View article

Pseudo-halide (PH) anion engineering has emerged as a surface passivation strategy of interest for perovskite-based optoelectronics; but until now, PH anions have led to insufficient defect passivation and thus to undesired deep impurity states. The size of the chemical space of PH anions (>10⁶ molecules) has so far limited attempts to explore the full family of candidate molecules. We created a machine learning workflow to speed up the discovery process using full-density functional theory calculations for training the model. The physics-informed machine learning model allowed us to pinpoint promising molecules with a head group that prevents lattice distortion and anti-site defect formation, and a tail group optimized for strong attachment to the surface. We identified 15 potential bifunctional PH anions with the ability to passivate both donors and acceptors, and through experimentation, discovered that sodium …