Harnessing ion tuning mechanisms for neuromorphic computing: from artificial synapses to dynamically reconfigurable architectures

Analog neuromorphic computing hardware, despite their application-specific energy efficiency, are not easily reconfigured for generality, thus impeding them from competing with entrenched general purpose digital processors. Here we address this fundamental limitation by enabling a single neuromorphic component to be functionally reconfigured to express neuronal, synaptic, interconnect and switching behaviors. Via precise voltage-controlled on-chip injection of oxygen vacancy defects into VO2, a Mott insulator, we nucleate and stabilize phase coexistence across various oxides of vanadium, much like spinodal decomposition of a water-oil mixture, and tune the overall phase transition properties. Such phase coexistence is not possible in purely electronic components, and has not been achieved electrochemically under normal chip operating conditions. Using electrochemically controlled phase coexistence, we demonstrate both functionally tunable computing, such as reconfigurable logic gates, and also unusually stable information retention, with 1% loss over 14 years in ambient conditions. On-chip defect-tuned phase coexistence paves the path for functionally dense and dynamically reconfig