Flexible Ferroelectric Multistate Information Storage And Neuromorphic Computing Memristor

With the development of the Internet of Things,artificial intelligence and big data,the flexible wearable electronic devices with the advantages of light-weight,softness,and large area manufacturing have been used widely in health monitoring,humancomputer interaction,etc.,presenting the huge market prospects.The explosive growth of the data puts forward more challenging requirements on flexible information storage devices.Memristor is a nonlinear resistor with the memory function,which can achieve the non-volatile multi-state storage and has been used widely in neuromorphic computing.However,the traditional flexible memristor has the problem of poor storage stability,facing great challenges in information processing and the application of artificial synaptic devices.Therefore,the researchers have devoted to researching and developing a new type of flexible memristor.In recent years,among the varieties of emerging storage devices,the ferroelectric memristor based on the mechanism of ferroelectric polarization reversal can achieve the stable and non-volatile information storage.Therefore,the flexible ferroelectric memristors with the advantages of low power consumption,high storage density and robust flexibility can be achieved via combining the ferroelectric memristors with flexible substrates.The flexible ferroelectric memristors can be used to construct the artificial synaptic devices andrealize the neuromorphic computing,which will bring the huge prospect for the applications of new flexible information devices in artificial intelligence.In this dissertation,we will take the flexible ferroelectric memristors as the research object,and conduct the in-depth study on their multi-state information storage and the prototypes of the neuromorphic synaptic devices.This dissertation is divided into five chapters,the main contents of each chapter are summarized as follows:In chapter 1,we mainly introduced the basic definition and different physical mechanisms of the flexible memristors;focused on the memory mechanisms and the research progresses of the flexible ferroelectric memristors;outlined the research progresses of the flexible memristors in artificial synaptic devices and the neuromorphic computing;and finally,concluded the development tendency and existing problems of flexible memristors,and provided the research contents and technical routes in this dissertation.In chapter 2,the Au/BiFeO₃（BFO）/SrRuO₃（SRO）/BaTiO₃（BTO）/mica flexible ferroelectric diode heterostructure memristor is constructed.The flexible ferroelectric heterostructure has the largest ferroelectric polarization among all the reported flexible devices,good flexibility and the resistance on/off ratio.The flexible ferroelectric heterostructure has the good memristive behaviors,which can be used to simulate the fully connected deep neural network with a high recognition accuracy of images.The high-performance（111）-oriented Au/BFO（～360 nm）/SRO（～25 nm）/BTO（～20 nm）/mica flexible ferroelectric diode heterostructure was fabricated on the flexible mica substrate through an appropriate design of the SrRuO₃（SRO）/BaTiO₃（BTO）double buffer layers,achieving the largest ferroelectric saturated polarization（100μC/cm²）and remanent polarization（97 μC/cm²）among all the reported flexible ferroelectric devices,and ferroelectric polarization is very stable after 104 bending cycles under 5 mm bending radius,highlighting the good flexibility of the devices.Based on the ferroelectric polarization reversal of the BFO,the flexible ferroelectric heterostructure can achieve the memristive behaviors with continously tunable resistance,and the resistance on/off ratio of 25.Therefore,the flexible ferroelectric heterostructures can be used to construct the artificial synaptic devices,and achieve the functionality of spike-timing-dependent plasticity.Based on the realistic performance of the flexible ferroelectric memristor,a fully connected deep neural network was simulated,and a recognition accuracy of>90%on MNIST handwritten digits was obtained through online supervised learning.Our work highlights the application potential of the flexible ferroelectric memristors for non-volatile information storage and neuromorphic computing.In chapter 3,the Au/Ti/ZnO/BFO/SRO/BTO/mica flexible ferroelectric tunnel junction（FTJ）memristor is designed and constructed.The flexible FTJ has the resistance on/off ratio of 10,good retention and robust flexibility.The flexible FTJ has the robust memristive behaviors,which can be used to simulate the fully connected artificial neural network and achieve the high recognition accuracy of images.The Au/Ti/ZnO（～120 nm）/BFO（～1.2 nm）/SRO（～20 nm）/BTO（～20 nm）/mica flexible FTJ memristor was fabricated on the inorganic flexible mica substrate.Through the ferroelectric polarization reversal manipulation of the Schottky barrier,the flexible FTJ memristor can achieve the non-volatile continuously-tunable tunneling resistance,and the resistive switchings of the tunneling resistance are robust after bending 103 cycles under the bending radius of 9 mm,demonstrating the good flexibility of the flexible FTJ.Therefore,the BFO-based flexible FTJ can be used to construct artificial synaptic devices,and the robust spike-timing-dependent plasticity（STDP）was realized under bending condition.The accurate conductance manipulation with good linearity（-0.24）,high symmetry（0.18）and low cycle-to-cycle variation（1.77%）was also achieved.Based on the realistic performance of flexible FTJ memristors,a fully connected artificial neural network was simulated in computer software for recognizing the MNIST handwritten digits and F-MNIST image datasets,and the recognition accuracies are 92.8%and 86.2%,respectively,which are close to the performances for ideal memristors.These results indicated that the flexible FTJ memristor with stability,non-volatility,multi-state storage and flexibility shows the application potential in neuromorphic computing.In chapter 4,the Pt/Hf_0.5Zr_0.5O2（HZO～7.6 nm）/TiN/mica flexible ferroelectric thin-film heterostructure with robust ferroelectricity and superior flexibility is prepared.The flexible thin-film heterostructure has the large ferroelectric saturated polarization（Ps～30 μC/cm²）and remanent polarization（2Pr～35 μC/cm²）.Meanwhile,the heterostructure can achieve the robust and stable ferroelectricity after bending 104 cycles with bending radius of 6 mm,highlighting its the superior flexibility.Based on the robust ferroelectricity of HZO,the Cr0.11In1.89O₃/HZO/TiN/mica flexible ferroelectric field effect transistors（FeFET）with good non-volatile storage and switching characteristics.The flexible FeFET shows the non-volatile manipulation of the channel resistance with the on/off ratio>103 at the low operation voltage of ±2 V,and a low subthreshold swing of 102 mV/dec is also obtained.In chapter 5,the obtained research results about the flexible ferroelectric multistate information storage and neuromorphic memristors are summarized.Meanwhile,some aspects which need to be improved in this study are analyzed and the future research interests of flexible ferroelectric memristors are also prospected.