Research Of True Random Number Generators Based On The Conductive-Bridge Threshold Switching Memristors

In the era of the Internet of Everything,true random number generators（TRNGs）based on emerging devices are expected to become the basic security component of edge devices,due to its high integration,low power consumption and intrinsic stochastic parameters.As a volatile device,conductive-bridge threshold switching memristors（CBTSM）avoid additional reset operations and can’t be easily physical-cracked,have great potential in the field of information security.However,the existing TRNGs based on CBTSM are still in their early stages,suffer from problems like massive penetration of metal particles,additional calibration and low throughput,which limit their further development.This dissertation starts from the optimization of CBTSM through process and structure improvement to provide reliable and stable device for TRNG.Then,aiming at the problems in the existing research,two TRNG with calibration-free characteristics and high throughput were proposed and verified respectively,which provided guidance for the optimal design of TRNG.The main research contents of this dissertation are as follows:Firstly,aiming at the unreliable device characteristics caused by metal particle penetration,based on structure and process improvements,two device design schemes were proposed and verified:in terms of structure,Cu/Ge Te/Al₂O₃/Pt devices were fabricated by stacking fast and slow ionic materials.The Ge Te layer acts as the buffer layer to solve the low operation current and poor endurance of single-layer devices,increasing the maximum compliance current by 5times;by changing its thickness of the Al₂O₃ layer,the volatile and non-volatile characteristics of the device can be modulated.Based on this scheme,the Ag/Ti N/Hf O_x/Hf O_y/Hf O_x/Pt device with three-layer hafnium oxide stack structure was further developed,which has a faster switching speed（60 ns/500 ns）.In terms of technology,through the via-hole structure and the electron beam evaporation,a simple three-layer Ag（EBE）/Hf Ox/Pt device has been fabricated and has a high on-state current density（1.27 MA/cm²）and a large operating current（800μA）.Secondly,in view of the problem that traditional TRNG schemes are susceptible to parameter distribution shifts and require additional probability calibration,a calibration-free TRNG based on threshold voltage fluctuations of Ag（EBE）/Hf O_x/Pt devices is proposed and verified:first,the tolerance of the traditional TRNG scheme based on the reference value to the parameter distribution shifts is analyzed.It can be found that a large deviation of 48%the probability will occur under a small distribution shift of 5%.Then,based on the principle of equal division method,a TRNG scheme with calibration-free characteristics is designed.The probability of this TRNG output will only have a small range of fluctuations（<10%）near the standard value and have good data independence（Chi-square value=0.5935）,when the distribution shifted occurs between cycles and devices.the outputs.According to the simulation evaluation,even under distribution shift of 50%,the TRNG output can still maintain great randomness（Shannon entropy=0.99998）,which further confirms the calibration-free characteristics of this TRNG.Finally,in order to solve the problem of large probability error in the existing stochastic computing scheme,a novel scheme that can precisely control the probability is proposed.The simulation results show that the probability error can be less than 0.6%,and the calculation result error will be less than 0.3%.At last,aiming at the problem of low throughput of existing CBTSM-based TRNG schemes,a high-speed TRNG scheme based on Ag/Ti N/Hf O_x/Hf O_y/Hf O_x/Pt device is proposed and verified:the random characteristics of the integration-and-firing process was studied and verified,and a true random number generator scheme with good randomness（passed 9 NIST tests）and data independence（Chi-square value=0.1035）has been achieved.The feasibility of the scheme was verified through the breadboard circuit.The optimal throughput of this TRNG is～108 kb/s due to the microscopic entropy source and the high switching speed of the device.In order to further improve the throughput,a new TRNG acceleration scheme combined with a linear feedback shift register is proposed,which can further increase the TRNG throughput by more than 10 times.The 1 Mb data obtained based on this scheme can pass all 15 NIST tests,and can realize the calculation ofthrough Monte Carlo simulation with an error of less than 0.8%.In conclusion,this dissertation provides a solution to the key problems like excess metal particle penetration,the need for additional calibration,and low throughput in existing studies,improving the performance of CBTSM-based TRNG and accelerating their further practicality.This process paves a new way for information security of the device under edge computing.