Back to Event List

Moore's Law

This is a PAST event. See "Meeting Notes" section for audio, video, documents and other information.

Original event date/time: Friday October 26th, 2007, 7:00 pm to 9:00 pm

Moore's Law is the doubling of computer processing power every 2 years or so. How long will it continue?

Abstract:

In 2002 Doug bought a 2.4 GHz PC. If speeds are doubling every 18 months, then he ought to be able to buy a 40 GHz machine today. Yet, he says, "I just clicked on the Dell webpage and looked at their Precision machines, and the fastest processor is 2.66 GHz. OK, it's a quad-core, which might give 2.66*4=10.6 GHz, but no operating system actually quadruples the throughput with a quad core." So apparently speeds have only about doubled in 6 years. He asks, "Have we finally hit the limit on Moore's Law?"

• John says Moore's Law was never about GHz, it was about transistor density. GPU polygon rendering times have continued to double, even though CPU clock speeds have not. The problem with multi-core is that parallel programming is hard.

• How Dense can transistors get? Will carbon nanotube transistors and other alternatives be ready when photolithography, our current manufacturing technique, runs out of steam?

• Dirk points out that assessing progress is fairly difficult using the metrics that companies typically use to market processors. Mark Horowitz at Stanford has developed methods to help break down the contribution of "circuits" or "process" vs. "architecture". Dirk also points out that higher clock rates lead to a lot of heat, and heat has began to limit the the CPU performance. (In silicon valley, people would joke that computer makers should put the CPU on the outside, so you could use it to warm your coffee).

• How much is real and how much is "mine is better than yours and the one you bought last year sucks now" marketing?

Mark your calendar for October 26th -- this promises to be a very lively discussion!

Videos

The person behind the law: 1-hour interview with Gordon Moore. This is a great "human-interest" video -- there is nothing technical, but it tells a lot about Mr. Moore and the people who created the early computer industry.

Charlie Rose - Gordon Moore
Gordon Moore, Co-Founder, Intel Corporation

Michio Kaku says: Moore's Law will collapse! ... and artificial intelligence will take 50 to 100 years to happen!

But quantum computers in 20 to 30 years ... ?

Michio Kaku on Artificial Intelligence

Ray Kurzweil expresses the contrary opinion...

Ray Kurzweil at The Singularity Summit at Stanford
Ray Kurzweil's keynote address at The Singularity Summit at Stanford put on by the Singularity Institute in May 2006 (3 parts, ~ 50 min total)

Will the end of Moore's Law scaling be the ultimate bug or the ultimate feature?

Neil Gershenfeld: Life after the digital revolution

A great graphic (literally) demonstration of Moore's Law:

Evolution

We are less than 30 days from the start of volume manufacturing of Penryn, Intel's new 45-nanometer quad-core chip:

Crysis on Penryn

45nm -- Biggest Change to Transistor in 40 Years

Read-Ahead

Ikko Tuomi: The Lives And Death of Moore's Law. Tuomi criticizes the idea that Moore's Law is a general "law of nature" and along with that the idea that we can expect it can continue and that the idea of "Moore's Law" can be generalized to other forms of information processing, not just transistor counts.

The Lives and Death of Moore's Law

Seth Lloyd estimates how far Moore's Law can go before running into limits imposed by the laws of physics.

Ultimate Physical Limits To Computation
PDF, 22 pages

Krauss and Starkman estimate what the limit should be anywhere in the universe.

Universal Limits on Computation
PDF, 3 pages

And here are some papers and articles discussing alternatives to silicon.

One alternative is carbon nanotubes, the technology mentioned by Ray Kurzweil.

Nanotube field-effect transistor

Carbon Nanotube Single-Electron Transistors at Room Temperature

There's also reports of work using DNA to self-assemble carbon nanotube transistors:

DNA used to create self-assembling nano transistor

Carbon nanotubes are not the only alternative. There's quantum effect transistors:

Molecular transistors based on quantum interference

The Quantum Interference Effect Transistor: Principles and Perspectives

There was talk years ago of transistors made with polymers (plastics), but I haven't heard of any recent progress.

Polymer Promise: One Day IT Will All Be Organic

High-Resolution Inkjet Printing of All-Polymer Transistor Circuits

John Smart posted this article about the graphene transistor earlier this year:

New graphene transistor promises life after death of silicon chip (Update)

And here's the most far-out one, all-optical transistors:

A single-photon transistor using nanoscale surface plasmons

Article from Doug on power consumption and parallel programming:

Computing in a Parallel Universe
The pace of change in computer technology can be breathtaking—and sometimes infuriating. You bring home a new computer, and before you can get it plugged in you're hearing rumors of a faster and cheaper model. In the 30 years since the microprocessor first came on the scene, computer clock speeds have increased by a factor of a thousand (from a few megahertz to a few gigahertz) and memory capacity has grown even more (from kilobytes to gigabytes). Through all this frenzy of upgrades and speed bumps, one aspect of computer hardware has remained stubbornly resistant to change. Until recently, that new computer you brought home surely had only one CPU, or central processing unit—the computer-within-the-computer where programs are executed and calculations are performed.

Google's MapReduce:

MapReduce: Simplified Data Processing on Large Clusters
MapReduce is a programming model and an associated implementation for processing and generating large data sets. Users specify a map function that processes a key/value pair to generate a set of intermediate key/value pairs, and a reduce function that merges all intermediate values associated with the same intermediate key. Programs written in this functional style are automatically parallelized and executed on a large cluster of commodity machines.

This is a past event.

Meeting Notes: