András Magyarkuti

We explore the merits of neural network boosted, principal-component-projection-based,unsupervised data classification in single-molecule break junctionmeasurements, demonstrating that this method identifies highly relevanttrace classes according to the well-defined and well-visualized internalcorrelations of the data set. To this end, we investigate single-moleculestructures exhibiting double molecular configurations, exploring therole of the leading principal components in the identification ofalternative junction evolution trajectories. We show how the properprincipal component projections can be applied to separately analyzethe high- or low-conductance molecular configurations, which we exploitin 1/f-type noise measurements on bipyridine molecules. This approachuntangles the unclear noise evolution of the entire data set, identifyingthe coupling of the aromatic ring to the electrodes through the pi orbitals in two distinct conductance regions, and its subsequent uncouplingas these configurations are stretched.

In this paper, we review the 1/ f -type noise properties of nanoelectronic devices focusing on three demonstrative platforms: resistive switching memories, graphene nanogaps and single-molecule nanowires. The functionality of such ultrasmall devices is confined to an extremely small volume, where bulk considerations on the noise lose their validity: the relative contribution of a fluctuator heavily depends on its distance from the device bottleneck, and the noise characteristics are sensitive to the nanometer-scale device geometry and details of the mostly non-classical transport mechanism. All these are reflected by a highly system-specific dependence of the noise properties on the active device volume (and the related device resistance), the frequency, or the applied voltage. Accordingly, 1/ f -type noise measurements serve as a rich fingerprint of the relevant transport and noise-generating mechanisms in the studied nanoelectronic systems. Finally, we demonstrate that not only the fundamental understanding and the targeted noise suppression is fueled by the 1/ f -type noise analysis, but novel probabilistic computing hardware platforms heavily seek well tailorable nanoelectric noise sources.

We study the vulnerability of single-molecule nanowires against a temporary disconnection of the junction. To this end, we compare the room and low-temperature junction formation trajectories along the opening and closing of gold–4,4′-bipyridine–gold single-molecule nanowires. In the low-temperature measurements, the cross-correlations between the opening and subsequent closing conductance traces demonstrate a strong structural memory effect: around half of the molecular opening traces exhibit similar, statistically dependent molecular features as the junction is closed again. This means that the junction stays rigid and the molecule remains protruding from one electrode even after the rupture of the junction, and therefore, the same single-molecule junction can be reestablished if the electrodes are closed again. In the room-temperature measurements, however, weak opening–closing correlations are found, indicating a significant rearrangement of the junction after the rupture and the related loss of structural memory effects.

We investigate gold-4,4'-bipyridine-gold single-molecule junctions with the mechanically controllable break junction technique at cryogenic temperature (T=4.2K). We observe bistable probabilistic conductance switching between the two molecular binding configurations, influenced both by the mechanical actuation, and the applied voltage. We demonstrate that the relative dominance of the two conductance states is tunable by the electrode displacement, whereas the voltage manipulation induces an exponential speedup of both switching times. The detailed investigation of the voltage-tunable switching rates provides an insight into the possible switching mechanisms.

A combined principal component and neural network analysis serves as an efficient tool for the unsupervised recognition of unobvious but highly relevant trace classes in single-molecule break junction data.

We demonstrate Nb₂O₅ memristive devices with tuneable low-bias resistances and optimized high-bias impedance for high speed neuromorphic operation.

We present a new automated method for structural classification of the traces obtained in break junction experiments. Using recurrent neural networks trained on the traces of minimal cross-sectional area in molecular dynamics simulations, we successfully separate the traces into two classes: point contact or nanowire. This is done without any assumptions about the expected features of each class. The trained neural network is applied to experimental break junction conductance traces, and it separates the classes as well as the previously used experimental methods. The effect of using partial conductance traces is explored, and we show that the method performs equally well using full or partial traces (as long as the trace just prior to breaking is included). When only the initial part of the trace is included, the results are still better than random chance. Finally, we show that the neural network classification method can be used to classify experimental conductance traces without using simulated results for training, but instead training the network on a few representative experimental traces. This offers a tool to recognize some characteristic motifs of the traces, which can be hard to find by simple data selection algorithms.

We investigate pi-stacked dimer molecular junctions using conductance, flicker noise and force measurements.