by Joel Hochstetter (University of Sidney).
The brain’s efficient information processing ability has inspired the development of neuromorphic systems constructed from nanoscale material structures exhibiting “brain-like” responses to electrical stimulation. Here we report on one such system, neuromorphic nanowire networks. Bio-inspired self-assembly of these nanowire networks confers a complex neural network-like structure, while nanowire-nanowire junctions exhibit synapse-like electrical switching due to ionic transport. Under electrical stimulation, adaptive collective dynamics emerges from the interplay between network connectivity and the non-linear interactions between junctions. We studied these emergent dynamics using simulated and experimental time-series data. We found that switching events are consistent with avalanches with size and life-time distributions that obey a power-law with exponents obeying the scaling relation consistent with critical dynamics. By driving these neuromorphic networks through different dynamical regimes, from ordered to chaotic, we quantified the dynamical states by calculating the maximum Lyapunov exponent. The computational capacity of these states was assessed through a learning task, non-linear wave-form transformation, which revealed that networks close to the edge-of-chaos were the most robust performers. Our results suggest that critical-like dynamics can be utilised to optimise information processing in neuromorphic systems.
Spotlight talk presented on October 6th 2020, at the Brain Criticality Virtual Conference 2020 (Plenz D., Chialvo D., de Arcangelis L. & Battaglia D. organizers)