A self-learning magnetic Hopfield neural network with intrinsic gradient descent adaption DOI Creative Commons
Chang Niu, Huanyu Zhang, Chuanlong Xu

et al.

Proceedings of the National Academy of Sciences, Journal Year: 2024, Volume and Issue: 121(51)

Published: Dec. 13, 2024

Physical neural networks (PNN) using physical materials and devices to mimic synapses neurons offer an energy-efficient way implement artificial networks. Yet, training PNN is difficult heavily relies on external computing resources. An emerging concept solve this issue called self-learning that uses intrinsic parameters as trainable weights. Under inputs (i.e., data), achieved by the natural evolution of intrinsically adapt modern learning rules via autonomous process, eliminating requirements computation Here, we demonstrate a real spintronic system mimics Hopfield (HNN), unsupervised performed process. Using magnetic texture-defined conductance matrix weights, illustrate under voltage inputs, naturally evolves adapts Oja's algorithm in gradient descent manner. The HNN scalable can achieve associative memories patterns with high similarities. fast spin dynamics reconfigurability textures advantageous platform toward efficient directly materials.

Language: Английский

Perspective on nonvolatile magnon-signal storage and in-memory computation for low-power consuming magnonics DOI Creative Commons

A. Nizet,

Mingran Xu, Shreyas S. Joglekar

et al.

Applied Physics Letters, Journal Year: 2025, Volume and Issue: 126(16)

Published: April 21, 2025

Magnons are the quanta of spin waves and transport angular momenta through magnetically ordered materials. They can be used to distribute control on-chip GHz signals without charge flow, thereby avoiding Joule heating. Beyond multiplexed signal processing, filtering, Boolean logic, they allow for hardware implementation neural networks exploiting cascaded magnon scattering on nanoscale. A game-changing boost is expected if nonvolatile magnon-signal storage in-memory computation schemes become realistic. We outline recent progress in experimental research micromagnetic modeling toward these goals before sketching remaining challenges.

Language: Английский

Citations

0

Decision making module based on stochastic magnetic tunnel junctions DOI

Yifan Miao,

Li Zhao, Yajun Zhang

et al.

Science China Physics Mechanics and Astronomy, Journal Year: 2024, Volume and Issue: 68(1)

Published: Nov. 6, 2024

Language: Английский

Citations

0

Targets capture by distributed active swarms via bio-inspired reinforcement learning DOI
Kun Xu, Yue Li, Jun Sun

et al.

Science China Physics Mechanics and Astronomy, Journal Year: 2024, Volume and Issue: 68(1)

Published: Nov. 21, 2024

Language: Английский

Citations

0

A self-learning magnetic Hopfield neural network with intrinsic gradient descent adaption DOI Creative Commons
Chang Niu, Huanyu Zhang, Chuanlong Xu

et al.

Proceedings of the National Academy of Sciences, Journal Year: 2024, Volume and Issue: 121(51)

Published: Dec. 13, 2024

Physical neural networks (PNN) using physical materials and devices to mimic synapses neurons offer an energy-efficient way implement artificial networks. Yet, training PNN is difficult heavily relies on external computing resources. An emerging concept solve this issue called self-learning that uses intrinsic parameters as trainable weights. Under inputs (i.e., data), achieved by the natural evolution of intrinsically adapt modern learning rules via autonomous process, eliminating requirements computation Here, we demonstrate a real spintronic system mimics Hopfield (HNN), unsupervised performed process. Using magnetic texture-defined conductance matrix weights, illustrate under voltage inputs, naturally evolves adapts Oja's algorithm in gradient descent manner. The HNN scalable can achieve associative memories patterns with high similarities. fast spin dynamics reconfigurability textures advantageous platform toward efficient directly materials.

Language: Английский

Citations

0