[extropy-chat] Re: A slow down in Moore's law?

[Dan Clemmensen]

The original Moore's law did not mention CPU speed at all. It was about 
the cost per transistor.
As a transistor shrinks, the effective cost per transistor decreases.

The reduction in transistor size is one factor in the decrease int he 
cycle time a  CPU. For many years, we have been accustomed to 
correlating the inverse of CPU cycle (i.e., CPU clock rate) with CPU 
capability. This is grossly over simplistic.

Moore's law (decreased cost per transistor) continues apace. The best 
direct gage of <Moore's law is not CPU speed, Look at cost per bit of 
RAM instead. This continues to lurch forward.

The most cost-effective way to convert transistors into effective 
computation has changed over the years.  In addition to brute-force 
speed increase made possible by smaller transistors, we use more 
transistors to create wider buses (64-bit rather the 32-bit pointers, 
etc.) We use much bigger caches, and we implement dual-core processors.

Back to the original Moore's law: Any given technology will eventually 
reach its physical limits. Moore's Law is abstract, but was first 
replicated in terms of Silicon photlithography. This family of 
technologies has sustained the "law" for the last forty years. For the 
last decade or so an industry stearing committee, SEMI, has attempted to 
provide an  ongoing road map for the industry. This worked well foe 
several "process nodes." However, the current projection is in the 
process of failing. The road map showed the next nodes were to be 
reaches by using a fluorine laser with a Calcium Floride (CaF) lens. As 
it happened, the CF lens was too hard to make, The industry is 
scrambling to shift to immersion lithography instead. This is the most 
serious bobble since Moore's law was first explicated.