AHC: Could Moore's Law be faster?

[VioletRiver]

I think I recall reading that, for at least the first decade after integrated circuits were invented in 1958, the number of transistors per square inch doubled every 12 months, prompting Gordon Moore to predict in 1965 that this trend would continue for another 10 years; he revised it in 1975 to every 24 months.

For me, this begs the question: is it possible that this trend could have continued on pace with its previous growth? I know the development cycle of a new chip takes awhile, but is there any technical reason (or reason besides $$$) that it couldn't be accelerated, and we could end up with OTL 2017 technology during the Clinton administration?

I feel like there's some precedence for microprocessor technology advancing faster than it has in our history, especially since the US Navy apparently developed a chip more powerful than the Intel 4004 two years before its release: https://en.wikipedia.org/wiki/Microprocessor#CADC

 

[EWHM]

Developing a new silicon manufacturing technology is done by starting from some starting point and doing experiment after experiment to improve its yield until you get to something that will actually yield some usable die. In a new process you generally start from zero--ie everything you manufacture in your first runs doesn't work. If you started somewhere other than zero, you weren't aggressive enough in your targets. From there, your speed of mastery of the process is driven by:
A) How fast you can get your experiments through the fab to get the yield learning out of them and
B) How many experiments you're running (limited by your manufacturing & metrology tools) and
C) How much you glean out of each experiment

You can improve factor A) to some extent by throwing more resources at it. You can certainly improve B) through more resources (buy more tools and more techs and more engineers to devise and analyze your experiments). C) you can improve through more experience and tribal knowledge.

Between technologies you're constrained by your scientific pathfinding (i.e., your guys in more research-y fabs busily cataloging and doing very preliminary work to identify exploitable physics for your next generation). You can improve this by getting a larger bite out of the high-end PhDs of your country.
The one big constraint is you've got to do all of this at a profit. If you loosen that a bit you may be able to run faster for a process generation or two. Yes I do work in the industry.

 

[Asp]

Get people/the government to start shoveling R&D funding into creating faster microchips.

The easiest way to do this historically is a major war. Those boost technological development like nothing else. Have there be a conventional war between NATO and the Warsaw Pact in the 1980s. Whatever the aftermath is, the government quickly decides we need more money going to defense technologies. Microchips and tech had already been designated through the 3rd offset strategy as the way of the future so just have them pore a ton of money into that on at least a NASA during the moon race level and have it trickle into civilian sectors like a lot of stuff we developed through NASA and WWII programs did.

 

[EWHM]

Asp, a major war like that might not be the best thing. See the semiconductor industry is really brittle against the loss of tribal knowledge. It'd be really tempting for the other side to strike Silicon Valley/Silicon Forest and the returns would be very high. There's a reason companies like Intel have very very low turnover rates historically.

 

[Electric Monk]

Realistically? Yes, but…. Intel has burned the torch of Moore's Law for 40 years. Indeed Intel is the main reason Moore's Law actually continued because Intel's x86 cash meant they could pour money into fab tech. On the other hand being stuck with x86 drastically slowed their processor design ability vs RISC in the 1990s. So I dunno, Windows jumps over to RISC fully perhaps?

I'd suggest Michael Malone's The Holy Trinity if you want to dig into potential computing PODs that may speed things up here and there. It's also a great book about Intel.