Discovery of New Fast Oxygen Conductors: Bi2MO4x (M= rare earth, X= halogen) Via Unsupervised Machine Learning

Abstract

Discovering new fast oxygen conductors is critical for developing many technologies, e.g., solid oxide fuel cells (SOFCs), solid oxide electrolysis cells (SOECs), proton ceramic fuel cells (PCFCs), and solid oxide air batteries, particularly for operation at lower temperatures (below 700 ˚C). By leveraging unsupervised machine learning techniques and large databases of computed materials data (e.g., Materials Project), we are able to rapidly explore and screen for new materials with high oxygen mobility. In this work, we used unsupervised machine learning analysis applied to features extracted from the structure of materials and discovered a new structural class of vacancy-mediated oxygen conductor. To do this, we applied the K-nearest neighbor (k-NN) algorithm on all the approximately 62,000 oxygen-containing compounds in the Materials Project database, using X-ray diffraction (XRD) and pair distribution function (PDF) data to represent the structural features. The neighbors of reference known for good oxygen conductors were screened with criteria relating to material stability, synthesizability, and electronic property dissimilarity (to ensure a novel material) compared to the reference material. Starting with the best-known oxygen conductor, the fluorite-type material, we examined the materials list of its nearest neighbors and identified the Bi2MO4X (M=Rare-earth element, X= halogen element) structural family as a novel group of fast oxygen conductors. In particular, Bi2LaO4Cl is a layered bismuth oxyhalide crystallized in the space group P4/mmm, adopting a “triple fluorite” layer [Bi2LaO4]+, balanced by the anion Cl- layer. The stability and mobility of oxygen vacancy in Bi2LaO4Cl were studied using Density Functional Theory (DFT) methods. 2+ oxygen vacancy is the most thermodynamically favorable defect type. Ab initio and machine learning-trained interatomic potential molecular dynamic simulations were performed to study the high temperature oxygen mobility in Bi2LaO4Cl. The simulation results show that Bi2LaO4Cl presents an extremely low migration barrier of 0.14 +/- 0.03 eV below 700 ˚C, suggesting 10 to 20 times higher ionic conductivity compared to yttria-stabilized zirconia (YSZ). Bi2LaO4Cl is an insulator with a wide band gap of 2.65 eV, indicating that Bi2LaO4Cl may be a promising electrolyte material with high oxygen ion conductivity and low electronic conductivity. This work demonstrates the power of combining unsupervised machine learning approaches with large materials databases for finding new fast oxygen conductors.