Is Moore's Law coming to an end?

[Peter Hewkin]

I think it is - because gates smaller than a single atomic diameter seem pretty difficult to me. But does it matter? If you believe in essentially serial computers with 16 or 32 bit words it probably does .. but if you are more convinced by massively parallel architectures (and new paradigms of programming) maybe not. But what do I know? I bet there are experts out there..

[Martin Kendall]

I think the developing capability to compute in parallel has reduced the importance of serial computing. In addition, the concept of the unified "personal computer" is now ad odds with the concept of dedicated "computing function".

[Richard Atkinson]

Moore's law actually states that the cost of a single transistor on a die goes down exponentially with each new process generation - and I think this aspect is set to continue for a few more generations yet. However what has already run out of steam is the notion that each new generation of smaller transistors can be clocked faster and/or at lower power.

[Hongwei Li]

For the cost of a single transistor, I believe it will still alive for another decade as big companies are talking 450 mm wafers already. With the development of multicore system for parallel computation, the speed of whole integrated system will still be going down for certain time, even the single device may be not. There will be no stop for technologies to going forward. At the same time, new nano-computation schemes are under development now in many universities and industry laboratories. Hopefully within another two decades, we will have something as cleaver as a simple brain with combination of large number of parallel computation theads and large amount of transistors.

[John Grant]

At a talk a while ago (can't remember who or where, might have been Ray Kurzweil in New York in '03, might have been a Computer Lab Wednesday seminar) the speaker suggested that ICs are about the 5th paradigm, starting with relays and going through valves, discrete transistors, etc, and if you plot complexity (log scale) against time (linear scale) they're all on the same straight line so a 6th paradigm (maybe involving 3-dimensional structures) will take over and continue the trend.

[Peter Hewkin]

John Grant said:

At a talk a while ago (can't remember who or where, might have been Ray Kurzweil in New York in '03, might have been a Computer Lab Wednesday seminar) the speaker suggested that ICs are about the 5th paradigm, starting with relays and going through valves, discrete transistors, etc, and if you plot complexity (log scale) against time (linear scale) they're all on the same straight line so a 6th paradigm (maybe involving 3-dimensional structures) will take over and continue the trend.

I have been asked to lay on (under the CULIL brand) a talk about this .. does anyone know of potential thought-provoking contributors in the Cambridge region?