Thinking with Nanomagnets

Jack C. Gartside is a Royal Academy of Engineering Research Fellow on Low-Energy Metamaterials for Neuromorphic Computing at Imperial College London. Jack works across a range of physical nanosystems, magnetic and optical, and focuses on designing and leveraging emergent functional behaviours in complex metamaterials, including neuromorphic computing, physics-based image processing & GHz reconfigurable magnonics.

Title: Thinking with Nanomagnets: Building nanoscale physical neural networks from dipolar magnetic nanoarrays

Artificial Intelligence is increasingly widespread in society, due in part to recent breakthroughs in generative AI & large language models. However, these models can contain billions of trained parameters - with a corresponding huge CO2 footprint. Training Chat-GPT consumed 550 tons of CO2.

In contrast, the brain consumes just 20 W. Neurons in the brain function very differently to CMOS hardware, they integrate processing & memory in a single unit and are connected in parallel to thousands of nearby neurons. 

Taking inspiration from the brain, we can leverage complex physical dynamics to compute 'neuromorphically', potentially bringing AI energy efficiency closer to biological brains. 

In our team at Imperial College London, we build nanomagnetic 'neurons' and network them together via dipolar magnetic field. We can read out their complex collective behaviour via the GHz spin-wave response of the system, providing a fast, low-energy platform. We demonstrate the computing power by challenging chaotic future prediction and nonlinear signal transformation tasks, and build powerful physical neural networks by interconnecting multiple nanomagnetic arrays. We will explore how building 3D artificial spin ice arrays can enhance processing power and enable access to Ultrastrong Magnon-Magnon Coupling regimes.

Gartside, Jack C., et al. "Reconfigurable training and reservoir computing in an artificial spin-vortex ice via spin-wave fingerprinting." Nature Nanotechnology 17.5 (2022): 460-469.

Lee, Oscar, Tianyi Wei, Kilian D. Stenning, Jack C. Gartside, Dan Prestwood, Shinichiro Seki, Aisha Aqeel et al. "Task-adaptive physical reservoir computing." accepted, Nature Materials (2023).

Dion, Troy, Gartside, Jack C. et al. "Ultrastrong Magnon-Magnon Coupling and Chiral Symmetry Breaking in a 3D Magnonic Metamaterial." (2023) Under review.

Stenning, K. D., Gartside, J. C., Manneschi, L., Cheung, C. T., Chen, T., Vanstone, A., ... & Branford, W. R. Neuromorphic Few-Shot Learning: Generalization in Multilayer Physical Neural Networks.(2023) Under review

Gartside, Jack C., et al. "Reconfigurable magnonic mode-hybridisation and spectral control in a bicomponent artificial spin ice." Nature Communications 12.1 (2021): 2488.