Friday, December 18, 2015

Article About Caeleste

Belgian local DSP Valley Newsletter publishes an article about Caeleste (see page 6). Few quotes:

"Caeleste was created in December 2006, by a few people around the former CTO of Fillfactory/Cypress. After an organic growth start, Caeleste is now 22 persons large and fast growing with a CAGR of 30-50%. It is specialized in the design and supply of custom specific CMOS Image Sensors (CIS). Caeleste is still 100% owned by its founders.

...In fact, Caeleste has almost the largest CIS design team in Europe, exclusively devoted to customer specific CMOS image sensors.

Some of Caeleste R&D projects:
  • Caeleste has the world record in low noise imaging: We have proven a noise limit of 0.34 e-rms; this implies that under low flux conditions (e.g. in astronomy) each and every photon can be counted and that the background is completely black, provided that the dark current is low enough.
  • Caeleste has also designed a dual color X-ray imager, which allows a much better discrimination and diagnosis of cancerous tissue than the conventional grey scale images.
  • Caeleste has also its own patents for 3D imaging, based on Time of Flight (ToF) operation. This structure allows the almost noise free accumulation of multiple laser pulses to enable accurate distance measurements at long distance or with weak laser sources.


  1. "a noise limit of 0.34 e-rms; this implies that under low flux conditions (e.g. in astronomy) each and every photon can be counted"

    To count every photon you need 100% QE. There is a difference between counting every photon and counting every electron generated by the incoming light.

  2. They definitely DO NOT hold the record for low noise imaging! Even for non-avalanche devices. And I think there 0.34e- number was for lower temperatures well below RT.
    That said, the Caeleste team is one of the best custom design houses in the world, right there with Forza and Brookman.

    1. How can you arrive to your conclusion please?

    2. which conclusion? For non-avalanche input-referred read noise it is easy, my group (still) holds the record with Kawahito's group close behind. In terms of being one of the best custom design houses, I just need to look at the published track record.

    3. Well, after the much-publicized departure of their lead designer, on what basis have you concluded on the current team? There is only one design expert now and the rest is a bunch of junior/just graduated guys. Don't dwell just on the past records and marketing bazooka.

    4. Well, I did not know my departure was "much-publicized", or at least not by me. I will let Caeleste answer this, but I am pretty sure there are still competent people at Caeleste. ;)

    5. Thanks Benoit! Without disparaging Benoit, anonymous does not know that we have 4 lead designers, none of which departed. Lift you visor, anonymous

  3. Is there a concrete product foe visible wavelength ?
    An actual instance is SONY IMX 224, a industrial sensor with very low noise of nearly 0.5 e- (even if not a counting photon sensor). And a very low price compared for example to EMCCD.
    I'm waiting for next generation of ultra low noise sensors : EM-CMOS ? Photon counting CMOS ? Intensified CMOS ? Hybrid ?


All comments are moderated to avoid spam.