April marks the 40th anniversary of Moore's law. Nobody in 1965 would have dreamed it would become so popular that even people who know nothing about semiconductor manufacturing would be quoting it in 2005. A quick Google search found over 248,000 hits for the phrase.
Unfortunately, most of those 248,000 references are wrong. Moore's law is one of the most widely misunderstood laws in history. Let's set the record straight.
Myth: Moore's law is a law. Fact: It's an observation, not a true scientific law. It doesn't actually govern the progress of semiconductors; it merely attempts to describe the pace of their progress.
Myth: Moore's law was created by Intel cofounder Dr. Gordon E. Moore. Fact: Although Moore did make his famous observation in the April 19, 1965 issue of Electronics magazine, he didn't call it Moore's law—indeed, the word "law" appears nowhere in his article—and he didn't explicitly state the law in a form easily recognizable today. Years went by before other people coined the term "Moore's law" and boiled it down to a succinct statement.
Myth: Moore's law says microprocessors get twice as fast every 18 months. Fact: Moore's law describes component integration on integrated circuits that are economical to manufacture. It doesn't describe clock frequency or other aspects of processor performance. Actually, Moore's article doesn't mention microprocessors at all, because they weren't invented until six years after his article appeared. And he wasn't talking about the maximum number of transistors that could fit on a chip, either. He was describing component integration on economical integrated circuits, not the most expensive chips manufacturable at any given time.
Myth: Moore's law predicts a new semiconductor generation every 18 months. Fact: Moore observed that component integration doubled approximately every 12 months. In 1975, he lengthened the period to 24 months. Since then, other people, not Moore, have split the difference to 18 months.
Myth: Moore's law is uncannily accurate. Fact: Moore's law is way off. If the original 12-month period had held true since 1965, today's chips would have more than 27 trillion transistors. The 24-month revision predicts 37 million. Actual progress: Intel's Prescott Pentium 4, a relatively economical chip, has 129 million transistors.
Myth: I'm attacking Moore's law and the eminent Dr. Moore. Fact: I admire Moore and his law. It has been a valuable way of describing the general progress of integrated circuits, even if it isn't infallible. I believe it remains popular because of its long-running optimism—at least something in the world is getting better at regular intervals brief enough to perceive within our lifetimes.