Research On Performance Modulation Of SrTiO₃ Memristor And Array Fabrication

With the development of artificial intelligence technology and the coming of big data era,the large-scale neural network attracts increasing attention and puts forward more stringent requirements for capability and efficiency of computing systems.However,with the end of Moore’s Law,it has been difficult for traditional silicon-based electronic devices to bring performance improvement through futher scaling down.The development of new devices is urgently needed.Memristor has the characteristics of small size,fast speed and low power consumption,and it naturally works as a synaptic device to build hardware neural network by two attractive virtues:the in-memory computation of single device and the high parallelism of computation in its crossbar array.This thesis focuses on the memristive device based on ternary transition metal oxide Sr Ti O₃.Two key parameters that affect network performance–resistive state retention and conductance modulation linearity are systematically investigated.The fabrication technology and resistive switching behaviors of crossbar array are also explorated.The main research contents are as follows:In terms of state retention,a non-volatile resistive switching with good state retention was achieved in the Pt/Sr Ti O₃/Sr Ru O₃ memristor based on oxygen vacancy conducting filament mechanisms.Noted that retention loss was generally observed in previously reported Sr Ti O₃ based device.The retention improvement in the prepared devices stems from limiting the migration of oxygen vacancy through the Sr-O rich interface layer in the bottom electrode of Sr Ru O₃ which was supported by the investigation of electrical transport mechanism,interface microstructure and resistive switching mechanisms of the devices.In the aspect of conductance modulation linearity,deposition parameters of Sr Ti O₃ thin film were modulated to adjust the conductance modulation linearity.The Sr Ti O₃ memristor with linear conductance modulation was obtained.The nolinearity of the device conductance modulation is 0.00008 and 0.049 in potentiation and habituation process respectively.And it was found that the device linearity was related to oxygen vacancy concentration in the Sr Ti O₃ thin film.On this basis,the electric performance of the prepared device was put into neural network,and the influence of conductance modulation linearity,learning rate and the number of resistance states on neural network training was analyzed.In the case of memristor array preparation,two fabrication methods based on plasma etching and Al O_x mask was studied.The advantages and disadvantages of these two methods were analyzed in detail.And the structures and resistive switching behaviors of the prepared arrays were characterized.