The famous Moore's Law, derived from a 1965 paper published by Intel co-founder Gordon Moore, may not be the most accurate model to predict the scaling of technology and technology development.
Researchers at the Santa Fe Institute (SFI) found that a much older similar prediction model is more accurate than Moore's Law.
In 1936, Theodore Wright authored a paper entitled "Factors affecting the costs of airplanes," which examines inflation-adjusted unit prices and the decrease of technology over time. Instead of Moore's prediction that the transistor count in a certain space doubles every 18 to 24 months, Wright's prediction is based on the emergence of volume production and can be translated to state that the cost of transistors is halved every 1.4 years. In conclusion, Wright's Law is slightly more accurate as a predictive tool than Moore's Law. The researchers also looked at Goddard's Law (drop in price due to greater productivity), Nardhaus’ Synthesis (combines elements of Moore's Law and Wright's Law) as well as Sinclair, Klepper and Cohen’s Synthesis (combines Wright's Law and Goddard's Law), but none of them was as accurate as both Wright's Law and Moore's Law.
Wright's Law has never been a popular way to describe the progress of technology, but since we know about the challenge of maintaining Moore's Law (that nature laws will more than likely trump Moore's Law), Wright's Law could become more valuable over time. One could question the value of those predictions and even Intel occasionally hinted that the doubling of transistor counts may not be as critical anymore in the future as new features and abilities take the spotlight. Still, Wright's Law could give the technology industry another guideline to hang on when new products are announced.
Are you sure you're on the Wright website?
Are you sure you're on the Wright website?
That happens far too often.
/too many puns
Also, a 'law' that merely describes 'nature' (assuming it is nature and not just a self-imposed schedule) is not particularly valuable. a law or theory that explains such a schedule would be extremely useful and valuable.
Secondly, what's the point of looking around in RETROSPECT for a "law" that provides the best prediction?
The best form of prediction is to use SEVERAL different forms of analysis for both the short and long term.
2.5d chip stacking on 22nm or 28nm will likely prove superior to 14nm transistors, unless you're self-esteem is directly tied to the process technology used on your CPU. Even then, that's not going to do much unless software can utilize 64 CPU cores stacked on top of each other, because thermal and electrical constraints aren't going away... That leaves GPU compute as the final frontier.