Developing new materials and electronic devices is a hugely time-consuming endeavor due to the large parameter space of the synthesis process. Our laboratory works on pulsed laser deposition of oxide thin films, where the process parameter space has around 10 dimensions (compositions, pressure, temperature, rate, etc.), making a complete optimization infeasible. We have therefore developed an autonomous synthesis workflow for the optimization of the structure of materials based on real-time electron diffraction feedback. We combine neural network image segmentation that enables automated analysis of diffraction images and the extraction of structural quality metrics for thin films (Fig. 1). The synthesis conditions are obtained from a Bayesian optimization algorithm that can rapidly find the best synthesis conditions even in a very multidimensional parameter space. Additionally, the optimization process provides a multidimensional map of the crystal quality in the synthesis parameter space (Fig. 2), which is an invaluable tool when integrating multiple materials in a device structure. Such quality mapping is practically impossible without a fully autonomous synthesis process.

• Publications
• Research Highlights