Single-molecule break-junction measurements are intrinsically stochastic in nature, requiring the acquisition of large datasets of “breaking traces” to gain insight into the generic electronic properties of the molecule under study. For example, the most probable conductance value of the molecule is often extracted from the conductance histogram built from these traces. In this letter, we present an unsupervised and reference-free machine learning tool to improve the determination of the conductance of oligo(phenylene ethynylene)dithiol from mechanically controlled break-junction (MCBJ) measurements. Our method allows for the classification of individual breaking traces based on an image recognition technique. Moreover, applying this technique to multiple merged datasets makes it possible to identify common breaking behaviors present across different samples, and therefore to recognize global trends. In particular, we find that the variation in the extracted molecular conductance can be significantly reduced resulting in a more reliable estimation of molecular conductance values from MCBJ datasets. Finally, our approach can be more widely applied to different measurement types which can be converted to two-dimensional images.
Graphene nanoribbons (GNRs) have attracted a strong interest from researchers worldwide, as they constitute an emerging class of quantum-designed materials. The major challenges towards their exploitation in electronic applications include reliable contacting, complicated by their small size (< 50 nm),as well as the preservation of their physical properties upon device integration. In this combined experimental and theoretical study, we report on the quantum dot (QD) behavior of atomically precise GNRs integrated in a device geometry. The devices consist of a film of aligned 5-atoms wide GNRs (5-AGNRs) transferred onto graphene electrodes with a sub 5-nm nanogap. We demonstrate that the narrow-bandgap 5-AGNRs exhibit metal-like behavior at room temperature and single-electron transistor behavior for temperatures below 150 K. By performing spectroscopy of the molecular levels at 13~K, we obtain addition energies in the range of 200-300 meV. DFT calculations predict comparable addition energies and reveal the presence of two electronic states within the bandgap of infinite ribbons when the finite length of the 5-AGNRs is accounted for. By demonstrating the preservation of the 5-AGNRs molecular levels upon device integration, as demonstrated by transport spectroscopy, our study provides a critical step forward in the realisation of more exotic GNR-based nano-electronic devices.
Unsupervised machine learning, and in particular data clustering, is a powerful approach for the analysis of datasets and identification of characteristic features occurring throughout a dataset. It is gaining popularity across scientific disciplines and is particularly useful for applications without a priori knowledge of the data structure. Here, we introduce an approach for unsupervised data classification of any dataset consisting of a series of univariate measurements. It is therefore ideally suited for a wide range of measurement types. We apply it to the field of nanoelectronics and spectroscopy to identify meaningful structures in data sets. We also provide guidelines for the estimation of the optimum number of clusters. In addition, we have performed an extensive benchmark of novel and existing machine learning approaches and observe significant performance differences. Careful selection of the feature space construction method and clustering algorithms for a specific measurement type can therefore greatly improve classification accuracies.
By studying transport through seven structurally related porphyrin derivatives with a machine learning algorithm we could identify and distinguish three different electronic paths.