That's a comment we hear often, but I have never seen it back up by technical explanations or measured data.
In terms of resulting images, what exactly is this difference?
I cannot answer the second question, and quality of implementation might overwhelm the structural differences between the two sensor types in question:
1) Full Frame CCD as from Kodak
http://www.kodak.com/global/en/business/IS...l?pq-path=14425 and Dalsa.
Here "Full Frame" is a sensor type, not a size description!
2) Active pixel sensors, usually CMOS but in Panasonic's case NMOS, and generically called CMOS.
The basic difference are embodied in the full description "active pixel CMOS"; there used to be passive pixel CMOS, but it sucked, so Eric Fossum and his colleagues at JPL invented the active pixel approach. All CCD's are all passive devices.
But to the first question, at some length:
The active pixel difference is that there is charge amplification done right at the pixel, or more precisely as part of the transfer of signal from each photosite directly to a "sense capacitor" at the edge of the sensor. In the basic design this is a fixed amplification so that the sense capacitor at the sensor's edge gets twice or more the charge that is in the photosite. Some recent designs have a variable charge gain, so that ISO speed can be adjusted at this stage(*). This amplification makes the signal less sensitive to noise during the rest of the analog processing, so at high ISO speeds in particular it can improve signal to noise ratios. And indeed, active pixel sensors, both CMOS and NMOS seem to do better at high ISO speeds.
A full frame CCD is a passive device and with no extra wiring to support video output, so the photosites are very simple, with a greater proportion of the photosite area available for detecting light and storing photo-electrons. This potentially improves well capacity and thus maximum signal strength, favorable for better S/N ratio when the sensor gets full exposure: at low, base ISO speed. The downside is that the charge must be transferred, unamplified, off the sensor, to off-board amplifiers, and this transfer is done in thousands of hops from photosite to neighboring photosite, first down to the edge and then along the edge to a read-put point. So more noise is likely to be introduced before the amplification, particularly to weak high ISO signals, and this "transport noise" then gets amplified along either the signal.
Note another difference:
- active pixel (CMOS) sensors have thousands of small amplifiers on-chip operating in parallel at relatively low speeds, processing thousands of pixels per second
- CCD's have a small number off-chip (and so potential higher quality) amplifiers than have to operate at far higher rates, processing millions of pixels per second.
All that leads to the prediction that a good FF CCD will be better at low ISO speeds, for DR in particular, while a good active pixel sensor will be better at higher ISO speeds. (Equal pixel size etc. assumed!)
However, my prediction is that as device fabrication moves to ever smaller feature sizes, with the latest 35nm being a tiny fraction of SLR photosite sizes, the space occupied by the active pixel wiring is becoming less and less significant, and so good SLR-sized CMOS sensors have now or soon will have no significant disadvantage in well capacity and DR, moving them towards clear performance superiority.
Why do FF CCD's stay around? One factor apparently is that CCDs are far easier to produce in small quantities, or at least small quantities of custom variations on the same basic photosite design. The sensors unique to the M9 and S2 come to mind, but all MF sensors are in effect low volume spin-offs of CCD sensor technology which Kodak and Dalsa sell mainly to other scientific, engineering and military customers. Look at the Kodak full frame CCD sensor product listing linked above: lots of models, many never used in any camera you have ever heard of; many not even offering color. Fans of big sensors and big pixels might drool over those 70mm diagonal, 50x50mm sensors with 24 micron pixels ... only 4MP though. (They are for X-rays, so ironically are for very high ISO work.)
(*) The Sony Exmor CMOS sensors even do A/D conversion at the bottom of each column of pixels, so clearly apply ISO gain at this early stage; Canon has described a similar approach in a publication about a research prototype sensor, but I do not know if any current product does it this way. My guess is "yes" for the 7D sensor at least.
