Research on image recognition using memristor crossbar

This study aims to develop a memristor-based crossbar circuit for implementing multilayer neural networks, targeting next-generation nonvolatile memory technologies, advanced logic operations and brain-inspired computing applications. The proposed system uses a single memristor array to achieve both positive and negative synaptic weights, enabling on-chip training and adaptive learning for energy-efficient, high-performance neuromorphic computing. This study proposes a memristor crossbar array implementation of multilayer neural networks, which uses a single memristor array to emulate both positive and negative synaptic weights. The system uses an adaptive back-propagation learning method to enable automatic synaptic weight adjustment and on-chip training capabilities. This study demonstrates that the proposed network achieves 95.6% accuracy in character recognition tasks, maintains strong robustness under 0%–30% noise levels and exhibits reliable tolerance to memristor drift variations up to 15%. This study highlights the potential of memristor technology for energy-efficient neuromorphic computing applications, providing important insights for next-generation Artificial Intelligence (AI) hardware design.