IBM Makes Revolutionary Racetrack Memory Using Existing Tools
Racetrack memory could someday supersede flash in terms of density and cost.
IBM has shown that a revolutionary new type of computer memory—one that combines the large capacity of traditional hard disks with the speed and robustness of flash memory—can be made with standard chip-making tools.
The work is important because the cost and complexity of manufacturing fundamentally new computer components can often derail their development.
IBM researchers first described their vision for “racetrack” computer memory in 2008. Today, at the International Electronic Devices Meeting in Washington, D.C., they unveiled the first prototype that combines on one chip all the components racetrack memory needs to read, store, and write data. The chip was fabricated using standard semiconductor manufacturing tools.
Racetrack memory stores data on nanoscale metal wires. Bits of information—digital 1s and 0s—are represented by magnetic stripes in those nanowires, which are created by controlling the magnetic orientation of different parts of the wire.
Writing data involves inserting a new magnetic stripe into a nanowire by applying current to it; reading data involves moving the stripes along the nanowire past a device able to detect the boundaries between stripes.
Earlier demonstrations of the technology employed nanowires on a silicon wafer in a specialized research machine, with other components of the memory attached separately. “All the circuits were separate from the chip with the nanowires on,” says Stuart Parkin, who first conceived of racetrack memory and leads IBM’s research on the technology at its research lab in Almaden, California. “Now we’ve been able to make the first integrated version with everything on one piece of silicon.”
The new racetrack prototype was made at IBM’s labs in Yorktown, New York, using a manufacturing technique known as CMOS, which is widely used to make processors and various semiconductor components. This proves that it should be feasible to make racetrack memory commercially, says Parkin, although much refinement is still needed.
The nickel-iron nanowires at the heart of the prototype were made by depositing a complete layer of metal onto an area of the wafer, and then etching away material to leave the nanowires behind.
The wires are approximately 10 micrometers long, 150 nanometers wide, and 20 nanometers thick. One end of each nanowire is connected to circuits that deliver pulses of electrons with carefully controlled quantum-mechanical “spin” to write data into the nanowire as magnetic stripes. The other end of each nanowire has additional layers patterned on top that can read out data by detecting the boundaries between stripes when they move past.
Dafiné Ravelosona, an experimental physicist at the Institute of Fundamental Electronics in Orsay, France, leads a European collaboration working on its own version of racetrack memory. He says IBM’s latest results are a crucial step along the road to commercialization for the technology. “It’s a nice demonstration that shows it’s possible to make this kind of memory using CMOS,” he says.
However, Ravelosona adds that the IBM work doesn’t yet demonstrate all of the key components that make racetrack memory desirable. “They have only demonstrated that it is possible to move a single bit in each nanowire,” he explains.
Much of the promise of the technology lies in the potential to store many bits—using many magnetic stripes—in a single tiny nanowire, to achieve very dense data storage. Ravelosona suggests that the material used to make the nanowires in the new IBM device lacks the right magnetic properties to allow that.
Parkin says that the intention wasn’t to target density but adds, “We’re focusing on exactly this question.” His group is currently working on how to fit as many magnetic stripes as possible into a nanowire and has begun experiments that suggest that wires made from a different type of material may do better.
The nickel-iron alloy of the integrated prototype is what’s known as a soft magnetic material, because it can be easily magnetized and demagnetized by an external magnetic field. Parkin is also experimenting with hard magnetic materials, which get their magnetic properties from their tightly fixed crystalline structure and as a result are not easily demagnetized.
“Using this different material, we have discovered we can move the domain walls [between magnetic stripes] very fast and that they are much smaller and stronger than in the soft magnetic material used in the integrated devices,” says Parkin.
That means not only that it should be easier to put many stripes into one nanowire, but also that nanowires fabricated with less precision will still work, which should make fabrication easier. “I call this racetrack 2.0,” he says.
AI is here. Will you lead or follow?
Join us at EmTech Digital 2019.