Spin-wave waveguides store and route signals without a bias field

A magnonic circuit can store information directly in the static magnetization of its spin-wave waveguides and then use those retained states to route signals without an external bias magnetic field. The demonstrated approach combines a low-loss magnetic insulator, yttrium iron garnet, with waveguide elements whose local magnetization directions determine how a spin wave travels. Experiments on a four-arm crossing showed that an incoming wave could be sent into a selected side arm with more than 90% of its incident amplitude, while simulations indicated that the same routing principle survives in waveguides only 200 nanometres wide.

The researchers also proposed and numerically assessed a selective switching method for programming individual waveguide states, giving the network a built-in, non-volatile control layer for guiding nanoscale information signals. The experimental building block consisted of two YIG strips crossing at right angles. Each strip was 700 nanometres wide, and the YIG layer was 120 nanometres thick, with the film grown by liquid-phase epitaxy on a gadolinium gallium garnet substrate.

Electron-beam lithography and argon-ion etching formed the waveguides, while a nearby gold antenna launched spin waves with a microwave signal. Before measurement, the waveguide magnetizations were initialized by briefly applying a 100-millitesla field at an angle of 45 degrees or minus 45 degrees relative to the horizontal axis. The field was then removed, and all subsequent routing measurements were conducted at zero global bias field, allowing the waveguide magnetization states to function as the retained configuration.

Spin waves were tracked by phase-resolved micro-focus Brillouin light scattering. A focused, 473-nanometre laser interrogated the magnetic dynamics through inelastic scattering, and a Fabry–Pérot interferometer resolved the scattered spectrum. By interfering that signal with a reference beam whose phase was modulated at the spin-wave frequency, the researchers could map both the local spin-wave amplitude and phase across the crossing.

The measurements also yielded the wavelength, wavenumber and dispersion relation of the modes. Experimental points agreed with micromagnetic calculations made using the open-source MuMax3 package. In the 700-nanometre-wide guides, the measured modes behaved as backward-volume spin waves, with frequency decreasing as wavenumber increased, and their frequencies remained above 1 gigahertz even without a bias field.

That zero-field frequency range arose from the quantization of the wavevector component perpendicular to the waveguide axis, which shifted the mode frequency upward as the guide width decreased. The agreement between the measurements and simulations indicated that the model captured the relevant properties of the fabricated structure. The experimental antenna was 500 nanometres wide and 150 nanometres thick and was positioned 10 micrometres from the crossing centre.

The phase-resolved maps were obtained by scanning an 8-by-8-micrometre area in 200-nanometre steps, while the dispersion relation was established from the wavelength or wavenumber measured at each excitation frequency. At an excitation frequency of 1.4 gigahertz, the crossing behaved like a magnetic traffic switch. A wave entering from the antenna was almost entirely deflected into one selected side arm, with forward transmission and excitation of the opposite side arm remaining minimal.

Reversing the angle of the initialization field reversed the preferred output: the diagonal ridge in the device’s internal effective field changed orientation, and the wave was sent into the other arm. The output amplitude in the selected “Out” arm exceeded 90% of the incident-wave amplitude. These observations provide the central experimental evidence that a non-volatile magnetization pattern can program the path of a spin wave without maintaining a static external magnetic field during operation.

Micromagnetic modelling explains the routing through the microscopic magnetic landscape around the intersection. Shape anisotropy aligns the static magnetization with each waveguide axis away from the crossing, but the magnetization rotates from the horizontal to the vertical direction in the central region. This rotation produces a nonzero effective magnetic field with a broad ridge along the diagonal perpendicular to the overall magnetization.

The ridge acts as a potential barrier for propagating spin waves, suppressing the forward and opposite-arm routes while reflecting energy into the selected arm. When the initialization field is reversed, the ridge aligns with the other diagonal, and the preferred output changes. The remembered state therefore determines signal flow through changes in the internal effective field rather than through continuous changes in an applied bias.

The researchers then tested the idea computationally at smaller dimensions. Their model used 200-nanometre-wide, 50-nanometre-thick YIG guides with rounded 100-nanometre corners, dimensions they considered compatible with practical nanofabrication without notable degradation from edge roughness. As the guides became narrower and thinner, dipolar effects weakened and the frequency interval supporting backward dispersion narrowed.

