Unsupervised Segmentation-Based Machine Learning as an Advanced Analysis Tool for Single Molecule Break Junction Data

Improved understanding of charge-transport in single molecules is essential for harnessing the potential of molecules, e.g., as circuit components at the ultimate size limit. However, interpretation and analysis of the large, stochastic data sets produced by most quantum transport experiments remain an ongoing challenge to discovering much-needed structure–property relationships. Here, we introduce segment clustering, a novel unsupervised hypothesis generation tool for investigating single molecule break junction distance–conductance traces. In contrast to previous machine learning approaches for single molecule data, segment clustering identifies groupings of similar pieces of traces instead of entire traces. This offers a new and advantageous perspective into data set structure because it facilitates the identification of meaningful local trace behaviors that may otherwise be obscured by random fluctuations over longer distance scales. We illustrate the power and broad applicability of this approach with two case studies that address common challenges encountered in single molecule studies: First, segment clustering is used to extract primary molecular features from a varying background to increase the precision and robustness of conductance measurements, enabling small changes in conductance in response to molecular design to be identified with confidence. Second, segment clustering is applied to a known data mixture to qualitatively separate distinct molecular features in a rigorous and unbiased manner. These examples demonstrate two powerful ways in which segment clustering can aid in the development of structure–property relationships in molecular quantum transport, an outstanding challenge in the field of molecular electronics.