This article is more than 1 year old

Scientists make spin ice breakthrough

Artificial spin ice with smallest features ever created could be part of novel low-power HPC

Researchers at the Paul Scherrer Institute and ETH Zurich in Switzerland have managed to accomplish a technological breakthrough that could lead to new forms of low-energy supercomputing.

It's based around something called artificial spin ice: think of water molecules freezing into a crystalline lattice of ice, and then replace the water with nanoscale magnets. The key to building a good spin ice is getting the magnetic particles so small that they can only be polarized, or "spun," by dropping them below a certain temperature. 

When those magnets are frozen, they align into a lattice shape, just like water ice, but with the added potential of being rearranged into a near infinity of magnetic combinations. Here the use cases begin to emerge, and a couple breakthroughs from this experiment could move us in the right direction.

The discovery, made by PSI physicist Kevin Hofhuis and PSI researchers/ETH Zurich professors Laura Heyderman and Peter Derlet, could pave the way for low-energy HPC applications with additional potential uses in reservoir computing, which involves using fixed linear systems of a higher dimension than the input for signal mapping. "There are many areas where reservoir computing can be applied including prediction of weather and financial markets, image and speech recognition, and robotics," Hofhuis said. 

Heyderman even speculated that a high-speed, low-power spin ice supercomputer could be akin to the human brain: "The process is based on the information processing in the brain and takes advantage of how the artificial spin ice reacts to a stimulus such as a magnetic field or an electric current."

The science behind spin ice

To be clear, this lays groundwork, but spin ice supercomputers aren't in the immediate future. That hasn't stopped researchers from speculating on how spin ice and the manipulation of phase transitions can be used. 

"Mag­netic phase trans­itions had been the­or­et­ic­ally pre­dicted for ar­ti­fi­cial kagome spin ice, but they have never been ob­served be­fore," said Hey­derman, who has been researching and publishing on spin ices for over a decade. 

For this experiment, a nickel-iron compound called permalloy was spread onto a silicon substrate, which was then lithographed in a hexagonal pattern, each connected via tiny bridges, which were key in enabling them to tune and observe the phase transition. 

As Hofhuis explains, each magnet in the kagome form (one ring consists of six magnets) has two alignments, which means 64 potential magnetic states. Two rings increases that to more than 2,000 possible states, and so on. "There is an unimaginably huge number of magnetic states available in our large arrays, which have several thousands of nanomagnets," Hofhuis said. 

The experimental team made two big breakthroughs: It built nanoscale magnet bridges between the magnets, making their responses more predictable, and verified how the magnetic states of the nanomagnets in an array evolve over time. That latter discovery required a special microscope and an x-ray synchrotron, but let them see the actual phase transitions in the spin ice. 

Those bridges were only 10nm wide (a human hair is around 70,000nm), and the researchers were able to capture videos of the interaction of the nanomagnets, but were unable to do anything beyond deduce the configuration of the magnetic "spins" that occur in the moment of phase transition. 

Hofhuis said that he needed simulations designed by Derlet to prove what he was recording was a phase change. "Only the com­par­ison of the re­cor­ded im­ages with these sim­u­la­tions proved that the pro­cesses ob­served un­der the mi­cro­scope ac­tu­ally are phase trans­itions," Hofhuis said. 

At the end of the day, the researchers produced and measured artificial kagome spin ice that was made with small enough features to do what spin ice is supposed to: only form through temperature-induced magnetic phase transitions. Supercomputing with it will take a bit more time. ®

Editor's note: An earlier revision of this article stated the bridge was 10 microns thick, not 10nm wide, and Laura Heyderman's name was incorrectly spelled. We're grateful to Kevin Hofhuis and Laura Heyderman for their help in clarifying this.

More about


Send us news

Other stories you might like