Autonomous exploration of non-equilibrium block copolymer assembly

Artificial intelligence and machine-learning methods hold enormous promise for accelerating the discovery of new materials. Autonomous experimentation (AE) leverages machine learning for decision-making within an experimental loop, allowing more efficient exploration of large and complex parameter spaces. However, several challenges remain in deploying these exciting approaches to realistic experimental problems. Here, we apply these methods to exploration of the multidimensional material and processing parameter spaces associated with non-equilibrium self-assembly of block copolymer (BCP) thin films, using AE to guide the execution of both small-angle x-ray scattering (SAXS) measurements, and molecular dynamics simulations of these materials. The experimental AE loop couples data acquisition, real-time data analytics, and a decision-making algorithm based on Gaussian Process modeling. The resultant multi-dimensional dataset is clustered using semi-supervised K-means in order to generate a catalog of structurally unique motifs. Molecular dynamics simulations elucidated the impacts of processing and material parameters on the diffusion of constituent BCP chains and the structural development of target nonequilibrium morphologies. Overall, this coordinated exploration enabled discovery of new non-native BCP architectures.