Intel and the Speed of Light

Have you ever thought why Moore's law is ending? Have you ever wondered what the limit to microprocessor speed is?

The answer is given by this equation:

lambda = c / f
Where lambda is the wavelength (m.) c is the speed of light (3.0e8 m./s.) and f is the frequency (in Hertz, this is, 1/s.).

Now let's apply this equation to a frequency of, say, 3.4 GHz. This is, 3.4e9 Hz. We have:

lambda = 3.0e8 / 3.4e9 = 0.088m = 8.8 mm.
This is, 8.8 millimeters. Pretty small, isn't it?

So what, you ask? Well, if the dimension of the circuit is around the dimension of the wavelength of the signals you want to propagate then you're in trouble. The circuit itself radiates, and generates interferences between different parts of the circuit, and you enter the world of distributed parameter circuits, or transmission lines.

This is, the higher the frequency of a microprocessor the smaller the wavelength (the closer to a distributed parameter circuit).

Or, as IBM people put it:

Raminderpal Singh, senior engineering manager at IBM Corp., agreed that signal integrity has become a significant and surprising problem at 90 nm. Part of his job is helping 90-nm customer designs get through IBM's fabs.

"As people push the density, and push the frequency, and voltage goes down, you just have a lot more happening and a lot less to live with," he said. "A whole series of effects becomes very real."

That simple equation, lambda=c/f, is the main reason why higher frequencies are not that important nowadays in microprocessors. Concurrence, multi-core and multi-processors, on the other hand, are getting more and more important.

So the time for higher and higher frequency processors is finished. Dual core systems (AMD) and multicore systems (Sun's Niagara) are the way to go. That's what we'll see in the coming months in the microprocessor area. If you're not ready for dual-core and concurrency you'd better rush, as Intel did.

Happy threading,
Antonio

Comentarios:

Um... 0.088 m == 8.8 \*centimetres\*, not 8.8 millimetres. Mind you, propagation speed on a chip is reduced by at least a factor of the dielectric constant of the chip material (at least 2, possibly as high as 3), and wires don't go in straight lines, either. You conclusion is still corrrect, despite the factor of 10 :->

Enviado por Russell Crook en agosto 22, 2005 a las 03:13 PM CEST #

Hi Russell. Thanks for the modifications. By the way, it seems AMD wants a dual-core with Intel.

Enviado por Antonio en agosto 23, 2005 a las 03:03 PM CEST #

Enviar un comentario:
Los comentarios han sido deshabilitados.
About

swinger

Search

Archives
« julio 2014
lunmarmiéjueviesábdom
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
   
       
Hoy