Moore's Law has been used for nearly 50 years to set targets for scheduled advances in semiconductor processes. A variety of criteria based upon interpretations of that law have been cited for use in grading these advancements. More or less, Moore's Law appears to have been followed in defining and measuring advances in this area.
There have been many references to Moore's Law in Seeking Alpha articles having to do with Computers, Mobile, and Semiconductor companies. It has been cited in articles about Mobile Processor market share and Servers. This has triggered some heated discussions about a subject that few have any real knowledge about. I am about to change that and tell you what it's really all about.
Why is Moore's Law so important?
It's not a law of physics, or of any other discipline we would call science. It's not something that will put you in jail if you violate it. It isn't based on deep philosophical reasoning that would compel followers to adopt it.
It is just a description of what Gordon Moore observed and reported in an article in Electronics magazine in 1965. It's important because we have adopted it as a measure of how well the semiconductor industry is progressing.
Moore's perspective was cost. And profit. Discrete circuits were expensive, and integrated circuits were even more expensive when he made his observations. He was thinking about what had to be done to make integrated circuits less expensive and more competitive with discrete circuits, and he hit the nail on the head. He must have thought "Cram more components on to that chip and it won't cost that much more than if there were only one component" - and he was right!
For emphasis, I repeat: Moore's Law is important because it has been used for about 50 years to set targets for what should be accomplished as a result of moving to a new manufacturing process for integrated circuits.
So it's really important - right? Yes, but ask someone - anyone - to quote it and you will get a response that is inaccurate. Because the world changes, and people look at what's happening now rather than what happened in the past or what will happen in the future; a variety of interpretations have been used to warp the meaning of Moore's Law. Every time there is a barrier to extension of the applicability of Moore's Law there is some change in interpretation that allows the continuation of the spirit of Moore's Law. Moreover, there have also been a number of interpretations that have had nothing to do with that spirit.
Moore's Law, the original statement, had to do only with the observation that the number of components on a chip was doubling about every two years. Moore comments on what he said in an article posted on the Web.
Moore said the number of components on a chip would double every 2 years. Period. And that has been more or less true. He didn't say "the density of transistors would increase," nor did he say "their speed would increase." He didn't even say they would be Silicon or FETs. He said what's in the commentary, and in the original article, which is still available on the web.
What are the classic barriers to conforming to Moore's Law?
1. You need to make the components (in CMOS, transistors are essentially the only components) smaller so that more will fit on the chip.
2. You need to get the heat out of the chip so it doesn't melt.
Those two are pretty obvious, and they are examples that we can point to and show where these have been done in order to extend compliance with Moore's Law.
However, that's not all we can do! We can make the chip bigger. But we can only do that if we can get the heat out and if we can achieve yields that are acceptable in the larger chip. That's because we can't allow the chip to melt, because yield is a critical part of the cost equation, and because cost is the underlying thrust of Moore's Law.
So now we know why we have these heat sinks and fans and other heat transfer mechanisms on CPUs; it's because of Moore's Law. And we also know why people, IBM and others, have developed multi-layer RAMs and other 3D integrated circuits, because it allows more components on the chip.
(I am going comment on the last few paragraphs: Please realize that there is really no difference between getting more components on the chip by making the chip bigger and getting more components on the chip by going 3D. The "chip" is really 2D, all the circuitry is pretty much planar - even with 3D gates it's still pretty planar. So the area of the individual layers in a 3D chip can be added together and compared to the area of a larger 2D chip for Moore's Law purposes. Note that this argument ignores some of the metal layers that are used to connect things together.)
So what else can we do? Well, short of utilizing time travel and alternate universes, there isn't anything. What about going to carbon nanotubes? That's part of item 1, above. "1" doesn't limit us to silicon, or even to FETs. It says "make the components (transistors in this case) smaller," and that's what carbon nanotubes do.
How about using superconductors? Well that could be a combination of 1 and 2; it gets the heat out and it could make the components smaller.
What about optical switches? Now we are dealing with photons, and that could help reduce the power dissipation and reduce the requirement for metal interconnects that take up space, are hard to define, and that limit the speed with which we can communicate with other locations.
But all of these things deal with making the components smaller and getting the heat out, and that's what it takes to make Moore's Law work.
There are many additional things we know about and there are people already working very hard on these things. There are even more things we don't know about that people are looking at with the hope that they will also contribute to this end, and some of them will fail and some will succeed. Some will succeed spectacularly, just as the first transistor succeeded. By virtue of these successes, Moore's Law will continue to be followed for many years, many more years than most readers will grant.
As a historical example of similar technological development, I cite the following:
In 1965, the rule of the internal combustion reciprocating gasoline driven engine as a motive force for automobiles was being written off. Chrysler was field testing a gas turbine. Wankel engines were being used in two production vehicles in Germany and Japan. People were talking about steam and electric drive.
The Chrysler gas turbine program was terminated. A variety of problems arose with Wankel engines, and the NSU and Mazda programs were scaled back or terminated. Forty eight years later, we are starting to see electrics gain a very small market share. Steam has gone nowhere.
The gasoline powered reciprocating internal combustion engine continues to prosper, increasing fuel efficiency by a factor of about 2X and increasing output power per cubic inch by a factor of almost 3x.
What happened with automotive internal combustion engines is in the process of happening with silicon integrated circuits. Moore's Law is alive and kicking!
Intel is building bigger chips with more transistors on them. All along, Intel has been building bigger chips with more transistors on them, but nobody has been talking about the "bigger" part of the program. In order to build bigger chips with more transistors on them, Intel had to develop a more energy efficient transistor. They had to get the power down to prevent the chips from melting.
They did, and that gave them the edge in the mobile market where the improved transistors resulted in longer battery life. They did, and that gave them the edge in the PC market, where the improved transistors resulted in lighter, higher performance Ultrabooks with longer battery life. They did, and that gave them the edge in the supercomputer market where the improved transistors translated into the ability to build bigger chips with more cores and better performance than any other chip manufacturer.
In fact, now they have the lead in all the microprocessor market segments, and this lead should last for several years - it will take two years for anyone else to get FinFETs into production.
Now you know. It's about money. It's always been about money, and it will continue to be about money. Intel will be coining money because they have the lead in the battle about Moore's Law, and anyone who says different had better take another look.
Additional disclosure: I have no plans to trade any stock mentioned in this article in the next 72 hours.