IBM's Faster, Denser Memory
Researchers led by Stuart Parkin are developing a new type of memory chip that combines the benefits of magnetic hard drives and flash.
Researchers at IBM have demonstrated the feasibility of an entirely new class of data storage, called racetrack memory, which promises to combine the data storage of a magnetic hard disk with the ruggedness and speed of Flash memory, at relatively low cost. In addition, racetrack memory wouldn’t degrade over time as Flash does. While still in the early days of research, these benefits could make racetrack memory an attractive replacement for both hard disks and Flash memory, leading to ever smaller computers and extremely inexpensive memory for iPods and other portable devices that now rely on Flash.
In this week’s issue of Science, the team, led by Stuart Parkin, a physicist at IBM’s Almaden Research Center in San Jose, CA, described a way to read and write multiple bits of data to magnetic nanowires, an important step toward making a prototype. Previous work by the group illustrated that the fundamental concept of racetrack memory was feasible, but the researchers hadn’t yet demonstrated the manipulation of multiple bits. “It’s a milestone in developing a prototype,” says Parkin.
Racetrack memory consists of an array of billions of nanowires on silicon; each nanowire is able to hold hundreds of bits of data. Because the nanowires are so small, racetrack memory has the potential to be many times more dense than Flash. Unlike Flash memory, in which bits are stored as electrical charges in a transistor, racetrack memory stores data as a series of distinct magnetic fields along the wire. Flash memory degrades over time as charges leak and memory cells wear out, but racetrack memory, which uses magnetic fields, doesn’t have this problem. And compared to the hard disks used in laptops and PCs, which store data on a bulky, spinning platter, racetrack memory has no moving parts and can be built in silicon, making it more robust.
Data is encoded onto racetrack memory by changing the magnetic properties along the wire, creating a series of magnetic barriers–called domain walls–and gaps between. Just as electrical charge represents a bit in a Flash memory cell, the gaps between two domain walls represent bits in racetrack memory. To read and write data from the nanowire, the domain walls move along the tracks, single file, past where stationary read and write heads are positioned.
That is, at least in theory, how it would work. But before the current research, no one had shown that multiple domain walls–essentially, data–could move along a nanowire without being destroyed. In order to move the domain wall down the nanowire, Parkin uses principles from spintronics, which takes advantage of the quantum mechanical property of electrons, called spin. He injects a small electrical current into the nanowire. As a result, the electrons in the current become “polarized,” so that their spins are uniformly oriented, and when they contact a domain wall, they transfer the orientation of their spin to the atoms in the wall. This hand-off changes the magnetic moment of the atoms in the domain wall, shifting it forward on the racetrack, and likewise shifts all the domain walls on the racetrack forward, explains Parkin.
“This is the first time that someone has demonstrated that you can move two or three of these domain walls without upsetting them or causing them to interfere,” Parkin says. Parkin notes that it could take four years before he has a racetrack memory prototype, and three more years to commercialize it.
The appeal of racetrack memory, says Igor Zutic, professor of physics at the State University of New York at Buffalo, is that it can “unify the best properties of inexpensive, high-density storage of magnetic hard drives with high-speed operation of random-access memory in a single device, while avoiding their main shortfalls, such as speed and cost, respectively.”
The next step, Parkin says, is to implement a device to read the bits of data. He suspects that this will be fairly straightforward, because he could use pre-existing technology. In 2004, Parkin developed the small magnetic device that reads data from magnetic disk drives, and these devices, called magnetic tunnel junctions, would be sensitive enough to read the tiny magnetic fields produced by the domain walls in the nanowires.