Re: quantized resistor doesn't save power
From: Uncle Al (UncleAl0_at_hate.spam.net)
Date: 01/11/05
- Next message: Ken S. Tucker: "Re: Androcles and Draper resume Einstein 1905"
- Previous message: John Baez: "Re: too much information!"
- In reply to: Zigoteau: "Re: quantized resistor doesn't save power"
- Next in thread: Franz Heymann: "Re: quantized resistor doesn't save power"
- Messages sorted by: [ date ] [ thread ]
Date: Tue, 11 Jan 2005 13:27:44 -0800
Zigoteau wrote:
>
> Hi, Fred,
>
> > thanks for the extended discussion. What you say is entirely
> > correct, and I knew all that. You also brought up some additional
> > angles which are also useful (distance-dependent T, spin-dependence).
> > The point of bringing it up is precisely to indicate what happens
> > when you scale down a metallic wire to smaller and smaller
> > dimensions (area and length).
>
> OK, perhaps I was teaching you things you already know, but
> (1) I have come across exactly the same line of reasoning from a
> surprising number of people, and
> (2) it's still not clear that you have accepted everything I said.
>
> > Eventually you hit the lowest unit of conductance (if not
> > zero) and this corresponds to around 13 kilo-ohms.
>
> ?? The lowest unit of conductance??
>
> First, let's get a few terms agreed. 13 kilohms is a resistance. The
> unit of resistance is the ohm. Did you mean to say quantum? There is no
> "quantum of resistance". It is only conductance which has a quantum,
> and then only in a certain sense. The unit of conductance is the
> siemens (S). Yhe quantum of conductance is 38.7 µS.
>
> Conductance quantization is not a problem at all, per se. If you want a
> resistance less than 13 kilohms, you can just use more modes. If you
> want a resistance greater than 13 kilohms, you put in a tunneling
> barrier. You can get any value you damn well please. For the engineer,
> quantum mechanics is just part of life's rich tapestry, providing
> constraints on the one hand and an extremely rich repertoire of
> behavior on the other.
>
> > The conductor is
> > already a resistor.
>
> Absolutely. Only superconductors have exactly zero resistance, and even
> then only at zero frequency.
>
> > Between power and ground, you would need enough
> > resistance across all paths to reduce power. You won't get this by
> > scaling things down.
>
> ??? I am not sure where you are coming from, and you have clearly not
> understood what I said. It is perfectly possible to reduce power and
> scale things down at the same time. I can reduce power to zero if I so
> desire, by turning all the switches off. Why do you think I can't?
>
> There definitely are power constraints on electronics at the nanoscale.
> These are already being encountered in modern microprocessors. You
> hinted at one when you said (in so many words) that, for reliable
> computation at 300K, the power supply voltage should be at least 0.25V
> = 10kT/e. In fact, for each bit reliably computed, it is necessary for
> at least one electron to go from the negative supply rail to the
> positive supply rail, thus dissipating at least 10kT = 40 zJ per
> computed bit. I would have said 100kT myself.
>
> I discount the possibility of reversible computation, which is supposed
> to get around this limit. Although it has been considered possible by
> many intelligent people, none of them has figured out a way in which it
> could actually be carried out.
>
> If you assume switching elements packed at a density of 1 nm^-3 and all
> switching at 1 THz, that's a power dissipation density of 40 GW.m^-3,
> and things get seriously hot before you have assembled very many logic
> gates. However if you want to keep the power dissipation low, all you
> have to do is to reduce the switching rate for most of the gates. The
> human brain is proof that you can perform an amazing amount of
> computation even with these restrictions. I think the brain dissipates
> a femtojoule or so per switching operation, orders of magnitude worse
> than the theoretical limit, and the switching rate is at most kHz.
> Cheers,
Low power density serious computation is approachable,
1) Physically discrete clusters, including multiple CPU cores/chip.
Rack-mounted Opteron 800-series blades are impressive. Ask Google
about managing heat dissipation.
2) Primary design. AMD knows how to build a cool-running CISC CPU
for serious number crunching throughput with x86, Win, and Linux
compatibility. Apple pulls it off in RISC processors. Intel is
strictly second rate - hot and low throughput. CPU frequency is a
hollow boast.
3) Don't erase. Zero-dissipation computation is possible, but it
doesn't reduce to practice. Ditto quantum computation with its world
record seven-bit demo.
4) Run the thing immersed in liquid nitrogen with forced liquid
cooling. That will get a Prescott Pentium4 to about 8 GHz. Kinda
pricey - and don't repeatedly cycle to room temp.
5) Build 3-D CPUs with internal forced coolant. That sounds a lot
better than it plays.
6) Human math prodigies and idiot-savants demonstrate that the
greasy mess we call a brain can have awesome minor variants. Building
the improved versions is the small problem. What does one do with the
failures?
-- Uncle Al http://www.mazepath.com/uncleal/ (Toxic URL! Unsafe for children and most mammals) http://www.mazepath.com/uncleal/qz.pdf
- Next message: Ken S. Tucker: "Re: Androcles and Draper resume Einstein 1905"
- Previous message: John Baez: "Re: too much information!"
- In reply to: Zigoteau: "Re: quantized resistor doesn't save power"
- Next in thread: Franz Heymann: "Re: quantized resistor doesn't save power"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
