Investigation Of The Novel La₂Ti₂O₇ Ferroelectric Memristor For Neural Computing

As the size of CMOS devices continues to shrink and approaches their physical limits,the continuation of Moore’s Law faces serious challenges.At the same time,the development trend of technologies such as big data and deep learning has put forward higher requirements for the computational power of traditional mainstream hardware platforms.In traditional von Neumann computing structures,computing cells and storage cells are separated,and the transmission loss between them leads to the von Neumann bottleneck.The human brain relies on neurons and synapses for efficient storage and computation,combining storage,and computation.Therefore,simulating neural computing in the human brain has become an effective way to solve current power consumption and computational power problems.Memristors are expected to become the core hardware units in neural computing due to their low power consumption,high integration,and ability to simulate synaptic plasticity.Based on the advantages of low power consumption and strong stability of ferroelectric memristors,the novel ferroelectric memristor based on lanthanum titanate（La₂Ti₂O₇）has been prepared in this paper.The performance of the device has been optimized by the adjustment of the structure of La₂Ti₂O₇based device.At the same time,the ability of the device to simulate synaptic plasticity has also been deeply studied.The specific work is as follows:1.Many ferroelectric materials have been investigated and it is found that LTO materials have excellent ferroelectricity,but few people use them for the preparation of memristor at present.New ferroelectric memristor based on LTO materials needs to be studied urgently,so LTO materials have been selected as the resistance layer of devices.To explore the substrate compatibility and film preparation process of LTO,LTO films were deposited on silicon wafers by Magnetron sputtering,and the films were characterized by AFM and PFM.2.Firstly,Ag/LTO/Pt devices were prepared by magnetron sputtering,and DC electrical tests were conducted on the devices.It was found that the devices exhibited both unipolar and bipolar resistance characteristics under different voltage ranges of scanning.Then,the Ag/LTO-Ag/Pt device was prepared by improving the thickness of the intermediate layer of the device and adding Ag during sputtering of the LTO intermediate layer.The device was subjected to the same DC electrical testing.Compared to the previous case,the device exhibited lower switching voltage,higher switching ratio,and multiple resistance state characteristics.Finally,the mechanism fitting of the I-V characteristic curve during Set process was performed,and the conduction mechanism of the device was verified through first principles calculations.3.The ability of Ag/LTO/Pt devices to simulate synaptic plasticity and the potential of devices as image recognition hardware units was studied intensively.On the one hand,the conductance modulability of the device was successfully realized through DC and different pulse tests.On this basis,the synaptic plasticity behaviors such as PPF,STP,LTP,STP to LTP transition,and STDP were simulated.On the other hand,on the basis of simulating synaptic plasticity,the device parameters were fitted with MATLAB,and the application potential of the device in image recognition was discussed with convolutional neural network.The recognition accuracy after learning is up to 96%.