Resistive Switching Devices and Their Applications for Computing Beyond von Neumann Architecture

Abstract

As the demand for processing artificial intelligence (AI), big data, and cognitive tasks increases, new devices and computing architectures that can reduce the cost of the memory bottleneck have gained significant interest. One of emerging device that can enable non-von Neumann architectures such as neuromorphic computing and in-memory computing, resistive random-access memory (RRAM), has been extensively studied due to its properties such as nano-scale feature size, low power, and inherent functionalities that allow it to emulate biological synapses and stochastic events. In this thesis, I will discuss optimization and development of RRAM devices as well as the application of RRAM devices for machine learning tasks and combinatorial optimization problems. Experimental demonstration of feature extraction by using tantalum oxide-based analog RRAM devices will be first introduced. To achieve robust operation of RRAM crossbar array, tantalum oxide devices are optimized to reduce the forming voltage. The optimized RRAM array is successfully used to perform principal component analysis (PCA), an unsupervised learning algorithm for feature extraction and dimensionality reduction, of a breast cancer dataset. In the second project, an RRAM structure that offers very low power and large on/off ratio is developed using copper active electrode and atomic layer deposited Al2O3 layers for low-power in-memory computation and digital version of neuromorphic computing applications. Desirable device performance such as self-current limiting, forming-free resistive switching, ultra-low current, and improved uniformity have been obtained. Beyond device optimizations, I will present two projects that aim at demonstrating the applications of RRAM devices, implementing RRAM-based hardware acceleration of simulated annealing of the two-dimensional spin glass problem, and stochastic learning of deep neural networks. At the end of this thesis, a practical application of RRAM array for combinatorial optimization like travelling salesman problem is proposed as a future work.