Article ID: 19.20220440
Inspired by the idea of memristor designed in standard CMOS process and based on TSMC 0.18μm process, an analytical model describing memristive characteristics was presented. Then an adaptive image recognition hardware neural network consisting of input neuron, CMOS-based memristor array and operation and feedback neurons, was proposed and was implemented using analog circuit. The synaptic weights are adaptivevly modified according to input image signal. The simulation results indicated that our image recognition circuit overcame the bottleneck of conventional von Neumann computing architecture, and successfully completed the training and recognizing for letter images of z, v and n without the assistance of digital circuit and software. Our proposed memristor and adaptive image recognition circuit have great application in the field of image recognition due to its good compatibility with standard CMOS technology.