Traditional CCDs vs. CMOS
Traditional Multi-port CCDs are better than CMOS Imagers due to the following issues:
High Quantum Efficiency (QE) of Back-thinned CCDs
CCDs can be back-thinned to make the entire detector surface light-sensitive. Standard silicon detectors have a peak quantum efficiency (QE) around 45% whether they are CCDs or CMOS imagers. Back-thinning CCDs can increase the peak QE to 90% or more. Although CMOS imagers can be back-thinned, they are generally not.
Low CMOS QE due to Fill Factor
Unlike CCDs, CMOS pixels have elextronic circuits inside each pixel. Since these circuits are light-sensitive, they have to be masked. This results in each pixel having a light-sensitive area and a light-insensitive area. The ratio of the light-sensitive area to the total area of the pixel is known as the fill factor and is often of the order of 65%. This means that it cannot be more than 65% as sensitive as an identically-sized CCD pixel.
MTF degradation with microlenses
The QE of CMOS imagers can be effectively increased with the use of microlenses to direct photons to the light-sensitive areas of each pixel. Microlenses can only increase the peak QE to 65 or 70%, but distorts the image while doing so. Microlenses inevitably have interfaces between them where the behavior of photons is uncertain due to the tolerance in manufacturing. Even worse, on-axis photons exhibit a different QE from off-axis photons.
Loss of dynamic range due to interpixel variability in gain and offset
Every pixel from the same segment of a CCD is read out through the same output port and is read out through the same amplifier. It therefore exhibits the same voltage response per electron of signal and the same voltage offset. This means that the CCD output can be tuned so that zero signal is near to digital zero for all the pixels in the same segment of the CCD and that they will all yield a similar response to incident light. Variations in response would be due to slightly different physical sizes of different pixels, but these variations are typically extremely small (<1%). Since CMOS pixels all have different amplifiers they typically have a wide variation in voltage offset and gain. This typically results in the loss of 20% to 30% of the dynamic range of the output.