I had a college student build an overclocked PC using phase-change technology. (This is essentially an air-conditioning unit with the evaporator attached directly to the motherboard.) He said that cold temperature made the CPU more stable at higher clock speeds (5.5 GHz) by affecting how the atoms and electrons behave. My question is, is this true? If so, in what way does cold temperature affect semiconductors?


Temperature certainly affects semiconductors.

As someone who is not specialized in semiconductors, I can think of at least two microscopic effects at play here. The first is that at lower temperatures there are less phonons (quantized vibrations of the atomic lattice). The effect of phonons is to scatter the electrons and lower the conductance of the semiconductor. Thus at lower temperatures the conductance should be increased (as in the case of metals).

The second effect is that as the temperature is lowered, the fraction of electrons on high energy states is lowered (and on low energy states increased) thus potentially changing the amount of electrons electrons above/below the band gap (carrier concentration). The effect of this is more complex than the effect of phonons.

  • $\begingroup$ Although the other explanations are good, they focus on the practical effects of CPU cooling rather than the underlying physics, the latter of which was the answer I was looking for. $\endgroup$ – LucasMcGraw Jul 24 '12 at 20:52

I am not really an expert on solid state physics, and I'm prepared to look an idiot here - but I don't think it's the low temperatures that help.

By overclocking a CPU 5.5 GHz you are almost doubling the power consumption and hence the dissipation. A large reduction in the temperature of the cold site of the heat sink helps it remove twice as much power while keeping the hot side (i.e. the CPU) at a reasonable temperature.

Resistance of semiconductors increases at lower temperatures and certainly once you get to cryogenic temps (below 100 K) you are very limited in the types of IC devices you can use and their performance.

I suspect that running at 5.5 GHz, putting out several hundred watts then even if the heat sink is in liquid nitrogen the CPU core is still near room temperature.

  • $\begingroup$ This is not true-- overclockers routinely manage to get the heat out efficiently, just by putting a liquid next to the chip instead of air. The heat conductivity of a liquid is higher than that of air, and it cools the solid chip more efficiently. Actually, I don't get what this answer is trying to say--- is it saying you can't reduce the heat in the chip? Or that it requires liquid nitrogen? Neither is true. $\endgroup$ – Ron Maimon Jul 24 '12 at 2:25
  • $\begingroup$ I'm saying that the impressive cooling systems (some even use LN2!) used to clock a CPU at 2x it's spec is in order to get the heat out rather than improving the semiconductor performance by having the CPU core operating at very low temperatures. $\endgroup$ – Martin Beckett Jul 24 '12 at 2:42
  • $\begingroup$ I see, that makes sense. $\endgroup$ – Ron Maimon Jul 24 '12 at 11:31
  • $\begingroup$ @Ron - just reread my post, I didn't mean to claim that you needed a lower T_cold. In fact you just need to maintain the previous T_cold with more power. I've rewritten the answer - thanks $\endgroup$ – Martin Beckett Jul 24 '12 at 16:22
  • $\begingroup$ The temperatures that the surface of the chip are experiencing are -40 C. I agree that the primary purpose is to remove heat, but CPU stability is also an issue here, right? I was told that stability is an issue of the materials that the CPU is made of, and varies from CPU to CPU. $\endgroup$ – LucasMcGraw Jul 24 '12 at 18:55

That is true. "Another reason that electronics has been operated at low temperatures is improved performance of the cooled electronics itself. Improvement upon cooling results from a combination of effects: in general, transistors (field-effect types) exhibit increased gain and speed and lower leakage; also parasitic resistances and capacitances in interconnections decrease, heat transfer improves, and many devices exhibit lower noise." (http://www.extremetemperatureelectronics.com/tutorial1.html)


Being too cold or too hot can cause problems in either case. Conductance at semiconductor junctions change with temperature. The junctions are more conductive at lower temperatures thus increasing switching speeds and less conductive at high temperatures thus decreasing switching speed.

It is this variation of switching speed that can cause different operating characteristics. Some parts of your circuit might start switching too fast or too slow for other parts. The idea of heating and cooling is to keep the entire circuit in the required temperature operating range. Also not to mention the obvious that if you are dumping 35 watts of power with nowhere for the power to go you are going to burn up your part.

This problem is first addressed at the design level by selecting parts with acceptable switching characteristics across the entire range of temperatures required for operation. Then it is the responsibility of the user to make sure that the design is kept in this temperature range during operation.


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