Title:Mastering processing-microstructure complexity through the prediction of thin film structure zone diagrams by generative machine learning models

Abstract:Thin films are ubiquitous in modern technology and highly useful in materials discovery and design. For achieving optimal extrinsic properties their microstructure needs to be controlled in a multi-parameter space, which usually requires a too-high number of experiments to map. We propose to master thin film processing microstructure complexity and to reduce the cost of microstructure design by joining combinatorial experimentation with generative deep learning models to extract synthesis-composition-microstructure relations. A generative machine learning approach comprising a variational autoencoder and a conditional generative adversarial network predicts structure zone diagrams. We demonstrate that generative models provide a so far unseen level of quality of generated structure zone diagrams comprising chemical and processing complexity for the optimization of chemical composition and processing parameters to achieve a desired microstructure.