Highly Efficient Modulation of a VO2/TiO2 Heterojunction Optoelectronic Synapse for Multifunctional Neuromorphic Computing

Abstract Optoelectronic synapses hold great promise for high-speed, efficient neuromorphic computing. However, achieving highly efficient response and relaxation modulation and multifunctionality in oxide-based optoelectronic synaptic devices remains challenging. Herein, we constructed a VO2/TiO2 heterojunction optoelectronic synapse through oxygen vacancy interface engineering. Benefiting from the trapping of photogenerated carriers by these oxygen vacancies in the dual-oxide heterostructure, the heterojunction increases the photoresponse current amplitude by approximately 2 orders of magnitude compared with that of pristine VO2, yielding an extended slow relaxation time. It enables the desired photoionization and electrical recombination under UV illumination and electrical bias, exhibiting stable bioinspired synaptic behaviors. A fully connected neural network incorporating the measured weight-update characteristics achieves 93.5% accuracy for handwritten digit recognition. Notably, the device exhibits frequency-dependent high-pass filtering for image edge extraction and sharpening. Our work provides a new strategy for multifunctional neuromorphic computing hardware based on oxide heterojunction synapses.