"Intel Delivers!"
Wannabmw writes, quoting me: "Tim, Re: "Intel's strength has always been its semiconductor prowess. This sounds obvious, but it bears repeating. It's why "so so" designs end up dominating: the sheer ability to crunch out 15 million Pentium 4s in one quarter makes up for a lot of architectural imperfections."
"Intel can make miracles out of disasters, in other words. That seems to be a common these throughout Intel's history. I hope it continues to hold true."
This is a terribly important point, one I agree with.
I titled this post "Intel Delivers!" This used to be Intel's marketing slogan in the 1970s. Some folks here may not realize how important a slogan (and philosophy) this was at the time...and still is.
The late 60s and early 70s was a time when a dozen or more small chip companies were jostling to get the business of the big computer companies (the Seven Sisters, as they were called, or sometimes, the Seven Dwarves). And to get the business of up-and-coming noncomputer businesses. Some of the chip companies, lost in the mists of time: Monolithic Memories, Advanced Memory Systems (AMS), Electronic Arrays, American Microsystems Inc. (AMI), Synertek, Mostek, and on and on. The big names then were TI, Motorola, National, and Fairchild, besides Intel, which was growing rapidly.
But a common event back then was to "lose the process." Some little trick or accidental process step would be forgotten, or some chemical would be changed slightly, and the fab would "lose the process." Folks at the time often referred to chip-making as a black art. (The problems got worse with larger chips with, obviously, lower yields. Slight alterations in the recipe could produce wafers with _zero_ yield.)
This was especially the case because a lot of the fabs were small (in tilt-up buildings in Mountain View, for example) and generally only one "line" existed for a particular product. So, for example, a bipolar counter chip might have been made by just one small line at, say, Motorola. And maybe by only a couple of technicians who handled the process. Lots of things could cause that line to start zeroing out in yield.
(A friend of mine designed the world's fastest frequency counters for H-P in the mid-70s. When a supplier--I recollect it was Motorola--couldn't deliver as promised, his product went into a tailspin.)
Losing a process could be devastating for a company dependent on a particular product. Many small companies failed when a vendor lost his process.
Intel spent a huge amount of effort to head off this situation. It replicated processes to second, and then third, fabs. It set up formal second-source agreements for products like the 1103A memory (one of the first such second-source agreements ever was for a Canadian company to also make the 1103A).
(There were also some technical reasons Intel was able to more consistently deliver. There were NMOS stability problems which AMS, AMI, Fairchild, and others were grappling with...I used to evaluate some of these turkeys! Intel had some very secret, very clever methods for reflowing pyroglass to get better step coverage, for gettering out ionic contaminants, and for doing "dip-back" oxidations. "Getter III" will ring a bell for a few of you reading this...at least one of you who read SI was one of the chief engineers on this kind of stuff.)
Intel also had a culture which ensured that fabs didn't just muck around with their own ideas of how to do things (often a cause for "losing a process"). Specs, meetings, and what Grove called the "McIntel" hamburger factory--emulating the way McDonald's could replicate their burgers everywhere. Today this is called "Copy Exactly" and Intel can build essentially the same chips in a dozen different fabs.
Back to the 70s. Intel began advertising "Intel Delivers!" This was crucial for the acceptance of RAM as replacement for core memory, for example. And for the acceptance of RAM in electronic switching systems (AT&T), etc.
This was probably a critical philosophy to have when IBM was looking for a CPU to power their new PC. The 68000 was more impressive architecturally than the 8086/8088, but was large in area and transistors (roughly 68,000 square mils and roughly 68,000 transistors, purely by coincidence) and Moto had laid it out on rubylith with X-acto knives! Intel had by then moved to CAD and was already proliferating variants of its CPUs while Moto was trying to take the rubylith and use it to make follow-ons...which were years late in arriving.
IBM reportedly looked at the production and design capabilities of each company and concluded that "Intel Delivers!"
(I'd surmise that a similar evaluation took place for Microsoft's pick of the Intel product for their X-Box.)
As the rest of the chip business got a bit better with being able to deliver and not lose processes so often, Intel stopped using the "Intel Delivers!" slogan. But it still mattered.
So, in closing, Intel has always depended critically on manufacturing abilities. This sounds obvious, as Intel is a manufacturing company. But I think many folks lose sight of this when they talk about how some other particular architecture is "better" in some way. Maybe the Digital Datawhack VLIW Forward-Pumped Hypercache 128-bit CPU is "better," but who will deliver?
--Tim May |
