Startup Claims 16.8M Pixel Camera Sensor
Reality Master 101 writes "ZDNet has an arti cle about a new start-up claiming they have a new CMOS-based sensor for digital cameras that produces 16.8 million pixels -- and cheaper than CCD, too. If you're like me and you've been disappointed with the performance of digital cameras compared to film, this sounds exciting! The only question is whether the color and shadow sensitivity will be as good as film, which has also been a limitation of digital cameras."
- Signal to Noise Ratio: The CMOS APS suffers from inferior SNR due to the fact that it must use a surface channel FET to "read out" the individual pixel. CCDs use only buried channels to transfer charge to the readout amp. One company (name fails me) has developed a CMOS "Active Column Sensor" that offers better SNR, but still not as good as a CCD. This property is very important for getting a good contrast in images that have both light and dark areas.
- Fill Factor: The photosensitive area of CCDs can cover 100% of the silicon, whereas in CMOS APS devices, some area of each pixel must be devoted to switching elements, resulting in less than a 100% fill factor.
- Pixel Size: This is related to fill factor. Since the switches can only get so small, the smaller your pixel size, the worse your fill factor. Thus, to maintain a reasonable fill factor (>50%), CMOS pixels have larger minimum pixel size than CCDs, about 18 microns, whereas CCDs are now sold with pixels as small as 6 microns.
- Modulation Transfer Function (MTF): This is a subtle concept, but anyone familiar with electronic filter design can relate to it - it's the transfer function of signal strength as a function of spatial frequency. In layman's terms, it's the ability of a camera to preserve the contrast in regular patterns (like stripes or arrays) imaged by the system. Fill factor comes into play again here - focal planes with less than 100% fill factors introduce aliasing into the MTF, amplifying weirdnesses like what happens when the sportscaster wears a plaid jacket...
Finally, there are some things that people don't take into account when they compare electronic sensors and film:
- Film has a logarithmic response to exposure to light. CCDs and CMOS have a linear response. Therefore, the electronic devices will never be able to match the dynamic range of film, or at least not with a generous dose of innovation. This is very apparent when you light a scene for film, and then light another for video. Professional video cameras use a few tricks to approximate a log response, but the result is far less than perfect.
- A close friend and colleague of mine worked on developing a 12 megapixel CCD intended for use in Cinematic cameras, going so far as fabricating and testing the device. When my friend's company showed the CCD to Dreamworks, the digital cinema folks weren't interested: they had determined that the needed resolution for acceptable cinema was only about 1280x1024 (I know the aspect ratio is wrong, but this is the example they gave). Thus, they weren't at all interested in a 3k x 4k CCD.
Counterintuitive, yes! But it becomes believeable when you remember the descriptions about how flaws and "noise" had to be added after early digital cinema trials resluted in audience dissatifaction about how the image looked too "perfect" or "fake."
One more thing of note: For the past couple of years, Sarnoff Labs and MIT Lincoln Labs have been working on CCDs built using CMOS foundries. MIT uses a SOI process that is very promising, and Sarnoff just uses a big giant P-well to build its CCD in. With these kind of devices, you can achieve same high levels of integration as a CMOS APS and still get the performance of CCDs. The only element that does not improve much is power. CCDs will always require a lot more power than CMOS APSes.
I can see the fnords!