Chip Weather Forecast: CoolChip Weather Forecast: Cool

Scientists at the University of Rochester have come up with a chip design that could lead to 3,000-GHz--that's 3-terahertz--processors that produce very little heat.

The "ballistic computing" design is a radical change, says Marc Feldman, a professor of computer engineering at the university. "There's a real problem for standard transistors to keep shrinking," he says. The Ballistic Deflection Transistor, as the new design is called, doesn't have the capacitance layer that current transistors have. That layer becomes problematic at very small scales.

Quentin Diduck, the graduate student at the university who came up with the idea, describes the design as the next step on the evolutionary track after relays, tubes, and semiconductors.

The transistors bounce electrons into their chosen trajectories, using inertia to redirect them, "instead of wrestling the electrons into place with brute energy," according to the university. A transistor functions more as an intersection for electrons than as a device that expends energy to stop and start them. Because of this approach, far less power is required.

The new design relies on a layer of a semiconductor material called a "2D electron gas," which facilitates the transit of electrons without the interference of impurities.

The transistors are expected to put out very little heat. "We don't have the mathematics to predict how small this is going to be," Feldman says. "But the currents it would take are very small. So the power has to be small."

The heat generation for early versions of the design should be around a few microwatts per transistor, orders of magnitude less than current high-frequency transistors, Feldman says. "That's without doing any tricks to cut down the power," he says.

Companies that maintain large, energy-hungry data centers--Google, Microsoft, and Yahoo, to name a few--are no doubt looking forward to that day.

The transistors have a way to go before they're in PCs and servers. "Up to now, this has been a one-graduate-student effort," Feldman says. "We don't have any transistor behavior yet, but probably soon."

The National Science Foundation is providing the University of Rochester with $1.1 million to develop a prototype.