A research team at Tohoku University has shown that a steady electric current can hold electron spins in positions that should be impossible to maintain, effectively forcing magnetization to point the wrong way against an applied magnetic field. The finding, reported in Nature Materials on March 4, 2026, introduces what the authors call a “spintronic Kapitza pendulum” and could reshape how engineers design low-power computing hardware that relies on spin states rather than conventional charge-based transistors.
Forcing Spins to Defy Their Natural Alignment
In ordinary magnetic materials, electron spins settle into the orientation that minimizes energy, aligning with whatever external field is present. Pushing them the opposite way is like balancing a pendulum upside down: the slightest disturbance sends the system crashing back to equilibrium. The Tohoku team found a workaround. By driving a spin-transfer torque through a cobalt-based experimental platform, they demonstrated that electric current can dynamically stabilize magnetization in an energetically unfavorable orientation, holding it steady against the field rather than letting it relax. A companion access page on the Nature site frames the work as a route to active control of non-equilibrium spin states.
The analogy to a Kapitza pendulum is precise. In classical mechanics, a Kapitza pendulum stays inverted because its pivot point vibrates fast enough to create an effective restoring force. Here, the spin-transfer torque from the current plays the same stabilizing role, continuously compensating for the tendency of spins to flip back. The result is a magnetic state that exists only because energy is being pumped in, not because the material’s own energy landscape favors it. The experiment shows that, under the right conditions, dissipation and drive can cooperate to hold spins in a configuration that would otherwise be catastrophically unstable.
Why the Kapitza Analogy Matters for Hardware
The Kapitza framing is more than a teaching metaphor. It tells device designers that spin states can be treated as dynamically programmable rather than fixed by material properties alone. Traditional spintronic memory, such as magnetic tunnel junctions in MRAM, stores data by switching between two stable orientations that the material already supports. The Tohoku result suggests a third option: maintaining states that the material does not naturally support, as long as current keeps flowing, and tuning the stability of those states on the fly.
That distinction has practical consequences. A press release from the WPI-Advanced Institute for Materials Research at Tohoku quotes the team as saying that electric current can actively stabilize spins in energetically unfavorable states. The institutional summary also ties the work to probabilistic computing models, including restricted Boltzmann machines, which require hardware elements that can fluctuate between states in controlled ways. A spin held at an unstable equilibrium by tunable current could, in principle, serve as exactly that kind of stochastic element, toggling between orientations when the current is modulated or noise is injected.
Because the stabilized state is inherently out of equilibrium, it is extremely sensitive to small perturbations. That sensitivity is usually a liability in memory devices, which must resist noise, but it becomes a feature in probabilistic hardware, where controlled randomness and rapid switching are essential. The Kapitza analogy therefore points to a new design space in which instability is not suppressed but engineered and harnessed.
From Isotropic Magnets to Helical Spin Textures
The cobalt system used in the Nature Materials paper is described as a nearly isotropic magnet, meaning its spins have roughly equal energy in all directions. That near-isotropy is what makes current-driven stabilization possible: without a strong built-in preference for one axis, even a modest torque can tip the balance. The researchers show that by carefully tuning current density and magnetic field, they can map out regimes where the “wrong-way” magnetization becomes dynamically stable.
The same principle of current-driven spin control is emerging in very different material classes. A separate study in Physical Review Letters demonstrated current-driven collective manipulation of helical spin textures in the van der Waals antiferromagnet Ni1/3NbS2. In that material, the competition between exchange interactions and interlayer Dzyaloshinskii–Moriya effects creates complex spiral patterns, rather than a single uniform magnetization. The researchers showed that passing current through the material could reorient and reshape those patterns collectively, effectively steering an entire spin texture with electrical drive.
A more detailed discussion of the mechanisms behind that behavior appears in a related arXiv preprint, which analyzes how spin–orbit coupling and layered crystal symmetry mediate the coupling between charge flow and the helically ordered spins. Although the microscopic ingredients differ from the cobalt system, the conceptual link is clear: in both cases, current injects angular momentum and alters the effective energy landscape, enabling spin configurations that would be difficult or impossible to realize in equilibrium.
The two results attack the same problem from different angles. In cobalt, current stabilizes a single magnetization direction against an opposing field. In Ni1/3NbS2, current reshapes an entire texture of intertwined spins. Together, they demonstrate that electrical control over spin configurations is not limited to one material family or one type of magnetic order, a point that broadens the engineering possibilities considerably and suggests that non-equilibrium design principles could apply across ferromagnets, antiferromagnets and more exotic spin phases.
What Current-Stabilized Spins Could Enable
Most reporting on spintronics focuses on memory and storage, where magnetic states encode binary data. The Tohoku result points somewhere different. If spins can be parked at unstable equilibria and released on demand, they become candidates for computation itself, not just data retention. The Tohoku University press office explicitly describes the discovery as opening a path to new computing architectures built around unstable points that can, in principle, point equally in any direction.
The connection to probabilistic and neuromorphic computing is direct. Restricted Boltzmann machines and similar probabilistic models depend on hardware that can sample from distributions efficiently. A spin balanced at an unstable point by current is inherently sensitive to thermal and electrical noise, which means it can act as a natural random sampler when the stabilizing current is slightly reduced. That is a very different design philosophy from deterministic logic gates, and it aligns with the growing demand for hardware optimized for machine learning inference and generative models rather than traditional arithmetic pipelines.
Separate work on altermagnetic materials adds another dimension to this landscape. A recent preprint on altermagnetic bilayers discusses how symmetry-broken antiferromagnets can host spin-split electronic bands without net magnetization, enabling electrical readout and control without stray magnetic fields. In such systems, current-driven spin torques might be combined with the Kapitza-style stabilization concept to create devices whose active elements are both field-free and dynamically unstable, yet controllable.
In a notional probabilistic processor, different material platforms could play complementary roles. Nearly isotropic ferromagnets like the cobalt device might provide strongly tunable unstable equilibria that serve as noisy neurons or stochastic bits. Helical antiferromagnets such as Ni1/3NbS2 could implement couplings between those bits via their collective textures, effectively encoding weighted connections in the geometry of the spin spirals. Altermagnets could supply efficient interfaces to charge-based circuitry, reading out spin states as voltage signals without large external magnets.
Significant challenges remain before any of this moves from lab demonstrations to deployed hardware. The current densities required to stabilize or reshape spins must be reduced to avoid excessive power consumption and device heating. Materials will need to be engineered so that the dynamically stabilized states are robust against fabrication imperfections and temperature variations, while still remaining sensitive enough to noise to be useful in probabilistic computation. Integration with CMOS and with existing memory technologies will require careful attention to device geometry, interconnect design and error-correction strategies.
Even so, the core message of the spintronic Kapitza pendulum is already changing how researchers think about magnetization. Instead of viewing spins as passive degrees of freedom that simply relax to the nearest energy minimum, the new work treats them as active, driven variables whose stability can be sculpted by current. That conceptual shift, from equilibrium to engineered non-equilibrium, could prove as important for future computing architectures as the original discovery of spin-transfer torque was for today’s MRAM. If engineers can learn to reliably balance spins “upside down” and let them fall only when it is computationally useful, the line between memory, logic and randomness in hardware may begin to blur.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
