What will replace today's hard drives and flash memory devices? The former tend to be slow, the latter unsuited to long-term use. Then there is RAM, many forms of which are volatile—turn off the power, lose your data. These shortcomings, as well as the same demands that drive most technological innovation—to make it smaller, faster, cheaper, less prone to failure—have produced a number of candidate data-storage technologies in recent years, all of which offer some combination of advantages over the devices in wide use today.

But those advantages come at a price: the new technologies usually involve added complexity or at least underexplored physics. That latter characterization is certainly true of racetrack memory, a proposed scheme in which data bits, encoded as magnetized regions on nanowires, move back and forth along the nanowire "racetrack" and past read/write heads. In a standard hard drive, the head moves over the data like a turntable's stylus over a record, selecting the part of the disk it needs to access. In racetrack memory, the head stays put, and the data comes to it. One attractive feature of the technology is that the racetracks could be arrayed in three-dimensional forests, leading to vastly greater densities of data storage than in current two-dimensional schemes. But there are several hurdles that must be cleared before racetrack memory can elbow its way into use, not the least of which is understanding just how such a system would work on a basic physical level.

Now a leading racetrack memory group reports a finding in the December 24 issue of Science that advances that understanding and makes racetrack memory look a bit more feasible. Physicist Stuart Parkin and his colleagues at the IBM Almaden Research Center in San Jose, Calif., set out to determine just how magnetized regions move along nanowires when driven by electric current. They found that the magnetized bits exhibit inertia—that is, they do not get up to speed instantaneously, nor do they stop the instant the current is removed. But, conveniently enough, the acceleration and deceleration phases of the bits' motions work together in a predictable, controllable way.

In a nanowire racetrack, what moves is not the wire itself but discrete packets of magnetization within the wire. Each magnetized packet constitutes a bit, and so moving bits along actually constitutes moving the boundaries, known as domain walls, that separate regions of opposing magnetization—essentially, the wall that separates a 0 from a 1. An electric current can do the trick; as an electron crosses a domain wall and feels its own magnetic pointing, or spin, flipped from one orientation to the other, it forces an atom within the nanowire to flip magnetic orientations as well to compensate. The collective flipping of large numbers of metallic nanowire atoms, forced by a large number of electrons in the electric current, moves the domain wall—and the data bits it separates—along.

"The new rung in the ladder is that we've done some experiments to reveal exactly how the domain wall responds when you try to manipulate it with spin-polarized current," Parkin says. The researchers traced the movement of domain walls in iron–nickel nanowires 20 nanometers thick and 200 nanometers wide, finding that it took roughly 10 nanoseconds to accelerate a domain wall to its terminal velocity of 138 meters per second (about 500 kilometers per hour). Deceleration, or coasting to a stop, took about the same time.

Some previous studies had found that the distance covered by a domain wall was directly proportional to the length of the electric pulse used to drive it, hinting that perhaps domain walls move without inertia—that is, they move at top speed or not at all. Parkin's group found that the reason for the tidy relation between pulse length and distance traveled was not that the domain walls had no inertia but simply that the acceleration and coasting phases of their motion balance out. "What we found, and what surprised us, is that the distance it's lagging is exactly made up for by the distance it moves under its own inertia," Parkin says. "Because of these two competing effects, it looks like the domain wall has no inertia—or no mass, if you like."

The significance of the finding is that a racetrack memory device could control the flow of its bits simply by varying the length of the electric pulse used to push the bits along. That is certainly convenient, but more importantly the research helps solidify researchers' understanding of a key physical aspect of the technology, bringing its implementation a step closer to reality.

Parkin says that the new research, along with other fundamental steps in the past few years, casts racetrack memory in a promising light. "We believe that we now understand pretty much how domain walls can be moved by current, and that the basic underlying physics for racetrack works," he says. "The main hurdles are really to do with integration: how to build lots and lots of these racetracks and integrate them with read/write devices."

Whether or not racetrack memory ever finds its way into handheld gadgets and desktop computers, it seems likely that some new technology will come along to supplant the memory devices now in use. "Eventually, and it will probably be sooner rather than later," Parkin says, "these existing technologies will reach their fundamental hurdles that will be very hard to overcome."