Research On Design Of Memristive Multilayer Neural Network And Its Application

Memristor,which was proposed by Professor Leon Chua in 1971,is the fourth basic component discovered in 2008.Its storage and calculation mergence can simulate the synapses of brain neurons,and a multi-layer neural network based on memristor can be designed and implemented,so that the brain-like neuromorphic computing will have the ability of self-learning and self-adaptation like the real brain.In this context,this thesis focuses on multilayer neural networks based on memristor and its applications,and proposed two improved memristive neuromorphic computing models: memristive cellular neural network and memristive binary neural network,and which can be used in image edge detection and classification recognition.The main work and contributions of this thesis are summarized as follows:(1)Design of memristive multilayer neural networkBased on the analysis of two existing memristive synaptic structures,an improved design method of memristive multilayer neural network is proposed based on the two memristive synaptic structures in combination with other circuit elements,and the circuit implementation mechanism is analyzed.A simple logic operation was implemented in the SPICE circuit simulation experiment,and the results proved the feasibility of this design method.(2)Memristvie cellular neural network and image edge detectionAn improved model of memristive cellular neural network was proposed,based on the synaptic structure of memristor crossbar combined with cellular neural network,and applied to image edge detection.By introducing a figure of merit to evaluate the performance of image edge detection results,the experimental results show that the merit figure of the model is significantly higher than that of other traditional edge detection operators,proving the feasibility of the model.(3)Memristive binary neural network and image classification recognitionBased on the modified design method of memristive multilayer neural network and binary neural network,an improved memristive binary neural network model is proposed and applied to image classification and recognition.Through the recognition of MINST handwritten digits and pigmented skin diseases,the experimental results show that its memory consumption and calculation speed performance are better than that of other traditional classification and recognition models with rare different in recognition rate,which proves the feasibility of the model.