A Magnetization Wave Survived 18 Microseconds Inside a Crystal. The Old Record Was 200 Nanoseconds. The Bottleneck Was Never Physics.

Two hundred nanoseconds. That was the wall. For more than a decade, every team working on magnon-based quantum computing hit the same ceiling: magnetization waves in solid crystals decayed too fast to do useful computation, lasting roughly 100 to 200 nanoseconds before thermal noise ate them alive. Physicists debated whether the limit was baked into nature itself, a consequence of the spin-wave interactions that define magnons at the quantum level, and several prominent review papers framed the problem in terms that implied exactly that.

It was dirty crystals. That is the punchline of a study published in Science Advances in May 2026 by a team led by Andrii Chumak at the University of Vienna, with doctoral candidate Rostyslav Serha as lead experimentalist and collaborators at the University of Colorado Colorado Springs and institutions in Germany, the United States, and Ukraine. They extended magnon lifetimes to 18 microseconds, which is 18,000 nanoseconds, by doing two things: exciting short-wavelength magnons that are inherently less sensitive to surface defects, and cooling ultra-pure yttrium iron garnet spheres to 30 millikelvin in a mixed-phase cryostat.

One hundred times longer, not by discovering new physics, but by buying cleaner glass.

What Magnons Are and Why They Were Stuck

Magnons are collective excitations in magnetic materials, analogous to sound waves but carried by the magnetic ordering of atoms rather than their mechanical vibrations. Picture a row of bar magnets glued to a table, each one tilted slightly from vertical. Push one and the disturbance ripples down the line, and that disturbance, that collective excitation of the magnetic lattice, is a magnon. In crystalline materials like yttrium iron garnet (YIG), these ripples propagate coherently across the lattice and can be manipulated with microwave fields, making them candidates for carrying quantum information.

YIG has been the magnon material of choice since the 1960s. Low magnetic damping, high Curie temperature, compatible with standard microwave electronics. But prior experiments using long-wavelength (uniform-mode) magnons suffered from surface scattering: any impurity, pit, or crystal lattice defect at the sphere's boundary hammered the magnon's lifetime. Short-wavelength magnons penetrate deeper into the crystal bulk, away from the surface, which is why Serha's team chose them. But nobody had cooled YIG spheres of varying purity to 30 mK and compared the results head-to-head until now.

Three spheres of different purity grades, and every single one beat the old record: purer crystal yielded longer lifetime in a clean linear relationship with no evidence of a fundamental floor. Chumak's conclusion, stated plainly in the paper: the path ahead is wide open.

The Comparison Nobody Ran

Coherence time is only half the story, because what matters for a quantum computer is how many gate operations you can execute before the qubit dies.

Platform	Typical Coherence	Best Demonstrated	Gate Time	Max Operations	Physical Scale
Superconducting transmon	~100 μs	~1 ms (Aalto, 2025)	~20 ns	5,000–10,000	cm-scale per qubit
Trapped ions	~1 s	Minutes (single qubit)	~1–100 μs	10,000–1,000,000	Table-top apparatus
NV centers (diamond)	~1 ms	Seconds (isotopically pure)	~10 ns	~100,000	Nanometer defect
Magnons (old record)	100–200 ns	~200 ns	~1–10 ns	10–100	nm wavelength
Magnons (Serha 2026)	18 μs	18 μs	~1–10 ns	1,800–18,000	nm wavelength

At 18 microseconds, with gate operation times in the 1-to-10-nanosecond range demonstrated for spin-wave logic gates operating at GHz frequencies, a magnon qubit can now support between 1,800 and 18,000 gate operations before decoherence. Superconducting transmons at 100 microseconds coherence with 20-nanosecond gates manage 5,000 to 10,000. Magnons are now within a factor of two to four of the dominant commercial platform in useful operations per lifetime, while occupying physical space measured in nanometers rather than centimeters.

The Size Math Gets Absurd

IBM's Eagle and Condor quantum processors pack roughly 1,000 superconducting qubits onto chips measuring approximately 500 square centimeters. Each qubit occupies a fraction of a square centimeter because the resonators, coupling buses, and readout lines that surround it all require centimeter-scale features to operate at microwave frequencies.

Magnonic logic gates, by contrast, operate at wavelengths that can be compressed to ~100 nanometers, yielding gate footprints on the order of 1 square micrometer. Place 1,000 such gates on a chip: 0.001 square millimeters. That is 50 million times smaller than IBM's layout. Fifty million. A magnonic quantum processor holding 1,000 qubits could, in principle, fit on a chip the size of Abraham Lincoln's nose on a penny.

There is an asterisk the size of a refrigerator, and we will return to it shortly.

