The work of Jim Allen, a physicist at the University of California, Santa Barbara, is so far removed from everyday experience that he has to explain it by analogy: a tabletop covered with refrigerator magnets. "They all interact with each other and do funny dances," he says.
But these are not ordinary magnets; they are nanoscale "quantum dots," and their dance is far from random. The tiny magnets can be choreographed, or programmed, to solve a logic problem. A tweak to one dot causes its neighbors to do "interesting things," Allen explains.
The magnets employ a property called electron spin. For decades, computer circuits have been based on the charge, or flow, of electrons. But electrons not only flow; they also spin up or down, offering a new way to store, manipulate and communicate information. Electron spin was discovered in the 1920s, though practical applications have been limited.
But Allen and his colleagues at Stanford University and the UC campuses at Santa Barbara, Los Angeles and Berkeley hope that's about to change. The four schools recently joined with Intel , IBM and four other companies to form the Western Institute of Nanoelectronics, which specializes in spintronics.
Although it encompasses several projects and goals, WIN primarily aims to find an alternative to conventional CMOS semiconductor technology. As CMOS chip features shrink below 65 nanometers, more energy is wasted as heat. Chip makers worry that their approaches have just a few more years to run effectively.
"The major reason for spintronics is clearly anticipation that there are really no solutions below 20 or 30 nanometers, particularly in terms of power dissipation," says Kang Wang, WIN's director and an engineering professor at UCLA. "Today we use electron charge, but we are looking for alternatives."
Spintronic circuits have other attractive properties. The spins can align like tiny bar magnets to create a kind of magnetism that can be retained even when powered off. Such nonvolatile memories have many potential applications in computers and elsewhere.
In July, Freescale Semiconductor announced the availability of a 4Mbit, spin-based, nonvolatile memory called magnetoresistive RAM. MRAM uses magnetic materials combined with conventional silicon circuitry to deliver the speed of static RAM with the nonvolatility of flash RAM. One use of such memory could be in computers that boot up instantly.
Near-term advancements in spintronics should result in faster and denser MRAM, says Wang. WIN is also working on harnessing electron spins in logic circuits, which could use the nonvolatile property to retain their state.
In fact, a single spintronic logic device may include a memory state within it, unlike conventional devices that have combinations of transistors and capacitors for logic and memory. The tight integration of memory and logic may greatly speed applications that require frequent and fast memory access, such as image processing, Wang says.
WIN has been funded for four years, with $14.4 million from chip makers, plus a separate $10 million equipment grant from Intel and $3.8 million from the University of California's Industry-University Cooperative Research Program. During that four-year period, WIN is to work in three broad areas: spin logic devices, such as magnetic quantum dots; spin circuits, which involves looking at how to communicate between spin devices and between spin devices and conventional circuits; and benchmarking and metrics. Reflecting the desires of its industry sponsors, WIN is expected to come up with prototype devices that have the potential to become commercial products. - - - PB - - -
In parallel with this applied research, some WIN scientists are doing more theoretical work that could have longer-term payoffs. For example, they are trying to find out how another spin phenomenon -- that of the atomic nucleus -- might be harnessed to make subatomic-scale memory and logic units. One notion is to use the spin of the electron as a "bus" to convey data to the nucleus, where it would be acted on or preserved by nuclear spin.
David Awschalom, a physics professor at UC Santa Barbara, says nuclear spins are long-lived, measured in days or weeks, depending on temperature. He says the angle and magnitude of a single spin can represent any of millions of storage states, and nuclear spin may someday be harnessed to make nonvolatile memories 100 million times smaller than conventional memories.
Awschalom says electron spins in semiconductors are "a perfect platform" for building a simple quantum computer, something that is far closer to reality today than could have been imagined just five years ago. And, Awschalom says, because spintronic circuits can easily be made to convert back and forth between electrons and photons, they offer a way to tightly integrate the two kinds of technologies. He says it's possible to imagine systems consisting of spintronic-based logic and memory units communicating via light at extremely high speeds and with low power consumption.
Awschalom says it's hard to guess where work at the frontier of physics may lead: "It's not clear what you could do with these really new types of phenomena. This is all so new, and the discoveries are happening so quickly."
Stanford University electrical engineering professor Jim Harris says the relatively short-term, practical orientation of much of WIN's work in spintronics is crucially important. "I always worry that if you are working on something that could be 20 years off, people lose their stamina and their funding," he says.
But it would be a mistake to see Harris as a short-term thinker. "My main contribution is maybe recognizing that this is an important thing to do and learning at least enough to help the next generation or two of students behind me," he says. "They are going to be the ones that really make the discoveries and make this happen. A large part of what we are looking at, nobody has ever looked at before."