Can Moore's law and its digital implications constrain us too much?
An interesting article in The Wall Street Journal discussed how the greatest asset of giant book retailer Barnes & Noble booksellers has become its greatest headache, "E-Books Rewrite Bookselling" In this case, B&N's claim to fame and market strength was based on its many huge stores with their enormous, in-depth, tangible selection. But the rise of e-books, the slow economy, the decline of book reading as a casual activity, and the proliferation of online bookstores even for physical books, have made these superstores expensive, high-overhead holdovers facing declining markets and new competition.
It's easy to look back in hindsight and say that they should have seen it coming, and smugly say what they should have done, but we won't; that would be unfair. Instead, we'll look closer to home, at our own Industry, driven by ever-shrinking feature size, increasing functionality on a chip, bigger and bigger wafers, and fabs which cost billions, have construction lead times of years, and must have full-capacity utilization rates and high yields to make sense. In response, many companies have gone to a fab-lite model, and make extensive or exclusive use of outside foundries. There is no sensible other choice, in most cases
Even with that model, the up-front cost to develop a sophisticated digital IC is enormous. To make a decent profit and ROI, you have to sell ever-more of these ICs, and their ASPs are often declining at the same time. It's a tough path to navigate. Aggravating the problem, the number of market opportunities that will support the volumes you need for that IC are fairly limited. Worse, the product life-cycle is so short that the chip had better get designed in quick and make it big fast, since it will be obsolete pretty soon. Result: the promise of Moore's law has turned into a runaway train and we're not sure how to slow it down, or if we should.
Then there's the analog IC world: what a difference. In general, analog components (which include power and RF ICs and some discretes) don’t depend on process shrinks to the same extent as digital ICs, they don’t try to pack more and more active devices and functions onto the IC, and they tend to have a much longer viable life in the market. Further, while many analog vendors have no fabs, some still do and have even invested heavily in new ones, for various technical and proprietary reasons.
Whichever way it is—fabless, internal fab, or a mix—the ground rules for analog IC business success are not defined to such an extreme extent by Moore's law. Thus, analog vendors have more flexibility and degrees of freedom in what they do and how they do it. They are not on that runaway train, but can benefit from some of its cargo and capacity, which is a good way to go.?