TiO2/Bi4Ti3O12 heterojunction optoelectronic synaptic devices for simulating associative memory and neuromorphic computation

Neuromorphic computing is regarded as an effective way to break through the bottleneck of traditional von Neumann architectures, and its hardware realization has attracted much attention. As the core component of brain-like chips, high-performance artificial synapses are the key elements to realize efficient neuromorphic computing. Based on the above, Au/TiO₂/BIT/ITO optoelectronic synaptic devices were fabricated in this study, which effectively simulate the functions of biological synapses, including short-term/long-term plasticity, paired-pulse fusibility (PPF), spike-time-dependent plasticity (STDP), and so on. Undethe stimulation of light pulses, non-volatile modulation and “learning experience” processes are successfully realized. In addition, Pavlov’s dog conditioned reflex experiment is successfully simulated using optoelectronic combination. Finally, a convolutional neural network (CNN) is constructed for the MNIST and Fashion-MNIST datasets with recognition accuracies of 96.74 % and 83.03 %, respectively. Reservoir neural networks are also constructed to evaluate the reliability of the memristor, and the recognition accuracy for the MNIST dataset reaches 92.5 %.