Spinning Electrons to Attain Low-Power Computing

Humble rust may provide the key to a new generation of low-power computing devices and memories. Its properties could overcome some of the problems of today’s attempts to harness magnetic fields for spintronic systems.

Electron spin, the source of magnetic fields, could prove a far more energy-efficient way to process data than today’s charge-based circuits currently dominated by silicon technology. Aside from the read-write heads in disk drives, however, progress in spintronics has been slow. This is partly because implementation has relied on inherently ferromagnetic materials that present serious drawbacks outside applications like those read-write heads.

Ferromagnets can generate strong magnetic fields that are easy to detect, but they are highly susceptible to external fields. They have also needed large electric currents to create and move magnetic domains around a circuit. These drawbacks have limited devices based on magnetic fields to niche applications. While they are relatively expensive and lack the density of flash-based designs, the magnetic memories used in satellites and space probes are less affected by cosmic rays than charge-based devices are.

Antiferromagnets, like the hematite in rust, may fare better. The neighboring spin centers in antiferromagnets align to cancel out any net field, which means external magnetic fields have little effect on them. That behavior made them seem like far less-promising avenues for hosting spintronics in the past than their ferromagnetic counterparts. Scientists in the field point to Louis Néel’s 1970 Nobel Prize speech, in which he described the target of his award-winning work as “interesting, but useless.” Yet attitudes have changed, largely because physicists’ ability to probe these materials has improved dramatically.

Despite the tendency to try to cancel out any net field, antiferromagnets can host complex microscopic eddies. Those same eddies appear in ferromagnetic materials, but they suffer the same kinds of side-effects seen with today’s magnetic memories, such as being thrown off course by external magnetic fields. Antiferromagnets have far better resistance to these fields.

Paolo Radaelli, professor of physics at the U.K.’s University of Oxford, said, “Antiferromagnets have much lower energy dissipation and have much, much faster dynamics, which is ideal for computing applications.”

The work to identify useful properties in antiferromagnets by Radaelli’s group focuses on hematite and similar metal oxides. Other researchers, such as Mathias Kläui, professor of physics at Germany’s University of Mainz, have focused more on synthetic antiferromagnets, created by making sandwiches of ferromagnetic alloys that mutually align their fields to cancel out.

The vortices that form in these materials are surprisingly stable, partly because the underlying lattice structure of the crystal controls how they form. In suitable crystals, the spins of the electrons do not align with a single plane. They need to align at subtly different angles to minimize the lattice’s overall energy. The web of interactions leads to a spiraling magnetic field that has a definite chirality or handedness. To flip the vortex in the opposite direction or destroy it entirely takes a large amount of energy. As a result, they behave more like a single particle, earning the name “skyrmion” after physicist Tony Skyrme, who developed theories about these kinds of dynamics.

Identifying these quasiparticles and determining if their behavior could work in computer processors and memories is painstaking work. It is only recently that scientists have been able to get detailed information on their structure. The primary tool used by Radaelli’s team to study these vortices has been the X-ray synchrotron at the Diamond Light Source (https://www.diamond.ac.uk/) 15 miles south of Oxford. Though open to all researchers, time on the ring is precious. The experiments demand repeated measurements at different polarization angles to find signs of the changing field directions that skyrmions and their variants possess. Even then, these measurements do not give up vital information on the chirality.

To get that extra data, the Oxford team turned to a table-top instrument based on one built by University of Cambridge scientist Anthony Tan for his Ph.D., along with fellow students Lucio Stefan and Michael Högen. The instrument applies a concept proposed by two researchers at Los Alamos National Laboratory 20 years ago that is beginning to become a feature of not just research into magnetometry, but also quantum computing and other areas.

The technique uses atomic-level defects in diamond fragments to measure the spin of electrons a fraction of a nanometer from the tip of a diamond inserted into an atomic-force microscope. Part of the key to the instrument’s performance is the nature of diamond itself and its highly symmetrical tetrahedral crystal structure. The inclusion of impurities such as nitrogen and vacancies disrupt this symmetry. A nitrogen and a vacancy lying next to each other results in one electron in this center being able to take on a spin that unconnected with the rest of the crystal. It also has a set of characteristic energy levels reflected in the wavelength of visible light it emits when excited by a laser.

