Study On Logic Circuit And Functionality Of Memristor-Based In-Memory Computing

In the era of big data,computers are urgently required to show more excellent information processing capabilities.In the traditional von Neumann computing architecture,the separation of memory and processor limits the computing performance in terms of power consumption and delay.In contrast,the in-memory computing technology based on nonvolatile memristor can directly process data in memory,which provides a choice to innovate the traditional von Neumann computing architecture,break through the computing bottleneck and continue to improve the efficiency of information processing in the era of post Moore's law.Logic computation is the fundamental means for modern digital computer to realize any information processing.In this thesis,two promising kinds of memristor-based in-memory logic computing methodologies are focused: resistance input-resistance output(R-R)logic and voltage input-resistance output(V-R)logic.On one hand,in the research of the R-R logic with fluent logic cascading,aiming at the fully memristive in-memory computing system,systematic research has been carried out from the design and optimization of logic operations at the bottom to the construction of in-memory computing architecture at the top.In this work,firstly,taking resistance as input and output signals,a universal,reconfigurable and non-volatile R-R logic method is designed.With less resources,16 kinds of Boolean logic,several arithmetic functions and varies of operation modes are realized efficiently in the memristor array.Secondly,corresponding optimization methods are proposed to reduce the influence of non ideal factors in devices and circuits on R-R logic calculation,ensuring the high calculation accuracy.Lastly,the memristive in-memory computing architecture and the computing array's structure are designed,satisfying the realization of basic data storage function and information processing function,laying the foundation for the hardware implementation of the novel in-memory computing architecture based on the memristive R-R logic.On the other hand,in the research of the V-R logic,considering the features of the device,circuit structure,and practical processing tasks,several logic methods with strong functionality are designed and demonstrated in various important information processing applications,broadening the application field of in-memory logic computation.In this part,firstly,by reserving both two voltage input signals,efficient V-R XOR logic implementation is realized in the anti-serially connected structure of memristor,and the specific logic functions such as digital comparator and reversible logic are realized in a single multivalued memristor,demonstrating the ability of memristive multivalued characteristics for binary calculation.Then the logic methods are applied in the important information processing fields including Hamming distance calculation,digital function realization and low power computation,showing the improvement on computing performance.Secondly,by reserving one voltage input signal,specific logic functions such as AND and XOR are designed in the 1T1 R structure with V/R-R implementation form.High-efficiency Hamming weight calculation and high-security data encryption are further demonstrated using the proposed logic methods,expanding the application range of in-memory computing.This thesis supports for the construction of the digital memristive in-memory computing system,and exploits the potential of the memristive logic methods for the specific information processing tasks at the same time,providing the direction for the development and application of the memristive in-memory logic computing.