Filamentary Resistive Switching Memory for Neuromorphic and High Density Memory Applications

Abstract

With continued semiconductor industry growth over the past few decades, electronic devices have significantly influenced our lives. Thanks to the recent explosive demand for mobile devices such as laptop computers, tablet PCs, and smart phones, the industry will continue to grow. Furthermore, the impending 4th industrial revolution will be characterized by a wide range of new applications such as smart cars, smart factories, smart sensors, wearable devices, and artificial intelligence. To accomplish this, an exponential growth of connected devices is required. The key technology driving continued electronic device growth is transistor performance improvement based on dimensional scaling benefits. The transistor-based logic and memory devices have also grown in terms of power consumption, storage capacity, and operation speed. However, in conventional von Neumann architecture, the performance gap between processor and memory has been widening owing to high performance multicore processor development. To mitigate the performance gap, storage class memory (SCM) is introduced as a new memory class. The SCM technology offers similar performance to DRAM and NAND Flash memory in terms of capacity and cost. The SCM is also classified as DRAM-like or NAND Flash-like performance. From a different computing system viewpoint, neuromorphic systems have been considered beyond today’s von Neumann architecture. The parallelism of this computing system can be a solution for processing significant amounts of data. Therefore, this dissertation focuses on the development and analysis of resistive switching memory (RRAM) devices that can meet SCM requirements or neuromorphic applications. First, to achieve the requirements of neuromorphic applications, important parameters of RRAM devices such as multi-level cells (MLC) and I–V linearity were investigated. Based on electrical characterization and simulation, an understanding of the switching mechanism and appropriate electrical characteristics of RRAM devices was developed. These results indicate that the conductive filament (CF) shape and thermal properties are crucial for obtaining linear I–V characteristics with MLC. By selecting appropriate materials and an optimized switching process, RRAM devices with the required properties for neuromorphic applications are designed. Second, I demonstrated a hybrid memory device showing both memory and selector characteristics for high density SCM applications. To obtain memory and selector properties in single devices, the electrolytic effects on the device characteristics were investigated. The CF stability related to interaction energy of atoms is important for achieving efficient memory and selector behaviors. These results enable hybrid memory device to achieve high density cross-point arrays without integrating additional selector devices.