Harnessing Machine Learning and Automation to Unlock High-Throughput Composition Analysis of Powder Materials

Self-driving labs hold the promise to substantially accelerate the process of material innovation by reducing the time required to synthesize novel materials [1]. We employ the A-lab, a robotic setup for inorganic solid-state synthesis [2]. Starting from solid powdered precursors, A-lab autonomously doses and mixes the powders, and fires them in a controlled atmosphere. The reaction products are then autonomously characterised via X-Ray Diffraction (XRD) and automated Rietveld refinement [3]. Hence, for each reaction, the A-lab identifies the crystal phases and their relative fractions. However, relying on a single characterization method has limitations. XRD cannot detect amorphous products, and closely overlapping peaks from different phases can make it difficult to identify the obtained reation product with a high confidence. Multiple complementary characterization methods are desirable to accurately determine the reaction products and strengthen the confidence in the characterisation results.<br/><br/>To this end, we have implemented Auto-SEMEDX, a framework for automated Energy-Dispersive X-Ray Spectroscopy (EDX) at a Scanning Electron Microscope (SEM) to measure the composition of the reaction products. EDX provides information about amorphous phases and can be employed to cross-check XRD results by offering a second method to determine phase composition.<br/><br/>Samples are (autonomously) prepared via a setup built in-house, where micron-sized particles are transferred from a crucible containing powder reaction products onto a carbon-taped SEM stub. The stubs are transferred to the SEM, and Auto-SEMEDX is initiated, proceeding as follows:<br/>- Single particles of few microns are located on the C tape via image processing.<br/>- EDX spectra are collected at the particles.<br/>- Machine-learning algorithms such as k-means clustering and non-negative matrix factorization are employed to determine the phases observed in the sample.<br/>- The measured compositions are cross-checked with the XRD-determined phases, and overall confidences for each phase are calculated using Bayes’ theorem.<br/>The process repeats iteratively until phases can be determined with high confidence.<br/><br/>Auto-SEMEDX rapidly and autonomously collects compositional data with statistical significance, allowing to determine the composition of phases present in the sample with a high confidence. By combining automated SEM EDX analysis and XRD refinement, this work showcases the power of complementary characterization techniques in a robotic synthesis lab, reducing human workload.<br/><br/><br/>References:<br/><br/>[1] Abolhasani, M., Kumacheva, E. The rise of self-driving labs in chemical and materials sciences. <i>Nat. Synth</i> <b>2</b>, 483–492 (2023).<br/>[2] N.J. Szymanski, B. Rendy, Y. Fei, R.E. Kumar, T. He, D. Milsted, M.J. McDermott, M. Gallant, E.D. Cubuk, A. Merchant, H. Kim, A. Jain, C.J. Bartel, K. Persson, Y. Zeng, G. Ceder, An autonomous laboratory for the accelerated synthesis of novel materials, Nature. 624 (2023) 86–91.<br/>[3] Szymanski, N.J., Bartel, C.J., Zeng, Y. <i>et al.</i> Adaptively driven X-ray diffraction guided by machine learning for autonomous phase identification. <i>npj Comput Mater</i> <b>9</b>, 31 (2023).