The Analysis Of Memristor Model With Application Research On Synapse And Controller

With the rapid development of electronic information technology, the requirements of the electronic component’s performance and the circuit’s integration are becoming more and more strict. Conventional transistors are faced with severe challenges not only in structure size, manufacturing process but also in function, and the development of other related fields is also a bottleneck. In particular, the neural networks which need a high integration degree and a large number of information to process face the above problems as well as the PID control systems which need to change the internal parameters in real time to adapt to the numerous uncertain factors. Memristor is considered as a kind of novel passive electronic component. Its memory function makes it rather suitable for a synapse applied in neural networks, and the characteristic that resistance can be adjusted continuously provides favorable condition for regarding it as a tunable parameter to replace the fixed one inside of the PID controller. In addition, its nanoscale dimension is very helpful to reduce the size of the entire neural network and control system, and to improve the integration.Firstly, beginning with the basic memristor models, this paper introduces the physical structure, expounds the principle, derives the mathematical equations, and displays the numerical simulation results of the two common memristor models,respectively. On the basis of this, taking into account the importance of the size to memristor, we take advantage of the Monte Carlo algorithm based on probability distribution to analyze the influence from slight change of the size on the two kinds of memristor models, and then make a simple comparison on them.Secondly, Focusing on the synapse function of memristor. This thesis takes the memristor with Palladium-tungsten oxide-tungsten structure as an example, derives its work principle and mathematical equations, and points out the shortcomings of the original model through comparing with the real synaptic function. The original modelhas been improved by changing the ion diffusion term, and the new model has the typical functions of synapse such as the long-term, short-term memory, forgetting characteristic and so on, and better matches the real characteristics of memristor synapse. In addition, since the temperature will also affect the value of the ion diffusion coefficient, this paper explores the changes of the synaptic weight with the different temperatures by building a memory attenuation function which is based on temperature.It is found that the higher the temperature, the faster of the decay of synaptic weight,conversely, the slower.Thirdly, in terms of another memristive device, namely meminductor, study its synapse characteristic. Taking a common meminductor model as an example, simulating the change curve of the state variable over time. By comparing with the actual forgetting curve, it is shown that the primary meminductor model does not have real synaptic properties as expected. Then, from the matching point of view, improve the equation of the change rate of the state variable by adding some additional dynamic parameters, we found out the long term, short term memory characteristics and the experience-learning phenomenon which do not exist in the primary model, and well fit the real synapse functions.Finally, as an application, this paper takes a PID controller based on an operational amplifier as object. By derivation, it is recognized as a parameter fixed linear PID control circuit. Then, the constant resistance unit of the PID control circuit is replaced by spin-memristor, and the nonlinear memristor based PID controller whose internal parameters are adjustable in real time is built. By simulation and comparison, it shows the superiority of the PID controller. In order to apply the memristor based PID in practice, a system function is constructed to simulate the controlled object, and the Simulink model of the whole control system is set up. Good simulation results show the effectiveness of the spin-memristor PID controller.