Machine learned metaheuristic optimization of the bulk heterojunction morphology in P3HT:PCBM thin films

We discuss results from a machine learned (ML) metaheuristic cuckoo search (CS) optimization technique that is coupled with coarse-grained molecular dynamics (CGMD) simulations to solve a materials and processing design problem for organic photovoltaic (OPV) devices. The method is employed to optimize the composition of donor and acceptor materials, and the thermal annealing temperature during the morphological evolution of a polymer blend active layer composed of poly-(3-hexylthiophene) (P3HT) and phenyl-C61-butyric acid methyl ester (PCBM), for an increased power conversion efficiency (PCE). The optimal solutions, which are in qualitative agreement with earlier experiments, identify correlation between the design variables that contributes to an enhanced material performance. The framework is extended to multi-objective design (MOCS-CGMD) to attain a Pareto optimality for the blend morphology, and enhance concurrently the exciton diffusion to charge transport probability and the ultimate tensile strength of the material. The predictions reveal that a higher annealing temperature enhances the exciton diffusion to charge transport probability, while a PCBM weight fraction between 0.4 and 0.6 increases the tensile strength of the underlying blend morphology.