[extropy-chat] diffraction limit

Eugen Leitl eugen at leitl.org
Mon May 31 20:49:44 UTC 2004


On Sun, May 30, 2004 at 02:39:07PM -0400, Dan Clemmensen wrote:

> Thanks! Very interesting paper. Assuming I understand it correctly, it 

Very interesting indeed. Due to LCDs I almost never make hardcopies of a
papers, but of that one I did.

> says that no matter how you build it or what you build it out  of, 
> digital electronics cannot be packed more closely that about 20 nm 
> cubed. Each such cube can contain a piece of interconnect, a gate, or a 

No. I haven't had time to read it thoroughly, but it definitely puts
"irreversible logic" and "separation of a single charge" on a 
disclaimer list. I don't understand why they at all mention von Neumann, 
because it is completely irrelevant to the logic of the argument.

To start with, they assume SNL, which is only valid for irreversible
computing. Patient: "Doctor, Doctor, it hurts if I erase bits!" Doctor: "Then
don't do that". I don't have an URL, but there's a paper on people
estimating on threshold when reversible logic is worthwhile, and while it isn't yet
worthwhile for 2d (for a given structure size/switching speed which I don't
recall), the threshold is something over >10 layers, or so. It is certainly 
a vademecum for any 3d integration. 

Coherent spin transport and spin valves are more energy efficient, precisely
because they don't require rapid tweaking of charge separation. 

> flip-flop. And yes, I'm making a very free interpretation of the paper.
> 
> The paper is almost entirely about 2D systems. For a 2D system, with 
> flip-flops of 20nm squared, the heat dissipation will be about 60W/cm 
> squared.

I thought we were past that power density in current flock of chips.
 
> To a first approximation, this paper concludes that the practical limits 
> of electronic computation coincide with the practical limits of CMOS, 
> and are consistent with the Moore's-law industry roadmap's goal for the 
> year 2017.

Do you really expect we won't have molecular components at least in memory
within 13 years?
 
> Note that these limits will are for CPUs (i.e, each cell changes state 
> frequently) as opposed to memory. A nanomechanical memory can be as much 
> as 1000 times denser than a nano-electronic CPU.

Cellular architectures make no difference between a memory cell and a CPU (so
mentioning von Neumann is a red herring). MRAM cells can be made to do 
arbitrary logic with two extra wires.

Mechanical oscillations can be used for switching in a quantum dot (a paper
from last couple of Science or Nature).

While Moore is doomed on the long run (any exponential process in real space
is) burying the fellow prematurely is not a good reality model, especially as
AI bootstrap is hardware-bounded.

-- 
Eugen* Leitl <a href="http://leitl.org">leitl</a>
______________________________________________________________
ICBM: 48.07078, 11.61144            http://www.leitl.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE
http://moleculardevices.org         http://nanomachines.net
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: not available
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20040531/3d2e5989/attachment.bin>


More information about the extropy-chat mailing list