On Tue, 25 Apr 2000 hal@finney.org wrote:
> I saw a pointer on slashdot yesterday to this article at EE Times:
> http://www.eet.com/story/OEG20000421S0024. It discusses some of the
> issues relating to future progress in miniaturization of electronic
> circuitry.
[snip]
> The other interesting thing about the article was their long term
> projection for shrinking feature size. It shows a decrease from 180 to
> 25 nm over 18 years, a factor of 7.2. If density is proportional to the
> square of the feature size, that would be a density increase of 52, about
> 6 doublings. 6 doublings in 18 years is one doubling every three years.
> That would represent a halving of the traditional rate of increase of
> Moore's Law, which has been quoted as a doubling of density every 1.5
> years (actually variants using 1 and 2 years also exist).
Not all of the improvement in Moore's Law has been due to simple
increases in circuit "density" (due to decreases in line width).
There have been changes in device architectures as well. There
are several of those in the pipeline that are designed to change
the architecture of DRAM cells as well as turning the transistor
architecture on its side.
Then of course we have copper interconnects slowly being implemented.
One has to wonder if we will one day see silver which has lower resistance.
There is a fair amount of work going on with software on compiling code into
gates. Given the progress Transmeta has made on the software side the
next logical progression (given the large chips that will be available)
will be a large chunk of gates into which the highly used code segments
can be "compiled".
Processor-in-Memory is coming as well as a some work on asynchronous clocks
and reversible computing. Reversible computing gets phased in when they hit
the limits of decreasing voltages and have to do something about power
consumption as the clock rates continue to increase. The adoption of
reversible computing architectures is the only place I expect there
might be a blip in the current trends. That is because you have to
take a fairly large hit in gate count to implement reversible architectures.
In general however, I'd not hurry to the window to place my bets against
Moore's Law slowing down anytime soon.
Robert
This archive was generated by hypermail 2b29 : Thu Jul 27 2000 - 14:09:49 MDT