PDA

View Full Version : Moore's Law



mugaliens
2005-Dec-10, 12:58 AM
For years they've been saying there's a limit to Moore's law. The first time I'd heard Moore's law had reached an impasse was 1984.

For years, technological breakthroughs have proven them wrong.

It's happened yet again:

http://news.yahoo.com/s/nf/20051209/**_nf/40040

When will it stop? Or will technological breakthroughs always make the difference?

Candy
2005-Dec-10, 01:18 AM
Moore's Law is the observation made in 1965 by Intel cofounder Gordon Moore that the number of transistors per square inch on integrated circuits will double roughly ever year, meaning that chips will continue to get smaller and more powerful.

While many industry observers have suggested that we are nearing the limits of chip miniaturization, breakthroughs like this could, at least in theory, help continue to increase processing power long into the future.

Good question.

Richard of Chelmsford
2005-Dec-10, 01:33 AM
Won't we reach the limit by round about 2020?

weatherc
2005-Dec-10, 01:50 AM
Well, it all has to stop when the logic gates/transistors reach the size of individual molecules. We're still a couple of decades away from that happening, though.

However, Moore's "law" (it's not actually a law, and I hate that people even call it that) only states that the number of transistors per square inch on a circuit will double every two years; it says nothing about the actual number of transistors used in an actual chip. If a method can be found to deal with the heat output of processors, you might be able to stack them one on top of the other, and keep adding layers. It would be a real pain to engineer and fabricate something along those lines, I imagine, so I don't know if that will happen soon, if ever. Once your transistors are molecule-sized, all you can do is add more of them, or figure out ways of making them operate faster.

Van Rijn
2005-Dec-10, 02:02 AM
We hit some pretty tough scaling issues with CMOS a few years back. My primary home computer (3.06 GHZ P IV) is, by the standards I'm used to, shockingly old yet still quite respectable. In the mid '90s I put together a far too expensive 133 MHZ Pentium system and it was quite obsolete inside of 6 months. There were also massive disk capacity increases around the same time, and that growth too has slowed way down.

There are other technologies, and ways of increasing complexity or speed, but it is unclear how that will proceed. We're currently seeing the move to dual core CPUs because of the scaling/speed problems. There are some advantages to multiple CPUs, but you would really prefer to double or triple the speed of one CPU if you could.

I'm guessing you'll see speed ups and slow downs in scaling as we move to new technologies, but I think we hit the limit of easy scaling.

Van Rijn
2005-Dec-10, 02:08 AM
However, Moore's "law" (it's not actually a law, and I hate that people even call it that) only states that the number of transistors per square inch on a circuit will double every two years;

Me too, it is more like "Moore's Observation." Even he has complained about how people treat it. It has taken on a life of its own, with people making interpretations and conclusions far beyond his original statements, much like some other things I can think of.

*cough* Fermi's Paradox *cough*

weatherc
2005-Dec-10, 02:57 AM
Economics may also play a role when it comes to how chips get scaled and when. Even if more transistors can be packed on a chip, if it can't be created at a cost that is less than or equal to the existing chips, then it won't be worth it for the chip companies to manufacture them. It's also important to remember that just because a processor may have twice as many transistors as its predecessor, it won't be twice as fast; there are also many other bottlenecks within a computer system to take into account, so even if the processor is twice as fast, the whole computer won't be twice as fast.


*cough* Fermi's Paradox *cough*:D

Taks
2005-Dec-10, 03:09 AM
actually, moore's law says several things, one interpretation IS indeed about total transistors... from wikipedia:


This 'law' can be interpreted with varying degrees of rigor and frequently is. The least rigorous interpretation of Moore's law is that computing power per unit cost will continue to grow exponentially. In this form, "Moore's law" has a factual basis and has proven useful for planning purposes.

