Memristive Neural Networks Based On STDP Rules With Its Applications

With the rapid development of the electronic technology, the integration of the CMOS transistor gets higher and higher; people are longing for the neuromorphic systems’more expensive development prospect. However, the neuromorphic system development meets obstacles because the size reduction of the transistor is coming to approach limit, and meantime the existing learning rules cannot solve the real-time and complex intelligent problems of the neuromorphic system, so it is more urgent that intelligent processing unit is added into the neuromorphic system. A new circuit element, memristor, and the new discovery of life science, Spike-Timing-Dependent Plasticity (STDP) learning rules, introduce new vitality into the development of the neuromorphic system. Memristor has small size and natural information storage capability, which is suitable to be the electronic synapse. On the other hand, combination STDP rules with memristive neuromorphic system and construction novel neuromorphic system will expect for characteristics of bionic intelligence and high integration, which is hoped to improve the intelligent information processing capability of the neuromorphic system.The basic theory and properties of memristor is studied thoroughly in this thesis, the memristive synapse is realized by the theoretical deduction and the memristor bridge synapse-based circuit and meanwhile the memristor-based perceptron is constructed to realize the logical classification; combined with the advantage of Chebyshev neural network in the function approximation, the memristor-based Chebyshev neural network is studied to be used for function approximation; the memristor-based Fourier network is established to be used for image restoration. The memristive neural networks are verified to own better information processing capacity and are easy to be realized by hardware circuit from the above work. Then the bionic characteristic of the artifical neural network is considered, the experiments have proved that the STDP in biology can be effectively realized by the memristive synapse, the memristive neural network of different structure is constructed, the STDP-based memristive neural network is presented to be used for image storage, and its applications in the binary, gray and color image storages are discussed. Finally, the memristive bridge neural network based on STDP is designed according to the validity of the neural network combined with STDP and memristor, the advantage of cellular neural network in image processing is considered, the different weight template is constructed and the different image processing function is realized. The new type of neural network has the simpler structure and the memristive synapse weight is very easy to be adjusted, and then the image processing of the different template is realized. The research of this thesis will promote the development of neuromorphic system, and it is also hoped to completely change traditional information processing mode and design new extra-high integration and bionic intelligent machines.