Bayesian Optimization of Colloidal Monolayer Formation in Langmuir–Blodgett System

The practical application of colloidal self-assembly in lithography is hindered by two main challenges: defect formation and sensitivity to multiple process parameters. Conventional empirical optimization, often referred to as the trial-and-error approach, is inefficient and often yields suboptimal results, thereby impeding scalability and commercial implementation. In this study, we overcome these limitations by employing a Bayesian optimization algorithm to accelerate the formation of large-scale defect-free hexagonal close-packed (HCP) arrays from polystyrene spheres (1.25 and 1.8 μm) using a custom-built automated Langmuir–Blodgett system. Our approach identified the optimal combination of six parameters in just 13 iterations, achieving a defect-free domain area of 16,000 μm2 and 23,000 μm2 for 1.25 μm and 1.8 μm particles, respectively. Multiple regression analysis revealed common key factors influencing domain growth: increased surface pressure and transfer rate, as well as reduced ethanol concentration. We found that pH is also one of the most significant factors, however, its effect strongly depends on particle size. This trend was in good agreement with zeta potential measurements. Our study demonstrates that Bayesian optimization serves as a powerful machine learning tool for the controlled fabrication of colloidal monolayers, enabling rapid development of lithographic masks with tailored properties.