Friday, March 18, 2011

PCO Presentation on sCMOS Sensor

PCO, Germany-based camera maker and Fairchild Imaging's sCMOS sensor development partner, published a nice presentation on sCMOS imager titled "Modern CMOS Image Sensors for Scientific and Industrial Camera Applications" from Workshop on Scintillating Screen Applications in Beam Diagnostics held on Feb 14-15, 2011 in Darmstadt, Germany.

sCMOS 5T pixel has two transfer gates per pixel (problematic architecture considering the possibility of PD-to-TG barrier mismatch between the two gates):



The chip architecture looks like two vertically mirrored sensors on a single die:


There is a very nice graph showing regular sensor's advantage over electron-multiplying detectors, such as emCCD - no excess noise at higher light levels:


Overall, it's a nice and easy to read marketing presentation with light technical content - recommended.

Update: As mentioned in comments, there is a recent Photonics Spectra article comparing EMCCD with sCMOS "EMCCD vs. sCMOS for Microscopic Imaging" written by Photometrics, Arizona-based supplier of CCD cameras. It's interesting to see their comparisons to be somewhat similar to PCO's.

22 comments:

  1. Of course, in EMCCD the gain can be turned off when there is more photons available, say 100, so no excess noise is generated. The graph is misleading.

    ReplyDelete
  2. thank you for posting this.

    I would think that back thinning the sensor would at least diminish if not completely destroy any frame shutter operation.

    ReplyDelete
  3. Tweaking of the EMCCD gain is a tricky business in the best of circumstances.

    In an image where the difference between the highest and lowest signal is small, turning the EM gain on and off would work.

    but for images where the low signal areas are at ~10e- and the high signal areas are at 1000e-, turning the gain off would effectively blind the sensor in the low signal.

    I suppose if the scene is static, the camera could take multiple frames, some with the gain turned on and some with the gain off and then fuse them into a "super DR" image.

    that probably won't work too well in the real science world where the scene is usually dynamic.

    ReplyDelete
  4. In one of life's "coincidences", there is an article comparing the sCMOS with the EMCCD's published in Photonics Spectra.

    it can be found here:

    http://www.photonics.com/Article.aspx?AID=46174&refer=spectraNewsletter&utm_source=spectraNewsletter_2011_03_10&utm_medium=email&utm_campaign=spectraNewsletter

    ReplyDelete
  5. Indeed a coincidence. Photonics Spectra article authors appear to belong to EMCCD camp, it's nice to see their view on that.

    ReplyDelete
  6. nowaday, the "scientific" marketing is shameful. I think that nobody trusts these curves. Only the real field test gives you the real performance.

    ReplyDelete
  7. totally agree with 10:53

    what little integrity existed in the past is long gone and forgotten with today's pressures on the marketeers.

    so to make themselves look good in spec papers, they are willing to shift assumptions, omit inconvenient results and stretch difficult to measure parameters. (QE seems to be the favorite these days)

    ReplyDelete
  8. EMCCD guys are touting large pixels that give better SNR for given photon flux on sensor. Isn't this just the normal trade off of resolution and pixel SNR? Seems like a funny kind of argument to make.

    ReplyDelete
  9. Because they don't know how to make a larger pixel with microlens. Look at the shape of their microlens :)!!!

    ReplyDelete
  10. The shape looks correct according to the fill factor. The central part should be flat.

    ReplyDelete
  11. The center should not be flat.

    ReplyDelete
  12. What is important is that the rays are directed to the photodiode. For large pixels with good fill factor, flatness in the central region is just fine.

    ReplyDelete
  13. NO, the ray doesn't arrive in parallel ...

    ReplyDelete
  14. Of course they don't arrive in parallel. Why don't you play with some ray tracing or just consider the limiting case of fill factor -> 100%

    ReplyDelete
  15. Also, I think this sensor is designed for high end optical systems so near-telecentricity can be expected in most applications.

    ReplyDelete
  16. telecentric doesn't mean that all the ray will be in parallel!

    ReplyDelete
  17. I spent the latter half of last week building the curves in the Photonics Spectra article based on the S/N equations (simple exercise in excel calculating flux per square micron for each of the sensor types and propagating through S/N ratios)- seems like where the rubber hits the road with low photon flux- they may be correct when it comes to EMCCD's outperforming sCMOS on microscopes.

    ReplyDelete
  18. I did calculate and prepare the graphs of the PCO presentation, by using a standard model that is used by Andor also. The main difference is that the dark current assumed by the Photonics Spectra article is the high dark current of the global shutter operation by sCMOS, which is mostly constant plus the exposure time and temperature corresponding part. The rolling shutter values are nearer to reality.
    For the graph in our presentation I assumed 1 second exposure an 20 °C for all sensors. As well in contrary to the Photonics Spectra article in the graph front illuminated sensors (sCMOS and emCCD) and back illuminated sensors (sCMOS and emCCD) are compared.
    In my eyes the articles comparison between large and small pixels is non-sense, because if enough small pixels with the same total area are compared to the same area large pixel, the resulting SNR will be similar, as well if the optics are changed to image the same image to a smaller total area for the smaller pixels.
    regards,
    gerhard holst, pco ag

    ReplyDelete
  19. I agree with the fact that accounting for pixel size, the numbers would be different and be a little more favorable for sCMOS. However changing optics is not cheap nor trivial on a microscope and there are always trade offs (including cost). I wish the sCMOS was capable of on chip binning - then there would be a very different outcome to this discussion.

    ReplyDelete
  20. The binning, I mean the summing of charge carriers before readout really helps at the absolute low end of the signal. If your signal just pops out of the noise, then, now doubt, you would be in the 1-10 photon range, where only a back illuminated emCCD can help. But above that, as I said, the averaging after readout, can do the same job to improve SNR, if the resolution is not of interest.
    I absolutely agree, that adaption of microscope optics is costly and not an easy one. But when I used in the 90's microscopes the slide cameras still used f-mount to connect to a microscope. Nowadays it is c-mount, but why can't the f-mount optics be used again?
    Further, I think the have compared in the Photonics Spectra article front illuminated sCMOS with back illuminated emCCD, which doesn't make a lot of sense to me.
    Regards,
    gerhard holst, pco ag

    ReplyDelete
  21. Hi there,

    When will the BI sensor camera be available? Will it be much more expensive? Should I wait on purchasing the current camera because I'm very interested in low flux sensitivity.

    ReplyDelete
  22. That's a tough one. We have a first sample in house for testing, but it'll take some weeks to do the tests. Best guess and expectation is probably by the end of the year, but maybe our colleagues in Belfast are faster.
    But don't nail me to the date,
    regards,
    gerhard holst, pco ag

    ReplyDelete

All comments are moderated to avoid spam.