Wednesday, June 17, 2009

sCMOS Sensor Presented

Bio Optics World Magazine: A whitepaper presented this week at the Laser World of Photonics Conference and Exhibition (Munich, Germany, June 15-18) describes scientific CMOS (sCMOS) technology. It is said to be based on next-generation image sensor design and fabrication techniques and capable of out-performing most scientific imaging devices on the market today. sCMOS is said to be able simultaneously offer extremely low noise, rapid frame rates, wide dynamic range, high QE, high resolution, and a large field of view.

The development is the result of combined resources of Andor Technology (Northern Ireland), Fairchild Imaging (United States) and PCO (Germany). The companies have opened sCMOS web site entirely devoted to the new sensors.

Current scientific imaging technology standards suffer limitations in relation to a strong element of 'mutual exclusivity' between performance parameters, i.e. one can be optimized at the expense of others. sCMOS can be considered unique in its ability to concurrently deliver on many key parameters, whilst eradicating the performance drawbacks that have traditionally been associated with conventional CMOS imagers.

Performance highlights of the first sCMOS technology sensor include:
  • Sensor format: 5.5 megapixels (2560(h) x 2160(v))
  • Read noise: < 2 e- rms @ 30 frames/s; < 3 e- rms @ 100 frames/s
  • Maximum frame rate: 105 fps for rolling shutter, 52.5fps for global shutter modes
  • Pixel size: 6.5 um
  • Dynamic range: > 16,000:1 (@ 30 fps)
  • QEmax.: 60%
  • Read out modes: Rolling and Global shutter (user selectable)
About the only omission in these claims is zero image lag. I hope they just forgot to mention it.

Talking about the implementation details, the sensor has split readout scheme, where top and bottom part of the sensor are read out independently. Each column coupled to two amplifiers with high and low gain and two ADCs, as on the figure below:

Update: As the whitepaper says, the sensor has <1% non-linearity, which can be corrected down to 0.2%. Antiblooming protection is stated as better than 10,000:1. The charge transfer time is said to be less that 1us, enabling very fast exposure times. This again rises the question about completeness of the charge transfer and image lag.

Update #2: Below is the chip view. The readout occupies a huge area, even knowing there are two separate readouts on top and bottom and double CDS and ADCs in each of them. Why it's so large? May be because the noise is suppressed in a bruteforce manner by using very large caps?


  1. This is a nice engineering achievement and I guess it is new to combine a lot of these existing features in one sensor. Certainly scientific CMOS APS devices have been a long time in coming.

    Still, I wonder why they compare their device to interline transfer CCDs? A scientific CCD is almost always a full frame CCD and often backside illuminated. (And often operated at cold temperatures.) I think comparison to a scientific CCD would still come out favorable for the Fairchild device.

  2. Hi Eric

    The choice to compare the new technology to interline CCD was more a question of market focus. As you said, sCMOS would compare favourably to back-thinned full frame CCDs, especially under very low light conditions (at higher photon fluxes, the lower read noise of sCMOS would be cancelled out by more favourable signal to shot noise ratio of the back-illuminated CCD) and also in experiements where fast dynamics are required.

    However over the past 10 years or so the interline CCD has been shipping in seriously large volumes to many applications in the bio-imaging sector, especially in microscopy. We felt a Sony Interline CCD was a sensible comparison to kick off with, especilly when considering CCDs with respectable frame rate capabilities.

  3. No problem. We are pretty excited about this new technology so all questions are welcome!

  4. Nice article and excellent results in terms of read noise and dynamic range. But calling it a completely new technology feels more like a marketing trick.

    I also have my doubts about the global shutter mode. I may have missed it in the article, but I wonder how the global shutter is combined with CDS in 5T pixel. Reset and signal from pixels are read out seperately and combined off-chip (halving of the frame rate)? Why is the read noise increased by sqrt(2)? Is this then a 'integrate then read' kind of global shutter? These are possibly important limitations. Further I would like to learn more about the shutter efficiency they achieve in the sensor. If the global shutter is operated as I imagine it is operated, low shutter efficiency may turn this mode useless for high contrast scenes, i.e. spurious charge from shutter inefficiency (and dark current leakage) may be order of magnitudes larger than read noise.

  5. What is new here? At best, this might be a good implementation of already existing technology.

  6. All questions are welcome .... Well, here is one ! You show interesting low noise figures for the rolling shutter, but can you give us also a noise value for the global shutter ?

  7. Actually, paragraph 2.4 of the whitepaper says that the global shutter read noise is 1.4 times of one of rolling shutter at half frame rate.

  8. Just to make myself more clear, part 2.4 of the whitepaper talks about the global shutter read noise being 1.4 times higher than the rolling shutter one. However, the frame rate at which 1.4 factor is measured is not clearly specified. There is a general statement that the global shutter frame rate is twice slower than the rolling shutter one.

  9. It is hard to believe that the total (!) noise in global shutter mode is just 1.4 times higher than in rolling shutter mode. What about kTC noise in global shutter mode ? What about dark current shot noise in global shutter mode ? What about dark current shading in global shutter mode ?

  10. I'd guess that the dark current is reduced by cooling. After all, it's a scientific sensor, so price of added cooling might not be an issue.

    As Anonymous said on 19/6/09 11:59, in global shutter mode the read might be composed of two frames: first reset level is read for whole the array, then pixels globally transfer their charges to floating diffusions, then a frame of the signal is read. Once the reset and the signal frames are subtracted, there should be no kTC noise.

  11. And what about 1/f noise ..... ????

  12. And the leak (dark current) in the floating diffusion?

  13. Well, CDS does not completely eliminate 1/f noise anyway. I would estimate that such a long time global reset CDS leaves twice higher residual noise than the normal short time CDS.

    As for the floating diffusion dark current, they might use cooling, as I mentioned before. Going to -40C makes it negligible.

  14. Are there plans to sell just the chip for development for specialty applications, or will it only be available as a complete camera?


All comments are moderated to avoid spam and personal attacks.