Deep Physical Reservoir Computing with Programmable Nanomagnetic Hierarchies
ORAL
Abstract
Nanomagnetic artificial spin-systems are ideal candidates for neuromorphic hardware. Their passive memory, state-dependent dynamics and nonlinear GHz spin-wave response provide powerful reservoir computation1,2 (RC).
Any single physical reservoir is typically forced to trade-off between performance metrics including nonlinearity and memory capacity, with the compromise typically hard-coded during nanofabrication.
Here, we present three artificial spin-systems and show how tuning array-geometry and magnetisation dynamics defines computing performance.
Beyond spin-system design, we engineer physical `meta-reservoirs'3 (deep RC) comprising multiple nanoarrays. Data is input through several reservoirs in parallel, hierarchical/`deep' and hybrid deep-parallel architectures. Meta-reservoirs show substantially enhanced performance beyond any single reservoir across a broad taskset. Crucially, we present a method for hierarchy-programming, allowing designer configurations of meta-reservoirs with any desired nonlinearity/memory-capacity. This solves the problem of reservoir performance compromise, providing efficient system reconfiguration and task-dependent optimisation.
1, Stenning, Kilian D. et al. Nature Nanotechnology 17.5 (2022): 460-469.
2. Jensen, Johannes H et al. ALIFE 2018: The 2018 Conference on Artificial Life. MIT Press, 2018.
3. Manneschi, Luca, et al. Frontiers in Applied Mathematics and Statistics 6 (2021): 76.
Any single physical reservoir is typically forced to trade-off between performance metrics including nonlinearity and memory capacity, with the compromise typically hard-coded during nanofabrication.
Here, we present three artificial spin-systems and show how tuning array-geometry and magnetisation dynamics defines computing performance.
Beyond spin-system design, we engineer physical `meta-reservoirs'3 (deep RC) comprising multiple nanoarrays. Data is input through several reservoirs in parallel, hierarchical/`deep' and hybrid deep-parallel architectures. Meta-reservoirs show substantially enhanced performance beyond any single reservoir across a broad taskset. Crucially, we present a method for hierarchy-programming, allowing designer configurations of meta-reservoirs with any desired nonlinearity/memory-capacity. This solves the problem of reservoir performance compromise, providing efficient system reconfiguration and task-dependent optimisation.
1, Stenning, Kilian D. et al. Nature Nanotechnology 17.5 (2022): 460-469.
2. Jensen, Johannes H et al. ALIFE 2018: The 2018 Conference on Artificial Life. MIT Press, 2018.
3. Manneschi, Luca, et al. Frontiers in Applied Mathematics and Statistics 6 (2021): 76.
*This work was supported by the Leverhulme Trust (RPG-2017-257) to WRB and the Engineering and Physical Sciences Research Council (Grant No. EP/W524335/1) to KDSAV was supported by the EPSRC Centre for Doctoral Training in Advanced Characterisation of Materials (Grant No. EP/L015277/1).
–
Presenters
-
Kilian Stenning
- University College London