Inverse-design topology optimization of magnonic devices using level-set method

Autor(en)

Andrey A. Voronov, Marcos Cuervo Santos, Florian Bruckner, Dieter Suess, Andrii V. Chumak, Claas Abert

Abstrakt

The inverse design approach in magnonics exploits the wave nature of magnons and machine learning to develop novel logical devices with unique functionalities that exceed the capabilities of analytical methods. Despite its potential in analog, Boolean, and neuromorphic computing, existing implementations are limited by memory usage, restricting computational depth and the design of complex devices. This study introduces a level-set parameterization approach for topology optimization, coupled with an adjoint-state method for memory-efficient solution of magnetization dynamics equations. The simulation platform employed is $\texttt{neuralmag}$, a GPU-accelerated micromagnetic software that features a unique nodal finite-difference discretization scheme integrated with automatic differentiation tools. To validate the proposed inverse design method, we first addressed a magnetic nanoparticle shape optimization task, demonstrating how additional constraints on the objective function can control the design solution space and govern the optimization process. Subsequently, the functionality of a magnonic demultiplexer was realized using a 300-nm-wide yttrium iron garnet conduit. This device achieves spatial frequency-selective separation of spin waves into distinct outputs. This task demonstrates the algorithm's efficiency in identifying local minima of the objective function across various initial topologies, establishing its effectiveness as a versatile inverse design tool for creating magnonic logic device designs.

Organisation(en)

Nanomagnetismus und Magnonik, Physik Funktioneller Materialien, Forschungsplattform MMM Mathematics-Magnetism-Materials

Publikationsdatum

11-2024

ÖFOS 2012

103017 Magnetismus

Schlagwörter

Link zum Portal

https://ucrisportal.univie.ac.at/de/publications/04590f4b-e336-470f-8370-8811cea3c23d