Machine-learning-assisted low dielectric constant polymer discovery

Abstract

Machine learning (ML) has excellent potential for molecular property prediction and new molecule discovery. However, real-world synthesis is the most vital part of determining a polymer's value. This paper demonstrates automatic polymer discovery through ML and an intelligent cloud lab to find new environmentally friendly polymers with low dielectric constants that have potential applications in high-speed communication networks. In the machine learning discovery, we use ML on SMILES from databases to identify ideal functional groups with reasonable solutions. Moreover, the solutions are sent to the cloud and synthesized via our intelligent system. A few of them can be successfully synthesized and two of them have excellent performance in low-dielectric-constant applications. This autonomous system enables reliable and efficient combinations of data-driven research and synthesis, reduces both the time and cost of polymer-discovery experiments, and accelerates the overall process for low-dielectric-constant polymer discovery.