Is Moore's Law a law?

daveleau · March 13th, 2003, 07:47 PM

I posted a comment in RobRich's recent article that talks about the future of processors. It discusses Moore's Law. Instead of starting a discussion there about the semantics of calling Moore's Law a law, I thought I would start one here.

I have always held the idea that a law in scientific terms is something that is proven and is unchanging, like the Law of Gravity or Law of Thermodynamics. For the past few decades (as SoopaStar pointed out int he discussion of RobRich's article) Moore's Law has held constant. I believe this is only the case in the processor market because AMD became a viable competitor to Intel. If Intel had no competition in the PC market as was pretty much the case in the late 80s and early 90s, I do not believe we would have a 3.06 GHz processor on the market right now. The speed of which processor have been produced, upgraded and released coincided with AMD's emergence on the scene. If AMD is in decline as some say (I do not hold this view), I believe Moore's Law will be broken in the coming months to years as Intel slows their R&D and production to pre-AMD levels.

What are your thoughts?

March 13th, 2003, 08:19 PM

Not many Law's are completely immutable. They are dependent upon the conditions which the Law assumes.

In this case, the market conditions are an important driver for CPU R&D. If the market slows, then money for R&D may well be reduced.

The same would apply if demand for faster CPU's decreased. Manufacturers could manage by continuing to produce current speed CPU's if there was no demand.

I'm not saying these things WILL happen. But if they did, they would impact Moore's Law.

IMO, anyway.

Cheers
Mick

sKiT · March 13th, 2003, 08:19 PM

How about we re-write as Moore's "Law"?

Tomteriffic · March 13th, 2003, 08:38 PM

I never really saw it as a "Capital L Law", it was a postulation that was based upon empirical observation, and the term "Moore's Law" was sort of coined in a lighthearted manner.

I've just kind of watched over the years to see how well it holds up, and at what point the immutable "Capital L Laws" of physics might stop it. Evidently a lot of the microprocessor industry watches it too, and spends a whole lot of time, effort and money to see to it that it holds true for as long as possible.

Obviously, there are good business reasons to do so, but I kind of wonder if there isn't a point of pride involved as well.

That said, the new Intel Mobile chip, which places low power consumption at respectable performace levels at a higher priority than raw clock speed may be an indicator of just this "slowdown" phenomenon.

March 13th, 2003, 09:08 PM

I think that if Intel and AMD drop the ball, nVidia and ATI will pick it up and run with it. Think of the current density of their chips and the fact that they, too are increasing with speed.

daveleau · March 14th, 2003, 01:35 AM

Good points, all of you. I agree with all of you.

sKiT, I was thinking Moore's theory, but...

**RobRich** · March 14th, 2003, 01:43 AM

Intel will still use the old Moore's Law of frequency scaling for atleast the next five years, as do I for right now, but the semiconductor industry is currently rewriting the basis of the "law."

http://www.techimo.com/newsapp/index.pl?photo=1653

cunokyle · March 14th, 2003, 02:01 AM

Quote:

I have always held the idea that a law in scientific terms is something that is proven and is unchanging, like the Law of Gravity

Actually, gravity fluctuates and the laws do not hold up when you get into Quantum Mechanics. Also, another bad assumption is assuming the speed of light is a constant. Light can be slowed down or sped up by passing it through different mediums...the most common being water. Just thought I would toss that out there.

PyroSama · March 14th, 2003, 02:42 AM

Also in the future though we will be seeing processors up to 4096 bits.

(My father is part of the development/reasearch of this)

So even with us being forced to stop the GHZ frenzzy processors will still continue to grow in processing abilities.

We will also see multi core processors like the 'CELL' being produced by Intel, sony, toshiba, and 2 or 3 others.

So processing abilites will continue to grow despite the roadblocks.

PyroSama

**RobRich** · March 14th, 2003, 02:59 AM

Sun and IBM are the leaders in development for multiple core on die processors, but we are atleast a decade away from such chips in the consumer segments.

Actually, current silicon semiconductos will not prove viable beyond 2016 according to most analyst. The IRTS (Intel Research) development roadmap indicates a massive roadblock for electronics based upon electric charge store state. We will be forced to move on to other technologies to store finite states for massive computing, regardless of parrallel core scaling for current or near future generation designs.

Quantum computing is a step in the right direction, but only time will tell if the research will actually pay off. A magnetic dipole interaction between adjacent atoms will be used to set a finite state, thus creating the fundemental basis for the qubit. The real problem with quantum computing right now is storage capacity, as quantum memory requirements scale with a 2^N factor. Given a standard quantum dot density of 10^11 per square centimete for current GaAs semiconductor estimations, it can be determined that we need a massive 2^10^9 qubits per square centimeter to maintain a quantum computing state. Work to use Hopfield neural networks is being undertaken, but such networks require 2n+1 qubits to store a finite data set. Accordingly, we would be limited to only "n" patterns per operation, which is not economically viable. Probabilistic QUAM is a possibility, as it is intended to scale exponentially n^2 for data patterns, but the work is extremely theorectical at this point.

Electron spin creation, manipulation, and transport are already well into development through research in ferromagnetic proximity polarization in GaAs materials, but actual data pattern storage is still presenting the most problem IMO. Then again, voltage switching is presenting another problem as well, as a variation over a 25 GHz range is required for maintaining a parabolic quantum well to drive the gate controlling mechanism. To arrive back at my article, research such as the Terahertz transistor project at Intel will take care of this issue, and hopefully QUAM research will be significantly increased in the upcoming years.

Robert Richmond