In a discussion about sensor yields- intimating (and certainly not disclaiming) that 35mm sensors (parts that exit for a long time- at least in terms of Canon's fab plant) on relatively large die sizes (and I'm assuming here that you consider a 35mm sensor to be a relatively large size) get greater than 90% yields. So either you're intimating that 35mm sensors should be getting 90% yields, or you're drawing a conclusion about the semiconductor industry that hasn't applied to imaging sensors so far and that none of the current imaging sensor manufacturers I've seen are trumpeting, telegraphing or making any noise at all about.
You keep missing the point. The argument that yield is related to die size is true, but die size is not the only factor, nor the most significant factor. The TI digital micro-mirror device (DMD) has a relatively large die size, but the most complex part of the process is fabrication of the tiny micro mirrors that actually twist. Most of the yield loss comes from this part of the process. Drawing a blind argument between die size and yield is naive. Like an image sensor, a DMD chip does not have redundancy which could be used to reroute electrical signals from faulty circuits to redundant or backup circuits. But TI has developed other techniques to recover yield. There are "correctable" faults even on a DMD.
My statement about 90% yield on long-life parts is exactly meant to assert that achieving high levels of yield on FX sensors should not be considered impossible. People who believe in the impossible are, in my view, useless people. They are the ones who thought a mission to the moon was impossible, and they would never have undertaken the challenges needed to make it a reality. The semiconductor industry has solved numerous very challenging problems, and is about to solve the challenges for 22nm. To think that they cannot bring FX sensor yields above 20% is lunacy.
Prove that it is not *representative*.
You don't understand the rules pertaining to burden of proof. It is incumbent upon you to prove your assertion.
I've looked at annual reports, patents, and everything else, and while nobody's pushing numbers like they were ~24 months ago, that's not a huge amount of time- and nobody's annual or even quarterly reports are talking about killer yield numbers. The relative numbers are the same, even if the absolute costs have gone down- and even if yields have gone up some, it's still a surface area problem.
Do you know why you cannot readily find yield numbers? Fab yield is one of the most closely guarded secrets.
The market for FX sensors is currently very small, which does not provide incentive for companies to pour resources into developing new process technologies. For general semiconductor devices, development of new process technologies is very expensive, and getting more so. Consortiums such as Sematech and IMEC are formed to spread the costs. Once the market for FX sensors grows, I see every reason for R&D expenses to rise to meet the demand for volume and price.
Finally, are you asserting that Canon replaces its steppers every 2 years for the same sensor line?
No.
Present a process improvement that will only benefit large sensors- the gain to smaller sensors may not be as material, but it will still apply
Explain clearly why the larger die size of a Full Frame sensor is so difficult to manufacture. What are the specific fabrication problems? Now explain why these fabrication problems don't affect smaller die sizes, but only affect large die sizes. Now if we find a solution for the large die size problem, it will have the greatest benefit for large die and maybe nothing for small die.
Now let's say we have 100% yield on small sensors because we've made them really small. And we have 20% yield on large sensors because, well, because they're large. Now if I increase my yield on large sensors from 20% to 50%, will I increase my yield on small sensors from 100% to 120%? I would really like to see that!
The point is clear: you have to understand the actual fabrication issues that limit yield. Citing only die size is naive and misleading.
- edge defocus issues have been solved, which is why it (reportedly) no longer takes more than one patterning step to produce a 35mm sensor- but multiple exposures didn't do anything to yield, it just spoke to process improvements- and it meant that you could expose more smaller sensors in a step- so it's not unique to larger sensors. It still doesn't address yields, but it's not a large sensor only improvement.
The Canon APS-H sensor used in the 1D (not 1Ds) body is, according to Canon's white paper, the largest die size that can be printed with one exposure. Their full frame sensor is printed with 3 exposures, but these are 3 smaller reticles than the single APS-H reticle because the FF sensor is not 3 times the size of the APS-H sensor. Canon solved the edge defocus issue by exposing the die 3 times, but the APS-H sensor is more sensitive to edge defocus.
Now are you saying that 3 separate exposures with smaller reticles did
nothing to improve Canon's yield on FF sensors? Wow. I really need a reference for that.
What part of "to achieve parity" don't you understand? Your mythical, but unknown "leap" isn't realistic as far as I can see.
Because I don't think you're really looking. You really seem fixated on the notion that nothing can be done to improve yield on FF sensors, so the industry might as well throw up its arms and give up. That's a really wonderful attitude...
But again, it does nothing to address yields, so it's a strawman- more 35mm sensors means even more APS-C sensors, larger wafers isn't a sensor yield issue, it's a sensors-per-process-step issue.
But it does reduce cost because the number of good die per wafer increases. After all, we are talking about bringing down the price of FF sensors. That can be done by yield improvement and by more die per wafer. I'm surprised you didn't follow this argument.
Unlike your assertion that alignment issues aren't a big deal, Canon's last annual report trumpets their new solution for alignment. They're also focused on inspecting for defects up front- which doesn't exactly scream "way cool improvements in yields by solving a problem with the wafer" now does it?
There are two types of stepper alignment. One is die-to-die alignment. The other is die-on-die overlay alignment for subsequent mask layers. Both are well-understood problems and have been for a while. By using 3 exposures for a FF sensor, Canon has to accomplish both die-on-die and die-to-die alignment in each pass, but even this isn't new.
Again, we have to know what exactly causes yield loss on large sensors to have a meaningful debate. Talking about die size alone is devoid of reason.
Once again, even if 35mm yields doubled, they'd still be 30% behind APS-C yields from two years ago. Nobody seems to be trumpeting a 100% increaase in 35mm sensor yields in the last two years, and since only Canon and the MF manufacturers were doing 35mm processes in volume it's not like there was a huge universe of folks attacking the problem.
Now here we agree. As the market for full frame sensors grows, so will the R&D investment needed to reduce costs.
Feel free to show some real data and anything other than blind hope that the "industry" is going to "make a leap" that has no current substance.
Blind hope? Is that what John F. Kennedy had when he commissioned NASA to put a man on the moon? Is that what the semiconductor industry had when it solved the lithography problem for 45 and 32 nanometer? Yield of a full frame sensor is just another technological problem. You seem to suffer from blind pessimism.