Research On Memristor-based Hamming Distance Calculation Methods And Realization

In the context of the era of big data,due to Von Neumann bottleneck and storage wall and some other issues,Von Neumann-based computers can no longer meet the explosive growth of data computing needs.In order to further improve calculation efficiency,in-memory computing technology is proposed in which information storage and processing are combined.In-memory computing technology is an important method to break through Von Neumann bottleneck.Memristor will be the basic candidate of the in-memory computing cause its high speed,low power consumption,simple structure,high endurance,easy to integrate and compatible with CMOS process.Under the background of in-memory computing technology,this article mainly studies two Hamming Distance calculation methods:voltage input and voltage-resistance mixed input methods.The fabrication and modification of memristors,the design and optimization of Hamming Distance calculation method have been studied.In addition,the applications of two Hamming Distance calculation methods in the field of in-memory computing are studied.This research lays the foundation for the practical application of in-memory computing technology.The main research contents of this paper are as follows:(1)W/HfAl_xO_y/Pt memristors are deposited.The devices exhibit high switching speed(100ns),low Set voltage(1 V)and Reset voltage(-1.1 V),high endurance(>10~5 cycles),good retention(>10~4s at 85?).In addition,dual-layer W/HfAl_xO_y/AlO_x/Pt memristors are deposited.The test results show that the dual-layer structure effectively reduces the cycle-to-cycle variation of HRS,but they are weaker than single-layer devices in terms of operating voltage and endurance characteristics.(2)The method of voltage input Hamming Distance calculation is studied.Based on1T2R memristive array,Hamming Distance calculation of any bit can be realized by three steps of initialization,XOR logic calculation and reading.The calculation results can be stored in-situ of the array.Hamming Distance calculation function is verified by W/HfAl_xO_y/Pt memristors.Simultaneously,the influence of the devices'non-ideal factors on the calculation accuracy is analyzed by simulation.In addition,it is proposed that the method can be used for the Hamming Distance calculation acceleration of traditional computing architecture and in-memory edge computing.(3)The method of voltage-resistance mixed input Hamming Distance calculation is studied.The calculation is completed through three steps of erasing,writing and calculation.The data written during the calculation will not change by which the in-memory image search scheme is proposed.The image can be retrieved directly in memory.The scheme shows extremely high robustness to non-ideal factors memristors through simulation by CASIA-FaceV5 dataset.Under 50%device-to-device variation and 10 times on-off ratio,the accuracy of image search can be guaranteed to be within 1%of the theoretical result.