Theoretical and computational study of voltage-controlled terahertz synaptic devices for neuromorphic systems

We demonstrate electrically tunable terahertz artificial synapses based on split-ring plasmon resonator (SRPR) arrays coupled with Schottky diodes for neuromorphic computing and single-pixel imaging applications in the form of simulation. By applying a gate voltage bias to the Schottky diode, we can modulate the transmission of the SRPR-based synapse from 36% to 85%, enabling precise control of the synaptic weight. We employ an array of these terahertz synapses to implement a ResNet-18 neural network, achieving an accuracy of 93.4% for image classification on the CIFAR-10 dataset. Furthermore, we utilize the synaptic array as a spatial light modulator in a single-pixel terahertz imaging system, significantly reducing system complexity compared to conventional schemes. The high-speed, low-latency nature of terahertz radiation, combined with the adaptability of electrically tunable resonators, opens up new, to our knowledge, possibilities for brain-inspired computing and intelligent imaging systems.