Developing and Optimizing an Ultra-Fast Force Field (UF3) for Modeling The Crystallization of Amorphous Silicon Nitride

Crystallization of amorphous silicon nitride, as observed experimentally, can alter material properties in microelectronics process conditions as a layer-stacked component. Understanding the crystallization mechanism mandates a multi-scale approach in which quantum simulations inform atomistic simulations, which further inform continuum scale simulations and subsequently, experimental work. This talk will detail the progress and challenges faced in developing an Ultra-Fast Force Field (UF3) specifically designed to bridge quantum and atomistic simulations for the Si-N material system. The UF3 integrates effective many-body potentials within a cubic B-spline framework with regularized linear regression, creating a fast and interpretable machine-learned potential (MLP).<b> </b>First, we will cover the nuanced requirements of the MLP's training data: it needs to be diverse enough to avoid overfitting to Si3N4 stoichiometry, while maintaining enough specificity to prevent unneeded generalizations across the entire Si-N compositional range. Following this, we leverage the interpretability of UF3's 2/3-body terms to understand the MLP's behavior and identify areas for improvement. The presentation will conclude by detailing the simulation results that were used to validate the MLP and an analysis of the simulated crystallization results of the final MLP.<br/><br/>Sandia National Laboratories is a multimission laboratory managed and operated by National Technology and Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA-0003525.