Physics Seminars
Controlling and Engineering Nanomagnetic States for Neuromorphic Computation
Speaker: Killian Stenning (Imperial College of London)
Date: Wednesday 15 November 2023
Time: 15:00
Venue: Queens Building N3.28
Modern AI and machine-learning continues to provide striking performance at an accelerating rate. However, this is accompanied by a rapidly accelerating energy cost due to the inefficiencies of conventional computing hardware. To alleviate this rising energy cost, new types of computing hardware are being explored. Neuromorphic computing [1] is a rapidly emerging field where the complex dynamics of physical systems are used for computation. The field has recently undergone an explosion in the range and sophistication of implementations, with rapidly improving performance. Nanomagnetic systems are attractive candidates for data storage and neuromorphic computing. The mathematics underpinning machine learning were originally invented to describe arrays of strongly interacting spins [2]. Each nanomagnet can store information in its magnetisation state and dipolar coupling between neighbouring information provides collective information processing. However, realising nanomagnetic array computing hardware proved challenging due to limited system dynamics and difficulties in inputting and reading out data for computation. In this talk, I will show how information can be encoded in individual nanomagnets via all-optical magnetic switching with a mW-laser for data storage [3] and how engineering nanomagnet shapes and geometries provides the complex dynamics required for neuromorphic computing [4]. I will then show how arrays of nanomagnets can be interconnected into network architectures to improve computing performance and to reach an `over-parameterised' regime [5], where the system no longer overfits to training data, allowing rapid learning of small datasets. Markovi, Danijela, et al. Physics for neuromorphic computing. Nature Reviews Physics 2.9 (2020) 499-510. Panchenko, Dmitry. The sherrington-kirkpatrick model. Springer Science Business Media, 2013. Stenning, Kilian D., et al. Low-power continuous-wave all-optical magnetic switching in ferromagnetic nanoarrays. Cell Reports Physical Science 4.3 (2023). Gartside, Jack C., Stenning, Kilian D. et al. Reconfigurable training and reservoir computing in an artificial spin-vortex ice via spin-wave fingerprinting. Nature Nanotechnology 17.5 (2022) 460-469. Stenning, Kilian D., et al. Neuromorphic Few-Shot Learning Generalization in Multilayer Physical Neural Networks. arXiv preprint arXiv 2211.06373 (2022).