GUIDO MEIER’S movie is unlikely to win any Oscars. Besides being less than 1 second long, it doesn’t look like much. Picture four triangular regions – each a slightly different shade of grey – sitting inside a grey square and swirling around in a vortex-like motion as if being stirred by an invisible spoon.
, Germany, has captured the lightning-fast movement of regions of magnetisation called domains – the phenomenon at the heart of a new breed of computer memory.
Meier captured “domain motion” in a minuscule square piece of metal, but the most promising aspect of this nascent technology is called “racetrack memory”. RM relies on the fact that magnetic domains – which store binary bits as magnetic fields pointing in one of two directions – can be made to “race” around a nanowire “track” in response to a current. Here’s the idea: first, the domains race in one direction past a magnetic memory writer – allowing data to be written – and afterwards a reversal in current makes the domains race back in the opposite direction, past a reader – allowing the bits to be read off.
Advertisement
Since the wires can be made of cheap magnetic materials and the data can be written and read very quickly, RM could replace hard-disc drives (HDDs), which are used for storage in PCs but are too slow for many applications. RM could even replace random access memory (RAM) in PCs, which is much faster than HDDs but much more expensive. The result may be smaller, cheaper PCs, cellphones and digital cameras.
“The movie shows that switching domains between states can be accomplished in picoseconds – that’s three orders of magnitude faster than state-of-the-art memories,” says Meier. Memories based on domain flow could also replace flash memory, which is used in digital cameras, USB memory sticks and MP3 players. Although flash reads data quickly, it can be slow to write and is prone to wear out.
“’Racetrack memory’ is orders of magnitude faster than state-of-the-art PC memory”
Despite these potential advantages, until now no one had built a working RM. For example, while Meier has coaxed domains to move, he has not used them to actually store and read out bits. Now Stuart Parkin at the in San Jose, California, and colleagues have written and read a short string of bits using RM. It’s a far cry from being put into PCs, but it’s a big step forward (Science, ).
The RM concept, which , borrows some ideas from HDD technology, which also exploits magnetism. HDDs consist of a rotating magnetic disc. To write a bit of data, a read/write head moves across its surface, inducing magnetic fields representing a “1” or a “0” on tiny patches of the disc. As each region of the spinning disc is filled with bits, the head moves to a fresh section. To read back the data, the same process happens, except this time the head reads the magnetisation of the patches on the disc’s surface, rather than inducing new ones. Waiting for the head and disc to finish moving is what makes HDDs relatively slow.
The time-saving idea behind RM is to write the magnetic domains to a wire instead of a disc and to move the written domains along the wire using current, which is potentially much faster than moving a read/write head. This way, the head could remain fixed in place and the disc disposed of.
To move domains through a wire, Parkin turned to a discovery made in the 1930s by Nobel physics laureate at the University of Cambridge. Electrons have a quantum property called “spin”. Mott found that when an electric current is injected into a magnetic material, the electrons orient their spin in a direction parallel to the magnetisation of a domain. As a result, when electrons move from one domain to a neighbouring domain that is magnetised in the opposite “direction”, the electrons’ spin flips. This flip creates a torque that imprints the direction of magnetisation of the first domain onto the next domain, effectively shunting the first domain over. Parkin reasoned that if he managed to take a wire and place patterns of domains pointing left and right onto it – to represent 1s and 0s – a current could make the patterns flow along the wire. One carefully timed pulse would nudge all the domains over by one unit, making space next to the read/write head for a new domain to be written. The process could then be repeated until the wire was full of bits.
Last year Parkin’s team that domains could be moved at high speeds, by flowing them through wires at around 110 metres per second (Physical Review Letters, ). But they did not use the domains to store a string of digital bits.
Now Parkin’s team has used “domain flow” to write, store and then read bits on a nanoscale data wire. The write head they used consists of a conducting copper wire placed close to the nickel/iron data wire and at right angles to it (see Diagram). When they pulsed current through the copper wire, it induced a magnetic field in a short section of the data wire, creating a magnetic domain. Next, a pulse of current through the data wire shunted the domain to one side, clearing space to write the next bit, and so on – all without moving the copper write head. The team managed to write a short sequence of 1s and 0s at a rate of about 30 megabits per second.
To read the data, they flowed pulses of current down the data wire in the opposite direction, shunting the domains back the other way. A magnetic-memory reader placed next to the copper writing head was able to read off the domains as they flowed past. The team found the optimum width for their data wire to be 100 nanometres, with current pulses lasting 1 nanosecond.
Before RM can replace HDDs and RAM in computers, however, there are several hurdles to overcome. One problem, which Parkin and other researchers are tackling, is preventing heat or stray magnetic fields from disrupting the domains. One way of doing this is by using “pinning sites” – notches carved along the edge of the data wire that encourage a domain to stay put until it is supposed to move.
Another challenge is making 3D arrays of the wires. Parkin’s demonstration was on flat nanowires, but to make the memory more compact, he would like to squash millions onto the surface of a silicon chip. “We need intelligent ways of using the third dimension so we can invent much more powerful storage and computing devices,” he says. To build into the third dimension, Parkin will make the wires U-shaped so that they resemble tall thin columns that could be placed next to each other on the surface of a silicon chip, in a “forest” of wires. “The fabrication of vertical racetrack nanowires will be a challenge,” he admits.
