Why Moore’s Law Works
Most industrial process get better over time and scale through an effect known ast the learning curve. It’s just that those processes based on semiconductors do so much faster and longer.
The term “learning curve” was introduced by the nineteenth-century German psychologist Germann Ebbinghaus to describe improvements he observed when people memorized tasks over many repetitions. But it soon took on broader meaning. The Wikipedia entry on the term explains it this way: “The principle states that the more times a task has been performed, the less time will be required for each subsequent iteration.” An early example of this was observed in 1936 at Wright Patterson Air Force Base, where managers calculated that every time total aircraft production doubled, the required labor time decreased by 10 to 15 percent.
In the late 1960s, the Boston Consulting Group (BCG) started looking at technology industries and saw improvements that were often faster than simple learning curves could explain. Where the learning curve was mostly about human learning, these larger effects seemed to have more to do with scale: As products were manufactured in larger numbers, the costs fell by a constant and predictable percentage (10 to 25 percent) with every doubling of value. BCG called this the “experience curve” to encompass institutional learning, ranging from administrative efficiencies to supply chain optimization, as well as the individual learning of the workers.
But starting in the 1970s, price declines in the new field of semiconductors seemed to be happening even faster than the experience curve alone could explain. The original transistors fell at the high end of the BCG rate and kept on falling. During one decade-long period, the Fairchild 1211 transistor’s sales increased four thousandfold. That’s twelve doublings, which experience-curve theory predicts would lead to a price decline of one-thirtieth the original figure. In fact, the price fell to one-one-thousandth that number. There was clearly something more going on.
What’s different about semiconductors is a characteristic of many high-tech products. They have a very high ratio of brains to brawn. In economic terms, their inputs are mostly intellectual rather than material. After all, microchips are just sand (silicon) very cleverly put together. As George Gilder, the author of Microcosm, puts it:
When matter plays so small a part in production, there is less material resistance to increased volume. Semiconductors represent the overthrow of matter in the economy.
In other words, ideas can propagate virtually without limit and without cost. This, of course, is not new. Indeed, it was Thomas Jefferson, father of the patent system (and a lot more), who put it better than anyone:
He who received an idea from me, received instruction himself without lessening mine; as he who lights his taper at mine, received light without darkening me.
The point: Ideas are the ultimate abundance commodity, which propagates at zero marginal cost. Once created, ideas want to spread far and wide, enriching everything they touch. (In society, such spreading ideas are called “memes.”)
But in business, companies make their money by creating an artificial scarcity in ideas through intellectual property law. That’s what patents, copyright, and trade secrets are: efforts to hold back the natural flow of ideas into the population at large long enough to make a profit. They were created to give inventors an economic incentive to create, a license to charge monopoly rent for a limited time, so they can get a return on the work they put into the idea. But ultimately, patents expire and secrets get out; ideas cannot be held back forever.
And the more products are made of ideas, rather than stuff, the faster they can get cheap. This is the root of the abundance that leads to Free in the digital world, which we today shorthand as Moore’s Law.
However, this is not limited to digital products. Any industry where information becomes the main ingredient will tend to follow this compound learning curve and accelerate in performance while it drops in price.
