Al Nanoparticle-Decorated Metal Oxide Synaptic Transistors for Ultralow-Energy Neuromorphic Computing with Wide...

Achieving ultralow energy consumption alongside high synaptic fidelity remains a key challenge in the development of practical and scalable neuromorphic hardware systems. Electrolyte-gated memtransistors (EGMTs), which enable low-voltage analog switching via electric double layer modulation, suffer from a fundamental trade-off between dynamic range and energy consumption. Here, a nanoparticle-engineered EGMT is reported that mitigates this limitation by incorporating aluminum nanoparticles at the interface between a solution-processed indium gallium zinc oxide channel and a solid polymer electrolyte composed of polyethylene oxide doped with lithium hexafluoroarsenate. This design yields 50 discrete conductance states at a drain voltage of 1 mV, achieving a dynamic range exceeding 78 and a synaptic switching energy of 0.62 pJ spike^-1, which ranks among the lowest reported for EGMTs. Neural network simulations (784  ×  60  ×  10), based on experimentally extracted conductance updates, predict energy savings of 99.7% during training and 91.4% during inference compared to digital complementary metal-oxide-semiconductor implementations.