Characterisation of Novel Resistive Switching Memory Devices

Abstract

Resistive random access memory (RRAM) is widely considered as a disruptive technology that will revolutionize not only non-volatile data storage, but also potentially digital logic and neuromorphic computing. The resistive switching mechanism is generally conceived as the rupture/restoration of defect-formed conductive filament (CF) or defect profile modulation, for filamentary and non-filamentary devices respectively. However, details of the underlying microscopic behaviour of the resistive switching in RRAM are still largely missing. In this thesis, a defect probing technique based on the random telegraph noise (RTN) is developed for both filamentary and non-filamentary devices, which can reveal the resistive switching mechanism at defect level and can also be used to analyse the device performance issues. HfO2 is one of the most matured metal-oxide materials in semiconductor industry and HfO2 RRAM shows promising potential in practical application. An RTN-based defect extraction technique is developed for the HfO2 devices to detect individual defect movement and provide statistical information of CF modification during normal operations. A critical filament region (CFR) is observed and further verified by defect movement tracking. Both defect movements and CFR modification are correlated with operation conditions, endurance failure and recovery. Non-filamentary devices have areal switching characteristics, and are promising in overcoming the drawbacks of filamentary devices that mainly come from the stochastic nature of the CF. a-VMCO is an outstanding non-filamentary device with a set of unique characteristics, but its resistive switching mechanism has not been clearly understood yet. By utilizing the RTN-based defect profiling technique, defect profile modulation in the switching layer is identified and correlated with digital and analogue switching behaviours, for the first time. State instability is analysed and a stable resistance window of 10 for >106 cycles is restored through combining optimizations of device structure and operation conditions, paving the way for its practical application. TaOx-based RRAM has shown fast switching in the sub-nanosecond regime, good CMOS compatibility and record endurance of more than 1012 cycles. Several inconsistent models have been proposed for the Ta2O5/TaOx bilayered structure, and it is difficult to quantify and optimize the performance, largely due to the lack of microscopic description of resistive switching based on experimental results. An indepth analysis of the TiN/Ta2O5/TaOx/TiN structured RRAM is carried out with the RTN-based defect probing technique, for both bipolar and unipolar switching modes. Significant differences in defect profile have been observed and explanations have been provided.