The human brain provides a prototype for how diverse collective dynamics and functionality can emerge from a complex network of relatively simple, non-linearly interacting components [1]. Neuromorphic networks formed by self-assembly of polymer-coated inorganic nanowires mimic these ingredients, with the non-linear switching of nanoscale junctions coupled by a disordered, recurrent network topology [2]. Experimental conductance time-series unveiled rich electrical switching dynamics and a phase transition between a low-conducting quiescent and highly-conducting active state. Near this phase transition, networks were found to exhibit cascades of activity with power-law distributed sizes and life-times. These scale-free avalanches are consistent with criticality, mirroring statistics of neural activity [1]. A physically motivated junction model, on a nanowire network topology, reproduced these features under constant stimuli. Even richer behaviour was found under periodic driving signals, with a controllable transition between periodic and chaotic dynamics. The role of network heterogeneity on dynamics was investigated by performing simulations on a randomly diluted lattice topology, with bond probability reduced towards the percolation threshold. In the ordered limit (regular lattice) the scale-free property of avalanches disappeared and irrespective of stimuli, only stable dynamical attractors were observed. As network disorder increased, the range of possible dynamical behaviours was maximised and the onset of critical-like, power-law distributed avalanches was observed. This suggests that the complex network structure may be crucial to the observed emergent collective dynamics of neuromorphic nanowire networks experimentally. These results may be utilised to optimise dynamics of nanowire networks for information processing applications in neuro-inspired frameworks, such as reservoir computing.
