Re: Moore's law [WAS MEDIA: Forbes ..]

Michael Nielsen (mnielsen@tangelo.phys.unm.edu)
Fri, 26 Jun 1998 18:01:56 -0600 (MDT)


This is a resend. I sent this message out 10 hours ago, but it never
seems to have made it to the list.

On Thu, 25 Jun 1998, Dan Clemmensen wrote:

> Michael Nielsen wrote:
> >
> > What are the possibilities for increasing computer power beyond that point?
> >
> Massive parallelism, "configurable computing", superconductors, reversable
> computing.
>
> The only new item in this list is "configurable computing." This is the
> most recent buzz phrase making the rounds of the chip designers. The idea
> is to make on-the-fly changes to the programs of the progammable logic that
> make up a portin of the computer, either to alter the instruction set or to
> alter the interconnectivity of various computing elements.

Interesting. Makes computer hardware more of a software problem. I
imagine it's quite difficult to do optimally, but it seems as though
significant gains would be possible merely by dynamically changing the
degree of connectivity.

> My list of workarounds is actually for computing power,
> not memory density. For memory density, just use nanotech to store one bit
> per hundred atoms or so, in three dimensions. This keeps Moore's law of memory
> density going to about 2030, I think.

Storing stuff in 3-d gives rise to a large dissipation problem.
Heuristically, the problem is that the heat generated goes like length
cubed (total number of components), while heat dissipated goes like
length squared (proportional to the surface area). As I've stated
elsewhere, the reversible compution solution to this problem has a large
potential hurdle to overcome, in the form of heat generated by error
correction, which suffers the same probelm re dissipation.

Michael Nielsen

http://wwwcas.phys.unm.edu/~mnielsen/index.html