Science: Liquid could power and cool mobile supercomputers
Getting microchips wet is normally best avoided. But a new type of chip that is both powered and cooled by fluid pumping through it could power the computers, smartphones and tablets of the future. If the design is successful, its inventors at IBM argue an entire supercomputer – like Watson, the firm’s natural-language-processing trivia savant – could one day be squeezed onto mobile devices small enough to fit in your pocket.
The idea is inspired by the way the human brain is powered, says Bruno Michel, who is leading the work at IBM’s Zurich Research Laboratory in Switzerland. “The human brain is 10,000 times more dense and efficient than any computer today. That’s possible because it uses only one, extremely efficient, network of capillaries and blood vessels to transport heat and energy, all at the same time,” he says.
Michel and his team’s idea is to stack hundreds of silicon wafers on top of each other to create three-dimensional processors. Between each layer is a pair of fluidic networks. One of these carries in charged fluid to power the chip, while the second carries away the same fluid after it has picked up heat from the active transistors – effectively creating a microscopic flow battery (see Rivers of vanadium below).
Chips in 3D have already been developed. Intel’s Ivy Bridge processors, which are expected to appear in consumer products next year, will use vertical transistors to allow more components to be crammed into a given area, yielding many times the processing power of conventional 2D chips. Intel has also dabbled with stacking microchips on top of each other, as have other chip manufacturers including Belgium-based IMEC and Tezzaron in Illinois.
“The use of liquid to cool 3D chips is not new,” says Bob Patti, chief technology officer of Tezzaron. “However, using the liquid as a power source as well as for cooling is a concept I haven’t seen before.”
Mark Zwolinski at the University of Southampton, UK, agrees it’s an interesting approach. “To get increases in high-performance computing it’s going to be necessary to move chips closer together,” he says, and that means stacking them. But powering them with liquid is uncharted territory, he says. “It’s not completely outrageous. I can’t think why it shouldn’t work, but it has never been done before.”
It had better work, says Michel, because the computing industry has for decades depended on computer chips to double in processing power approximately every two years – the phenomenon known as Moore’s law. But as transistors on conventional chips have shrunk, so too have the wires connecting them, increasing their resistance and making them less energy-efficient.
It takes about 85 kilowatts to run Watson, for example – enough to heat a dozen homes. And the machine’s servers take up the same amount of space as 10 large refrigerators.
Using this biologically inspired approach to combine the electrical and cooling systems into one should make it possible to reduce that power consumption considerably. Michel says he and his colleagues have demonstrated that it is possible to use a liquid to transfer power via a network of fluidic channels, and they plan build a working prototype chip by 2014. If successful, we could end up with Watsons in our pockets, powered by a battery akin to that found in a cellphone.
|Rivers of vanadium|
|To power chips using liquids, Bruno Michel of IBM has chosen a redox flow battery, which exploits the energy released when the oxidation state of a chemical changes. In this set-up, two electrolytes supplied by tanks outside the chip are pumped into the device in parallel channels. These fluids contain different types of vanadium ions, and electrons will flow from one to the other in an external circuit to create a current. Recharging the battery involves applying a voltage to reverse the process.
Flow batteries are ideal for chips because of their high power density, says Michel. The design is usually bulky, but Michel miniaturised it by lining the microfluidic channels with an electrode catalyst.
Duncan Graham-Rowe, New Scientist