For 15 years around the turn of the 21st century, the speed and capacity of computer memory increased at a stunning rate. Between 1990 and 2005, the areal bit density — the amount of information that could be stored per square inch of disk space — of magnetic drive systems grew by an average of 60% each year. By the 2010s, however, growth had leveled out. An increasing demand for storage capacity, along with the miniaturization and power-efficiency needs of portable devices like smartphones and laptops, made solid-state “flash” memory the dominant medium for digital storage.

Flash memory is fast and compact, but it’s also expensive. For decades, computer scientists have imagined another kind of memory based on magnetic spin at the quantum level — a new system of digital storage that’s as fast as flash memory but as cheap as magnetic disks, with storage capacities 100 times larger.

It’s called racetrack memory, and scientists at IBM’s Almaden Lab in San Jose have achieved a series of breakthroughs that could one day make it a reality.

In September 2004, IBM researchers at Almaden announced a major leap forward in quantum physics: They had measured the energy required to reverse the magnetic orientation of a single atom. The discovery tantalized the field of spintronics.

Spintronics is the term researchers use for theoretical computing and storage systems that exploit the quantum property known as “spin,” the magnetic orientation of an electron or other subatomic particle. All such particles have one of two spins: up or down. These two values can provide the basis for a binary system that computers use to store digital information — a system of subatomic ones and zeroes that would be millions of times more efficient than the memory and processors we use today. The possibility is thrilling to computer scientists, who generally agree that the power of conventional systems faces built-in limits.

“Some time in the next couple of decades, it will be impossibly difficult to continue improving transistors and other traditional microelectronic circuit elements by simply shrinking them,” Gian-Luca Bona, manager of science and technology at Almaden, observed in 2006. “We will then need alternative structures and, perhaps, altogether different ways of computing.”

Traditional computing systems keep track of where data is stored on a given memory device and then access that location — in the case of magnetic storage devices, by manipulating the storage medium to position the desired location under a reader. That makes the speed of the moving disk or tape a limiting factor in how fast the information can be accessed. Flash memory does away with this problem by creating a faster, solid-state drive — but it’s far more expensive than magnetic storage.

Racetrack memory promises the best of both worlds: a storage medium that’s as fast as flash but as inexpensive as magnetic, with theoretical storage capacities that outstrip both. Racetrack memory works by moving not the disk but rather the data itself. Bits are stored in the subatomic “domains” between areas of magnetic charge on a nanowire 1/100,000th the width of a human hair. These domains are then moved along the wire like cars on a racetrack, positioning them under the detector at speeds 1 million times faster than magnetic disks.

The trick is to move them very precisely, given that the domains in question can be as small as a single atom. Reliably manipulating domains was prohibitively complex and expensive until 2010, when IBM scientists became the first in the world to measure acceleration and deceleration of domains at different voltages. Along with the writing and reading techniques developed earlier at Almaden, this breakthrough in domain movement provided the third component theoretically necessary to make a working racetrack memory device.

Engineers have yet to develop a prototype for consumer use, but Stuart Parkin, who began researching racetrack memory at Almaden in 2004, believes that such devices will eventually replace solid-state drives in the same way SSDs replaced magnetic hard disks.

“One can have a solid-state memory with the same low cost of a disk drive but with a performance 10 million times better,” Parkin said. “It could replace disk drives; it could replace flash; it could replace most solid-state memories. And it would enable much simpler computers in the future.”

In 2021, a team of New York University researchers led by Professor Andrew Kent announced another step toward the future Parkin imagined: They discovered that materials called ferrimagnets, which produce smaller magnetic fields than conventional materials like iron and nickel, worked better to create the stable magnetic objects on which racetrack memory relies. This discovery represents another breakthrough in the ongoing project of bringing racetrack memory out of the lab and into consumer markets.