| In recent years,with the development of artificial intelligence,the construction of artificial neural networks has become a research hotspot.Synapses are the parts of the brain where functional connections between neurons occur,and also the key parts to transmit information in neural network.Therefore,the development of electronic devices that can simulate the function of biological synapses is an important basis for the construction of artificial neural networks.As an emerging electronic component,the memristor has potential application in the field of synaptic emulation due to its high degree of similarity with the synapses in structure and transmission mechanism.Memristors can be classified into analog and digital types according to the behavior of resistive switching.Among them,the analog memristor is widely used in the simulation of synapses rely on its continuously adjustable resistance states and non-linear transmission characteristics,and is considered as one of the candidates for artificial Synapses.The digital memristor has certain application value in achieving synaptic diversity and accurately simulating synaptic function with its large resistance change rate and fast resistive speed.However,the discrete resistance and the inability of the resistance states to be continuously adjustable with the voltage applied limit the development of digital memristor in the simulation of synapses.In order to solve the above problems,this thesis obtained continuous adjustable resistance state under the specific signal regulation based on the digital memristor with structure of Ag/AIST/a-C/Pt through the simple circuit design,further realized synaptic emulation of basic and higher-order learning functions.The detailed work is as follows:Memristive device with the structure of Ag/AIST/a-C/Pt was prepared,and electrical measurements showed that the memristor has stable resistance switching characteristics.Through adjusting the limiting current and reset voltage,multi-level resistive switching behavior is realized,which lays a foundation for the subsequent continuous adjustment of the resistance.Through connecting the Ag/AIST/a-C/Pt device in series with nMOS field effect transistor,the continuous adjustment of the resistance states was achieved by taking advantage of the effect of transistor's limiting current and the multi-stage storage characteristics of the memristor.On the basis of this circuit,we successfully achieved the spike-timing-dependent plasticity(STDP)by properly designing presynaptic and postsynaptic signalsThrough optimizing the circuit and designing the signal,we successfully simulate the associative learning behavior.Based on the signal design,the physiological process that the dependence of associative learning on stimulation time interval is further reproduced.The realization of these behaviors indicates that the device has the ability to simulate synapses,which lays a foundation for the construction of artificial neural networks in the future. |
PDF Full Text Request