Electrical resistivity as a descriptor for classification of amorphous versus crystalline phases of alloys

Abstract

Discovering new metallic glasses, non-crystalline alloys with unique combinations of mechanical and chemical properties, is a challenging endeavor because it requires exploration of a vast composition space. High-throughput experiments have greatly enhanced the efficiency with which composition-dependent properties of potential glass-forming alloys can be measured, but phase identification remains a bottleneck because slow or expensive techniques such as table-top or synchrotron-based X-ray diffraction measurements are required. In this study, we developed machine learning (ML) models that can classify amorphous and crystalline phases of alloys using electrical resistivity as a primary descriptor. Artificial neural networks were constructed to correlate the electrical resistivities and the X-ray diffractograms of a broad range of combinatorially synthesized alloys. The ML models are found to classify amorphous/crystalline phases in both thin-film libraries and bulk alloys with high accuracy.