Machine Learning Guided Prediction of Thermal Properties and Identification of Crystallizable Organic Semiconductors

Crystalline organic semiconductors feature improved exciton diffusion length and charge carrier mobility compared to their amorphous counterparts. Certain organic molecular thin films can be crystallized at large-scale via annealing of initially prepared amorphous layers. These films ideally crystallize as platelets with long-range order on the scale of tens to hundreds of microns, but have also been seen to crystallize as spherulites or to resist crystallization entirely. Molecules that can transform into a platelet morphology feature high melting point and crystallization driving force (ΔG_c). Here, we employ machine learning to identify candidate organic materials that were predicted to crystallize into large-area platelets by estimating the aforementioned thermal properties. Six materials identified by the machine learning algorithm were evaluated for their bulk thermal properties using differential scanning calorimetry, and their crystallization behavior via thin film crystallization. Of these six materials, three crystallized as platelets, one crystallized as a spherulite, and two resisted crystallization, displaying a successful application of machine learning in the scope of organic thin film crystallization and reinforcing the principles of melting point and ΔG_c as metrics that govern the crystallization behavior of organic thin films.