Why This Is a Materials Problem Now

Serha's most consequential finding is negative: no fundamental physics limit appeared anywhere in the data. In every YIG sphere tested, the magnon lifetime tracked material purity in a clean, monotonic relationship, and even the least pure sphere exceeded all prior published records. Impurities scatter magnons, and removing impurities makes magnons last longer, with no unexpected decay channels, no exotic quantum effects truncating the lifetime, and no wall.

Compare this to superconducting qubits, where coherence improvements stalled for years because researchers were fighting multiple loss mechanisms simultaneously: dielectric surface losses, quasiparticle tunneling, cosmic-ray impacts, stray infrared photons. Each mechanism required a different fix, and improving one sometimes worsened another. At its core, the magnon result says the problem is mono-causal: clean up the crystal, and you get longer lifetimes. That is a manufacturing challenge, not a physics puzzle, and manufacturing challenges have a track record of yielding to sustained industrial investment in ways that physics puzzles do not.

Growing high-quality YIG crystals is not trivial, requiring liquid-phase epitaxy or Czochralski growth in controlled atmospheres, with rare-earth dopant levels held below parts-per-million thresholds. But the semiconductor industry solved comparable purity problems for silicon decades ago, driving defect densities from millions per square centimeter to near zero. There is no reason in principle why YIG cannot follow the same trajectory, given sufficient demand and investment.

The Strongest Case Against Getting Excited

Nobody has built a magnon quantum gate, not a good one and not a bad one, because Serha's paper measures coherence time and nothing else. No entanglement between magnon qubits, no gate fidelity data, no error correction protocol, no demonstration that two magnons can interact in a controlled way that constitutes a logical operation at quantum-level fidelity. Superconducting qubits have had decades of gate engineering, thousands of researchers, billions of dollars in corporate and government funding, and a mature ecosystem of cryogenic control electronics. Magnonic computing has a 100x lifetime improvement and a lot of theoretical proposals.

Put differently: a car engine that runs for 100 times longer is impressive, but it is not a car. You still need a transmission, wheels, brakes, and a steering column. In quantum computing, that transmission equivalent is the two-qubit entangling gate operating at fidelities above 99%, and it has never been demonstrated with magnons. Until it is, the 18-microsecond lifetime is a necessary condition for magnon quantum computing, not a sufficient one.

About that refrigerator: the 50-million-fold size advantage applies to the chip, not the system. Magnons at 18 microseconds require cooling to 30 millikelvin, the same temperature regime as superconducting qubits. A dilution refrigerator maintaining this temperature is a room-filling, multi-hundred-thousand-dollar apparatus. Shrinking the chip from 500 square centimeters to a fraction of a square millimeter does not shrink the fridge. If the value proposition is miniaturization, the honest comparison is system-to-system, not chip-to-chip.

What We Did Not Prove

Our gate operations calculation in the comparison table uses estimated magnonic gate times of 1 to 10 nanoseconds, derived from published demonstrations of spin-wave logic operating at GHz frequencies. These demonstrations used classical magnons, not quantum-coherent single-magnon states. Extrapolating classical gate speeds to quantum operations assumes that the switching time does not degrade when operating at the single-magnon level, which has not been verified experimentally. Our 50-million-fold size reduction is a theoretical maximum assuming magnonic gate footprints of 1 square micrometer with 100-nanometer-wavelength spin waves, but no integrated magnonic circuit of any complexity has been fabricated. We also note that the YIG spheres used in Serha's experiment are macroscopic objects (millimeter-scale), not integrated components on a chip. Bridging the gap between sphere-in-a-cryostat and lithographically patterned magnonic circuit is a fabrication challenge that remains unsolved. Finally, the paper tests only three YIG samples, and while the purity-lifetime correlation is clear and consistent across all three, three data points do not constitute the kind of exhaustive materials characterization that would let manufacturers specify a target purity for a given lifetime guarantee.

The Bottom Line

If you work in quantum computing hardware: the magnon coherence bottleneck just moved from "probably fundamental" to "definitely materials engineering," and that matters because materials engineering scales with industrial investment in ways that fundamental physics breakthroughs do not. Watch for follow-up papers demonstrating two-magnon entanglement in high-purity YIG, because that is the next gate. If you run a crystal growth company or a materials science lab specializing in magnetic garnets: the commercial demand signal for ultra-pure YIG is about to intensify, and being able to quote defect densities at semiconductor-grade precision will be a competitive advantage. If you are a quantum computing investor evaluating the landscape: magnons just entered the coherence regime where they compete with superconducting transmons on gate operations while offering nanometer-scale circuit density, but they lack the gate engineering maturity that makes superconducting and trapped-ion systems commercially viable today. Position accordingly. And if you are following quantum computing as a curious observer wondering when any of this will matter to your daily life: not yet, and the team that ran this experiment would tell you the same thing. But the question changed: it used to be whether magnons could live long enough, and now it is whether we can build gates fast enough. That second question has a much better track record of getting answered.