Discovery of Solubilizers to Enhance the Swelling Kinetics of Epoxy SU-8 via Virtual Screening

The slow swelling kinetics of SU-8 epoxy resist hinders its removal in semiconductor fabrication processes. To accelerate solubilizer discovery for swelling, we developed a virtual screening framework integrating molecular dynamics (MD) and machine learning. Using the diffusion coefficient within SU-8 as a key descriptor for solubilizer efficacy, we generated an MD-derived data set to train a graph convolutional neural network (GCN), enabling diffusivity predictions across >20 million compounds. This pipeline identified 4,6-decadienal (4,6-DDA) as a promising candidate. Experimentally, a 10 wt % aqueous solution of 4,6-DDA achieved 80.16% SU-8 swelling within 60 min, which was 3-fold greater than that of propylene carbonate. Simulations confirmed rapid 4,6-DDA penetration and enhanced performance in water, which is consistent with experimental measurements. This work demonstrates a rational simulation-guided strategy for the discovery of functional molecules for industrial processes.