Investigations On Ferroelectric Heterostructures And Imformation Storage Prototype Devices

In the era of Industry 4.0,the rapid developments of artificial intelligence,big data,Internet of Things,and cloud computing require advanced information storage and processing technologies.However,China has encountered "bottleneck" problems in advanced memory chips.Besides,conventional computing hardware has to face "von Neumann bottleneck".Extra energy consumption and time latency are generated by the data transmission between the memory and the processer.Thus,to break through the bottlenecks,on the one hand,it is necessary to improve performances of traditional memory devices by introducing new materials and physical mechanisms.On the other hand,new types of nonvolatile memory devices with a fast operation speed,low energy consumption,multistates and memristive behaviour need to be realized for neural network computing,so as to deal with the "von Neumann bottleneck" problems.Among a variety of emerging memory,ferroelectric tunnel junctions(FTJ)have attracted much attention,because of advantages of high speed,low energy consumption and multistate resistive switching.The FTJs are promising for developing high performance nonvolatile memories and electrical synapses,which is significant to construct neural network computing and break through the "von Neumann bottleneck".Furthermore,the doped HfO2 ferroelectric materials show high dielectric constants and low leakages,promising for advanced dynamic random access memory(DRAM)capacitor materials.This dissertation is divided into six chapters,and the main contents of each chapter are summarized as follows:In Chapter 1,we introduce the developments and challenges of emerging memories including memristors.The focus includes the following two aspects.First,the resistive switching mechanisms,designs of various electrical synapse and applications of FTJs in neural network computing are described.Second,the design methods for morphotropic phase boundary(MPB)in doped HfO2 ferroelectric films and application in DRAM devices are discussed.In Chapter 2,the preparation processes of the Ag/PbZr0.52Ti0.48O3/Nb:SrTiO3 ferroelectric tunnel junctions and TiN/Al2O3/Hf0.5Zr0.5O2/Al2O3/TiN ferroelectric capacitors are described in detail.The working principles of various characterization techniques on crystal structure and ferroelectric domain are introduced.The design scheme for the ultrafast transport property test circuit is described.These contents are the foundations of the following experimental investigations.In Chapter 3,based on the design of band structure and crystalline orientation,we constructed a high-quality Ag/PbZr0.52Ti0.48O3(1.2 nm,(111)-orientation)/Nb:SrTiO3 ferroelectric tunnel junction as an emerging nonvolatile memory.The PbZr0.52Ti0.48O3 with a low coercive field and the Ag with a low work function were selected as the barrier layer and top electrode,respectively.Therefore,the operation voltage of the device was reduced obviously.Subnanosecond(630 ps)multistate switching can be achieved by applying an operation voltage of less than 5 V.And,the fastest resistance switching speed of 300 ps in FTJs is obtained.In addition,the device shows a lot of advantages,including high storage density with 4 bits nonvolatile resistance states,low operating energy consumption(5.3 fJ/bit),high endurance(109),long retention and good scalability.It shows excellent comprehensive performance as a nonvolatile memory device.In Chapter 4,by designing the multidomain metastable state in PbZr0.52Ti0.48O3,the Ag/PbZr0.52Ti0.48O3(1.2 nm,(111)-orientation)/Nb:SrTiO3 ferroelectric tunnel junction memristor shows high performance in terms of conductance manipulation with 256 states,good linearity(nonlinearities are 0.77 for potentiation and-0.94 for depression,respectively),and low cycle-to-cycle variation(2.06%)under pulsed voltages of 10 ns.As an electrical synapse,the FTJ meets all the target specifications.When the pulse duration decreases to 630 ps,as many as 150 states can still be achieved.Based on the performance of FTJ electrical synapse,the two-layer fully connected and ResNet-18 neural networks were simulated.The recognition accuracy based on 256 states is almost the same as the result based on floating point software,and the high accuracy is robust when recognizing noisy pictures.These results show the great potential of ferroelectric tunnel junctions as electrical synapses in building ultrafast neural network computing hardware systems.In Chapter 5,for the development of high-performance DRAM(volatile)capacitors,the MPB structures and dielectric properties of Hf1-xZrxO2 ferroelectric films are investigated.By manipulating the Zr content,the crystal and electronic structures,as well as the dielectric and ferroelectric properties of HZO films,were systematically analyzed.The results reveal that the MPB is located at a composition of x～0.5 in～6.5 nm-thick HZO films.Based on this,by inserting Al2O3 interlayers and designing heating treatments and electric field cycling processes,the leakage current was reduced,and the dielectric constant was improved.The measurement results of 10 samples show a high average dielectric constant of～46.7(device-to-device variation<1%),a low average equivalent oxide thickness(EOT)of～0.51 nm,and a low average leakage current density of<1×10-7 A/cm2 at～±0.5 V.In addition,the high dielectric constants show good stability with operating voltages and cycling times.These advantages will support the application of HfO2-based ferroelectric materials in advanced DRAM capacitors.In Chapter 6,we summarize the work of high-performance PZT ferroelectric tunnel junction-based memristors for electrical synapses and neural network computing simulations,and high dielectric materials based on HZO films for DRAM capacitors.In view of the existing problems for current investigations,we look forward to the future development of FTJ and their application prospects in future neural network hardware systems and ferroelectric heterostructures in realizing 3D DRAM capacitors.