Molybdenum Nano-Island Embedded Optoelectronic RRAM With Enhanced Linear Synaptic Plasticity for Neuromorphic Visual Perception

초록

Here, we develop optoelectronic resistive memory based on semiconducting nitrogen-doped indium gallium zinc oxide (IGZO) and thermally annealed molybdenum nanoislands. By engineering the Mo/MoOx/IGZO(N) interface, visible-light absorption is enhanced through optically induced ionization of oxygen vacancies, enabling reliable linear conductance modulation under optical stimuli similar to those encountered in biological vision. The existence of a large number of oxygen vacancies at the Mo/MoOx interface and in bulk MoOx, along with the presence of a valence band offset between n+Si and MoOx, increases the relaxation time of ionized oxygen vacancies. The devices operate at low voltage and low optical intensity while exhibiting key synaptic characteristics such as intensity-dependent plasticity, spike-number-dependent plasticity, and spike-rate-dependent plasticity. Frequency-dependent current response supports dynamic learning and forgetting behavior of the memristor, which is essential for neuromorphic visual processing. Device arrays further demonstrate synaptic functions, including visual image memory, Pavlovian associative learning, and Morse code implementation. Using the memristor-based reservoir computing system and incorporated into a fully connected convolutional neural network using experimentally obtained short-term memory, the system achieves recognition accuracy above 85% for handwritten digits. These CMOS-compatible artificial optoelectronic synaptic devices offer strong potential for constructing a hardware foundation for neuromorphic vision systems and next-generation intelligent human-machine interfaces.

키워드