Preparation Of Silicon-based Memristor And Research On Its Resistance Switching Mechanis

With the rapid development of artificial intelligence,the explosion of data information has caused the so-called"von Neumann bottleneck"in data transmission.The development of new devices and computing architectures beyond the Moore’s era has become an inevitable trend to break through the chip computing power bottleneck in the field of technology.As a promising non-volatile storage material with outstanding properties such as high-speed read and write,low power consumption,and high-density integration,resistive random access memory（RRAM）can be used in the field of neuromorphic computing to simulate the neural morphology.Two-dimensional（2D）materials,due to their atomic-level ultra-thin thickness,low power consumption,small volume,and excellent flexibility,have become a research hotspot.Conductive filaments can be confined to a limited atomic layer region,and their atomic thickness allows for low operating voltage,ultra-low power consumption,and fast switching speed.However,the preparation of 2D materials usually requires complex deposition techniques.For devices relying on the Electrochemical Mechanism（ECM）of active metal conductive filaments,their resistive switching mechanism is based on external ions.The continuous inflow and outflow of these external ions may cause changes in the dielectric film and even irreversible damage,resulting in conductivity fluctuations and drift.In contrast,considering the full compatibility of silicon-based RRAM materials with traditional CMOS technology and the larger activation energy of nitrogen and oxygen vacancies,the intrinsic resistive switching based on nitrogen and oxygen ion migration can show better compatibility and long-term reliability.Therefore,this paper aims to prepare high-performance and multifunctional silicon-based RRAMs and explore their applications in the field of neuromorphic computing from multiple perspectives,including intrinsic resistive switching mechanisms,device structures,preparation processes,and performance modulation.The related research achievements are as follows:Au/Si N_x/p⁺⁺Si structured memristor was fabricated using magnetron sputtering and thermal evaporation techniques.The electrical switching characteristics of the Si N_xmemristor were investigated at different temperatures.The electrical property testing results indicated that the Si N_xdielectric layer exhibited more defects at room temperature,and the memristor’s electrical switching function gradually disappeared with an increase in the number of cycles.The device was difficult to reset to the high-resistance state after entering the low-resistance state.As the temperature increased,the defects in the dielectric layer improved,and the switching performance of the device significantly increased with temperature.The device exhibited distinct resistance switching properties,and the preliminary research on the device’s conduction mechanism indicated that it belonged to the nitrogen vacancy conduction filament type.The carrier transport mechanism in the high-resistance state and low-resistance state conformed to the space charge limited current（SCLC）mechanism.Sn/Si O_x/p⁺⁺Si structured memristor was fabricated using electron-beam thermal evaporation and photolithography techniques to create lattice and vertical cross-grid patterns.The ultra-thin two-dimensional（2D）Si O_xmemristor with a non-layered and amorphous structure was used as the dielectric layer,which was formed through natural oxidation at room temperature.This 2D-like memristor achieved an electrically stable and gradually variable resistance switching,and small devices fabricated using photolithography exhibited similar resistive switching behavior.Moreover,the Si O_x-based device achieved low switching variability（3.7%）,nanosecond-level switching speed（<15 ns）,good durability（>10⁶cycles）,and high retention characteristics（>10³s at 85°C）.The resistive switching mechanism in different resistance states was analyzed in-depth using X-ray photoelectron spectroscopy（XPS）,and it was confirmed that the resistive switching behavior was related to the migration of oxygen vacancies under an electric field,while the Sn electrode acted as an oxygen storage layer.In addition,the Sn/Si O_x/p⁺⁺Si device exhibits low variability in its resistive switching behavior.By continuously tuning the conductivity of the Si O_xlayer,synaptic plasticity behavior of biological neurons was mimicked,and the synaptic function was emulated.A convolutional neural network（CNN）based on Si O_x-based memristor synapses was created using the Py Charm programming environment and the Tensor Flow framework.In the CNN simulation,the non-ideal state with a variation coefficient as low as 1.3%was taken into consideration.The simulation results of the MNIST image classifier based on ultra-thin Si O_xmemristors showed a recognition accuracy of up to 98%,indicating its potential in implementing non-layered 2D material neural network inference accelerators.