A Research on RRAM and ECRAM Devices for High Density Memory and Neuromorphic Applications

Abstract

Today, further advance in deep learning performance has been largely fueled by exponential increase of the depth and size of neural networks. To train such neural networks, the demand in computing usage has grown 10x/year since 2012, which was originally 2x/year prior to 2012. Therefore, this trend poses a significant challenge, necessitating not just faster but also more energy-efficient computational solutions to support and propel the continued growth of deep learning applications. In this context, several emerging memory technologies present promising opportunities. RRAM (oxygen-based resistive RAM), ECRAM (electrochemical RAM), PCRAM (phase change RAM), and MRAM (magnetic RAM) are at the forefront of this shift. While the specifics of device performance are heavily reliant on the type of active material used, these technologies can have several beneficial characteristics: non-volatility, compatibility with back-end-of-line (BEOL) processing, and simple fabrication techniques. However, to fully integrate emerging memory technologies alongside mainstream memory technologies, several critical issues must be addressed. These challenges encompass a range of technical and practical aspects, from material properties and device architecture to compatibility with existing technologies and scalability for mass production. Addressing these issues is crucial for the successful transition of these emerging memories from research and development to widespread commercial use. This thesis centers on the exploration of RRAM and ECRAM technologies. RRAM and ECRAM devices have distinctive advantages for high memory density and neuromorphic computing application, respectively. However, there are limitations on currently proposed 3D RRAM structures for high density memory applications. Also, there is a practical limit to use ECRAM as neuromorphic computing applications due to CMOS incompatible material using in ECRAM device. Thus, in Chapter II, the author focuses on experimental research into the feasibility of filament-based RRAM within a V-NAND structure, where the electric field is manifested as a fringing field. The author addresses a significant issue in the V-NAND design that hinders optimal switching performance, especially when compared with the conventional Metal-Insulator-Metal (MIM) structure. To demonstrate the feasibility of RRAM switching, material engineering strategies, including the careful selection of materials and the optimization of their composition, are examined. Furthermore, the chapter explores structural engineering solutions to overcome the structural limitations of the V-NAND architecture. In Chapter III, the author presents the development of a CMOS-compatible, all-solid-state H+-ECRAM device with a ZrO2 electrolyte, fabricated using a low-temperature atomic layer deposition (ALD) process. The impact of the ALD temperature on the ECRAM's switching speed is analyzed using electrical measurements, ellipsometry, and X-ray Photoelectron Spectroscopy (XPS) analysis. The electrical characterization concludes with the validation of a large on/off ratio, excellent weight update symmetry, and robust cycling endurance, underscoring the device's potential for practical applications.