| In the light of the big data era,emerging data-intensive applications such as cloud computing,the internet of things and artificial intelligence are flourishing.Limited by the bottleneck of the traditional von Neumann computing architecture with separated computation and storage units,the end of Dennard Scaling together with the saturation of Moore’s law,leads to a fundamental limitation,which is known as the Memory wall.The thermodynamic limits of general digital computing cause energy inefficiency,delay,and power loss issuses.In-memory computing based on emerging memristive devices is a promising non-von Neumann approach that allows processors to compute in memory.With the advantages of nonvolatility,high speed,low power consumption,high integration density,compatibility with CMOS technology,etc.,memristor can pave the way for the development of a highly energy efficient and reconfigurable computing-in-memory chip.In order to enhance the reliability and address sneak path current,together with the compatibility concerns and the thermodynamic limits at the device and logic level,this thesis focus on the study of performance enhancement,novel logic-in-memory methods,and reversible in-memory-computing.The major contributions and key finding are summarized as follows:Firstly,in order to solve the problem of forming process,reliability,endurance and thermal stability,a novel method based on defect concentration modulation is proposed and the W/Al Ox/Al2O3/Pt memristive device showing a stable bipolar switching behavior in the temperature range from 78 K-475 K is experimentally prepared.With this approach,the device shows superior resistive switching characteristics,including forming-free,high speed,large on/off ratio,low operating voltage,reliable endurance,and stable retention.A1 kb memristive array chip with the via-hole structure is developed to verify the feasibility of the performance defect concentration modulation scheme.Secondly,in order to solve the problem of high switching current and unexpected incorrect operations due to the leakage current,the self-selective V/Al Ox/Al2O3/Pt memristive device showing low power and highly reliability is developed on the Al Ox/Al2O3-based resistive switching.The introduction of the bipolar self-selective switching significantly improves the low resistance of memristors with the leakage current flowing through unselected cells suppressed.Furthermore,the impact of the built-in selectivity feature on the integrated size of the passive memristor array is quantitatively evaluated:on the premise of 10%read margin,the size of the passive memristor array is increased from 2×2 to 52×52.To suppress the conductance drift and the reading noise,two-step write-verify scheme and negative pulse sequence scheme are introduced to enable highly precise and stable conductance modulation in the range form 100 kΩto 10 MΩ.Thirdly,based on the AlOx-based memristor,non-volatile combinational and sequential logic functions are realized,and the applications in the field of reversible computing are investigated.To achieve voltage compatibility of CMOS,a novel approach based on resistive nonvolatile logic is proposed,both the combinational and sequential logic functions are performed in the same memristive array,which demonstrates its superiority in terms of seamless conversion and reconfiguration.The influence of temperature including400 K and 100 K on the logic operations is evaluated on the W/Al Ox/Al2O3/Pt memristive devices to experimentally verify the feasibility of proposed method.To address the thermodynamic limits of general digital computing,a novel reversible in-memory-computing approach based on memristive nonvolatile logic is proposed,both the basic and complex reversible logic functions are performed on the V/Al Ox/Al2O3/Pt memristive devices. |