Improvements in computing are typically compared against the predictions of Gordon E. Moore. His 1965 paper detailing the doubling of transistors on integrated circuits has faired remarkably well even to this day. However, a new study suggests that Moore’s law may not be as accurate as another technology prediction model.
An IEEE Spectrum article followed a working study by the Santa Fe Institute titled Statistical Basis for Predicting Technological Progress. The project vetted six technology forecasting models using historical cost data for 62 different technologies.
While the researchers made sure to track transistors and memory modules, they included data for low-tech goods like beer and consumer appliances during periods of technological evolution. Information was used as early as 1930 all the way to 2009.
The data showed that low-tech goods shared similar evolutionary performances as their high-tech counterparts. In fact, the study’s authors found that a similar set of rules were integral to accurate prediction of all 62 areas studied.
Because of variations between goods within their own segments and over time, the study chose to vet the predictive models using inflation-adjusted unit prices. This method translated Moore’s law from “transistor counts double every 18 months” to “transistors drop in price by 50 percent every 1.4 years”.
Along with Moore’s law, the researchers tested the predictions of four models including:
- Wright’s Law – Tracking the drop in unit prices as part of cumulative production.
- Goddard’s Law – Tracking the drop in unit prices due to increased productivity
- Nardhaus’ Synthesis – A Combination of Moore’s and Wright’s laws to track unit prices vs. time and cumulative production
- Sinclair, Klepper and Cohen’s Synthesis – A Combination of Wright’s and Goddard’s laws to track unit prices vs. the rate of production and cumulative production.
The study found that while Moore, Wright and Goddard made predictions based on different variables, their models produced similar data. It was also noted that the Sinclair, Klepper, Cohen Synthesis broke down over long spans.
In the end, Wright’s law had the highest accuracy with Moore’s law coming in a close second place. Surprisingly, the study’s authors found that Moore’s law was fairly poor at predicting transistor prices. This was due to rapid increases in chip density and production, factors that were better accounted for by Wright’s law.