Magnetic Logic Attracts Money
DARPA funds spintronic and nanomagnet research teams to create low-power nonvolatile logic
Photo: Getty Images
4 January 2011—The U.S. Defense Advanced Research Projects Agency (DARPA) wants a new type of computer logic. It will rely on magnetism instead of electricity to do its job, and its developers say this difference could one day allow computers to run on a fraction of the energy now required. Some even predict that the change will make booting your computer a near instantaneous affair. The defense agency has doled out US $8.4 million for a four-year ”spintronics” project led by the University of California, Los Angeles, and $9.9 million for ”nanomagnet” research led by the University of Notre Dame. Both groups aim to build a basic magnetic logic circuit.
Right now, it’s impossible to have true instant-on computers, in part because today’s volatile processors forget what they’re working on as soon as they lose power. The chips forget because they depend on the flow of electric current. When the power disappears, the flow of charge stops, as does any progress on the processing task. When the power returns, the circuit must essentially start from scratch, using information stored separately from the CPU in nonvolatile memory—and that takes time.
Nonvolatile magnetic-based logic, however, could remember even when you unplug it and could get going the moment that power returns. Pedram Khalili, the project manager for the UCLA team, says this trait could lead to faster-booting machines.
Alex Khitun, an electrical engineer at UCLA, and coleader of that team’s project with colleague Kang Wang, instead looks at nonvolatile logic as a means for conserving power. He says the logic’s stability could allow a chip to use energy only when it’s computing and not waste it while waiting for its next instructions. DARPA has set energy efficiency as the major requirement for the research, Khitun notes, and both teams aim to develop a 2-bit adding circuit that will use a mere 10 attojoules per logic operation. With traditional complementary metal-oxide semiconductor (CMOS) logic it would take orders of magnitude more energy, says Khalili. ”Plus, CMOS is volatile,” he says.
Instead of depending on transient electric currents, the UCLA-led team is developing a spintronic device that would perform logic operations by colliding what Khitun describes as ”waves of magnetization,” or spin waves, and storing the aftermath—the waves’ interference pattern—in a new type of magnetic memory. Spin waves start by ”tipping” an electron’s ”spin”—a quantum mechanical measure of the particle’s angular momentum, which has an associated magnetic field. Spin has no real equivalent in our macroscopic world, but tipping it might be imagined as tilting a spinning toy top to start it precessing.
Tipping an electron’s spin, which researchers can do by applying a small voltage in special multiferroic materials, will upset its neighbors’ spins as well. Those neighbors in turn disturb other nearby electrons’ spins in a quantum game of whisper down the lane. Khitun says his group has already successfully set these ripples in motion along what they call spin-wave buses—channels for the waves to flow in, made up of magnetic nanometer-size wires. Since 2005, the UCLA group has practiced adding the undulations together to perform logic operations. ”I strongly believe that the coming generations of logic devices will be wave-based devices,” Khitun says.
The tricky part, Khitun believes, will be storing the spin-wave computations, because spin waves, just like electric currents in today’s logic circuits, fade over time. To make the new logic truly nonvolatile, his team is ”intensively” developing magnetoelectric cells that can record the output of the spin waves’ operations as they happen. So if the power vanishes, the circuit’s inchoate computations will not.
The University of Notre Dame group, headed by Wolfgang Porod, director of the university’s Center for Nano Science and Technology, believes there is no need for this two-step method—first performing logic operations and then storing the results. Instead of propagating spin waves, which Porod says will take considerable energy to launch, the Notre Dame nanomagnet logic would shuttle magnetic fields between diminutive magnetic ”dots”—some 60 to 90 nanometers wide and separated by 20 nm.
Just as one kitchen magnet can deflect another, one nanomagnet, made from ferromagnetic materials, can alter the magnetization inside others nearby. Using simulations comparing the nanomagnet logic’s performance to that of traditional CMOS, Porod says his team has found that by placing many of these dots in deliberate designs, they can form basic logic circuits that will use about one-hundredth the power and are inherently nonvolatile, given that the means for computing is the same as for storing the results. ”We knew there would be such advantages to using magnetic phenomena,” he says, ”but we had never really been bold enough to do it. One day we bit the bullet—and it worked.”
But UCLA’s Khalili says that spin-wave-based logic has its own advantages, noting that it could prove faster and easier to scale down in future circuits compared with the discrete ”magnetic islands” required by the nanomagnet approach. Spin waves can have a wavelength measuring only a few nanometers, he says, and can propagate across a continuous medium. Khalili believes that these features make them perfect for relaying messages across the 1-square-micrometer circuit that DARPA desires.
Although he believes that magnetic logic will find fundamentally new applications aside from lower-power, faster-starting gadgets, Khalili says that his group’s design was a natural evolution from the devices used today. ”You have logic. You have memory,” he says. ”We’re just bringing these two together in an energy-efficient manner.”