ニューロモルフィックaiハードウェアを指向した銀ナノ粒子抵抗変化デバイスのシナプス記憶機能に関する研究
Year: 2023
Type: dissertation
Abstract: The aim of brain-inspired computation has increased interest in the development of artificial synapses that mimic the intricate functionalities of biological synapses. This thesis presents a novel approach to realizing artificial synapses using silver nanoparticle atomic switch networks (Ag NPNs) for brain-inspired hardware and computing applications. The thesis begins by introducing the background of traditional computing architectures and neuromorphic computing paradigms. We highlight the limitations of traditional computing architectures in solving increasingly complex and data-intensive tasks, necessitating the exploration of alternative approaches such as in-memory computing. The theoretical foundations of atomic switches and their potential applications in brain-inspired technology are reviewed. Further, we explore the significance of biological synapses in neural information processing. The core of this research lies in the design, fabrication, and characterization of the Ag NPN-based artificial synapses. Through a series of experiments, the electrical properties of Ag NPNs are investigated to achieve memristive behavior, resembling the synaptic plasticity observed in biological synapses. The Ag NPN's ability to adapt its conductance values based on input stimuli allows it to learn and store information, laying the foundation for brain-inspired computing paradigms. The thesis also investigates the capability of the Ag NPN and its potential applications in in-materio reservoir computing (RC) systems. A thorough analysis of the multiple nonlinear forms of the Ag NPN reveals its viability for enhancing computing performance. The findings presented herein contribute to the growing wealth of knowledge in neuromorphic engineering and may pave the way for more efficient and adaptable computing systems, bridging the gap between conventional computing and the remarkable processing capabilities of the human brain. The thesis consists of six chapters in total. Chapter 1 introduces the topic of neuromorphic computing and a comprehensive review of the relevant literature. It reviews neuromorphic algorithms and hardware components, with a focus on memristors and their resistive switching phenomena. The chapter also delves into atomic switches, exploring their basic principles and potential for brain-inspired computing. Synapses and their plasticity mechanisms, including how atomic switches can realize synaptic plasticity, are discussed. The problem statement, research objectives, scope, and thesis outline are presented. Chapter 2 outlines the methodology used in the research. It covers the synthesis of silver nanoparticles (Ag NPs), specifically the chemicals and materials involved, including the Two Phase Brust-Schiffrin method. The chapter then discusses the fabrication process of Ag NPNs devices, followed by electrical measurements such as I-V measurements and pulse measurements. Chapter 3 provides a detailed analysis of the characteristics of Ag NPs and Ag NPN-based switching devices. The chapter explores the morphological and chemical properties of Ag NPs and investigates the electrical behavior of Ag NPN-based switching devices. It includes an examination of the typical I-V characteristic, the switching mechanism, the effect of compliance current, and the impact of electrode spacing on the devices. These findings lay the foundation for understanding the fundamental properties of the materials and devices used in the research. Chapter 4 investigates the synaptic plasticity and learning ability of Ag NPNs artificial synapse. The chapter begins with an introduction, followed by an analysis of the device structure and switching behaviors. It investigates STP and the learning ability of Ag NPNs, showcasing their adaptive behavior similar to biological synapses. The transition from STP to LTP and related memory processes are also examined. The chapter provides valuable insights into the potential of Ag NPNs in neuromorphic computing. Chapter 5 explores the use of Ag NPNs devices for implementing RC. Standard benchmark tasks for RC are performed. The effect of different nonlinearity forms on RC performance is highlighted. Finally, Chapter 6 concludes the thesis and proposes potential future research directions. Overall, the thesis aims to explore the application of Ag NPN-based memristors in neuromorphic computing.
DOI: 10.18997/0002000367
Author: Srikimkaew, Oradee
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7143664524