Thomas Unold3,Christian Utama1,Leo Choubrac2
Freie Universität Berlin1,Université de Nantes2,Helmholtz-Zentrum Berlin3
Thomas Unold3,Christian Utama1,Leo Choubrac2
Freie Universität Berlin1,Université de Nantes2,Helmholtz-Zentrum Berlin3
Multinary chalcogenide thin film technologies are interesting for various energy conversion related applications, such as photovoltaics, photocatalysis or thermoelectrics. Due to the large degrees compositional and structural degrees of freedom of this multicomponent materials class, the establishment and control of single-phase deposition routes and material homogeneity is a challenging task, while the detection of secondary phases with X-ray diffraction, electron microscopy and Raman spectroscopy can be tedious and time-consuming. Using the model system of combinatorially-deposited Cu2ZnSnSe4 thin film libraries we demonstrate how rapid hyperspectral optical reflection measurements in conjunction with machine-learning can be used to predict the presence of secondary phases in these material libraries covering a large region of the Cu2Se-ZnSe-SnSe2 ternary phase diagram with high accuracy. The advantages and limitations of different machine-learning algorithms for secondary phase prediction will be discussed. Extending this approach to include hyperspectral photoluminescence measurements, we show how photovoltaic performance can be predicted on a 575 sample library within a 2-hour measurement/analysis cycle.