Materials Design through AI Composition and Process Optimization: Poster Session
Program Organizers: Noah Paulson, Argonne National Laboratory; Tiberiu Stan, Asml; Brandon Bocklund, Lawrence Livermore National Laboratory; Arun Kumar Mannodi Kanakkithodi, Argonne National Laboratory
Tuesday 10:00 AM
November 3, 2020
Room: Poster Hall
Location: MS&T Virtual
A Physics-informed AI Assistant for Atomic Layer Deposition: Noah Paulson1; Angel Yanguas-Gil1; Steven Letourneau2; Jeffery Elam1; 1Argonne National Laboratory; 2ASM
Atomic layer deposition (ALD) is a method for depositing thin films for applications including microelectronics, energy storage, and biomedical implants. Depositing conformal films with atomic precision relies on a self-limiting reaction between two precursors. For a new set of precursors or reactor, determining the values of parameters that result in a high quality film is a laborious manual process that relies on the experience and intuition of the operator. In this work, we present a physics-informed AI assistant that automatically tunes these processing parameters. This software, based in Bayesian optimization, leverages readings of mass-gain per cycle to sequentially suggest parameter settings and drive towards an ALD process with uniform film thickness and low precursor consumption. We demonstrate this approach for a variety of target chemistries including Al2O3, TiO2, and W.
AI-driven Discovery of Novel High Entropy Semiconductor Alloys: Arun Kumar Mannodi Kanakkithodi1; Xueying Li2; David Fenning2; Maria Chan1; 1Argonne National Laboratory; 2University of California San Diego
High-entropy alloys in semiconductor chemical spaces resulting from arbitrary mixing at cation or anion sites can help enhance the stability, optical absorption, electronic properties and performance of materials for applications such as solar cells, infrared and quantum sensors, and electronics. In this work, we develop a general AI-based framework for the on-demand prediction of the structure, formation energy, band gap, optical absorption and defect behavior of high entropy alloys belonging to semiconductor classes of interest in photovoltaic and related optoelectronic applications. This framework is powered by high-throughput quantum mechanical computations, unique descriptors ranging from atomic coordination environments to elemental properties to low-fidelity computational outputs, and the rigorous training of advanced neural network-based predictive and optimization models. AI-based recommendations are synergistically coupled with targeted synthesis and characterization, leading to the successful validation and discovery of novel compositions for improved performance in solar cells.
Enabling Process Optimization Using High-throughput Machine Learning-based Image Analysis: Tiberiu Stan1; Peter Voorhees1; 1Northwestern University
One of the advantages of modern materials processing techniques (such as additive manufacturing) is the ability to correct for defects during part fabrication. To be successful, the images obtained through in-situ monitoring must be rapidly collected and analyzed. We showcase the use of convolutional neural networks (CNNs) as an efficient way to accurately evaluate large materials imaging datasets. Novel approaches to CNN training using experimental and synthetic datasets will be presented, as well as techniques for comparing and determining the success of different machine learning methods. Future steps to incorporating artificial intelligence into process optimization and anomaly detection will also be discussed.