Tuesday, August 31, 2010

Canon Developed 8" by 8" CMOS Sensor

Imaging Resource, CNET: Canon announced that it has successfully developed the world's largest CMOS image sensor, with a chip size measuring 202 x 205 mm2. The newly developed CMOS sensor is among the largest chips that can be produced from a 12-inch wafer, and is approximately 40 times the size of Canon's largest commercial CMOS sensor. By ensuring the cleanroom environments during the production process, Canon minimizes image imperfections and dust.

Canon said it has solved large sensor speed problem through an innovative circuit design, making possible the realization of a massive video-compatible CMOS sensor. The sensor makes possible the image capture in one one-hundredth the amount of light required by a 35 mm full-frame CMOS sensor, facilitating the shooting of 60 frame-per-second video with a mere 0.3 lux of illumination, about one-half the brightness of a moonlit night.

Potential applications for the new high-sensitivity CMOS sensor include the video recording of stars in the night sky and nocturnal animal behavior.

No word is said on the sensor resolution or its pixel size. Also, it's not clear what fab Canon used to produce that large sensor. The picture below compares the new sensor with a full-frame 35mm one:

Update: Tech-On presents more details on the Canon's giant sensor. Its pixel pitch is 160um and the resolution is just 1.6MP.

Update #2: Imaging Resource published Canon Expo 2010 report mentioning a huge 300mm wafer-sized sensor. I'm not sure this is the same sensor, as report mentions 600um pixel size, 1MP resolution - the numbers do not make sense. Even assuming 300mm per side (exact size is not mentioned in the report), there would be only 500 x 500 pixels, much below the stated 1MP. The minimum illumination is 1 Lux. The sensor is currently used in a telescope in Japan. Here is the picture from the report:


    If you carefully look then you can see that the device is three-sides buttable. Taking that into account it might be clear what the main application is going to be ....

  2. But does it actually work - notice they showed the die, but no images....

  3. They don't have lenses to use the sensor with, thus no images...:)

  4. even if it will be used in direct-imaging like x-ray applications or whatever like Albert is suggesting, large format lenses cover up to 50 cm of image circles, there are sheet films in analog photography that are bigger than this device...

    example of existing lenses: http://www.largeformatphotography.info/lenses/LF11x14in.html

  5. What happens on the buttable edge ? Is there some image loses ?

  6. It appears to consist of 10 sub-arrays. You can see narrow vertical lines.

    I'm pleased that they're using cleanrooms for the production. That might be an idea that will take off, possibily even replacing building semiconductor devices in cow sheds!

  7. "In the past, enlarging the size of the sensor resulted in an increase in the amount of time required between the receiving and transmission of data signals, which posed a challenge to achieving high-speed readout. Canon, however, solved this problem through an innovative circuit design, making possible the realization of a massive video-compatible CMOS sensor. "

    Any idea what innovative circuit design that is?

  8. I am guessing the narrow vertical lines contain some row addressing circuitry, which may be part of the "innovative circuit design".

    it is also what allows the device to be 3 side buttable.

    if they are willing to sacrifice substantial image area for larger focal plane size, most likely they will be taking multiple images while dithering the sensor (piezo's) and combining the resulting images to create one "perfect" image filling in the missing sections.

    that also will work wonders to improve yield of the sensors, since they can fill in any bad pixels/rows/columns from the 2nd and 3rd images.

    if this is for direct Xray imaging, then the pixels are probably quite large (30um to 100um) and the number of pixels would be fairly small.

    my $0.02.

  9. Is this the first stitched sensor from Canon?

    While it looks a lot like some x-ray sensors, I would be surprised that Canon would enter that field.

    If was for Japanese Defense Forces (satellite, recon) I would be surprised if they could show a picture.

    If it was for astronomy, then I am surprised they don't mention that.

    All in all, I am prepared to be surprised!

  10. Behind the scenes...

    Marketing Guy #1 - "Yeah, the boss wants me to get something out on this today."

    Marketing Guy #2 - "Boring! Does that mean you're not going to make it to the game tonight?"

    MG #1 - "Probably not."

    MG #2 - "Too bad."

    MG #1 - "Wait a minute... I have an idea!"

    MG #2 - "What?"

    MG #1 - "You know that ISW blog is going to pick this up, right?"

    MG #2 - "Yeah."

    MG #1 - "And they always pan press releases, right?"

    MG #2 - "I so hate that. Stupid engineers."

    MG #1 - "What if we...well, what if we have fun writing this release?"

    MG #2 - "What do you mean?"

    MG #1 - "Well, what if we don't give them any of the specifications?"

    MG #2 - "You mean...nothing? Oh... Oh! That's brilliant!"

    MG #1 - "We'll just make a picture where it's three-side buttable."

    MG #2 - "Very excellent!"

    MG #1 - "What should we give them as applications?"

    MG #2 - "Hmm... How about...how about nocturnal animal observation?"

    MG #1 - "OMG! That's epic!" [Note that they're young, hip marketing guys.]

    MG #2 - "This is going to be the best press release ever! You go write it up. I'll go log on so we can post it."

    MG #1 - "Yes! I'm already on it."

    MG #2 - "Pwnage, and we'll make it to the game early!"

  11. Canon is already in the x-ray imaging market (sold about 20K aSi flat-panel sensor to date). 8x8 size is a starting point for medical x-ray applications (other CMOS startups with similar size CMOS sensors for this market as well). So it could very well be for future x-ray products.


  12. Anyone has any idea what's the resolution and pixel size? This could answer also the application question. Very interesting!

