Monolithically Integrated IGZO Synaptic Transistors on a Silicon Nitride Waveguide for Optoelectronic Neuromorphic Computing

We report the first demonstration of monolithically integrated amorphous indium–gallium–zinc oxide (IGZO) synaptic transistors on a silicon nitride (Si3N4) waveguide, enabling direct on-chip optical programming by using guided 532 nm light. In this architecture, the evanescent field of the guided mode modulates the IGZO channel conductance, while the three-terminal transistor configuration enables selective electrical access and readout of the programmed synaptic states. Unlike previously reported IGZO optoelectronic synapses that rely on externally aligned free-space illumination, the waveguide serves as a lithographically defined optical programming path through which a single guided optical signal interacts with multiple synaptic transistors distributed along the same waveguide. The monolithically integrated guided optical input and electrical output configuration therefore provides a hybrid interface between integrated photonic signal delivery and transistor-based neuromorphic hardware. The waveguide geometry was designed to maintain evanescent coupling to the transistor channel while limiting propagation loss, and guided-light-induced modulation was experimentally observed in five transistors integrated along one waveguide. The IGZO TFT additionally exhibited photoresponse under free-space illumination from 405 to 635 nm, confirming the visible range photosensitivity of the active layer. Under pulsed guided-light excitation, the devices exhibited persistent photoconductivity, paired-pulse facilitation, and tunable short- and long-term synaptic plasticity. An ANN simulation using experimentally extracted conductance update characteristics achieved a handwritten digit classification accuracy of 98.02%, demonstrating the applicability of the measured synaptic response to neuromorphic learning tasks. These results establish a monolithically integrated waveguide–IGZO synaptic platform that combines guided optical programming, persistent electronic weight storage, and device-selective electrical readout for optoelectronic neuromorphic computing.