Leveraging Atmospheric-Pressure Spatial Atomic Layer Deposition and Machine Learning for Nanomaterial and Device Design

Abstract

The deposition and design of nanometre-scale oxide films is an integral component of the ongoing nanomaterial revolution, from cell phones, to batteries, to photovoltaics. Atmospheric-pressure spatial atomic layer deposition (AP-SALD) and chemical vapour deposition (AP-CVD) are two techniques that show great promise for commercialization, due to their speed and the vast array of materials and stoichiometries they can produce. However, this flexibility comes at the cost of complexity; the presence of oxygen during film deposition induces defects which may have advantageous or deleterious effects dependent on the application of the film. Machine learning is a statistical technique that allows us to make sense of such complex systems of interaction, without the need for expensive ab-initio simulations. Through this work, I demonstrate our capacity to deposit entirely new materials with our lab-scale AP-SALD/CVD system, and develop several characterization methods that will permit us to leverage the flexibility of that system to maximum effect. I developed a system for measuring the resistance of our films in-situ, demonstrating the effects of atmospheric oxygen on film properties, as well as implementing machine learning into our in-situ reflectance system providing accurate real-time measurements of film thickness and band gap. Next, I designed a gaussian process regression tool to assist researchers in finding accurate optical model parameters with our spectroscopic ellipsometer, much more quickly than previously possible. Lastly, I implemented a basic computer vision tool for tracking the degradation of perovskite and calcium thin films in real-time. To my knowledge, this work represents the first time that machine learning has been leveraged to improve the deposition of films by AP-SALD and to enhance the characterization of their properties and performance.