  13. See the update at the end of the post.

  14. With 160um pixel and 1.6MP pixels, this is for x-ray for sure. It is mainly targeted for x-ray fluoroscopy applications (Canon and others have already shown several aSi sensor for this type applications in last year's RSNA). A couple Korea companies (affiliated with Samsung and LG) are also working on large format CMOS sensors for x-ray applications and it is very interesting to note that Canon is entering this field as well.

  15. This could be the result of someone trying to make the physically-largest possible CIS array capable of 60 fps video. A fab process would set the maximum array size. Then a transmission structure would have to be selected taking into account the long travel distances for signals. This, in turn, would determine a maximum pixel count that could be supported. The pixel count and the array size would then determine the maximum pixel size.

    Once you'd made such a chip, you'd want some recognition, so you'd have to have a press release, at which point you'd need to come up with a practical reason reason for making the chip in the first place. (Other than just to make the largest one that can do 60 fps, that is.)

  16. 160 um is a huge pixel size, almost as large as a soccer field ! But most probably the pixels will not be a pinned photodiode. On the other hand there is plenty of room to put some nice and intelligent circuitry in the pixels itself. Curious to learn more about it !

  17. These speculative blog threads are tremendously fun. My wife is slightly annoyed that my image sensor interests are rather more a hobby than a profession, but that's what I get for living in a place where there aren't (m)any tech companies, and no image sensor labs. Also, she's taking the kids to the airport at the moment to pick up her parents, so I'm temporarily completely free to do some image sensor goofing-off.

    Anyway, some further questions...

    Do we actually know this chip is three-sided buttable?

    To me there seem to be three edges without the reddish-brown rectangular features and one with, but the chip itself is mounted in a carrier frame of some sort. So, maybe, some of the parts are hidden.

    Also, does anybody have any thoughts on what those reddish-brown rectangular features are?

    They seem remarkably large - each maybe 1/10 to 1/6 of the area of the whole smaller chip. If the smaller chip has 21 Mpixel (or whatever), and the larger chip has only 1.6 Mpixel, you'd think the non-pixel circuitry wouldn't need such a large area. On the other hand, the innovation to solve the problem of transmission distance might use devices that can source or sink really high currents.

    I still think this chip represents someone conducting a little prestige design-and-fab on the side and then having to scramble to justify it to whomever was responsible for the design-and-fab budget.

  18. @CDM, please keep it technical. No need to tell us about wife, kids and dog.

  19. If the wafer size used is 300mm, then the size of the sensor can not be 300mm. In that case the chip is maximum 210mm in width/height, which I still consider as EXTRMELY large. The maximum size of 210mm confirms the data mentioned in the first announcement : 202mm x 205mm.

  20. Yes, I knew that 300mm per side is a gross overestimation. I just said that there is no way to get 1MP resolution with 600um pixels, no matter what sensor size we assume.

  21. What if there is no wafer?

    I've been advocating deposited silicon (amorphous, polycrystalline, or nanocrystalline) on glass as a way of making large sensors for a decade.

  22. 160 um is about the lower limit for resolution (3 lp/mm) for medical fluoroscopy but 200 x 200 mm is the right size to replace nominally 9-inch image intensifier tubes. We designed a 200 x 250 mm fluro panel with 127 um pixels at Varian in the late 90s that is still on the market (dPix a-Si:H panel). There were a couple of 200 um panels in prototype at the time but the resolution was considered too low (2.5 lp/mm).

    The problem then (and now) with amorphous silicon is the noise due to the long read lines. We all knew that big silicon wafer arrays would be better but no one could make them. Canon likely put in some sort of buffers and maybe local read circuitry.

    The question is - what happens to this array under x-ray exposure - does it make white spots when the radiation is on? Does it die from threshold shift early? CMOS is only about 100 Gray rad-hard, not 50 kGray like amorphous silicon.

  23. @ @CDM: Well, technically I did not tell you about wife, kids and dog. :D

    The irreverent - maybe you think irrelevant - color commentary is my personal antidote to the anonymous animus that appears so often here. This should be a place which is welcoming, but too often it's one filled with boasting and bashing. I'm just doing my part to fight the good fight, tongue-in-cheek style.

    @ dimension comments

    The two pictures look like they're of the same product to me. Probably some of the details get lost in translation when the device leaves the lab and goes on the convention circuit.

    @ hardening comments

    The radiation hardening issue is intriguing, and I have a follow-up question.

    Are the radiation figures you give on threshold shift for hard failure, or are they for smaller shifts that make the subject transistors not meet some speed or matching specification?

  24. I did some scouting around and found a similar chip described in a paper from IISW 2009 by Reshef et al., "Large-Format Medical X-Ray CMOS Image Sensor for High Resolution High Frame Rate Applications".

    Reported specifications for the chip are 120 mm by 150 mm for the imaging area, 150 um pitch, and 964 by 786 pixels. There's 12-bit column-parallel single-slope ADC at 60 fps with binning and 30 fps without.

    The radiation lifetime dose is 100 Gray, and the "X-ray maximal linear dose" is 40 uGray. However, the paper shows actual images at 5 uGray and 160 nGray. With these levels, the lifetime would be reached at 20 million and 625 million images respectively, or at the full 60 fps rate, about 93 or 2800 hours.

    I'd say with these figures, the chips wouldn't be good for a long space mission, but would be quite adequate in a hospital setting.

  25. So, I'm still puzzled - is this Canon chip made for astronoly or for medical x-ray applications?

  26. I don't think there is enough public information out there to say one way or the other. The x-ray suggestions here are basically speculation based on the reported chip size and pixel pitch.

  27. I have an application for this sensor. Is it "real", where would you get one and what would be the price?


All comments are moderated to avoid spam and personal attacks.