Synaptic learning behavior and neuromorphic computing of Au/MXene/NiO/FTO artificial synapse

A traditional von Neumann structure cannot adapt to the rapid development of artificial intelligence. To solve this issue, memristors have emerged as the preferred devices for simulating synaptic behavior and enabling neural morphological computations. In this work, Au/NiO/FTO and Au/MXene/NiO/FTO heterojunction memristors were prepared on FTO/glass by a sol-gel method. A comparative analysis was carried out to investigate the changes in electrical properties and synaptic behavior of the memristors upon the addition of MXene films. Au/MXene/NiO/FTO artificial synapses not only have smaller threshold voltage, larger switching ratio, and more intermediate conductivity states but also can simulate important synaptic behavior. The results show that the Au/MXene/NiO/FTO heterojunction memristor has better weight update linearity and excellent conductivity modulation behavior in addition to long data retention time characteristics. Utilizing a convolutional neural network architecture, the recognition accuracy of the MNIST and Fashion-MNIST datasets was improved to 96.8% and 81.7%, respectively, through the implementation of improved random adaptive algorithms. These results provide a feasible approach for combining MXene materials with metal oxides to prepare artificial synapses for the implementation of neuromorphic computing.

Automated summary

MXene materials combined with metal oxides were utilized to prepare Au/MXene/NiO/FTO heterojunction memristors. These memristors exhibit improved electrical properties and synaptic behavior simulation compared to traditional von Neumann structures. Furthermore, they have better weight update linearity and conductivity modulation behavior, as well as long data retention time characteristics. By implementing improved random adaptive algorithms, the recognition accuracy of the convolutional neural network using these memristors was significantly improved on the MNIST and Fashion-MNIST datasets.