Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Is there a theoretical limit to the speed of a processor?

Author: Roy Eassa

Date: 11:26:38 06/07/02

Go up one level in this thread


On June 07, 2002 at 14:10:39, Jesper Antonsson wrote:

>On June 06, 2002 at 20:12:36, Robert Hyatt wrote:
>
>>On June 06, 2002 at 19:40:56, Roy Eassa wrote:
>
>>>In any case, how many more years can we keep doubling speed every 1.5 - 2 years
>>>before we run into the laws of physics?  I can't imagine it's more than a few
>>>decades more.
>>
>>
>>I find it amazing that we have not already run into this.  Clock frequencies
>>above 2 gigahertz are way into the microwave range, and electrical properties
>>of things change way up there...  How electrons propagate.  How atoms physically
>>move around.  Etc.
>
>I attended a university course in Modern Physics some years ago where one of the
>professors argued that the physical limit using extensions of current technology
>lie around 20 nm; less than that nothing could keep electrons from tunneling
>around too much. Today, P4 processors are manufactured using a 130 nm process
>(0.13 micron). If 20 nm holds, since the number of transistors are  proportional
>to the inverse square of the line width, we could get about 42 (that is:
>(130/20)^2) times the transistors of today in the same space. That means the
>chip limit for RAM would lie somewhere around there; 256*42 Mbit is somewhat
>more than 8 Gbit, so barring a technology breakthrough, don't expect much more
>per chip. 42 is between 5 and 6 doublings, so given 2 years for each doubling,
>we should hit the barrier in ten to twelve years.
>
>Processor speed, then? Well, that depends on what you use the transistors for,
>but clock speed alone has traditionally scaled *better* than the inverse square
>of the line width, so 100 GHz for a processor should be a conservative estimate
>of the limit. Then the extra transistors enable us to get even more speed, so
>200 million nodes/s should be very possible before we hit the physical wall.
>That's Deep Blue capacity on a chip, folks!
>
>One problem is that the cost of plants seem to increase almost as fast as speed,
>so that may stop development before the actual physical limit does. Lets hope it
>won't, though. :-)


Thank you for a thorough answer.  So the "decades" region is a better guess than
the "centuries" region for how much longer Moore's Law can hold, eh?

BTW, when I first read your name, I mis-read it as Jennifer Aniston.  (Sorry!)



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.