Heterostimuli chemo-modulation of neuromorphic nanocomposites for time-, power-, and data-efficient machine learning

Heterostimuli-modulated neuromorphic devices were created to emulate chemo-modulated biological associative learning with concurrent volume and wiring transmission. The zinc oxide (ZnO)/polyvinylpyrrolidone nanocomposites sandwiched by silver and indium tin oxide were used to mimic synapse wiring transmission through electrically induced resistivity switching. Broad photostimuli to emulate volume transmission and photoactivated ZnO provided additional modulation of the electric conductive path for synaptic weight adjustment through photo-/electric-associated redox chemistry. The resultant associative learning memristor (ALM) demonstrated rapid learning (1 ms), reliable memory operations, and extended memory retention (>1 day), with a learning efficiency 1,000 times better than the prior ones. A 5 × 5 crossbar of ALM was incorporated into an artificial neural network (ANN) algorithm, demonstrating data-efficient machine learning with 90% accuracy in small training datasets, while conventional ANN shows 76% accuracy. This biomimic network is also >1,000 times more power-efficient than existing models, indicating time-/power-/data-efficient neuromorphic artificial intelligence.