Optimization of Superconductors Fabrication by High-Throughput Experimentation and Machine Learning

Recently, strong effort is being devoted to develop new strategies that aim to achieve efficient, fast and informed materials development and overcome the limitations of the commonly used Edisonian approach. Special attention is addressed in current and future challenges in fields such as superconductivity, batteries, renewable energies, etc. In this sense, high-throughput experimentation combined with machine learning is gathering much attention due to its ability to explore a broad range of experimental parameters for an accelerated discovery and optimization of materials. This strategy consists of the fabrication of samples with different compositions in a discrete or gradient fashion by using physical or chemical deposition methods. Chemical solution deposition and, particularly, drop-on-demand inkjet printing is an ideal approach being cost-effective, versatile and industrially scalable to produce high-throughput combinatorial samples with high spatial resolution. These samples are then used for parallel investigations on morphological, structural and functional properties using characterization techniques that allow fast acquisition of large amounts of data that can be later used to develop machine learning models.<br/>The use of machine learning for materials science has increased in recent years with the goal to find hidden patterns within computational and experimental data that contribute to the optimization and discovery of materials and processes, thanks to the development of new algorithms, the increase of computational capabilities and the spread of big data tools.<br/>In this work, we present the methodology based on high-throughput experimentation and machine learning that we developed to speed up the optimization and fabrication of cuprate superconducting films through the innovative transient-liquid assisted growth (TLAG) CSD process which can reach growth rates up to 1000 nm/s by employing non-equilibrium reaction paths.<br/>We deposited compositional gradient films by drop-on-demand inkjet printing using different rare-earth cuprate precursor solutions. These solutions were mixed through the combination of specific printing positions in the right proportion to achieve the desired discrete/gradient compositions, enabling to study the influence of rare-earth and liquid phase compositional changes on the epitaxial growth of cuprate superconducting films.<br/>Data from deposition and growth experiments was obtained from characterization tools such as optical microscopy, scanning electron microscopy, electron X-ray spectroscopy, laboratory and synchrotron X-ray diffraction, and later used to develop machine learning models. These models were used with the purpose to optimize: a) the deposition process and investigate the homogeneous merging of precursor inks during inkjet printing, b) the epitaxial growth by identifying the best sample processing conditions as a function of the different compositions.<br/>Ultimately, the goal was to find patterns in the deposition and epitaxial growth characteristics of the different REBCO compositions that allow us to plan future experiments and help us optimize these processes.<br/>References:<br/>[1] Soler et al. Nat. Comm. 11, 344 (2020).<br/>[2] Rasi et al. J. Phys. Chem. C 124, 15574-15584 (2020).<br/>[3] Queraltó et al. ACS Appl. Mater. Interfaces, 13, 9101 (2021).