Dear all,
Here's what I heard from a friend who works in the IT industry, with some clarifications on CCD vs CMOS I found it extremely helpful in understanding the issues regarding digital photography.
---------------------------------------------------------------------------- -------------------------------
Kyocera pulled back N Digital to upgrade the firmware and add in better cooling devices. I have not heard anything regarding to what sensor they will use for the next high end digicam though. With the lesson learned from N Digital and Kodak's 14n, I think Kyocera will be super conservative this time.
The N1/ND was a very ambitious project. Probably it was too ambitious. Kyocera underestimated the engineering efforts around the 6MP Philips sensor. The Philips sensor specs look very good on paper; however, for real world application, it has two major problems: 1. It eats up lots of power. 2. It can generate lots of heat which will increase the noise level. Kyocera had lots of troubles with these two issues. The Philips sensor is Ok for med format digiback because it has the space to install good cooling devices. This is not an option for N1. At the end, Kyocera had to lower the ISO rating to reduce the noise. Also, Kyocera was short of software and firmware engineers. Many things could not implemented correctly and on time. Another minor issue is the corner performance. Even with larger N mount and larger image circle of N lenses, the angle which light strikes the micro-lense at the border of CCD is still off too much. This causes some performance degradation with the 17-35 lens.
CMOS sensor indeed has more noises than CCDs. Another problem with CMOS sensor is that it does not have the idea of shutter. I.e. CMOS sensor is always ON. The good news is that the CMOS process allow the designer to put lots of extra circuits to the same chip. A s&ling circuit can be added beside a sensor to monitor its value all the time. When the camera is not doing the shot, the s&ling circuit will record down the noise level in the sensor. When the camera is taking a picture, the s&ling circuit will act like an artifical shutter, record down the sensor value at the end of the shot then decrease the recorded value with the noise value s&led last time to get the real picture value. With this, you will see a very clean image with little noise. The penality is that this will reduce the dynamic range of the sensor. It hurts the tonal gradation as well. This is the main reason you can see why pictures from Canon D60 and 10D all look very clean; however, they look just more fake than S2Pro and D100's. Anyway, almost all CMOS sensor are using this multi-s&ling noise reduction circuit now. Canon is just the first one which used it on consumer digicams.
One more good thing about CMOS sensor is that the sensor pixel is shaped like a tile, more or less like an area sensor. The light which hits any spot on the tile will be recorded. On the other side, CCD sensor pixel shaped more like a well. A micro lens has to be added to the top of a sensor to focus the light to the right spot. This is the main reason that with the Nikon F-mount, Nikon will never be able to produce a full-frame digicam with CCD sensors. The angle of the light path to the film boundary is too narrow for a micro lens to focus light to the sensor.
With CMOS sensors, Canon has more design freedom. The 1Ds uses bascially a sensor made up by two D60/10D sensors.
Kodak's 14n uses a 14MP CMOS sensor from FillFactory. The sensor is actually more hi-tech than Canon's; however, since it has more pixels, each pixel has to be smaller than Canon's. Small pixels means that each pixel will have less area to accept light. With the smaller Nikon F-mount, the pixels around the border have the light angel issue again. Kodak had to add in complex exposure compensation function to correct this problem. Another issue is that the FillFactory chip was noisy... something Kodak had not expected. On paper, the sensor should have a dynamic range around 11.5bit(69db?) but in reality, it is less than that... I think probably FillFactory has tried too many fancy things with their noise reduction circuit which may not work well all the time. On the other side, Canon just relies on some proven brute-force apparoach.
Sure, Kodak also has lousy software engineers such that the post imaging processing is simply not as good as Canon's. It will take a while for things to improve.
Here's what I heard from a friend who works in the IT industry, with some clarifications on CCD vs CMOS I found it extremely helpful in understanding the issues regarding digital photography.
---------------------------------------------------------------------------- -------------------------------
Kyocera pulled back N Digital to upgrade the firmware and add in better cooling devices. I have not heard anything regarding to what sensor they will use for the next high end digicam though. With the lesson learned from N Digital and Kodak's 14n, I think Kyocera will be super conservative this time.
The N1/ND was a very ambitious project. Probably it was too ambitious. Kyocera underestimated the engineering efforts around the 6MP Philips sensor. The Philips sensor specs look very good on paper; however, for real world application, it has two major problems: 1. It eats up lots of power. 2. It can generate lots of heat which will increase the noise level. Kyocera had lots of troubles with these two issues. The Philips sensor is Ok for med format digiback because it has the space to install good cooling devices. This is not an option for N1. At the end, Kyocera had to lower the ISO rating to reduce the noise. Also, Kyocera was short of software and firmware engineers. Many things could not implemented correctly and on time. Another minor issue is the corner performance. Even with larger N mount and larger image circle of N lenses, the angle which light strikes the micro-lense at the border of CCD is still off too much. This causes some performance degradation with the 17-35 lens.
CMOS sensor indeed has more noises than CCDs. Another problem with CMOS sensor is that it does not have the idea of shutter. I.e. CMOS sensor is always ON. The good news is that the CMOS process allow the designer to put lots of extra circuits to the same chip. A s&ling circuit can be added beside a sensor to monitor its value all the time. When the camera is not doing the shot, the s&ling circuit will record down the noise level in the sensor. When the camera is taking a picture, the s&ling circuit will act like an artifical shutter, record down the sensor value at the end of the shot then decrease the recorded value with the noise value s&led last time to get the real picture value. With this, you will see a very clean image with little noise. The penality is that this will reduce the dynamic range of the sensor. It hurts the tonal gradation as well. This is the main reason you can see why pictures from Canon D60 and 10D all look very clean; however, they look just more fake than S2Pro and D100's. Anyway, almost all CMOS sensor are using this multi-s&ling noise reduction circuit now. Canon is just the first one which used it on consumer digicams.
One more good thing about CMOS sensor is that the sensor pixel is shaped like a tile, more or less like an area sensor. The light which hits any spot on the tile will be recorded. On the other side, CCD sensor pixel shaped more like a well. A micro lens has to be added to the top of a sensor to focus the light to the right spot. This is the main reason that with the Nikon F-mount, Nikon will never be able to produce a full-frame digicam with CCD sensors. The angle of the light path to the film boundary is too narrow for a micro lens to focus light to the sensor.
With CMOS sensors, Canon has more design freedom. The 1Ds uses bascially a sensor made up by two D60/10D sensors.
Kodak's 14n uses a 14MP CMOS sensor from FillFactory. The sensor is actually more hi-tech than Canon's; however, since it has more pixels, each pixel has to be smaller than Canon's. Small pixels means that each pixel will have less area to accept light. With the smaller Nikon F-mount, the pixels around the border have the light angel issue again. Kodak had to add in complex exposure compensation function to correct this problem. Another issue is that the FillFactory chip was noisy... something Kodak had not expected. On paper, the sensor should have a dynamic range around 11.5bit(69db?) but in reality, it is less than that... I think probably FillFactory has tried too many fancy things with their noise reduction circuit which may not work well all the time. On the other side, Canon just relies on some proven brute-force apparoach.
Sure, Kodak also has lousy software engineers such that the post imaging processing is simply not as good as Canon's. It will take a while for things to improve.
