Optimized Materials in a Flash (newscenter.lbl.gov)

Berkeley Lab researchers unveiled AutoBot, an integrated robotics + machine-learning platform that autonomously synthesizes, characterizes, and optimizes advanced materials—demonstrated on metal halide perovskite thin films. Over a few weeks AutoBot explored a ~5,000‑point fabrication parameter space and, by actively selecting the most informative experiments, sampled under 1% of combinations to locate high‑quality recipes that would have taken up to a year with manual trial-and-error. The system’s rapid learning and closed-loop decision making point toward scalable, autonomous optimization labs that can dramatically accelerate materials discovery and process development. Technically, AutoBot varied four synthesis parameters (timing of a crystallization agent, heating temperature, heating duration, and chamber relative humidity), and characterized films with UV–Vis spectroscopy, photoluminescence (PL) spectroscopy, and PL imaging. A multimodal data‑fusion pipeline converted disparate outputs—including image homogeneity metrics—into a single quality score used by active‑learning models to maximize information gain per experiment. Key findings: high‑quality films are achievable at 5–25% relative humidity (relaxing strict environmental controls) while humidity above 25% destabilizes deposition. The approach, published in Advanced Energy Materials and developed at Berkeley’s Molecular Foundry, is generalizable across material systems and could lower barriers to industrial-scale manufacturing.