Heard at SemiCon West: local active cooling for opto-electronics, power devices and SoCs—not to mention genetic engineering--moves closer to reality

Solid-state active cooling, by way of thermoelectric piles or films, has looked promising for a long time. And it is being used in fairly large-format applications—for instance in solid-state refrigeration and as a replacement for heat sinks on IC packages. But Nextreme Thermal Solutions, a company with technology from both RTI and JPL, is suggesting that thermoelectric cooling—and heat-scavenging—can be done on a much more local level, and at substantially lower cost than would be necessary for complete large-scale coolers.

Nextreme’s technology rests on a thermoelectric thin film with very high heat-transfer capability: typically in the range of 100 W/cm-squared, and in very local areas under ideal conditions up to six times that figure. The thin film is uniform enough to be useful in very small patches. So, borrowing copper bump technology from the flip-chip packaging business, Nextreme’s designers came up with a novel approach: a sheet of the thermoelectric thin-film covered with an array of copper bumps, with a total height—film plus bumps–of 100 microns. In this case, the bumps act not as electrical connections but as thermal connections to carry heat from the object to be cooled onto the surface of the film. Applying a voltage to the film transfers the heat from one surface of the film to the other, and hence away from the hot device.

The neat part here is that once you have a sheet of film with an array of bumps on it, you can scribe and break it into pretty much any size you need. Nextreme is supplying what appears to be a several-square-inch patch for heat redistribution in one notebook computer. At the other end of the scale, you can carve off pieces as small as a few hundred microns on a side. This means that individual coolers—with the ability to pull Watts of power away from an active device–can be the size of a single MOSFET, optoelectronic component, or hot-spot on an SoC.

According to company vice president of marketing and business development Paul Magill, Nextreme has already demonstrated 20 C cooling on an LED for lighting applications, for instance. At this level the device would allow a lighting manufacturer to cut supply current in half—net of the current to the cooling device—while maintaining the same luminosity. Similar studies are underway for power MOSFETs.

There are other, not-so-obvious applications as well, Magill says. For example, the ability to rapidly alter the temperature of a very local area by, say, 10 C could be extremely useful in electronics test applications. And temperature cycling at a 12 C/second rate is necessary for the Polymerase Chain Reaction used in gene splicing. Nextreme might enable nano-scale gene processors on a chip, perhaps. One study is even investigating the use of extremely local cooling inside the brain (with an invasive probe) to halt seizures in epileptic lab animals.

Reversing the energy flow, the film can of course be used to convert waste heat into electrical current. It’s not very efficient in that direction—about 5 percent at a 200 C temperature difference. But that may already be sufficient for some energy-scavenging applications, and with study the efficiency can perhaps be doubled, Magill says.

For our purposes, it’s interesting to speculate about the use of the little micro-coolers in SoCs. Increasingly of late attention has shifted from the overall heat dissipation of advanced SoCs to a much more difficult problem: local heating. Circuits in the 65 and 45 nm domains can generate a great deal of local heat because of their very high switching frequencies. And that heat in turn increases leakage current, causing even more localized heating. Some high-duty-cycle circuits within an SoC can be much hotter than average figures for the chip would suggest.

This effect is not easy to model without very precise understanding of worst-case activity scenarios and excellent thermal tools. So hotspots can show up during production rather than during design. In some cases even if you know about a hotspot, there may be no way to avoid it without substantial architectural changes. In cases such as these, a local cooler that could chill a small, specific portion of a die could mean the difference between production and re-spin.

But is it practical? With discrete cooler devices, probably not. But Magill says that if an SoC required several areas of local cooling, the job could well be done by taking only about ten pins on the package to power the coolers. And if the tiny film-and-tower devices were built into the package assembly instead of added on later, the cost could be very plausible. Needless to say this is an area of active research for Nextreme.