Understanding Design Rules of Colloidal Self-Assembly Using Autonomous Phase Mapping

Automation in many experimental pipelines at laboratory and central facilities has shifted the bottleneck to autonomously driven data analysis and decision-making. Exponential growth in tools available for data-driven modeling resulted in the advent of self-driven laboratories (SDL) that aim to automate and accelerate the entire workflow starting from synthesis to characterization and device integration for emerging technologies and energy needs. Platforms based on solution-processible materials (polymers, colloids, and nanoparticles) are amenable to automation both at the synthesis and characterization levels. Techniques such as scattering and spectroscopy provide faster high-throughput alternatives to capture a signal of the underlying structure allowing us to construct composition-structure phase maps. However, one of the common goals in ‘phase mapping’ a system is to accurately identify phase boundaries that can have materials with interesting structures and properties. In the realm of SDL, the problem of mapping phase boundaries is tackled using a combination of Bayesian active learning and data clustering. These techniques however cannot predict the phase map by ‘filling’ in the unexplored space that can provide information about the type of phase transition a boundary represents. We address this problem by reformulating the closed-loop phase mapping as a Bayesian active learning of a surrogate model that predicts measurement curves (spectroscopy, diffraction, or scattering). We apply the proposed approach to several classes of nano-scale colloidal and polymeric materials to learn phase maps that can be effectively used in understanding design rules to engineer colloidal self-assembly.