jonjt said:
Could you expound on what "so far ahead" means? I'm curious why Canon's sensors are dead last in terms of performance and innovation.
See below.
zlatko said:
What is a fact and what is relevant is not always the same. It depends on the individual and the situation. It is a fact that the Sony sensor is better in certain situations. But that fact is not relevant for all photographers. Obviously it is relevant to some such as yourself — I am not disputing that.
You are still missing the point.
You keep going back to perceptual factors...subjective factors...things that change from person to person. Some people may only use ISO 400-1600 for their photography, and the DR they get from a Canon sensor is more than enough for their needs. Great. But that has nothing to do with what I'm usually talking about.
You keep injecting these external subjective factors, when they simply don't matter when it comes down to the technology. You get annoyed that I say Canon sensors are far behind everyone else. Why? This has nothing to do with whether Canon sensors are good enough for you personally. If they are, GREAT! That's wonderful.
However just because a sensor is good enough for you, or good enough for John and Jane Doe over here, doesn't change the facts. The simple fact of the matter is, Canon sensor technology is really old. They have made minor evolutionary changes here and there, but fundamentally, it's still developed on an ancient process, with old 200mm wafers with lower quality silicon, on a very very old 500nm process, resulting in lower quantum efficiency, lower energy efficiency, lower yield rates, etc.
These are simple facts. It simply is what is...that is the state of Canon sensor technology. They have pushed the 500nm process REALLY freakin far...which in one sense is quite admirable from a "Getting your money's worth" from a particular fabrication process. However, at this point, it's become their Achilles' Heel...the technology is presenting limitations, and while they could keep pushing things, doing so is more likely to hurt performance even more than not. For example, on a 500nm process, the transistors and wiring take up a certain amount of die space. Why hasn't Canon shot strait to 36mp yet, or 47mp? Fill factor would become a concern...they might get more megapixels, but their older technology is probably going to hurt their IQ with smaller pixels on a FF sensor more than benefit it...relative to the competition. The competition is using 180nm and 130nm processes (maybe even smaller, 90nm), which require far less die space, resulting in fewer fill factor problems, making it a lot easier to push pixel count without losing overall light gathering capacity (which impacts overall performance.)
Canon has done an admirable job with the 500nm process, but given the design decisions they have been making the last few years, it seems clear to me that the large transistor size is holding them back. I mean, we are talking about 2000-era fabrication technology here...it's REALLY old. I don't know why Canon hasn't moved to a more advanced process, with larger 300mm wafers made from better silicon. They have 180nm fabs capable of laying down copper interlink wiring (which is much more efficient), which do use 300mm wafers and which are made from better silicon. Those fabs have already produced high efficiency sensors with 59% Q.E. (which is much closer to the top of the line Sony sensors, which are topping 60% these days). Canon has better technology...they simply are not employing it in their APS-C and FF sensors. They are still employing very dated fabrication technology and techniques for these sensors...and the impact of that older technology is quite apparent when you start digging down into Canon RAW data.
I do astrophotography, have for about eight months now. For astrophotography, working the linear signal is very important. Processing astro "sub frames" is very different from processing a RAW normally for regular kinds of photography. You work the data in linear space a lot first, before "stretching" it with an MTF curve. When your working the linear data, it's easy to see the differences. Canon data is very, very noisy, and lacks in color fidelity in the lower echelons of the signal. Exmor data (I've only worked with it from D800 cameras) is far cleaner, with richer data way down near the noise floor. CCD data is the cleanest, with extremely clean, random noise and ultra low dark current (which means the darkest parts of the signal are very smooth, without hot pixels and dark current noise intruding into the signal.) The difference between Canon RAW data and CCD data is huge. The difference between D800 RAW data and CCD is a lot smaller. Some astrophotographers who have applied the black point hack have been referring to D800 data as "near-CCD quality", which is something no one has ever said about Canon data.
The differences are real. Hence the reason I refer to this stuff as fact.
Whether the differences affect your particular kind of photography is a whole different thing. That is not concrete, factual information. It's a matter of perception. It's a matter of subjective judgement. That isn't incorrect, however, it doesn't change the facts either. Not everyone needs to use every part of the sensor. Some people simply don't care about shadow falloff quality. Some people prefer high contrast, in which case they are going to tighten up that contrast curve and crush the blacks anyway. Some people shoot at an ISO range where there are minimal differences between different sensor types. If you don't utilize the RAW signal information in a way that reveals the problems with Canon sensors, GREAT!
However, just because you don't utilize the the RAW signal that way doesn't mean no one does. Again, that doesn't change the FACTS. Canon sensors are fabricated with old technology...a very old, very large 500nm process, on 200mm wafers (most companies stopped using those period years ago, and the ones that still do are not manufacturing huge sensors on them...everyone else uses higher grade 300mm wafers). Canon's readout system is also getting rather dated, and is still utilizing high frequency ADC units (and only a few of them) to handle all the pixel data coming off the sensor, which is largely where their very high DR-killing read noise comes from (many manufacturers have moved to on-die ADC units, one per column, that are able to operate at much lower DR-preserving frequencies that don't generate nearly as much noise), as well as even combining all the image processing into a stacked sensor+DSP package, which shortens transfer bus from centimeters to microns, which also reduces the places where noise can seep into the signal. Canon uses really old sensor technology for their APS-C and FF sensors.
If they change that (they should, at some point...personally I think it's long overdue, but there is probably some logistical or budgetary reason why they haven't yet), and move to their newer fab that uses 300mm wafers and a much higher end 180nm fabrication process, Canon sensor tech would rocket forward. By the time Canon actually does that, 180nm will already be "last generation"...companies are already moving from 180nm and 130nm processes to 90nm processes...but at least the 180nm process is much newer and more advanced. It won't limit Canon's ability to shrink pixel size at the cost of fill factor. It will probably make a move to BSI for APS-C and FF parts much more viable with much smaller pixels. A LOT of good, on the purely technological front, could be realized simply by moving off their 500nm process onto their 180nm process.
Don Haines has a theory that Canon is waiting for P&S sensor production to ramp down on their 180nm fab before they crank up production for APS-C and FF sensors there. I think that's a highly plausible explanation for why they haven't moved to a better, more modern fabrication process and more modern sensor designs yet. Personally, I have to figure that with a 50% or larger contraction in the P&S market the last couple of years, Canon should already have the ability to start manufacturing newer APS-C sensor designs there at the very least...but there could be other limitations...maybe they have to completely ramp down the P&S manufacture before they can ramp up manufacture for something else, I don't know. Anyway, at some point, it should happen. When is the real question...and everyone has differing opinions on that.