The most rigorous interpretation of Moore's law is that the total number of transistors on the cheapest CPU will grow exponentially at a constant rate and that this constant rate produces a doubling every 12 (or 18, or 24) months. This variant oversimplifies a complex history.
the first interpretation is the "law" that i've followed over the course of my career, and it most certainly encompasses economic factors, but not as a reason, but as a result. i.e. they're getting cheaper per unit of processing power.

either way, the size of transistor gates is limited by the size of atoms in semiconductors (the word molecules really shouldn't be used as that implies some compound structure of various atoms). however, regular semiconductors are nearing their useful limits well before they get down to atom sized gates. there's a few device physics experts running around in here that can explain better.

carbon nanotubes, however, may be next. :)

taks

LurchGS
2005-Dec-10, 08:57 AM
or optical gates...

when the copper traces between the transistors get too small, they can't withstand the heat of electrons flowing down them, and break. Sure, dropping the power requirements go a ways to alleviate this, but at some point you have to stop and say 'we need SOME power to operate this gizmo'

Carbon nanotubes... interesting idea.. aren't they using a variant of that in the new SED monitors?

A molecular transistor - now, there's a keen idea.

At some point, so long as we stay in the physical universe, we WILL reach a limit on the number of binary switches we can cram into a given volume.

One thing that might stretch things out is a multi-state gate... tri-istor, quadistor, etc. No idea if such a thing is possible, but it would make programming a hoot!

Taks
2005-Dec-12, 01:55 AM
multi-state is already possible with tri-state (low, high, high-impedance), but that's not what you're thinking of i bet...

i believe quantum gates would be the answer you're looking for.

taks

mugaliens
2005-Dec-12, 01:59 AM
Quantum gates and states, now there'd be something!

LurchGS
2005-Dec-12, 04:01 AM
actually, tri-state was what got me thinking along those lines... but a mutli-state would ideally be...oh, 10 states ( 10 is such a nice number)

And Quantum gates .. I don't know enough about them, sadly.

ASEI
2005-Dec-12, 04:10 AM
However, Moore's "law" (it's not actually a law, and I hate that people even call it that) only states that the number of transistors per square inch on a circuit will double every two years; it says nothing about the actual number of transistors used in an actual chip. If a method can be found to deal with the heat output of processors, you might be able to stack them one on top of the other, and keep adding layers.

This would amount to a parralel processor, or more processors than one to operate your task. Parralel processing has no theoretical limit. You can just keep adding more. However, some tasks have to be accomplished using serial processing. You can't do step 10^13+1 until you do step 10^13. It's obvious how to accomplish parralel tasks on a serial processor. It's not so obvious if you can in general accomplish serial tasks on a parralel processor.

Most of the stuff the average consumer is interested in - games, large 3D simulations, rendering, ect, can probably be accomplished in parralel, so Doom MMCXXXVIII is in no danger of running up against a limit in graphics performance. Certain engineering simulations that currently take a week to run . . . might still end up taking a week to run though, until some mathematical savant figures out how to solve a generalized serial problem in parralel.

TheBlackCat
2005-Dec-12, 05:10 AM
but a mutli-state would ideally be...oh, 10 states ( 10 is such a nice number)
Heresy! Ten is the bane of all computer programmers. Everything has to be a power of 2 (binary, octal hexadecimal).

LurchGS
2005-Dec-12, 05:31 AM
exactly why I like it. (I use Python to irritate my partner because he uses PERL :) )

I suppose 11 states would be even more fun...

Swift
2005-Dec-12, 06:02 AM
Me too, it is more like "Moore's Observation." Even he has complained about how people treat it. It has taken on a life of its own, with people making interpretations and conclusions far beyond his original statements, much like some other things I can think of.

*cough* Fermi's Paradox *cough*
To me it sometimes seems like it is The Gospel According to Moore. The computer industry seems to take it not as in a law of physics, but as in a law that MUST BE OBEYED! The hardware people seem to be driven by some preceived need of the software people, and the software people just write bigger and bigger programs that must take advantage of the new hardware.

The computer I have now, which is several years old and ancient by current standards, doesn't really do anything for me that my previous one couldn't really do. But if I want even my basic software to be supported: e-mail, virus protection, word processing, then I have to keep getting the updated software, which means updating my operating system (it won't run otherwise) which means upgrading my hardware. I feel like George Jetson ("Jane, how do you stop this crazy thing!").

LurchGS
2005-Dec-18, 07:38 AM
I agree - I use a Mac B&W at the office - I've upgraded it from a G3 300 to a G4 500.. but the elevator stops there (with 10.2.8). It does everything I need, in a timely manner.

The windows box, on the other hand, is the ... 3rd I've had in as many years..

on the gripping hand, the Linux box is the same one I've been using since Moby Dick was a guppy. (though, I admit, I HAVE updated the OS a bit, here and there)