What happens to the behavior of a conventional circuit using conventional components as it is cooled towards absolute zero?

To make this question more relatable: What would happen as we cool a modern CPU and its power circuits?

In general resistance decreases with temperature, so I assume that initially we would see the device's power consumption drop for the same cyclic frequency. (Or would the clock mechanism change its frequency as it is cooled also? If so, let us assume for the rest of the question that the clock is external and not altered.)

Do transistors and other semiconductors begin to fail below certain temperatures? If so, by what mechanism?

At some very low temperature I assume that various conductors would begin to superconduct. How does this alter the behavior of the power supply and computational circuits?


Anything based on semiconductors would simply stop working, because they depend on electrons kicked into the conduction band by their own thermal energy.

Metallic conductors might become superconductors, but that doesn't mean anything if the first thing you hit is a chunk of inert silicon.

There are types of logic that are designed to work at cold temperatures and using superconducting materials, for example, Josephson junctions.

  • \$\begingroup\$ Do all known semiconductors depend on a thermal mechanism to alter their conductivity? I thought there were a wide range of conductivity triggers for semiconductors, though it would be useful to know that they reduce or equate to a thermal mechanism. \$\endgroup\$ – feetwet Sep 8 '19 at 19:51
  • 1
    \$\begingroup\$ Well, you asked about "conventional" components, and modern CPUs specifically. But yes, that's pretty much the definition of what separates a semiconductor from a metal. The latter has free electrons at all temperatures. \$\endgroup\$ – Dave Tweed Sep 8 '19 at 19:56
  • \$\begingroup\$ Can logic circuits be made with josephson junctions? I was of the impression that they weren't the sort of thing that could be used for any sort of switching. \$\endgroup\$ – Hearth Sep 8 '19 at 20:03
  • 1
    \$\begingroup\$ @Hearth: They can, but it doesn't look anything like the kinds of circuits we use every day. \$\endgroup\$ – Dave Tweed Sep 8 '19 at 20:12

First it would stop working because everything is designed to work in the commercial temperature range (0 to 70 degrees C). One would hope it works down to 0 degrees, although if the designers were not dillegent it may not make it down that far.

It would stop working because the various bits would fall out of specification, and each would do so differently. It's hard to say what would fail first. It would be one of:

  • There would be so much timing skew between the signals that the various digital bits were receiving and what they required at that temperature that interfaces (i.e., between memory and processor) would stop to work, or logic would get scrambled.
  • Oscillators would stop working because the gain necessary to maintain oscillation wouldn't be there.
  • Voltage references would generate the wrong voltage (and not predictably, at least not up front).
  • Transistors in the power supply circuits would fail to conduct properly.

There's a vanishingly slim chance that it'd still work below 0 degrees C. I'm starting to guess here, but there's yet more that can go wrong:

  • Anything spinning, such as disks and fans, will seize up because of differential thermal contraction. They may be OK when they warm up again, but depending on how the disks stop working they may damage their surfaces.
  • If the temperature goes down far enough, differential thermal contraction between various bits (i.e., the traces and the PCB, the leadframes in the IC's and their epoxy packaging, the silicon chips themselves and either the leadframes or the packaging) will cause things to crack. This damage would be permanent.
  • If this is done in atmosphere, water vapor (followed by CO2, and then the various other constituents of air) would condense on the PC. Oxygen would condense out before nitrogen, which may cause interesting chemical reactions.

Basically, long before anything superconducts, the 'puter will have stopped working, and will very probably be permanently damaged.

Note that you can design electronics to work at cryogenic temperatures -- I used to work in IR imaging, where the imaging array was running at 77K, with CMOS circuitry. But that was circuitry that was specifically designed for the task, and which did not work well (or at all, sometimes) at higher temperatures.

  • \$\begingroup\$ Actually water will start condensing above 0 degrees C, this is called a dew point and is often mentioned in the context of extreme overclocking of commercial PC components. \$\endgroup\$ – Jan Dorniak Sep 9 '19 at 1:06
  • \$\begingroup\$ @JanDorniak: yes, I missed that. If you're worried about CO2 deposition, then water would be an earlier concern. I did a bit of design of electronics for -40 to +85C -- if you don't intentionally make it work over that range, it won't. \$\endgroup\$ – TimWescott Sep 9 '19 at 2:17
  • \$\begingroup\$ that's for sure, although some single units might work. And after writing that comment I realised that you would know what dew point is having worked with sub zero electronics. \$\endgroup\$ – Jan Dorniak Sep 9 '19 at 2:24
  • 1
    \$\begingroup\$ Actually it wasn't much of an issue, because the stuff at cryogenic temperatures was in a vacuum flask, and the support electronics was in an enclosure that was purged with dry nitrogen. But yes -- it's an issue. \$\endgroup\$ – TimWescott Sep 9 '19 at 2:34
  • \$\begingroup\$ pretty much every PC overclocking guide has a big warning: never cool under dew point until you know what you are doing. \$\endgroup\$ – Jan Dorniak Sep 9 '19 at 2:36

Convex Computer Corp exploited the speedup of CMOS logic for some of their machines.

The transconductance (delta_amps out for delta_volts in) got bigger, and the parasitic gate and flipflop and buss capacitances were charged and discharged ....... faster, as temperature dropped. I think they used liquid nitrogen.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.