Tuesday, August 24, 2010

Canon Announced 120MP APS-H-sized Sensor

Imaging Resource: Canon announced that it has successfully developed an APS-H-size CMOS image sensor that delivers an image resolution of approximately 120M (13,280 x 9,184 pixels). Its imaging area is about 29.2 x 20.2 sq. mm, suggesting 2.2um pixel size - quite an achievement assuming that Canon process is 0.25um. The sensor's speed is 9.5fps continuous. The sensor can output Full HD video from any approximately one-sixtieth-sized section of its total surface area.

Just over a year ago Canon announced 50MP APS-H sensor. The highest resolution Canon DSLR camera is still limited by 21MP, to the best of my knowledge.


  1. use for lightfield imaging?

  2. Enormous, but the partial 1080p readout seems very unpractical. Converts a wide angle still lens to a video super tele.

  3. Did the 50 MP sensor ever landed in a product ? As far as I know I did not. So I am wondering whether this 120 MP sensor is intended for any photographic product, I do have my doubts.

  4. I am technically impressed by this announcement although in some sense it is similar in difficulty to UDTV (32 MPixels at 60 fps with smaller optical format). Wonder what the yield is like! Also, I wonder what the largest single-chip CCD size is nowadays. Last I heard it was around 111 Mpixels.

    It is interesting to note that output data rates have hovered around 1 Gpix/sec since the late 1990's. Photobit did 1.3 Mpixel at 1000 fps, 4 Mpixel at 200 fps and over the past 10 years the pixel count has gone up significantly and the pixel size down, but the data rate remains more or less at the same level.

  5. These products are well-suited to the toymakers of the defense and intelligence communities.

  6. DALSA announced in April 2010 a CCD of 140 Mpixels (see their NEWS page on www.dalsa.com). Is there anything else on the market with more pixels ?

  7. "It is interesting to note that output data rates have hovered around 1 Gpix/sec since the late 1990's. "

    There's cmos sensors out there running at least 10 times faster.

  8. Photobit has resolved all the problems, gentleman!

  9. "There's cmos sensors out there running at least 10 times faster. "

    So, are they secret? Perhaps they are just buried in this blog somewhere. Let's just for fun try to identify biggest and fastest.
    CMOS: 120 Mpix 1.2 Gpix/s Canon 2010
    CMOS: 25 Mpix 1.3 Gpix/s Cypress 2010
    CCD: 140 Mpix? ?Gpix/s 2010

    Some CMOS to be identified >12 Gpix/s (# bits/pixel?)

  10. 2.4GPix /s is currently being done even in 0.18um Technology.

  11. Eric,

    Well, I remember:


    Since these are already camera's being published, I'm sure the next generation sensors will probably already exist and run at the 10X factor the other anon suggested

  12. Fair enough comments...

    The Photron product is 2 to 4 Mpixel at a pixel throughput of 4.19 Gpixel/sec and the Vision Research Product is 0.9 to 4 Mpixel at 2.7 to 6.4 Gpixel/sec.

    Also, ARGUS-IS is buried in this blog somewhere, at 1.8 Gpixel with internal data rates of 400 Gbit/sec. It's a 368-imager system, not a single chip though.

    Eric's observation is maybe more apt if restricted to the fastest chips just at the highest available resolutions.

    Without too many details available, it seems to me that the imaging chips in the Photron and Vision Research products probably devote vast resources of chip space and power to high-performance ADCs. I'm a little bit skeptical about the pixel depth figures they give, and would guess that to get 12 bits the ADCs are actually something like a pair of 6-bit flash converters, one low-gain and one high-gain.

    Anyway, the Canon products seem more likely to have pitch-matched per-pixel ADCs. With the given figures, and assuming top-half, bottom-half single analog bus lines, single-slope ADCs would be operating at around 44 kHz.

    I think with this particular architecture, the roughly Gpixel/sec throughput may well be a pretty firm limit because you're relying on linearity of a shared periodic ramp signal and can't just bump the frequency up when you want to boost the throughput.

    Without being too much longer (I have to wake a 7-year-old for day 2 of the school year) I'll note that if you forego the ramp linearity in favor of a precision nonlinear waveform, it might be possible to increase the throughput considerably (say, by the popular factor of 10) without changing the pitch-matched ADC circuits much, if at all.

  13. PS My last paragraph above is an open research question, if anyone is interested.

  14. Yes, these are good examples. Some more fast cameras with custom-designed sensors:

    Photron SA5, 1MP @ 7,500fps = 7.5Gpix/s

    Vision Research Phantom v710, ~1MP @ 7,530fps=7.5Gpix/s

    I have not heard about 12 Gpix/s sensors though.

  15. Wow, CDM already posted some of those!

  16. Seems more useful to me to get a Gpix/s/pin number since most of those high Gpix/s have a rather high pincount.

  17. With PCIe 3.0 gearing toward 8Gbps/pin, one can reduce the number of pins using the readily available IP. The only problem is that PCIe 3.0 IP is only available in 45nm and beyond processes, while most CIS processes are not there yet. Sounds like a good business opportunity for IP vendor.

  18. RED Digital Cinema announced that a pre-release version of their 15 MP 120 FPS camera (1.8 Gpix/s) will be used to shoot a feature film next month ("Contagion"), just like they did a few years ago with their 9 MP 24 FPS camera (0.3 Gpix/sec).

  19. what is the main difficulty here ? I think that this is fully dictated by applications.

  20. From another site....
    This development is a step towards Canon’s long-term plan, which was showcased as the Wonder Camera Concept at the Expo 2010 being held in Shanghai.

    According to the demonstrated concept, the future of photography involves extremely high-resolution sensors and video-only capture, from which selected areas and frames may be cropped to result in still images.

  21. This link from one of the companies on the ISW Image Sensor Companies list - http://www.imagerlabs.com/productsHighSpeed.html- claims "up to 20 Gpix/sec".

  22. http://www.imagesensors.org/Past%20Workshops/2013%20Workshop/2013%20Papers/11-5_076-cremers.pdf

    This is 14-bit...


All comments are moderated to avoid spam and personal attacks.