Katherine, >>>Sure, bad processing is bad processing, and yield and reliability are related on
individual wafers. But that's not the same as implying, as you did, that an entire
fab's output of *wafers that passed final test* is unreliable.<<<
It's not my implication, it's IBM's words (hate to make statements based on another company's words, but that is what they said at a recent conference). Doesn't it make sense that a particular wafer, which could have had poor registration WRT masks or reticles, or some other anomaly, would have marginal, some "good", some bad, die all over the wafer? And, IBM is so very extreme WRT reliability goals for their enterprise servers (not hours or years, but decades of MTBF), that it makes sense to me that they will "eat" a few potentially good but marginal die for reliability sake.
Tony |
