From: Billy Brown (bbrown@transcient.com)
Date: Thu May 04 2000 - 08:29:15 MDT
Robert J. Bradbury wrote:
> Regarding Billy's comments, I went and got some numbers.
> Covalent atomic bonds are usually in the range of 0.5-1.7 aJ.
> von Neuman's classical limits on computer switching are kT ln 2
> ranging from 4.1x10^-3 aJ at room temps to 5.9x10^-5 aJ at lHe temp.
>
> So, if you represent a bond by a bit in a switch that can
> flip at something close to the theoretical limits, you can
> simulate that bond using 3-4 orders of magnitude less energy.
> This makes sense because one is probably representing the bit
> as a single electron, atomic magnetic alignment, etc.
> Brillouin [1962] got a similar result using photons to detect holes
> in punched tape (presumably related to the probability that
> the photon can reliably cause a change in the energy level of
> an electron orbiting an atom. This makes sense when you
> consider that only UV photons have enough energy to break
> atomic bonds.
I could nitpick by pointing out that you can't represent a bond with a
single bit if you want a simulation with atomic-scale fidelity, but that's
really beside the point. The key issue here is that you can simulate the
bond using less energy that the actual bond takes to operate, but you can
not get a further level of improvement by constructing a simulation of the
simulation. Once you've built the most efficient simulation you know how to
make, any further abstraction will increase your costs instead of reducing
them.
The implication is that the most efficient way to simulate anything is going
to be a specialized simulation program running directly on physical
hardware - IOW, something roughly analogous to existing simulation software.
The data models used by the simulation may be extremely abstract and
specialized, but they need to come as close as possible to being directly
implemented in hardware. Adding anything else to the system (like a VR world
running virtual hardware) would simply impose additional overhead and slow
it down.
> Now, if you can represent that bond by the position of a very
> low energy photon in a photon "gas", you can probably drive
> the cost of the simulation even lower.
Probably so, but we're still talking about a direct simulation here. What
you're looking for is a way to make a simulation of a photon computer more
efficient than an actual photon computer. I don't think you're going to find
one.
> Anyone know the energy required to flip the magnetic alignment
> of an atom?
I think Frietas gives the equations in Nanomedicine, in the chapter on
sensors, but I don't have my copy handy.
Billy Brown
bbrown@transcient.com
http://www.transcient.com
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:28:24 MST