Chip Chiller

Wiesel Werkstätte writes "Nature (just call me Natureboy) reports that a team at the University of California/Santa Barbara have developed a a silicon nano-scale superlattice that can be fabricated directly on chips to cool individual components thermoelectrically."

Comments

Re:More important than it may seem

[peccary]

While it's true that there are some spots on the chip that run hotter than others, keep in mind that this actually consumes additional power to provide that localised cooling. Therefore, it cools hot spots at the expense of heating up the whole chip. So it's not a pancea for, eg, the heating problems facing a rackmounted server. (And since this has to be designed into the chip, it's obviously of little use to the home overcoo^H^Hlocker.)

This technology would be useful for creating that 2GHz P4, but the thing would be hotter than a Kenner EZBake Oven, and it would be very difficult to get that heat out of the case before it cooked some other components.

More important than it may seem

[martyb]

IANACD (I am not a chip designer) but it strikes me that this technology has the opportunity to provide a rapid increase in processor speed. As the article notes, there are localized areas on chips that run hotter than the rest of the chip.

Emperical Method: Thermal imaging of a running chip would point out where the hot spots are. Move a few circuits around, refab using some of these on-chip chip-chillers, repeat.

Theoretical Method: Chip designers surely must know where the power ebbs and flows. I would expect that they could estimate even during the design stage, where the likely hot spots are going to be. Drop in a chip-chiller (or two) and you can design more aggressively than would otherwise have been possible.

Other applications: I don't know if this would pan out, but I would suspect that this technology would also permit the design and development of chips that could handle much higher ambient temperatures. Could be handy in harsh environments (cars, industry, space).

This has lots of uses

[human bean]

Consider creating a "cell modifier" with one of these. A fluid channel is etched into the wafer, cells float by and are examined by some means (size, imaging, fluorescence, etc.) "bad" cells are frozen in spot by tiny thermolectric cooler and destroyed by electrical current or heat. Now make thousand of them on a wafer. put multiple wafers together. Now you have a device that can filter enough fluid to be useful in treating patients. Maybe you don't even happen to make channels, you just create spots on the wafer and immerse the working surface, and let brownian motion work it out.