Resistive Modulation Of Ferroelectric Memristor Based On BaTiO₃ And Its Application In Neuromorphic Computin

Neuromorphic computing has a great possibility to replace the traditional computer system and become a new generation of computer system,and neuromorphic computing has a unique advantage in promoting the development of artificial intelligence.Due to its unique resistive characteristics,memristor is expected to become the hardware basis of neuromorphic computing.At present,various types of memristor have been widely studied.Among them,ferroelectric memristor has made a series of research progress and shown certain advantages and characteristics.In this work,we prepared two BaTiO₃-based ferroelectric memristors by physical vapor deposition（including pulsed laser deposition technology,magnetron sputtering technology and vacuum thermal evaporation technology）:Pt/BaTiO₃/（La,Sr）MnO₃and Au/BaTiO₃/Nb:SrTiO₃memristors are designed to develop memristors with excellent resistive properties and high performance neuromorphic computing applications.We mainly studied the film properties,resistive properties,synaptic plasticity simulation and artificial neural network image recognition of these two memristors.Its main contents are as follows:Part Ⅰ:This part is the performance study of Pt/BaTiO₃/（La,Sr）MnO₃memristor.First of all,the surface morphology of BaTiO₃film is very flat,and the roughness is～400 pm.The ferroelectric domain inversion image measured by PFM has clear domain boundaries,and the phase difference of hysteresis loop can be 180°,so it has good ferroelectric property.The resistive characteristic of the device is that the resistance increases under positive voltage and decreases under negative voltage.Then,the device successfully realizes the simulation of long-term plasticity,short-term plasticity,excitatory postsynaptic current,double pulse facilitation,pulse frequency dependent plasticity and pulse time dependent plasticity under different voltage amplitudes and voltage pulse widths.Meanwhile,we use this device to complete the experimental"Pavlov’s Dog"simulation.Finally,we combined the continuous resistance modulation data measured by this device with the artificial neural network to obtain high precision handwritten digital image recognition results.The recognition accuracy of 68（8×8）pixels is 96.1%,and the recognition accuracy of 784（28×28）pixels is 96.7%.The main resistive mechanisms are ohmic conduction,space charge-limited current（SCLC）conduction and Pool-Frenkel（P-F）emission conduction.Part Ⅱ:This part is the study of the performance of Au/BaTiO₃/Nb:SrTiO₃memristor.The ferroelectric domain flip image of BaTiO₃thin film also has clear domain boundaries,and the phase of hysteresis loop can be different by 180°,so it has good ferroelectric property.The device has a very large resistive modulation range,and the ratio of high and low resistive states can reach more than four orders of magnitude.The resistive characteristics show that the resistance decreases under positive voltage and increases under negative voltage.Then the device successfully simulated long-term plasticity,short-term plasticity,excitatory postsynaptic current,double pulse facilitation,pulse frequency dependent plasticity and pulse time dependent plasticity.Finally,the recognition results of handwritten digital image combined with the continuous resistance modulation data and artificial neural network are as follows:the recognition accuracy of68 pixels is 95.0%,784 pixels is 92.0%.The main resistance mechanism is Schottky emission conduction mechanism.