The spin configuration of the surrounding diamond helps protect this induced spin, which in turn leads to long coherence times. That proves useful not just for these microscopes, but potentially for the quantum networks that are now being used to transfer data with near-unbreakable encryption. Several years ago, a group led by Ronald Hanson at the Technical University of Delft in the Netherlands entangled states across three of these diamond-based elements in a network 30m across.

The Cambridge instrument provided insight into the ability of hematite to host a kind of eddy that may be more useful, but has proven harder to identify with traditional instruments. Whereas skyrmions tend form tall tubes, much like water spinning and descending through a pipe, merons confine themselves to a flat plane.

“Merons are in-plane analogs of skyrmions. We expect skyrmions to be easier to identify in our images because the background magnetization would be out-of-plane,” Tan said.

In a skyrmion, the way electron spins tilt as the field rotates around the center of the vortex leads to an effect akin to the Magnus force that soccer players apply when they want to curl the ball into the back of the net. When a current is applied along a conductor, this variant of the Hall effect causes them to move gradually to one side, where they often self-destruct. Merons are natively less prone to this effect. The protection comes at the expense of some stability, but Kläui’s team has found carefully selected antiferromagnetic sandwiches will stabilize merons.

Researchers are keen to see merons and skyrmions exploited in future logic devices and memories that support similar operations to today’s transistors. But as behavior is intimately tied to crystal structure, defects that inevitably find their way into the lattices also have an effect.

In principle, the energy needed to move a skyrmion using an electric current is five orders of magnitude lower than for the magnetic domain walls of the kind used in spin-torque memories available. Radaelli says he hopes to see the merons that appear in hematite used in future memories that use a “racetrack” design, replacing the more power-hungry magnetic domain walls that have been used in experiments on these structures so far. These serially accessed memories could be wrapped into dense 3D structures to deliver low-power, low-cost devices that behave similarly to today’s solid-state disks. They work by using electricity to push data encoded by merons along a ribbon of magnetic material at high speed past a fixed sensor.

As with today’s spintronic devices, the energy needed to manipulate merons and skyrmions may in practice be far higher than predicted by theory. Crystal defects and subtle properties that arise as the lattice distorts can lead to the skyrmions being pinned in place. Energy needs to be supplied to dislodge them.

Some theoretical designs for skyrmion logic circuits aim to exploit this sensitivity. Circuit designers would draw features on the surface of the device to trap the eddies at junctions and only allow them to pass under the right conditions. That behavior makes it possible to synchronize two of the quasiparticles before letting them either annihilate or combine, and in doing so perform logic operations. But it is not clear how efficient or reliable these processes will be. They are more complex to organize than conventional logic circuits.

Much more detailed data needs to be collected on their behavior to work out how devices might operate in practice. And NV-center magnetometry is not a complete answer for determining how these eddies behave under different conditions, and how they might deliver new devices.

“The technique is mostly limited to room temperature and static imaging,” Kläui said. So, the group is using both X-ray and scanning electron microscopy to study dynamic behavior over a wider temperature range.

To avoid the complexity of trying to manage individual merons or skyrmions, future circuit designers may turn to unconventional techniques that rely on statistical properties. Kläui, Radaelli, and others see reservoir computing as a potential target for this kind of application.

The reservoir is a random network of nodes that evolves over time in response to inputs. The reservoir’s big advantage is that it can take complex low-dimensional data such as speech, and produce an output in a high-dimensional space, akin to the data embeddings used by large language models to encode words and phrases. Downstream neural networks can find it easier to train on large numbers of simpler structures mapped over many dimensions than the original signal.

The problem, as with other neuromorphic architectures that may suit skyrmion-based devices, is the lack of maturity of reservoir computing itself. Kläui pointed out the further issue that there is stiff competition coming from other technologies, such as resistive memories and photonic crystals, even if the technique gains a foothold. “We are trying to explore an implementation where magnetism is the natural solution,” he said.

A long road remains ahead before these applications of magnetic spin can get a starting position on the commercial computing racetrack.