IMX Talks - Magnonic Hardware for Pattern Recognition

Neural networks are powerful tools to learn patterns and make inferences in complex problems. However, they rely on a massive number of neurons and interconnecting weights which require extensive training using a large dataset. To compensate for this, reservoir computing is based on recurrent neural networks with randomly fixed weights. Thereby, only the output weights require training for a particular task, reducing the training to a simple linear regression. Recently, there has been a shift towards physical reservoir computing, offering potential
advantages in speed, energy efficiency, and hardware simplicity. Physical reservoir computing utilizes the inherent nonlinearity of physical systems to map the input into a higher-dimensional space in which different input patterns become linearly separable. New advancements and experimental implementations use diverse physical substrates, including mechanical structures, optical systems, and spintronic devices.
In our work, we take advantage of the rich nonlinear dynamics inside magnetic vortices. Their eigenmode system comprises the gyrotropic motion of the vortex core as well as magnon modes with well-defined radial and azimuthal quantization in the vortex’s skirt. Harnessing the nonlinear interactions between these different vortex eigenmodes in reciprocal space, it is possible to perform temporal information processing and pattern recognition without relying on information transport in real space [1]. This presentation will give a comprehensive overview of experimental results and numerical simulations demonstrating the capabilities and advantages of magnon reservoir computing. Additionally, the mutual nonlinear interactions between the magnon modes and the vortex core gyration lead to a much richer set of eigenstates, extending the dimensionality of the magnon scattering reservoir.
ACKNOWLEDGMENT
This work has received funding from the EU Research and Innovation Programme Horizon Europe under grant agreement no. 101070290 (NIMFEIA).
REFERENCES
[1] L. Körber, et al., “Pattern recognition in reciprocal space with a magnon-scattering reservoir” Nature Communications, 14, 3954 (2023).

Bio: Dr. Katrin Schultheiss received her PhD in Physics (Dr. rer. nat.) on spin-wave transport in two-dimensional microstructures from the Technische Universität Kaiserslautern, Germany in 2013. Since 2015, she works at the Institute of Ion Beam Physics and Materials Research at the Helmholtz-Zentrum Dresden-Rossendorf (HZDR) in Dresden, Germany. As a member of the Department of Magnetism, her research is focussed on the study of linear and nonlinear magnetization dynamics in micro- and nanostructure as well as in spin textures using Brillouin light scattering microscopy. For her achievements on "Nonlinear magnonics as the foundation of spin-based neuromorphic computing" she received the HZDR Forschungspreis in 2022.