Two-dimensional materials offer potential for developing optoelectronic memristors and neuromorphic visual systems, benefitting from their atomically smooth surfaces and tailorable optoelectronic characteristics. However, the rapid recombination of photogenerated carriers within the two-dimensional materials-based optoelectronic memristors degrades the linearity of conductance updates, impairing the image recognition accuracy of neuromorphic visual systems. This work introduces a defective h-BN interlayer into HfS2-based optoelectronic memristors to improve nonvolatile conductance modulation. The fabricated HfS2/h-BN heterojunction optoelectronic memristor exhibits linear conductance tuning and emulates key synaptic functions, including excitatory postsynaptic current, paired-pulse facilitation, and transition from short-term to long-term plasticity. When used in an artificial neural network for image recognition, the device achieves an accuracy of 94%, higher than the 51% obtained with a pure HfS2 optoelectronic memristor. Mechanism studies reveal that the enhancement arises from efficient trapping and release of photogenerated carriers at the HfS2/h-BN interface. This work reveals that van der Waals heterojunction-interface defects-mediated carrier dynamics enable linear conductance modulation, offering a physical design principle for high-performance neuromorphic vision systems based on two-dimensional materials.
Meng et al. (Mon,) studied this question.