Regensburg 2025 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MA: Fachverband Magnetismus

MA 45: Computational Magnetism

MA 45.10: Vortrag

Freitag, 21. März 2025, 12:00–12:15, H18

Memory-Efficient Inverse Design for Advanced Magnonic Devices Using Level-Set Optimization — •Andrey Voronov^{1, 2}, Marcos Cuervo Santos^{2, 3}, Florian Bruckner^{1, 4}, Dieter Suess^{1, 4}, Andrii Chumak¹, and Claas Abert^{1, 4} — ¹Faculty of Physics, University of Vienna, Vienna, Austria — ²Vienna Doctoral School in Physics, University of Vienna, Vienna, Austria — ³Faculty of Sciences, University of Oviedo, Oviedo, Spain — ⁴Research Platform MMM Mathematics - Magnetism - Materials, University of Vienna, Vienna, Austria

Inverse design in magnonics utilizes the wave nature of magnons and machine learning to develop logic devices with unique functionalities. However, existing methods face memory constraints, limiting the exploration of complex systems.

To address this, we integrate a level-set parameterization approach with an adjoint state method for memory-efficient simulations of magnetization dynamics. Implemented in neuralmag, a GPU-accelerated micromagnetic software, this framework enables efficient optimization of device topologies.

We validate the approach through two tasks: optimizing the shape of a magnetic nanoparticle to control hysteresis behavior and designing a 300-nm-wide yttrium iron garnet demultiplexer for frequency-selective spin-wave separation. These results showcase the algorithm’s robustness and versatility in enabling the design of advanced magnonic devices for computational logic technologies.

Keywords: Magnonics; Inverse design; Level-set method; Adjoint-state method; Topology optimization