Phys. Rev. Applied 22, 024072 (2024) – Published 29 August 2024
Harnessing spin waves for high-speed computing: In this work an innovative spin-wave reservoir chip, utilizing ferromagnetic permalloy thin films, demonstrates exceptional capabilities. By strategically manipulating spin-wave interference, the authors achieve a multi-input–multi-output reservoir capable of memory retention, nonlinearity enhancement, and accurate magnetic field estimation. This spintronic hardware paves the way for high-speed applications in reservoir computing and signal processing.
Phys. Rev. Applied 22, L021001 (2024) – Published 1 August 2024
The ability to simulate XY models of classical spins is rather important, since e.g. it is related to solving NP-hard optimization problems. This study uses a photonic simulator to realize XY Hamiltonians with arbitrary spin connections and coupling strengths. The key unit is an optical system for vector-matrix multiplication that can perform arbitrary transformations of complex matrices. The Berezinskii-Kosterlitz-Thouless transition and ground-state search of several XY models are demonstrated experimentally. Thus this Letter provides an effective alternative approach for investigating such models and solving continuous quadratic optimization problems with optical systems.
A. López, J.D. Costa, T. Böhnert, P.P. Freitas, R. Ferreira, I. Barbero, J. Camarero, C. León, J. Grollier, and M. Romera
Phys. Rev. Applied 22, 014082 (2024) – Published 31 July 2024
A promising branch of neuromorphic computing aims to perform cognitive operations in hardware, leveraging the physics of efficient and well-established nanodevices. This work presents a reconfigurable classifier, based on a network of magnetic tunnel junctions, that can learn to classify spoken vowels. In this task the hardware network surpasses multilayered software neural networks with the same number of trained parameters. These results, obtained using the same devices and working principle employed in industrial spin-transfer-torque magnetic random-access memory, constitute an important step toward the development of large-scale neuromorphic networks based on established technology.
Haotian Li, Liyuan Li, Rongxin Xiang, Wei Liu, Chunjie Yan, Zui Tao, Lei Zhang, and Ronghua Liu
Phys. Rev. Applied 22, 014027 (2024) – Published 11 July 2024
Despite being formidable tools in artificial intelligence, artificial neural networks consume substantial energy during their training phase. This study introduces hardware-based artificial neural networks that utilize artificial and natural noncollinear spin textures, significantly reducing energy consumption and enhancing operational efficiency. The authors demonstrate two such spin-texture-based physical reservoirs, which exhibit robust information-processing capabilities in two nonlinear benchmark tests. Additionally, they implement a direct-feedback-alignment algorithm within hardware, further advancing the efficiency of deep neural networks.