>How much power can be got from a thermocouple, heated by the cooking
>CPU to help cool it off?
Every once in a while this goes around the mailing list, so let's do
the math. Assume the processor is hot enough to boil water at 373 K
(100 C or 212 F) and the immediate environment is 293 K (or 20C or
68F). Both of these assumptions are pretty ridiculous
in today's usage. However, continuing the analysis, Carnot efficiency
(the physically best we could do) is
\[\frac{T_{body} - T_{ambient}}{T_{body}} = \frac{(373 K - 293 K)}{373 K} = 21.4\% \]
of the heat, which is pretty good. However, the best materials that
couple heat to electricity today are < 10% of Carnot (this ignores
Strachan and Aspden's 1990 result of 73% which no one has been able to
reproduce) . Thus, the actual recovery would be < 2%. For a 10W
processor, you get .2W back maximum. To do this you need to maintain
the low temperature at the far end of the thermopile, which is
non-trivial without water convection, forced air, or a heat
pipe. Using this power for a fan is unlikely (fans take much
more W than this) and self defeating (if it can't
remove enough heat from the cool side to generate enough power to
start the fan, it can't remove enough heat ...).
Thus, its much more beneficial to optimize your compiler for
efficiency (either for code efficiency or with respect to heat - good
thesis project BTW) than to try to recover energy from processor heat.
Thad Starner
MIT Media Laboratory
Wearable Computing Project
From Wear-Hard Mailing list Archive (WH)
Maintained by R. Paul McCarty
Archive created with babymail