Remarkable progress has been achieved in metal halide perovskite (MHP) photovoltaics after more than a decade of intensive research by thousands of researchers on a handful of lead-based MHP compounds. However, it should be noted that even champion materials are not always thermodynamically stable at room temperature and most alloys suffer from intrinsic photodegradation phenomena. Perhaps for these reasons, the stability of highly performing devices remains marginal and further progress is incrementally made by fine-tuning existing MHP compositions with formulation, passivation, interfacial engineering and processing improvements. MHPs and their alloys are highly diverse materials with a vast chemical parameter space that provides numerous alternative candidates for discovering thermodynamically stable compounds that are likely to be stable under a variety of external stresses. However, it can be daunting to explore this parameter space without a clear framework and the toolkit to translate materials to thin film devices. Existing MHPs have well-tread formulation, passivation and processing pathways that have been hard-won with years of communal learning, whereas new MHP materials require re-learning how to process thin films.

We propose a streamlined approach to material discovery in the context of thin film photovoltaics. First, we have designed a robotic workflow (RoboMapper) that parallelizes and accelerates the screening of materials. In one example, we identify whether a alloys are cubic perovskite, achieve the desired bandgap and are light-stable. We show that RoboMapper reduces time-to-solution of new material candidates by an order of magnitude and commensurately reduces cost, energy and environmental impacts associated with exploring MHP compositional space. Through a physics-based framework linking defects-diffusion-stability, we can further predict which compositions yield light-stable, low-hysteresis and efficient operation in solar cell devices. In a second example we combine RoboMapper workflow with in situ heating of MHP alloys and implement AI identification and classification of phases to automate and accelerate the search for thermally stable perovskite alloys. Finally, with material candidates in hand, we demonstrate the implementation of a self-driving coater (RoboCoater) which autonomously develops and perfects coating recipes for MHP compounds without prior knowledge. Robocoater leverages in-line sensing, ML/AI and decision-making to accelerate 10-100 fold the translation of new materials into thin films and streamlines integration of new materials into device workflows.