Darpa Wants Teeny-Tiny Fluids to Cool Down Next-Gen Microchips


The Pentagon’s mad scientists have concocted a plan to keep the miniature, stacked brains of tomorrow’s advanced computers cool enough to power next-gen technological advances. It involves the world’s smallest bath.
Advanced new microchips are now stacking up like pancakes. This new turn toward stacked chips promises huge improvements in computing power for everything from advanced cameras to new smartphones. But the Pentagon is concerned about these new stacks of chips being too powerful — that is, they risk melting down because they get too hot.

Darpa’s plan: embed them with tiny fluid channels to circulate really, really small blobs of water. This month, the agency released a solicitation asking the industry to come up with designs for “microfluidic” cooling systems which can be embedded into microchip stacks, called ICECool. The specifics are painfully complicated, but the project would involve using tiny “microgaps” between “chips in three-dimensional stacks” (more on this in a minute) that can be used to pump “naturally-circulating flows as well as directed liquid jets” to keep the microchips cool.
To describe why Darpa is interested in this, in a crude and simplified way, we should start with Moore’s Law.

Microfluidic channels within a 3-D microchip stack. Illustration: Darpa
According to the “law,” the number of transistors — which transmit information inside a computer — doubles about every 18 to 24 months, thereby doubling computing power. It’s really more of a rule of thumb, but one that’s largely held up since the 1960s. Its further development is also necessary for efficiently and rapidly building more advanced computers like smartphones — with battery lives that last long enough for them to remain practical — and for ever-larger server farms and data centers. The maximum number of transistors stored on a single chip may also be coming to an end, requiring ever more strange and creative ways to get more computing power in less space.
To keep up with Moore’s Law, engineers have for years made individual components on microchips smaller. But for most memory chips, individual capacitors are still stacked next to each other like buildings on a city street, one after another. But a 3-D stacked version would place the capacitors vertically, like a skyscraper, giving the stack a lot more room and allowing a computer to multiply its processing power with less overall space. Even better than that would be stacking whole microchips on top of each other, a process called “3-D chip packaging,” with super-thin silicon wafers pancaked together and connected by hair-thin electrical wires.
They’re horribly complex to make. But it’s been done. And for your next smartphone battery and the Pentagon’s data centers, it’s almost perfect. You have more computing power in less space, with less latency thanks to the shorter wires, and with lower demands on electrical power compared to what you were using before — which means longer battery life. IBM is even working on 3-D microchip adhesives that could potentially make microprocessors that compute at 1,000 times the speed that they do now.
The problem is that 3-D stacked chips can get really, really hot — too hot for cooling fans to chill. That can damage or outright destroy the microchips, and at the least slow down their computing power. (There goes the next generation of smartphones you were waiting for.) More worrisome, the lack of a cooling mechanism for the stacked chips threatens to freeze their promise, inhibiting future technological leaps. “These thermal limitations have compromised the decades-long Moore’s Law progression in semiconductor technology,” the Darpa solicitation warns, “and threaten to derail the technology engine which has been responsible for much of the innovation in defense and commercial microelectronic systems.”
Enter Darpa’s extremely small liquid spritzes.
Obviously, developing microfluidic cooling systems won’t be easy. The scale is extraordinarily small, with droplets circulating through these channels at the microliter and nanoliter levels. To stop the water from interfering with the electrical flow of the chips, there will likely need to include an insulator coated with water-repellent material. The microchips’ electrodes also have to be insulated from the drops in order to keep up a steady flow.
There’s the question of how to keep the pressure steady in order to prevent the water from drying or burning up, and how to transfer excess heat away from the microchips. In theory, microfluids may work much better than current air-cooled systems at evening out temperature across the whole chip, allowing heat to dissipate fairly quickly. One way, according to Duke University Professor Krish Chakrabarty, is to automatically switch off electrodes in areas that get too hot (.pdf). Water near those electrodes is then dropped on an indium tin oxide plate between the electrode and the fluids channels, which then absorbs the heat and dissipates it away.
Darpa doesn’t specify what kinds of military systems it wants to cool with water. But there’s no shortage of them. Some experimental cameras are now in the 50-gigapixel range, and with growing demand for ever-more-powerful smartphones, smaller and smaller devices are being required to capture enormous gobs of data. For the military, that means figuring out how to get enough bandwidth to pull down enormous streams of data from cameras placed high in the sky.
“More generally, proposed approaches should be crafted to be scalable and adaptable to the environment of a modern military electronic system,” the solicitation noted. If it works, it could mean that a lot of those advanced military projects will have a lot more power to get going, and not have to worry about burning up. But left unsaid is how the military’s taste for advanced projects hasn’t cooled at all. LINK