'Racetrack' memory demonstrated
Apr 11, 2008 2 comments
Physicists in the US have demonstrated for the first time that data can accessed from forest-like arrays of 3D nanowires or “racetracks”. The demonstration indicates that so-called racetrack memory, which should be faster and cheaper than other forms of information storage, is on its way to becoming a commercial reality.
Stuart Parkin and colleagues at IBM’s Almaden research centre in San Jose, US first came up with the concept of racetrack memory back in 2004. The idea is that arrays of U-shaped nanowire racetracks are planted on a bed of silicon wafer. Along the length of each racetrack are domains that are magnetized in one of two directions, with each domain boundary or “wall” acting as a single bit — either a 1 or 0 — of information. By sending a current of spin-polarized electrons into one end of a racetrack, these domain walls can be shifted to and fro.
In theory, both reading and writing information should be possible at the base — in other words, the curve of the “U” — on the silicon wafer. For example, a read head could sense the configuration of a domain wall by measuring how the wall’s resistance changes in the magnetic field (a property known as magnetoresistance). A write head could consist of a perpendicular nanowire, which would switch the configuration of the domain walls as its own domains are shifted.
Putting this theory into practice has been tricky, but Parkin’s team have now shown that it is possible, at least when the racetracks are positioned flat on the silicon wafer rather than upright. They have been able to write and read three bits in a matter of nanoseconds (Science 320 209).
“It looks like very good progress,” says Yongbing Xu at the Univeristy of York, UK. “In principle there are no moving parts, and that’s a great advantage.” Xu is also impressed that the reading and writing can be perfomed so quickly — at a conference last year the IBM team suggested it would take microseconds. Still, he notes that manipulating many well defined domain walls will be challenging, which might mean a commercial prototype is three to five years away.
When racetrack memory does hit the shelves, it seems likely that it will quickly supplant existing memory types. Unlike random-access memory (RAM), which is based on 2D arrays of transistors, it should be cheap to mass produce. And unlike hard disk storage, which contains magnetic bits spread over a rotating metal disk, it should be fast.
“It's certainly a clever idea and has been executed very elegantly,” says Mark Blamire at the University of Cambridge, UK. “However, the substantial gap between this and any sort of commercialization is whether an appropriate technology can be developed to turn what is a linear device occupying a large amount of wafer area into a vertical structure, which occupies much less. The paper doesn't address that question but it would definitely be extremely challenging.”
About the author
Jon Cartwright is a reporter for physicsworld.com