Just so. Intel's manufacturing has ALWAYS been its key asset.
Intel never had the best designs, the best architectures. (Well, maybe some of them were pretty darned good, but there were usually other CPUs, other memory chips, which some thought were better.)
But what Intel had was the ability to deliver. In volume. And at a price that still allowed them to stay in business. The slogan when I started at Intel in 1974 was "Intel Delivers." A comment about how many other companies often "lost the process" on products. For a refresher, look at some 1970s magazines in the library and read the ads and product announcements for companies like Monolithic Memories, Mostek, AMS, Signetics, Fairchild, even Texas Instruments. Most are now long gone.)
Intel had a massive resurgence in manufacturing after the Japanese Scare of the mid-1980s, when the Japanese DRAM makers looked to be utterly dominant. Intel got out of the DRAM market (though we hear it may be re-entering, and certainly the partnership with Micron and the new stacked packaging is crucial).
Craig Barrett re-emphasized manufacturing, with pellicles, "copy exactly" (kudos to Andy Grove on this as well), and yields went sky-high. (I was long gone by then, so have no inside information, but apparently the yields approached "700 ISO," which basically meant nearly zero defects. (The "ISO" stood for "isodefect density," the curves for the yields of good dies at the same defect levels at various die areas, and, as has long been publicly described, was a deliberately-obfuscating term for "number of good dies of a certain size on a certain size of wafer," with about 700 of these 100 mils by 100 mil dies on a 2-inch wafer (!!) essentially representing a perfect yield. That is, a perfect 3-inch wafer would give 700 of these very small dies. Larger dies or higher defect densities would drop the yields by the usual Poisson distribution calculation. This is more accurate than talking about 60% yield, or 90% yield, whatever, because the die area plays a major role (via the Poisson distribution of course, the yield being the dies which have no defects given the probability of having one or more defects via the defects per unit area). The "obfuscating" part is that one did not want idle chatter about how many defects per square centimeter were being seen, so instead the talk was of "The 600 ISO Project" and the like.
I was astounded in the early 90s to hear reports that Intel was hitting close to "700 ISO" on key microprocessors. I never publicly commented (where would I?), but this was one reason I held on to my Intel shares and was not surprised to see Intel resume the manufacturing lead over the Japanese. (Who dey? Are the like Samsung or TSMC?)
Manufacturing has always been the key to Intel's success. "Intel Delivers."
--Tim May
AI
