Investigation of Resistive Switching Device for Memory and Neuromorphic applications

Abstract

Typical charge-storage based memories such as NAND or NOR Flash memory have faced limitations in a device scaling respect. To overcome the limitations, several future non-volatile memories: phase-change memory (PRAM), magnetic memory (MRAM), and resistive memory (ReRAM) have been proposed as alternative techniques. Among those candidates, especially, the ReRAM has been extensively investigated on the basis of its various advantages (simple metal-insulator-metal structure, low power consumption, fast operation, and CMOS compatibility). However, one main obstacle is the variability of switching parameters, which can be a critical drawback for the future non-volatile memory applications. The randomly generated conducting filament, during switching operations, is considered as a dominant origin of the switching variability. Thus, in this thesis, as the first section, the author investigated several approaches such as a defect engineering, device optimization, and structural engineering to maximize the advantages and minimize the disadvantages (mainly switching variability). In addition, for a feasibility of high density integration of the ReRAM, novel two terminal switch device was developed to prevent the leakage-current which can lead to a misoperation in cross-point array structure. Furthermore, in the second section, new application of the ReRAM: neuromorphic application that is a novel computing process for complex and huge amount of input was researched. To realize the neuromorphic application, the ReRAM needs to be modified for achieving a brain-inspired synaptic characteristics such as non-volatile behaviour, analogue, and symmetric conductance change. To achieve the synaptic characteristics from the ReRAM, the author optimized Pr0.7Ca0.3MnO3 (PCMO) based ReRAM, then the author demonstrated the feasibility of neuromorphic application by an identification of rat’s fear-state neural signal. To identify the rat’s neural signals, the neuromorphic system composed of a multi-layer artificial neural network and the optimized PCMO based ReRAM was simulated on the basis of normal- and fear-state neural signals. Consequently, the rat’s neural signals were clearly identified by proposed compensational circuit which can improve the synaptic characteristics of the PCMO based ReRAM. All of the results, in this thesis, obviously demonstrated the feasibilities of ReRAM not only in the memory application but also in the neuromorphic application.