The Fundamental Characteristics Of Memristor And Study Of Biological Synapse Based On Memristor

Large scale integrated circuit enters a rapid development period with the advent of intelligence information age. However, the development of IC is restricted by the limit of transistor sizes, and the thermal problem induced by the higher density of chips. Thus, the discovery and application of a new circuit element is urgent. Memristor has attracted much attention due to its simple structure, small size, low power, ease of integration, and compatible with traditional CMOS process. Synapse is an important component of neural network, and also the key in the simulation of the neural network. Simulation of conventional synapse adopts transistor devices, limited by the size and function of transistor, the density of traditional synapse is much lower than the density of synapse in biological neural network. Memristor with a size of nanometer scale can work like synapse, and its function is most analogous to that of a biological synapse, which provides the basis for constructing artificial neutral network. The use of memristive synapse can realize the simulation of the artificial neutral network with approximate density comparable to the biological neutral network, achieving low-power, highly integrated analogous neutral network.In this paper, the basic theory, physical mechanism, and the model of memristor have been investigated. Memristor plays an important role as the fourth basic circuit element in developing the electric circuit industry. The application of memristor in the field of non-volatile storage, analog circuit, digital circuit, and artificial neutral network is analyzed and then an electrostatic protection circuit based on memristor is proposed. Based on the analysis of the biological synapse structure, operating principle, important features, the memristor is applied to simulate synapse. Spike-Timing-Dependent Plasticity(STDP) learning rules of synapse is achieved with weight adjustment circuit. The neural network consist of synapse and neuron is further simulated. In order to realize the associate memory function, the model consist of three neurons and two synapses is applied to simulate Pavlov’s dog. Synaptic weight adjustment circuit and associative memory circuit are designed in PCB level. Resistive characteristics of Ni O and Hf O2 memristors are tested by Keithley 4200-SCS semiconductor parameter tester. The simulation and test results of the circuit included Ni O memristor confirm the completion of the required functions. In this paper, the simulation of synapse and neural network can be achieved through hardware-circuit based on memristor, rather than stay in the simulation, which provides a new method for building a more complex neural network, and is expected to further achieve ultra-high density of neural network in the near future.