On 8/30/06 3:07 PM, "Toralf Lund", <[EMAIL PROTECTED]> wrote:
>> OK, I finally found where these descriptions were. [ ... ] >> >> Anyway, here is what the Japanese article says; >> >> 1. yield from 8" wafer is 200/APS-C, 46/APS-H, 20/FF >> >> 2. number of LSI's on a single wafer is 1000~2000. If for example, >> there were 20 defects but rather uniformly distributed over the wafer, >> it could be possible that "ALL" FF sized sensor could be defective. >> > OK. I was referring to an article someone posted when the 5D was > discussed just after its launch, but of course I can't find it now > (haven't tried too hard, though.) I think it included figures similar to > the ones included above, but claimed that this was the (average) yield > after defective units was taken into account (or counted out, if you > like.) The number of units that can actually fit on a wafer was quoted > as something like 250 for APS-C and 80 for FF. You may notice that the > ratio between those numbers is already larger than the actual difference > in area because you're bound to loose more material near the edges with > larger chips (the wafer is round.) What's worse, however, is that > (according to the article), out of those 80 FF units, as much as 60 on > average are going to be useless. You'll loose some with the smaller > size, too, but "only" about 50 out of 250. > > But the article went on to say (or perhaps began by saying) that it used > to be worse - around the time of the introduction of the first 1Ds only > 5 to 10 of the 80 units would be usable. I don't know enough about IC > production to know for sure how Canon has managed to reduce the number > of defective units, but I'm assuming that the industry has learned over > time how to utilise lower-quality silicon (thus reducing the effects of > "bad" areas on the wafer) and also minimise the number of errors in the > production itself. And I've also been assuming that the production > quality will continue to improve, which will of course be most > beneficial for the larger chips, thus narrowing the gap... I do not know if the yield I quoted (20) already counted out the defects (by definition, yield means that after taking out the defects, right?), however, since they are talking about possible "zero" yield if the 20 wafer defects are distributed evenly, I would assume that 20/FF is a "pre-defect" yield. But calculating the yield count on zero defect wafer of a given size is not a rocket science and can be easily calculated (you do, please :-). They are talking about a 8' wafer and now they are pretty much going for 12". I think the article was merely trying to illustrate the problem associated with producing larger chips like photo sensors when the relative count of wafer defects have a significant impact on the yield compared with normal IC chips which could be produce by thousands. And Canon probably have no control over the wafer defects rate as they do not produce wafers (do they?). They may be exaggerating this effect in order to justify their high price of FF models but the zero yield could be theoretically true. It's a simple arithmetic with a bit of probability theory that the larger the chip is, the more impact by wafer defects. As you say, the improvement of wafer mfg should naturally increase the yield. But it is true that, when we are talking about the yield of 20 or even 80 as you say, it is not like 1000s as in case of IC chips and larger sensors are far more susceptible to a dramatically lower yield, even including near zero. That's the way I think anyway.... :-). Cheers, Ken -- PDML Pentax-Discuss Mail List PDML@pdml.net http://pdml.net/mailman/listinfo/pdml_pdml.net