Molecular Dynamics Simulation and Machine Learning Prediction of Stability and Structure of Hybrid Organic Inorganic Perovskites

Perovskites have shown great promise for high efficiency solar cells but are often plagued by instability. Hybrid Organic Inorganic Structures (HOISs) encompass a wide array of potential chemistries, leading to immense property tunability. However, the number of possible structures means that it could take years to synthesize a significant number of HOISs that are both highly efficient and stable. Classical molecular dynamics (MD) using the Interface Force Field (IFF) can significantly increase the rate at which these structures are checked, so that hundreds or thousands of structures can be simulated in a fraction of the time for laboratory synthesis. These simulations can help elucidate a variety of structural properties relevant to device performance, as well as formation energies and stability. The results obtained from IFF simulations of HOISs can be used to fill in the gaps of existing materials databases. In addition to directly calculating HOIS properties, this MD data can help train machine learning models to predict HOIS properties without directly creating the structures. With a proper dataset and model, the user can first screen predictions using ML models, simulate a subset of the predicted structures, and synthesize a handful of the most promising candidates.