From trial-and-error to intelligent workflows: machine learning and simulations for scalable 2D material synthesis

Abstract Wafer-scale growth of two-dimensional (2D) materials remains an unsolved challenge. Current synthesis still depends on trial-and-error, limiting reproducibility and slowing progress toward practical technologies. This perspective argues that machine learning (ML) and atomistic simulations can provide the foundation for intelligent synthesis workflows. Gradient-free optimization methods, including Bayesian optimization and adaptive Monte Carlo, already demonstrated that high-quality growth conditions can be learned efficiently from experimental feedback. At the same time, atomistic simulations deliver physical priors, synthetic data, and mechanistic insights that focus the search and mitigate data scarcity. We propose a synergistic framework where experiments, simulations, and ML form a closed loop: experiments provide validation, simulations constrain exploration, and ML uncovers patterns and proposes new recipes. Embedding this intelligence into growth protocols can turn 2D materials synthesis into a reproducible and scalable methodology, moving beyond intuition-driven experimentation, accelerating their translation into wafer-scale technologies. While current demonstrations remain system-specific, this perspective provides a pathway toward wafer-scale technologies and data-driven discovery.