I fjor intervjuade jag en av världens främsta sensorgurus för att reda ut lite olika begrepp kring CCD och CMOS. Han heter Albert Theuwissen och är knuten till Dalsa.
Q: Today the tendency is: CMOS in mobile phones, CCD in compact cameras, CMOS in SLR and CCD in mid format cameras. Why? Will we see CMOS in compacts and mid format in a near future as well?
A: I wished I could predict the future .... CMOS in mobile phones because they are compact, low voltage, low power, a lot of integration on the chip itself, CCD in mid format cameras mainly because of the market penetration and image quality.
Q: The charge in each pixel is converted to a voltage before A/D conversion. Is the ISO setting taken in consideration before the digital conversion? I mean, will for instance a medium exposed (zone V) area of the subject get the same binary code if the ISO is set to 100 or 1600?
A: Yes, the ISO setting is nothing else than an extra gain stage in the analog domain BEFORE the ADC.
Q: I have been teaching the zone system and black and white photography for 25 years in Sweden, and like to look at sensors in a sensitometric way. A high lit area at ISO 100 represents many fotons and many electrons, near the full well capacity of the sensor. A low lit area at ISO 1600 is in the opposit end of the scale, near read noise threshold. How many volts do these end points typically represent?
A: Saturation corresponds (typically) to 1 V, read noise threshold is about 500 uV, so factor of 2000 between the two levels.
Q: What is the advantage in using a 14 bit system as compared to 12 bit when well capacity devided with read noise is less than 12?
A: Never ever let the dynamic range of the camera being defined by the number of bits of the ADC ! On the other hand, the number of bits at the end of the camera output is always LOWER than the number of bits anywhere else in the system. So if you get 12 effective bits out, the sensor itself delivers more than 12 bits, the ADC needs to have more than 12 bits, because during the processing of the signals you will loose 1 or 2 bits.
Q: What is a typical fill factor of a CMOS with 6–7 µm pixels? What is a typical well capacity of such a sensor?
A: Fill factor is very much depending on the technology used. For a conservative technology the fill factor will be around 50 %, for an aggressive technology, the fill factor can go up to 75 %.
Q: Is it correct to assume that the extra circuits on a CMOS pixel takes the same space on a small and a big pixel? Will fill factor increase with bigger pixels?
A: Yes, if you stay with the same fabrication technology, the circuit area remains the same.
Q: Is there a direct correlation between well capacity and area of light sensitive silicon?
A: Yes, the open area of the photodiode defines the fill factor, but it exactly the same area that needs to store the charges.
Q: When will we see backside illuminated sensors in low or medium cost still cameras (compacts and SLR:s)?
A: Very hard to predict, the technology is ready but it comes at a too high price. Several companies are working to reduce the fabrication cost. But for very small pixels (exactly where back-side illumination is an advantage for consumer), cross-talk plays a major role.
Q: According to pre release information about the new Sony sensor in Nikon D3 the performance in high ISO is outstanding, even compared to Canon’s high end cameras. Is this the result of better noise reduction and/or better QE? I have heard that there are very few lost fotons from the low lit areas of high ISO subjects. If so, how did Sony/Nikon manage to do this?
A: I do not know all the tricks, and I can not tell all the tricks neither, but ISO is a matter of signal-to-noise. So if you can increase the signal for the same amount of photons in, and if you can decrease the noise floor, you will increase the signal-to-noise and consequently the ISO.
Q: There are many different types of CMOS sensors around. Conventional photo diod, pinned photo diode, photo gate and charge shared. Are all of these beeing used in SLR cameras today.
A: Mainly pinned photodiodes are used, standard photodiodes seems to be old-fashioned, photogates are completely gone.
In mobile phones (or when the pixels come below 2.5 um), shared pixels are being used, but still relying on the pinned photodiodes concept.
Q: Olympus are talking about Live MOS in their latest SLR E3. What is the difference between CMOS and Live MOS?
A: Marketing ! It is the same technology, same concept.
Q: Are there any CMOS sensors with A/D-conversion on pixel level today? If not, will we see them in the near future?
A: Yes, mainly for machine vision applications. See
www.pixim.com
Q: There is a direct correlation between temperature and read noise. Is it possible to see this difference in pictures shot outside in a cold Swedish winter and in the Sahara desert in summertime?
A: Not for short exposure times (e.g. less than 100 ms), but if you take pictures at night with extremely long exposures you will see the differences.
Q: Everybody wants higher ISO. This can be achieved with bigger silicone area, lower read noise level and higer QE.What is the theoretical achievable limit in ISO speed terms? How far are we today?
A: Good question ! In QE we can gain maximum a factor of 1.5 to 2.
In noise we still can gain more ! Especially the source follower transistor in the pixel is the bad guy, and several institutes/companies are working on improving the noise performance of the pixels. That is also part of the research I am doing with my PhD students at the University of Delft.
Q: The Foveon three layer sensors cought a lot of attension when they were new. What is the main disadvantage with the this type of sensor?
A: Too much overlap between the various colour spectral curves. This is translated in too much cross-talk. Theoretically cross-talk can be perfectly cancelled out by means of "colour matrixing". Unfortunately this is only possible in the case there is no noise. The colour matrix acts as an amplification for noise and makes the three-layer structure very bad in S/N performance when it comes down to low-light level imaging.