Data-mining-aided mapping of structure-property relationships for combinatorially generated Co-doped ZnO thin films

Abstract

We demonstrate the use of multivariate analysis and highdimensional visualization to uncover and exploit nonobvious quantitative structure-property relationships or processing-property-performance relationships in vast arrays of multidimensional photovoltaic (PV) datasets. Under the research framework of PV informatics, the critical role of these mapping techniques, for design of transparent conducting oxides in particular, is discussed in the context of combinatorially generated Co-doped ZnO thin films. Multidimensional maps are generated using principal component analysis, a dimensional reduction technique in data mining. We present high-dimensional information visualization techniques such as parallel coordinates for mapping relationships that exist in the huge amounts of heterogeneous high-throughput data of thin films.