Bend the signal and it dies—until now.
As the appetite of artificial intelligence and data centers for electricity grows into a civilizational concern, a team from Tohoku University, Shin-Etsu Chemical, and EPFL has quietly solved one of the most stubborn obstacles in low-energy computing: how to bend a spin wave without destroying it. By laying a copper film perforated with a hexagonal lattice of holes atop a magnetic garnet, the researchers created a structure that guides spin waves around sharp corners with 5,000 times greater efficiency than conventional designs. The discovery does not merely improve an existing technology — it removes the barrier that has kept spin wave computing theoretical, and points toward a future in which data centers consume only a fraction of the electricity they do today.
- Data centers and AI systems are consuming electricity at a pace that conventional electronics, with their heat-generating electron flows, cannot sustainably support — the pressure for an alternative has become acute.
- Spin waves have long promised a cooler, more efficient path for computing, but a fatal flaw persisted: the moment a signal was asked to turn a corner, it collapsed, making practical circuits impossible to build.
- Researchers inverted their own 2024 design — placing a hole-perforated copper film atop magnetic garnet rather than carving into the substrate — and simulations revealed it produced a complete magnonic bandgap, reflecting spin waves from every direction with minimal loss.
- A Z-shaped test path carved through the crystal by removing a line of holes transmitted spin waves 5,000 times more powerfully than conventional waveguides, which failed to deliver any signal at all.
- A patent has been filed for the waveguide structure, and the team is now moving toward building physical devices to confirm whether the real world matches what the models so dramatically predicted.
For years, spin wave computing has carried a quiet promise: ripples of magnetization moving through magnetic materials generate almost no heat, offering a potential escape from the thermal crisis that shadows modern data centers and AI infrastructure. The obstacle was equally quiet but absolute — bend the path of a spin wave around a corner, and the signal dies. No corner, no circuit. No circuit, no practical technology.
A research team spanning Tohoku University, Shin-Etsu Chemical, and EPFL has now broken that barrier. Their approach came from reversing a concept they had explored in 2024. Instead of carving grooves into a magnetic garnet substrate, they placed a copper film on top of it — perforated with a hexagonal array of holes connected by thin slits. Three-dimensional electromagnetic simulations showed that this geometry produces a complete magnonic bandgap: a range of frequencies at which spin waves are reflected regardless of the direction they arrive from. It was the first demonstration of such a bandgap in a two-dimensional magnonic crystal on magnetic garnet.
To test the design, the team created a Z-shaped channel through the crystal by removing a line of holes — a line defect. Spin waves sent through this path arrived at the other end. The conventional ridge waveguide, run as a comparison, produced nothing. The new structure outperformed it by a factor of 5,000 — not an incremental gain, but a categorical difference in what the physics allows.
A patent application for the waveguide structure has already been filed, signaling the team's confidence that the design is ready to leave simulation behind. The researchers describe their work as opening a practical route toward integrated spin wave circuits — devices that could allow data centers to operate on a fraction of their current electricity draw. The next test is whether physical devices confirm what the models predicted, and whether a laboratory breakthrough can become the architecture of a cooler, quieter computing future.
The problem that has stalled spin wave computing for years is deceptively simple to state: bend the signal and it dies. A team of researchers from Tohoku University, Shin-Etsu Chemical, and EPFL has now found a way around it—and the improvement is staggering. By routing spin waves through a carefully engineered metal film perforated with holes, they achieved transmission efficiency 5,000 times greater than conventional waveguides, even when the path bent sharply.
Spin waves are ripples of magnetization that travel through magnetic materials. Unlike electrons, which generate heat as they move through circuits, spin waves carry information with far less thermal waste. As data centers and artificial intelligence systems consume ever-larger amounts of electricity, and the heat from conventional electronics becomes an increasingly serious problem, spin waves have emerged as a promising alternative for low-energy computing. The catch: they weaken rapidly as they travel, and the moment you try to bend their path around a corner, the signal collapses. This signal loss has been the fundamental barrier preventing anyone from building practical circuits based on spin waves.
Associate Professor Taichi Goto from Tohoku University's Research Institute of Electrical Communication described the challenge plainly: bending a spin wave without losing it has been one of the hardest problems in the field. The team's solution came from inverting an earlier concept they had developed in 2024. Rather than cutting grooves into the magnetic garnet substrate itself, they placed a copper film on top of it—but not a solid film. Instead, they perforated it with a hexagonal array of holes, connected by thin slits running between neighboring holes.
Three-dimensional electromagnetic simulations revealed what this structure could do. The perforated pattern creates what physicists call a complete magnonic bandgap—a range of frequencies at which spin waves are reflected regardless of which direction they approach from. This was the first time researchers had demonstrated such a complete bandgap in a two-dimensional magnonic crystal built on magnetic garnet. The implications were immediate enough that a patent application for the core waveguide structure was filed.
To test the concept, the team created a Z-shaped path through the crystal by simply removing a line of holes, forming what they called a line defect. Spin waves sent through this path behaved entirely differently from those in conventional ridge waveguides. The conventional design failed—the signal did not reach the end. The new design succeeded, transmitting spin waves over 5,000 times more powerfully than the conventional approach. The numbers suggest not an incremental improvement but a fundamental shift in what becomes possible.
What matters now is whether this laboratory result can scale into actual circuits. The researchers framed their work as opening a practical route toward integrated spin wave circuits that could one day allow data centers to run on a fraction of today's electricity consumption. That is not hyperbole born from enthusiasm—it is the logical endpoint of a technology that loses almost no energy to heat. The patent filing suggests the team believes the design is ready to move beyond simulation. The next phase will be building and testing actual devices, and watching whether the real world matches what the models predicted.
Citas Notables
Bending a spin wave without losing it has been one of the hardest problems in this field. By turning the problem inside out—placing a patterned metal film on the magnetic garnet instead of cutting the garnet itself—we found a way to guide spin waves around sharp corners with very little loss.— Associate Professor Taichi Goto, Tohoku University
La Conversación del Hearth Otra perspectiva de la historia
Why does bending a spin wave break it so badly? What's happening physically?
Spin waves are ripples of magnetization traveling through a material. When you try to bend their path, the wave energy scatters outward instead of staying confined. It's like trying to guide water through a sharp corner in a pipe—most of it splashes out. The conventional approach was to carve grooves into the garnet itself, but that only made things worse.
And the new design solves this by putting holes in copper instead of cutting the garnet?
Exactly. The perforated copper film creates a magnonic bandgap—a kind of invisible wall that reflects spin waves back into the path you want them to follow, no matter which direction they're coming from. It's like building a fence that only lets certain frequencies through.
A 5,000-fold improvement is extraordinary. Is that the kind of number that actually translates to real-world applications?
It should. If spin waves can now travel around corners without collapsing, you can build circuits instead of just straight lines. That's the difference between a prototype and something you can actually use in a data center.
What happens next? Is this ready to move into production?
They've filed a patent, which suggests they think it's ready. But there's always a gap between simulation and reality. The next step is building actual devices and seeing if they perform as the models predicted. If they do, this could genuinely change how we think about computing efficiency.