The Model Of Cellular Neural Networks Based On Memristor And Its Applications

At the end of1990s, sensor technology revolution promotes the development of electronic and computer industry. Thousands and millions of analog signals obtained by sensors need to be processed. So a new computation device is expected to implement these tasks. Analogic cellular computer paradigm may be the acceptable and effective candidate. The core of this computer is a cellular nonlinear/neural network (CNN), an array of analog dynamic processor or cell, which is invented by Leon O. Chua and Lin Yang in1988. CNN can accept and generate analog continuous signals. However, cellular neural networks have complex dynamical characteristics which require higher performance of CMOS circuit elements. In addition, with the cost of intensive lithographic technology increasing and the devices having their own physical defects, complementary metal-oxide semiconductor (CMOS) technology will reach the limit of development and Moore’s Law will meet great challenge. The development of cellular neural networks based on CMOS technology also will reach its development bottleneck. All of these factors restraint the development and application of cellular neural networks. Fortunately, The theory discovering and physical implementation of memristor have brought a novel promising hope for the development of cellular neural networks. Memristor is a kind of nonlinear, passive and nano-scale device which has similar characteristics of synapse and memory characteristics. Employing the characteristics of memristor, memristor-based cellular neural networks (MCNN) are proposed, which have more simple structures, flexible weight value templates, higher integration level.In this thesis, we study the structures and operating principles of cell cellular neural networks and cellular neural networks universal machine and analyze the characteristics of memristor. Based on these work, new cellular neural networks with memristive templates and universal machine with memristive storage pattern are proposed. Moreover, it argues the applications of memristive cellular neural networks in color image processing and pseudo-random bit generator. The main work can be given as the following sections:(1) We introduced the structure and the operating principle of standard CNN, and analyzed its mathematical model and dynamic characteristics as well. Meanwhile, the structures and operating principles in analogous and logic signals processing of cellular neural network universal machine are also studied. Based on the features of two systems and the problems they may met with, we discussed how to improve their performance.(2) The physical and mathematical model and the basic characteristics of HP memristor are introduced, the transient behavior of memristor is simulated with different signals are given as external inputs. All of these work have laid the groundwork for further research.(3) The memristance changes with memristor’s external input signals. Making use of this characteristic, we designed memristive templates and memristive cellular neural network. This new network has faster processing ability, less energy consumption and higher degree of integration. Based on the formal work, employing memristor’s natural advantage in analogous signals storage and processing, we proposed cellular neural network universal machine based on memristive storage model.(4) The principles and the main applications of cellular neural networks in image processing are discussed. The effectiveness of memrsitive cellular neural networks in image processing of license plate recognition is confirmed by a series of computer simulations. Furthermore, after introducting the principle of pseudo-random bit generator based on chaotic circuit, we proposed a pseudo-random bit generator based on chaotic memristive cellular neural network system.Finally, we summarize the content of this paper and prospect the further research and practicing.