All-in-One Metal-Oxide Heterojunction Artificial Synapses for Visual Sensory and Neuromorphic Computing Systems

ABSTRACT

An all-in-one artificial synapse integrating central nervous and sensory nervous functions utilizing low-dimensional metal-oxide heterojunction is demonstrated in this work. With an ion-electrolyte gate, synaptic emulations modulated by electrical and photonic stimulus have been integrated into one high-performance three-terminal artificial synapse. Various long-term and short-term synaptic plasticity functions have been achieved by altering the electrolyte-gate stimulus amplitude/width/frequency/number. The linear potentiation behavior and maintained states enable artificial synapses for neuromorphic computing. Simulated artificial neural network based on the artificial synapses achieved Covid-19 chest image recognition (>85%). The photo-sensing metal-oxide heterojunction enables the synaptic functions mimicking the biological visual sensory functions responding to optical and UV stimulus. Photonic synaptic plasticity modulations responding to photonic stimulus wavelength/power/width/number are investigated, and short-term/long-term synaptic plasticity transition was achieved. Dual-mode modulation combining photonic stimulus and gate stimulus was examined, along with a ‘AND Gate’ demonstration with electrical and photonic inputs. Finally, an artificial neural network was demonstrated based on the synapses with dual-mode synaptic weight modulation, indicating the potential of the artificial synapse for compact artificial intelligence systems combing neuromorphic computing and visual sensory nervous functions.