The Design And Study On The Associative Learning Circuits Of The Memristive Neural Network

Currently, artificial neural networks(ANN) have been a major issue which has attracted many researchers’ attention. Its development is mostly based on the principles and rules of neural information calculus. Utilizing Von Neumann computer, the theoretical model of biological nervous system has been studied. But the real target of the ANN is, by studying the principles of biological nervous system’s intelligent behavior based on the level of nerve cells, to construct a smart machine which has essentially different computer systems with Von Neumann computer. It can be defined as the fifth-generation intelligent computers.For the realization of artificial neural networks with hardware,the magnitudes of integration and connectivity indicate that building a brain-scale system will be challenging. With the development in the field of new components, the emergence of nanoscale elements has brought a new breakthrough. Nano-crystalline silicon thin film transistor(nc-Si TFT) greatly reduces hardware’s integration of the original CMOS. At the same time it brings faster response and lower power consumption. Nc-Si TFT can greatly reduce the size of neuronal circuitry, and it also can provide the possibility for a higher degree of integration and linking. Synaptic is an important component to connect two nerves. It is responsible for the transmission of information between neurons. In order to achieve the function of nervous system memory and learning, the synapses need to have high synaptic plasticity to represent connecting efficiency. The appearance of memristor provides the basis for solving this situation since it memristor is a dynamic resistance. Its resistance can be changed with respect to the excitation voltage. Simultaneously memristor is nanoscale device which has low energy consumption, long-term memory and other features. All features signify that memristor has similar characteristics of biological synapse. Therefore, memristor can be used as the basis component of synaptic circuits whose conductance determines the synaptic weights.Firstly, the background and the development of memristor are described in detail, and the HP memristor is demonstrated in SPICE simulations. Secondly, the development of artificial neural networks and associative learning algorithm are introduced. Finally, the SPICE model of Nc-Si TFT is established. Then the simulation experiment of electrical environments is done. On this basis, for the realization of associative learning of synaptic circuits, an "Integrate-and-fire" neuron in SPICE simulations of spiking neuron circuits is established. Each component’s characteristics of the neurons circuit are analyzed, and the original Mead neuron circuits are modified. The process of generating spiking activity is demonstrated in SPICE simulations. Combined with the characteristics of TFT and memristor, the synapse connection consists of four nc-Si TFT and a memristor circuits are designed. The novel circuits of neurons and synapses which can achieve HEBB learning are proposed here. So it makes the synaptic circuitry more coincident with the real biological neural synapses. At the same time, it improves the extensibility and flexibility of the synapse circuits. The HEBB learning and average excitation rate learning included two neurons have been realized. In addition, the Pavlov experiment included multiple neurons is realized, it proves that this structure of nervous system is available. Meanwhile, in order to confirm that the neuron circuits can provide the same function of traditional comparison which can be done with logic gates circuits, the simulations of input signal coincidence detection based on self-learning are shown. The basic function’s realization of traditional circuits is very important for neural computing intelligent computer.