Wavelet Transform-Based Clustering Decoding the Dynamical Properties of Single Molecular Junctions

The ability to precisely identify the dynamic properties of tunneling molecular junctions is crucial for establishing functional molecular devices. However, the minor signals and events are normally obscured by the dominant signals/events. To this end, we put forward an unsupervised clustering method, which achieved 100% classification accuracy with a simulated data set. Applying it to analyze the conductance traces of dithiothreitol molecular junctions, we uncovered that the molecular dimer junctions can highly likely be formed via both S-S bonding and hydrogen bonding, which is extremely difficult to be distinguished by only experimental approaches. For aromatic molecules, we revealed that stable trimer junctions might be formed, and such trimer junctions tend to change configuration upon the reversed bias, a property that was completely masked by the traditional conductance histogram analysis. Beyond molecular electronics, this work provides an automated method to catch small-probability events and reveal the unique properties hidden in a large amount of data.