1
\$\begingroup\$

I was reading this pdf regarding computation from MIT's opencourseware and on p.8 through some obscure mathematical manipulation they basically reach the conclusion that the cooler you make your chips, the less heat is dissipated and the bigger the maximum integration density can get. This seems intuitively logical to me, but their last statement piqued my interest:

At lower temperatures, the power dissipation on chip is decreased, but the overall power dissipation actually increases due to the requirement for refrigeration.

For their actual calculation they used room temperatures, so I'm assuming they mean that if we try to cool the chip (to any temperature lower than room temperature), it would be obselete because the cooling itself would increase the overall power dissipation.

But I don't understand this argument. You could say the same for refrigerators. What does it matter that the overall power dissipation increases? If you cool the chips, you can increase the integration density which would lead to stronger computers. What does it matter that the heat in other parts of the system increases? I might be wrong about this, but this is how refrigerators work too, no? It's just in the laws of thermodynamics..

So an answer to this question would clarify the above statement (or say it's indeed stupid). Why shouldn't we cool chips to low temperatures in order to increase the maximum integration density?

\$\endgroup\$
1
\$\begingroup\$

Cooling is a vital part of most computer system design.

Air at room temperature is the usual coolant, because it's available everywhere* and you can move it easily with a small fan.

Some enthusiasts and large systems are liquid cooled. All the plumbing and pumping systems can't be miniaturized, and add cost, so this is usually done only on "tower" size systems: http://www.pcworld.com/article/227964/pc_liquid_cooling_system_do_it_yourself.html

The extreme is liquid nitrogen cooling: http://www.corsair.com/blog/setting-up-your-pc-for-liquid-nitrogen-overclocking/ , which is even more expensive and impractical. It's nearly always easier to stick with conventional cooling and buy more computers to achieve the same computational throughput.

(There are also issues with hold timing when running extremely cold - signals may arrive too early if the chip has not been designed to run at that temperature. If you move the minimum design temperature down to -200 that may force you to move the maximum temperature down as well, leaving you with a product that requires a liquid air supply and thus a very limited market)

  • Except in space, where ultimately all your cooling must be radiative
\$\endgroup\$
  • \$\begingroup\$ Interesting, thanks. So I'm guessing cooling is one of the areas of research in electronics? From your answer, it seems to be a field on which we could improve quite a bit. \$\endgroup\$ – Phaptitude Jan 12 '14 at 12:18
  • \$\begingroup\$ I'm not sure there's much scope for improvement - the behaviour of all fluid-based cooling systems is entirely limited by thermodynamics. You might be able to achieve something with miniature turbopumps for high speed coolant circulation. Materials science might provide better conducting substrates; the ideal would be a small square of synthetic diamond (very high thermal conductivity). \$\endgroup\$ – pjc50 Jan 12 '14 at 12:25
  • \$\begingroup\$ (Reducing the power consumed by the silicon per gate is of course a massive area of research) \$\endgroup\$ – pjc50 Jan 12 '14 at 12:26
  • \$\begingroup\$ Thanks. So one final question. How is it justified that they used room temperature for their calculation? \$\endgroup\$ – Phaptitude Jan 12 '14 at 12:47
  • 1
    \$\begingroup\$ "In practice, using water cooling of a uniformly heated Si substrate with embedded micro channels, a power density of 790 W/cm has been achieved with a substrate temperature near room temperature." == that was the temperature achieved under particular lab conditions in a particular experiment. Normally chips are designed for e.g. 80C junction temperature as a worst case. \$\endgroup\$ – pjc50 Jan 12 '14 at 13:15
0
\$\begingroup\$

It really depends on the design goals. If a goal is to maximise instructions per unit energy then refrigeration may not be worthwhile. If a goal is to maximise instructions per unit time without regard to the energy cost then refrigeration may well be a worthwhile option.

\$\endgroup\$
  • \$\begingroup\$ But why isn't cooling used to make computers stronger? I'm guessing that using more energy wouldn't be dramatic when testing for stronger computers? \$\endgroup\$ – Phaptitude Jan 12 '14 at 11:58
  • \$\begingroup\$ It is, when appropriate. \$\endgroup\$ – Scott Seidman Jan 12 '14 at 12:07
  • \$\begingroup\$ @ScottSeidman When is/isn't it appropriate? \$\endgroup\$ – Phaptitude Jan 12 '14 at 12:11
  • \$\begingroup\$ When the benefit isn't worth the cost, which is the case for most consumer targeted computers \$\endgroup\$ – Scott Seidman Jan 12 '14 at 12:16

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.