3 ms·
> CPU heat output is limited by the resistance of the semiconductors. Technically true but not applicable... Super conducting CPUs already exist (at below room
by tomxor 3y ago
> CPU heat output is limited by the resistance of the semiconductors.
Technically true but not applicable... Super conducting CPUs already exist (at below room temp), and they don't use semiconductors. They don't have an exact transistor equivalent but use digital logic circuits such as Quantum Flux Parametron (QFP) and rapid single flux quantum (RSFQ), which are comprised of Josephson Junctions [0] which are made from the same super conducting material. Relying on quantum effects of a gap between two conductors rather than specific material properties.
So while it is true that CMOS are limited by the resistance of both the conductor interconnects AND the semiconductors. Super conducting circuits only appear to need a conductor... There are other drawbacks, such as larger feature sizes due to the nature of super conducting materials, and QFP and RSFQ might be more complex? but this is potentially offset by huge advantages: zero resistance buys not only power efficiency and minimum heat generation (as required by landauer limit), but also allows circuits to be driven at a higher clock, so far demonstrated to 770GHz for one CPU [1]
[0] https://en.wikipedia.org/wiki/Josephson_effect https://en.wikipedia.org/wiki/Josephson_effect
[1] https://spectrum.ieee.org/superconductor-logic-goes-lowpower https://spectrum.ieee.org/superconductor-logic-goes-lowpower
- throwawaylinux 3y ago> so far demonstrated to 770GHz for one CPU That's not a CPU, it's a simple circuit. From the article, "Simple superconducting logic circuits have been shown to operate at speeds of up to 770 gigahertz." Exotic semiconductor transistors can switch that fast, record being over 800GHz I think. Regular transistors of the type in your phone can probably do around 200-300GHz. Not to say there could not be revolutionary designs opened up with superconducting wires or switches, just the switching speed itself doesn't necessarily tell us one way or the other.
- SomeoneFromCA 3y agoI doubt that regular cellphone transistor can _switch_ at 200Ghz; their GBW might be in that range, but for reliable switching you need clock frequency at most 1/10 of GBW
- throwawaylinux 3y agoThey can, as calculated by switching delays of simple circuits like inverters and NANDs (so simpler than a FO4). Actually it's probably more like 200 than 300, I haven't seen 3nm models but transistors are still getting faster.
- tomxor 3y agoGood catch, I misread that. Clock frequency is ultimately dependent on the longest path of propagation, so it does depend on the circuit, or CPU. I'd be interested to know from any experts whether there is potential for fundamentally higher switching speed in super conducting circuits. As you suggest, they are not necessarily directly indicative of the clock speed of a specific real world circuit or CPU, especially considering super conductors don't even use transistors and need to use different building blocks for equivalent logic... but as a rough indication based on order of magnitude of switching speed would be interesting.
- dumdumchan 3y agoIt might be possible to create a new kind of transistor that turns on/off by changing the superconductivity of the conducting material.