Morphology prediction for polymer blend thin films using machine learning

Abstract

When two immiscible polymers are spin-coated from a common solvent, they undergo phase separation, resulting in a mesoscale morphology that depends on a host of parameters. The phase-separated morphology plays a pivotal role in determining the potential applications of blend thin films. As a guide to experimentalists, a machine learning-based classification framework is proposed that can predict the morphology of PS/PMMA blend thin films. Different experimental parameters like weight fraction of PS, molecular weight of PMMA, concentration, and substrate surface energy were used as inputs based on which the morphology type, i.e., column, hole, or island, was predicted using a multi-class classification model. Several machine learning algorithms were used to develop the proposed classifier. Support vector machine (SVM) algorithm resulted in the highest accuracy of 93.75%. An explainable machine learning algorithm was also implemented to extract valuable insights from the proposed SVM model. These insights were found to be in excellent agreement with experimental observations, thus not only enhancing the reliability of the predictive model but also the understanding of phase separation in PS/PMMA blends. Based on these insights, several guidelines are recommended to further aid in the experimental design of specific morphologies. An easy-to-use web tool is also developed so that the proposed model can be accessed freely, which is expected to expedite the design of application-specific thin films.