Klaus Raab1, Maarten A. Brems1, Grischa Beneke1, Takaaki Dohi1, Jan Rothörl1, Fabian Kammerbauer1, Johan H. Mentink2, Mathias Kläui1
1 Institut für Physik, Johannes Gutenberg-Universität Mainz, Staudingerweg 7, 55128 Mainz, Germany
2 Radboud University, Institute for Molecules and Materials, Heyendaalseweg 135, 6525 AJ Nijmegen, The Netherlands
Reservoir computing (RC) [1,2] has been considered one of the key computational principles beyond von-Neumann computing, for which magnetic systems can provide a suitable platform based on the intrinsic non-linearity, dynamics on a wide range of timescales and easy integration with non-volatile magnetic memory. Magnetic skyrmions, topological particle-like spin textures in magnetic thin films are particularly interesting in this respect, since the magnetic textures in which they are embedded respond strongly non-linearly to external stimuli. However, while several theoretical proposals exist for skyrmion reservoir computing, experimental realizations are missing so far.
We demonstrate experimentally a conceptionally new approach for reservoir computing, that leverages the thermally activated diffusive motion of skyrmions [3] in a confined geometry [4]. By confining the gated and thermal skyrmion motion, we show that already a single skyrmion in a confined geometry suffices to realize all Boolean logic gate operations including the non-linearly separable XOR operation that cannot be realized using a conventional single layer perceptron. An effective potential well created by the confinement allows for a natural reset mechanism, which does not rely on pinning effects, but is instead enabled by the thermal fluctuations of the skyrmions. We demonstrate that the training costs are low and our ultra-low power operation with current densities orders of magnitude smaller than those used in existing spintronic reservoir computing demonstrations is enabled by thermally activated Brownian motion dynamics. Our proposed concept can be easily extended by linking multiple confined geometries and/or by including more skyrmions in the reservoir, suggesting high potential for scalable and low-energy reservoir computing.
References
[1] Romera, M. et al. Vowel recognition with four coupled spin-torque nano-oscillators. Nature 563, 230–234 (2018).
[2] Pinna, D et al. Reservoir Computing with Random Skyrmion Textures. Phys. Rev. Appl. 14, 054020 (2020).
[3] Zázvorka, J. et al. Thermal skyrmion diffusion used in a reshuffler device. Nat. Nanotechnol. 14, 658–661 (2019).
[4] Song, C. et al. Commensurability between Element Symmetry and the Number of Skyrmions Governing Skyrmion Diffusion in Confined Geometries. Adv. Funct. Mater. 31, 2010739 (2021).