am not sure it will ever get to mass production but you can buy a camera today with a Gigajot sensor in it (from Gigajot) so it is hardly far from production, and maybe it will get to large volume, and probably faster than Canon gets to large volume.
DeleteI can't say that the photon-counting image sensor games have actually begun in earnest. But Canon with a SPAD-QIS and Hammamatsu with a CIS-QIS (who actually makes that device, I wonder?) and of course Gigajot, are all entering this new game. And an image sensor less than 2Mpixel is probably not even a game player even at this early stage. So in response to this post, I will say welcome Canon to the photon-counting image sensor games.
Given that Eric referred to small pixel pitches indeed I implied Gigajot. Of course, Hamamatsu is by no means a startup. Albeit mostly focusing on niche and not really being one of the huge players REALLY pushing mass production to millions of devices per year. Also from what I see the pixel pitch at Hamamatsu is at 4.6um pretty comparable to recent SPADs (http://image-sensors-world.blogspot.com/2021/05/prnewswire-hamamatsu-photonics-releases.html). SPADs surely are much more widely deployed than QIS today. So Eric's comment was just off. I presume deliberately provocative given that QIS promises smaller pitches.
Deletehi Kazu, thanks for your comment and also congrats to you for spearheading this new possible product line at Canon based on your successful dissertation work with Edoardo at EPFL and the team at Canon. The always-on high gain of a SPAD, practically simultaneous with creation of a photo-electron, is a huge indisputable advantage over QIS in some applications, and dual-mode operation (time stamping & 1bQIS) at the multimegapixel level could be quite interesting. Good luck to Canon!
DeleteMaybe I missed it, but what is the SPAD pixel pitch on this sensor?
ReplyDeletehttps://global.canon/en/news/2021/20211215.html
DeleteThe SPAD Imager has a large power consumption, I think this type of sensor is suitable for scientific application, but not for always on application.
ReplyDelete