CMOS
From BioImagingUKWiki
Chair: Andrew Vaughan (UCL)
The following notes were prepared after a discussion at the Facility Managers’ Meeting in Newcastle on the 9th and 10th January 2012. Since that time there will have been new developments in scientific CMOS camera technology, and some or all of the issues below may well have been addressed.
Presentation
I gave a talk briefly explaining the difference between CCD and CMOS technology, before presenting some comparative tests I performed with the Hamamatsu Orca Flash2.8 scientific CMOS camera, the C9100-13 EMCCD camera and a C4742-95 Orca-ER CCD camera. There then followed a discussion of scientific CMOS camera technology involving scientists and industry representatives. The following points were discussed.
sCMOS vs. scientific CMOS
sCMOS is a trade name used exclusively by those cameras with specifically fabricated Fairchild Imaging chips (Andor, PCO). Other scientific CMOS cameras use Sony chips that have not been specifically fabricated for scientific imaging, although they do have lower read noise than conventional CMOS cameras and are faster than CCD chips. The pixel size of the Sony chip is small (3.63 microns) because high resolution is normally what is required in most market places. The Fairchild chip has 6.5 μm pixels to optimise photon collection and dynamic range.
Camera Bias vs. Baseline Clamp
Most scientific CCD cameras have a fixed bias that acts as a baseline for such things as signal to noise ratio measurements. CMOS chips may not have a fixed bias and may instead clamp to the baseline and rescale accordingly, which makes conventional gain and SNR calculations difficult because a fixed bias cannot be subtracted from the signal.
Noise Sources
The read noise profile is rather broad for scientific CMOS cameras. There are additional sources of noise in CMOS chips that contribute to more outlying pixels and an imperfect Gaussian noise curve. Hot pixel fixedpattern noise can be removed from CCD images by background subtraction but trapped charge, which can appear to be similar because it lingers in the same pixel for a while, cannot be subtracted in this way because it is dynamic. CMOS cameras exhibit random telegraph noise (aka 1/f noise, or burst noise), which produces discrete spikes of noise caused by charges being trapped and released at defects in the oxide.
-Ian Dobbie (Oxford University) reported that the Andor Neo still has some fixed pattern noise that can only be seen at low light levels.
Binning
Binning of pixels in CMOS cameras does not result in the same improvement of signal to noise ratio as in CCD chips. For example, the Andor Neo bins pixels after readout so the signal to noise ratio is improved by 2 rather than 4 in 2x2 binning, 3 rather than 9 in 3x3 binning etc., since noise is √signal.
Rolling vs. Global Shutter
The Fairchild sCMOS chip in the Andor Neo can be operated in rolling or global shutter mode. The former is not ideal for rapidly moving specimens, but has the advantage of lower read noise (down to 1 electron at 200 MHz). Read noise is in the region of 2.5 electrons in global shutter mode. The Hamamatsu Orca Flash2.8 can only be operated in rolling shutter mode.
Can scientific CMOS chips be back-thinned?
This would be technically challenging, but not necessarily impossible. Manufacturers’ representatives did not think it was likely to happen in the near future.
===Scientific CMOS in PALM, STORM and the other molecular localisation microscopy techniques=== On the face of it EM-CCD cameras that can reduce read noise to <1 electron would still seem to be the best choice for these techniques, but the higher resolution of CMOS together with the low (but non-negligible) read noise means that these cameras could also be useful. A paper has already been published where an sCMOS camera was successfully used for PALM imaging1, and Ian Dobbie also reported that localisation microscopy had been performed successfully with scientific CMOS in Oxford. Brighter probes are best when using CMOS because of the read noise floor, but CMOS could be useful in reducing acquisition times because of its faster readout frequencies.
Systems Integration
James Francis of Photometrics predicted that the following year would see more systems manufacturers integrating scientific CMOS cameras into their devices.
High Dynamic Range Imaging
Martin Spitaler (Imperial College) asked whether there was any way that the individual amplifiers and/or rolling shutter of the CMOS chip could be used to acquire images with different gains or exposures per pixel to facilitate acquisition of high dynamic range images. Gareth Sloan of Andor said something like this could be done with Andor’s dual amplifier mode.
Problem with data transfer from Andor Neo to computer
The Andor Neo has two Camera Link connectors to link the camera to a computer, but only one is currently used. The bandwidth of this bus is not sufficient to transfer image data to the computer when the camera is running at top speed, so data has to accumulate in the onboard memory, which is of limited capacity (4GB).
References & Links
1. Huang, ZL., Zhu, H., Long, F., Ma, H., Qin, L., Liu, Y., Ding, J., Zhang, Z., Luo, Q., Zeng, S., “Localization-based super-resolution microscopy with an sCMOS camera” Optics Express 19(20), 19156-19168 (2011)
