Application of machine learning to accelerate materials research and discovery

Machine learning (ML) has emerged as a transformative tool in materials research, enabling the accelerated discovery and optimization of novel materials. By leveraging advanced algorithms and data-driven models, researchers can predict material properties and behaviors with greater accuracy and efficiency, reducing the need for extensive experimental trials. We explore the use of ML in accelerating two areas of computational materials research.

(1) We demonstrate the use of ML to accelerate the discovery of stable alloy systems for use in electronic packaging. A considerable knowledge gap exists in understanding the structural, thermodynamic relationship at the atomic level of the solid solutions, especially for Sn-based alloys. We attempt to bridge this knowledge gap by using first principles-based DFT calculations with machine learning based on genetic algorithms to search for the most mechanically and thermodynamically optimal binary/ternary alloys. (2) We demonstrate using the Gaussian Approximation Potential (GAP) model and ‘on-the-fly’ learning to accelerate first principles-based molecular dynamics simulations. This allows for extended electronic transport simulations in mechanically controlled break junctions, achieving longer timescales and more extensive system sizes. These approaches have enabled us to accelerate our computational materials research while enabling our collaborators to capture a vivid atomistic understanding of their experimental work.