The demand for data storage – from mobile devices to enterprise applications – has been driving the explosive development of non-volatile memories (NVMs). However, as Moore’s Law approaches its fundamental limit, the aggressive scaling of Flash memory is faced with challenges from technical issues and economic concerns. Therefore, the next-generation NVM solution is urgently desired. Among various candidates, resistive switching memory (RRAM) has attracted broad interest due to its simple structure, high speed, long retention, excellent endurance and low power.
In this dissertation, we focus on RRAMs and related reconfigurable devices. Through simulation, we systematically discuss the sneak-path issue of passive RRAM arrays and benchmark the selector devices for large-scale crossbar integration. Next, we develop a tantalum oxide (TaO x ) selector device showing high nonlinearity and good uniformity. To meet the requirement of energy-constrained applications, we further develop ultralow-current RRAM cells with self-rectifying characteristics and high performance. In addition, we explore coupling the ionic migration process in resistive switching with the transistor operation, and demonstrate nonvolatile conductance modulation on a novel reconfigurable device based on the LaAlO 3 /SrTiO 3 heterointerface. Finally, an energy-efficient in-memory computing architecture using crossbar RRAM arrays has been proposed to break the boundary between computing and memory and offer high parallelism.