Neuromorphic computing with magnons

Within the last decade, spintronics and magnonics have demonstrated an impressive development in the experimental realization of Boolean logic gates. However, the exponential growth of data and the rise of the internet of things are pushing the deterministic Boolean computing of von-Neumann architectures to their limits or are simply to energy consuming. Moreover, it is accepted commonly that conventional Boolean computer architectures are likely to remain inefficient for certain cognitive tasks in which the human brain excels, such as pattern recognition, particularly when incomplete or noisy data are involved. One of the most generic and abstract implementations of brain-inspired computing schemes is reservoir computing, where the nonlinear response of a physical system is used to separate patterns hidden in a temporal data stream into distinct manifolds of a higher dimensional output space. In this presentation, I will demonstrate the experimental realization of pattern recognition based on reservoir computing using magnons. Recently, we reported on the nonlinear scattering of magnons in vortices in micron-sized Permalloy discs [1] which we also learned to control and stimulate by means of other magnons [2]. Now, we utilize these phenomena to employ magnons for pattern recognition without actually relying on magnon transport in real space. I will present a comprehensive overview of experimental results and numerical simulations demonstrating the capabilities and advantages of magnon reservoir computing in reciprocal space. [1] K. Schultheiss, et al., Physical Review Letters 125, 207203 (2020) [2] K. Schultheiss, et al., Physical Review Letters 122, 097202 (2019)