Cellular Neural Networks Based On Hybrid Memristor/RTD Structure With Its Applications

With the rapid development of information technology, more and more information need to be processed. However, because the decrease in size of transistor approaches the utmost and all the fields depending on IC technology meet a bottleneck. The cellular neural network is a kind of nonlinear network which used widely and it although has excellent advantages in the aspects of large-scale real-time information processing and IC realization, it has the same problem. Using the advantage of memory effect of memristor and the negative differential resistance property of resonant tunneling diode, then a novel cellular neural network was proposed. The network has advantages of simple structure and flexible application, meanwhile, the memristor and the resonant tunneling diode both are the nanoscale components, so the size of circuit will be greatly reduced and the integration density of system will be significantly improved.In this paper, a detailed study of memristor, resonant tunneling diodes and cellular neural networks are provided, then the effective combination mechanism of the three is investigated and the novel cellular neural network is presented, and meanwhile the application in image processing is investigated. This paper includes four parts mainly as following:(1) The dynamic characteristics of the standard cellular neural network are introduced and the stability and fault-tolerance of cellular neural networks are analyzed, then the effectiveness and feasibility in image processing are proved. Based on the structure of the cellular neural network, we discuss the why and how to improve its performance.(2) According to the theoretical model (Chua,1971) and physical model (HP Labs,2008) of the memristor, and we found that the memristor model which is suitable for the cellular neural network, and had carried on the MATLAB simulation and SIMULINK simulation and analyzed the transient behavior and response of memristor under various incentives. On this basis, we investigate the combination mechanism of memristor and the cellular neural network.(3) According to the memory feature of memristor, a new memristive cellular neural network (type MCNN-Ⅰ) is established by replacing the intercellular connection weight with the memristor, and then the working mechanism of this network is studied. We know that the application theory of cellular neural network depends on different templates realize different functions, and the different templates represent the change of connection weights. The standard cellular neural network realizes the weights of templates by multiplier which is not only too bulky, but need to be changed when modifying the templates. However, type Ⅰ memristive cellular neural network realizes the connection weights by using nanoscale device-memristor, so when we need to replace the template, only need to change the value of memristor by changing voltage source. Thus the template of the network model is more flexible, easier to change and its applications are more diverse.(4) According to the negative differential resistance property of memristor, we replace the resistance in cellular circuit with memristor, and construct type Ⅱ memristive cellular neural network model. There is no need to add a feedback mechanism into system if memristor is blended into it, because memristor is a kind of element with negative differential resistance property. In this paper, we present the stability analysis and computer simulation for the above network, and prove its application in image processing.(5) Because the negative differential resistance characteristic of memristor is particularly clear only under the hard-switch state and it’s not an ideal element to replace the resistance in the cellular circuit but anther nanoscale element-the resonant tunnel diodes (RTD) has the ideal negative differential resistance characteristic. The basic characteristics of resonant tunneling diode are analyzed and RTD cellular neural network based on memristor is built up by utilizing negative differential resistance property of RTD and auto-memory function. This network combines the flexibility of type Ⅰ memristive cellular neural network template and the simplicity of RTD cellular neural network circuit, it is hopeful to make the size of network become smaller and its application become more flexible. After that the stability and fault-tolerance of the network are analyzed, and the image is processed by using the appropriate template, and the validity of the network for image processing is verified, which provides the theoretical foundation for the subsequent work.Finally, the work we have done in this paper is summarized and the further discussion is forecasted.