University of Rochester

Researchers Lay Groundwork for Next Generation of High-Speed Computers

January 10, 2001

Though Intel recently unveiled its one-gigahertz Pentium chip to great fanfare, inside a frozen canister of liquid helium in a small laboratory at the University of Rochester, a prototype of a new kind of circuit silently cruises along 50 times faster. Powering the chip is a new kind of "transistor" capable of a searing 750 gigahertz.

The near-absolute-zero-temperature chip is a crucial link to the future of computers because by 2010 the computer industry will hit a brick wall. Transistors-the building blocks of computer chips-will have reached a size limit that will prevent computer scientists from pushing chips any faster. The smaller a transistor, the faster it can run, but within a decade engineers will have made transistors as small as they possibly can. At that point, the only way to faster speeds is to adopt a new technology.

Mark Bocko is looking ahead at one of the possible new technologies. "Superconductors are a realistic alternative to semiconductors for future electronics," he says, and there is a lot of evidence to back him up. Bocko, professor of electrical and computer engineering, has a superconducting circuit in his lab that has been clocked at 50 gigahertz. "And that's without a billion-dollar fabrication facility to make it."

Bocko and his collaborators, professor Marc Feldman and doctoral student Jonathan Habif, are testing the speed of these new chips, which at their heart contain "transistors" known as Josephson junctions. These junctions switch between "on" and "off" states like traditional transistors, but they do so in just two trillionths of a second-thousands of times faster than conventional transistors.

That furious speed creates a mountainous hurdle that Bocko and his collaborators have overcome. Until recently, the clock running on a 50-gigahertz chip ran so fast that researchers couldn't keep up to measure its stability-its ability to consistently keep proper time. With no way to govern the clock's stability, the rest of the chip was gridlocked. To understand how Bocko and his team have made superconducting chips feasible, imagine a chip as Manhattan Island: a web of interconnected streets packed with people wanting to get to specific destinations. At each intersection a traffic light keeps cars moving in the most efficient manner from street to street. If a traffic light changes a couple of seconds late, there's really no difference in the traffic flow.

Now replace all the bustling cars with bullet trains.

With everyone moving at 300 miles per hour, that same two-second delay in a red light could mean the difference between a successful trip across town and a series of trains T-boning each other in half. At extreme speeds, a little slop in timing can bring the entire system to a standstill.

This was the problem Bocko had to solve. The clocks, themselves built of extremely fast Josephson junctions, ran at such a pace that trying to track down any imprecision in them seemed impossible. But an answer came from an unlikely source: gravity.

Bocko worked for years on gravity wave detectors, devices meant to sense the warping of space created by binary black holes and colliding neutron stars. Such waves are tremendously hard to detect. While developing gravitational wave detectors, Bocko and his colleagues designed extraordinarily sensitive transducers to detect the vibrations of a cryogenically cooled multi-ton bar of aluminum. The sensitivity of the transducer depended critically on the stability of a clock. If a gravity wave passed through the antenna, it would contract and expand by a fraction of a proton's width, creating a signal from the transducer. If the clock was less than perfectly stable, it could cause a misreading of the transducer. Much of Bocko's contribution to the project was in detecting the tiny irregularities in the clock, and tracking down whether the gravity wave detector signals were due to a passing gravity wave, or just to clock noise.

To keep tabs on his new superconducting clock, Bocko inserted a cascade of Josephson junction toggle switches that allowed him to monitor the clock every 2, 4, 8, 16, etc. cycles. He matched those clock cycles to a quartz clock-essentially the same as in a wristwatch-kept in a temperature-controlled oven to prevent any change in its regularity.

"We have designed a simple tool that allows us to directly measure the variation of the time interval separating the ticks of a superconducting clock," says Bocko. "If it fails to be as accurate as we need, this diagnostic tool guides our design changes so we can fine-tune the design of the clock until we have exactly what we need."

After designing an accurate clock, the Rochester group worked with their collaborators at HYPRES Inc., a small superconducting electronics company in Elmsford, N.Y., which is looking to apply the group's work to superconducting wireless communications equipment. Together, they worked to figure out a way to uncover any irregularities in the network that distributes the clock's "ticks" around the chip. By putting the quartz and chip clocks in sync, Bocko can measure the superconducting clock signal as it works its way through junctions, switches and splitters of the chip network. By comparing the known time the clock sent its signal with the time it emerged from any point on the network, Bocko can tell what quirky roadblocks and irregularities litter the network.

"This information is crucial to building superconducting circuits," says Bocko. "Engineers need to know exactly how the clock, basic building blocks such as gates, and the clock distribution network perform before they can even begin designing an entire microchip."

Bocko mentions one possible design that is all but impossible for today's technology. Right now, circuits are laid out on a flat board, which puts some components far away from others and leads to longer communication times and slower speeds. Engineers would love to change the flat boards to cubes, building components on top of as well as beside one another. This would give more connections between transistors and shorter distances for communication to travel around a chip. Today's transistors, however, produce so much heat that if they were built into a cube configuration, the trapped heat would lead to a meltdown. A chip based on Josephson junctions, on the other hand, would create one thousandth as much heat, which could easily dissipate without causing harm in even the most densely packed circuitry cube.




Facebook