From Edisonian Optimization to Predictive Molecular Design: AI-Driven Discovery and Engineering of Advanced Semiconductor Interfaces
Physical models cannot currently predict or explain the behavior of semiconductor interfaces, which makes improving device reliability difficult because the potential combinations of chemistry to explore are too vast to screen by trial and error.
This project, led by the University of Kentucky, develops a closed-loop AI framework for generating mechanistic hypotheses to discover the origin of mixed-molecule self-assembled monolayers (SAM) outperforming single-component SAM at metal-oxide/perovskite interfaces. Specifically, it treats the gap between model predictions and experimental characterization as a signal, using physics-informed machine learning and a molecular foundation model to generate new hypotheses, then Bayesian experimental design to test them. A central target of the platform is to improve the pace of resolving open mechanistic questions and improve predictive accuracy. The approach is substrate-agnostic and positioned to accelerate design across future semiconductor interfaces.
