The atomistic structure of hybrid halide perovskite solar-cell materials

The energy transition required to reach net zero emissions by 2050 necessitates the fast development of renewable energy technologies and infrastructure. Now recognised as revolutionary materials for thin-film photovoltaics, hybrid perovskites are at the forefront of the so-called 'third generation' of solar cells with record- breaking efficiencies, overtaking those of traditional silicon panels that make the majority of solar technologies at the moment. Currently, perovskite photovoltaics are in the commercialisation stage, being released into the market. These materials are more flexible than silicon cells, much more easily modified for specific applications due to their tunable structures, and their synthesis is simple and cheap. However, in order to unlock the full potential of these materials, the electrons which act as charge carriers must be efficiently collected and used in the external electric circuit. For this to happen, preventing the loss of the charge carriers through competing processes such as charge recombination is crucial. The optical and electronic properties of a photovoltaic material are tightly linked with its atomic structure (such as the boundaries between the grains of the powder material, the surface and the interface it forms at the contacts with other materials). Experimentally, the microstructure of different types of perovskites have been imaged before, using very specialised techniques. Nonetheless, the processes which hinder charge carrier transport, and the microstructure of the material are challenging to visualise and map precisely using only experimental techniques. This is due to the instability of the material under electron irradiation, which is required for atomic microscopy experiments. Furthermore, an amorphous phase has been identified in some regions of the material, and due to the fact that such a phase has no long-range order, it cannot be studied crystallographically. As such, some of the properties inferred from different experiments in the literature are found to be contradictory. As a solution, we propose using high accuracy electronic-structure calculations to train a machine learning model that can be used to study the perovskite material, exploring new and efficient methodologies for creating the model. Machine-learned interatomic potentials approximate the potential energy surface of a material, which determines its most stable structural arrangements, and further calculations can retrieve other properties of the material from the obtained structures. This model can be optimised to more specifically study the regions of interest which are poorly understood, or where the loss of charge carriers is thought to happen. These regions of interest (grain boundaries, interfaces etc.) are akin to a brick wall: the "structure" of each brick is well understood, but the way they can be arranged can be quite varied and must be done properly. In the same way, modelling the whole perovskite structure, with its grain boundaries and different phases can present quite a challenge. The field of machine learning for material modelling has seen very fast development, reaching ever larger scales, which can produce simulations of structures up to a few tens of nanometers thick, such that the simulations can now broadly represent thin-film materials within a device. Using this approach for modelling hybrid halide perovskites could not only explain observed phenomena, but also inform future experiments and aid in the challenge of improving perovskite-based devices even further.