Property Phase Diagrams for Compound Semiconductors through Data Mining

This paper highlights the capability of materials informatics to recreate property phase diagrams from an elemental level using electronic and crystal structure properties. A judicious selection of existing data mining techniques, such as Principal Component Analysis, Partial Least Squares Regression, and Correlated Function Expansion, are linked synergistically to predict bandgap and lattice parameters for different stoichiometries of GaxIn1-xAsySb1-y, starting from fundamental elemental descriptors. In particular, five such elemental descriptors, extracted from within a database of highly correlated descriptors, are shown to collectively capture the widely studied bowing of energy bandgaps seen in compound semiconductors. This is the first such demonstration, to our knowledge, of establishing relationship between discrete elemental descriptors and bandgap bowing, whose underpinning lies in the fundamentals of solid solution thermodyanamics.