Yujia Wang, Guoyan Li, Xiaoning Jin, Swastik Kar
We present a machine learning (ML) guided approach for the accelerated optimization of chemical vapor deposition (CVD) synthesis of 2D materials towards the highest quality, starting from low-quality or unsuccessful synthesis conditions. Using 26 sets of synthesis conditions as our initial training dataset, we could systematically progress towards optoelectronic-grade monolayer MoS2 flakes with A-exciton linewidth (σA) as narrow as 38 meV after only an additional 35 trials (reflecting only 15% of the full factorial design dataset for training purposes). This translates to an 85% reduction in wasteful “trial-and-error” experiments. This remarkable efficiency, without any domain knowledge intervention, was accomplished by formulating a constrained sequencing optimization problem solved via a combination of constraint learning and Bayesian Optimization. We provide a clear visualization of “sweet spots” for a CVD reactor to an experimentalist. Our method is scalable to a higher number of synthesis parameters and target metrics and is transferrable to other materials and types of reactors.