Proceedings of MATSUS Spring 2025 Conference (MATSUSSpring25)
DOI: https://doi.org/10.29363/nanoge.matsusspring.2025.137
Publication date: 16th December 2024
New materials of interest for photovoltaics or other applications are invariably multi-cation compounds or alloys. Therefore, to properly explore their properties, synthesis experiments must be carried out over a broad parameter defined by several compositional variables, core thermodynamic variables of temperature and pressure, and kinetic variables such as reaction time and heating rate. Each of these can affect the outcomes of synthesis: the phases formed and their microstructural and defect characteristics that are so important for functional properties. A special challenge in the context of inorganic materials such as multinary chalcogenides is that they can be grown over a very wide range of temperatures and pressures as well as off-stoichiometric compositions. Thus, the parameter space to be explored is particularly extensive.
In this contribution, we will present our concept for automated exploration of new inorganic chalcogenides in a self-driving lab. This combines a powerful and rapid PVD-based synthetic method with automation and machine learning, to rapidly map the phase space and functional properties of new materials, without needing prior information. The developments to-date will be presented, starting with our approach for automated generation of co-sputtering processes. This involves two machine-learning stages coupled to a geometrical model of the sputter flux, fitted in real time using input from a trio of QCM sensors. Outputs of the trained models can be used to produce co-sputtering recipes that yield a specified composition, while allowing other parameters (e.g. pressure) to vary. In addition, compositional maps for each sample are obtained directly without time-consuming mapping in an external system. We will preview coming developments in automation of the subsequent process stage – rapid thermal sulfurization – and some initial work in image-based high-throughput characterisation combined with simulations, to derive optical properties of the target materials. Perspectives for the application of self-driving labs in development of new device materials will be discussed.