Devil's Advocate: More to Moore's Law

Most of the world of computing and communications depends on constant improvement in semiconductor technology. But, as Martin Brampton explains, those who invented the microprocessor have been misunderstood.

Everyone thinks they are familiar with Moore's Law - you know, the one that tells us something about the power of microprocessors repeatedly doubling. Over a period of years this results in quite remarkable change and people have often expected progress to slow at some point. Yet there is no sign that it will do so for at least another couple of decades.

But familiarity disguises the extent to which the story of Moore's Law is far more tortuous than we commonly imagine. There have been a good many different formulations of laws of this kind. The earliest was made by Intel co-founder Gordon Moore as far back as 1965 but was never described by him as his law.

It relied on the notion of there being a most cost effective semiconductor device. Building chips with too many components is uneconomic because it is excessively difficult and costly. But it is also ineffective to build chips with too low a density. There is thus an optimum. Moore's 1965 claim was that the number of transistors in the most economic chip would double every year.

By 1975 the law had not, in fact, held in the form it was stated. A new formulation was offered that referred not to the most economic device but to the best achievable. Moore had also pushed the doubling period out from one year to two. Despite popular belief, he has never referred to it being 18 months.

Yet the 18-month doubling has become a strangely powerful folk belief. So much so, in fact, that a figure published in a Physics Today article (Birnbaum and Williams, 2000) is illustrated with a graph showing a doubling time of about 26 months. At the same time, the graph has a caption that "Moore's Law is the empirical observation that the number of transistors on a single integrated-circuit chip increases by a factor of four every three years", which indicates that the data shows doubling time of 18 months.

Another way to look at processor developments has been to talk of computing power. This is complicated by the fact that it is not just the processor but the complete chip-set that determines actual processing power. There is also no agreed way to measure processing power. Still, perhaps we should be satisfied with a much less specific claim: computers have got a lot faster over the years.

Sadly that is not the end of the story. Moore pointed out that software has been a drag on faster processing when he said: “A lot of things take longer than they used to on the slower machine. There's just that much more software, I guess." Here we are up against another law, Wirth's Law, which states: "Software gets slower faster than hardware gets faster."

So it is interesting to hear that Intel's engineers are confident they can continue to build faster and faster processors for the next 20 years. But unless software writers are about to mend their ways we will not be seeing our systems running any faster. Perhaps the software will be doing more.

If we take an optimistic line, we can look at the new ways in which increasing processing power has been used and hope for similar changes in the future. Many of the quite dramatic gains in data communications stem directly from the availability of cheap and powerful processors. Ideas for moving data fast have existed for much longer than the practical ability to implement them.

Likewise, while one can wonder how much real gain comes from a better word processor, there are whole classes of new applications that were once impossible. Films and television, for example, are full of computer-generated effects that make many old science-fiction movies look pretty tame. Let's hope that the next 20 years will see equally interesting departures to take advantage of the efforts of the Intel engineers.