Conclusions

In this study, we have provided a new data-driven / AI framework for environmentally conscious selection of amine chemistries used in the synthesis of hybrid organic-inorganic perovskites. The selection strategy is based on exploring high dimensional data capturing structure-function- toxicity driven by molecular-scale information. To the best of our knowledge, this is the first such study to critically explore AI methods to rank toxicity impact from the perspective of molecular descriptors; and to harness this information to identify safer alternatives that also have been shown to be preserving the functional performance of such perovskites for photovoltaic applications. By coupling new probabilistic-based molecular descriptors with advanced data dimensionality such as UMAP, we have also established a database resource to explore other families of yet unexplored amine chemistries that may be used for hybrid perovskite structures. The need for searching and identifying alternative and safer chemistries for establishing a “benign-by-design’ has long been recognized, our work provides an example of how AI coupled to foundational materials chemistry principles can actually facilitate an a priori approach to select chemicals for materials synthesis that meet the structure-function and sustainability metrics.