The simulations also revealed frequency degeneracy below 2.05 gigahertz, a condition that can promote two-magnon scattering from inhomogeneities and edge imperfections, leading to additional attenuation. The researchers therefore selected 2.2 gigahertz in an exchange-dominated region without that degeneracy, corresponding to a spin-wave wavelength of about 250 nanometres. At that operating point, waves were routed into different arms depending on the initialized state, with deflected-to-incident amplitude ratios reaching 0.9.

Although the larger experimental guides carried waves with negative group velocity, the smaller simulated guides supported efficient routing in an exchange-dominated regime with positive group velocity. The researchers expanded the model into a network of three horizontal and two vertical waveguides, with adjacent intersections separated by 1.5 micrometres. By changing stored magnetization directions, they sent an input wave from the middle horizontal guide into a vertical branch after three turns, back toward its starting point after four turns, or into another horizontal branch after reversing the wave’s direction by 180 degrees.

Each turn reduced wave amplitude by about 10% because of reflections, a loss the authors said could be compensated using spin-wave parametric amplification. Controlling individual branches requires changing one waveguide’s magnetization without disturbing its neighbours. The simulations used a global field of 12 millitesla, slightly below a calculated coercive field of 12.6 millitesla, to expand domain walls from the intersections without immediately switching the guides.

A local field of 1.0 millitesla applied over a 500-by-500-nanometre region then unpinned the walls, allowing them to travel along one selected vertical guide and reverse its magnetization while leaving the other vertical guide unchanged. The authors relate this process to local magnetic-field generation in magnetic random-access memory, in which currents through metallic leads write individual cells. Under the simulated conditions, currents on the order of 1 milliampere were estimated to initiate switching.

Selective writing could also be enhanced with spin–orbit torque, but that extension was not demonstrated in the reported work. Short-pulse simulations showed that the routing could also operate beyond continuous-wave excitation. Gaussian spin-wave pulses with a 5-nanosecond full width at half maximum were launched through a miniaturized crossing at 2.2 gigahertz and zero bias field.

The selected output pulse had a temporal width of 5.27 nanoseconds, compared with 5.24 nanoseconds for a pulse traveling the same 5-micrometre distance in a straight waveguide, indicating almost no routing-induced distortion. Its slight amplitude reduction and shorter delay were attributed to partial reflection and propagation effects. Increasing the pulse amplitude entered a nonlinear regime in which output width fell until it matched the input width at a normalized dynamic magnetization of approximately 4.2 × 10⁻², consistent with nonlinear self-phase modulation counteracting dispersion-induced broadening.

Taken together, the experiments and simulations present a magnonic architecture in which the same YIG waveguides serve as signal paths and non-volatile memory. The measured zero-field routing, efficient deflection in submicrometre structures, programmable network paths, numerically demonstrated selective switching and pulse propagation address several requirements for nanoscale information routing. Important qualifications remain: selective switching, the complete miniature network, the alternative operating regime and the nonlinear pulse effects were assessed computationally, while the central zero-field routing result was measured in the larger fabricated crossing.

The results establish a concrete route toward self-contained magnonic nano-circuits for nanoscale transmission and processing, including data multiplexing and neuromorphic computing, without a continuously applied bias field.

Subject of Research: Experimental and simulated zero-field spin-wave routing using non-volatile magnetization states in YIG waveguides

Article Title: Zero-field spin-wave networks with integrated non-volatile memory

Article References: Nikolaev, K. O., Raskhodchikov, D., Bensmann, J., Niehues, S., Bessonov, V. D., Lomonte, E., Jin, L., Schmidt, R., Kern, J., Michaelis de Vasconcellos, S., Bratschitsch, R., Demokritov, S. O., Pernice, W. H. P., & Demidov, V. E. (2026). Zero-field spin-wave networks with integrated non-volatile memory. Nature Materials. https://doi.org/10.1038/s41563-026-02778-1

Image Credits: AI Generated

DOI: 10.1038/s41563-026-02778-1

Keywords: spin waves, magnonic circuits, non-volatile memory, yttrium iron garnet, magnetization switching, Brillouin light scattering, micromagnetic simulation, waveguide routing, nanomagnetism, pulse propagation, Zero-field, spin-wave