High Throughput Characterization of Combinatorial Library Based on SPM Based with LLMs, Self-Tuning and Reward-Driven Automation

Combinatorial libraries offer a high-throughput way to explore different compositions and growing conditions on a single substrate. However, materials discovery based on combinatorial libraries is bottlenecked by the speed of material characterization. Here we present a fully automated SPM system for high throughput characterization of combinatorial libraries. This SPM system can re-tune itself in both imaging mode and spectroscopy mode at different locations of the library to make sure the measurement conditions are the same across the library. We also implemented an image filter system with the help of Large Language Models (LLMs). This system can filter out good areas in the scan so that all the measurements are performed in sample regions with comparable qualities. We will also show implementation of automated workflows with measured physical properties as rewards on real SPM instrument. Finally, we will present the automated discovery of real binary ferroelectric combinatorial libraries of SmBFO and MZO, and ternary library of AlBN.