Dissecting Time-Evolved Conductance Behavior of Single Molecule Junctions by Nonparametric Machine Learning

Abstract

Improved understanding of charge transport in single molecules is essential for utilizing their potential as circuit components at the nanosize limit. However, reliable analyses of varying tunneling current acquired by break junction experiments remain an ongoing challenge to find molecular feature structure−property relationships. In this work, we report on an unsupervised learning approach for investigating molecular signatures in conductance traces. Our hybrid machine learning algorithm compares grids of data in conductance−time domains and judges the similarity without any researcher-crafted parameters to identify fine molecular components that may otherwise be obscured by background fluctuations. We demonstrate its ability for classifying Au−alkanedithiol−Au conductance traces acquired with microfabricated mechanically controllable break junctions. The unbiased procedure was able to not only judge the presence or absence of the carbon chains in the electrode gap but also to identify multiple conductance states of the molecular tunneling junctions with different conformations. This finding may offer a useful tool for studying single-molecule properties using break junction methods.

Using Weakly Supervised Deep Learning to Classify and Segment Single‐Molecule Break‐Junction Conductance Traces

“…Recently machine learning has been introduced into molecular electronics, [10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27] as a powerful tool to analyze break junction data. Supervised and unsupervised learning are the two main classes of machine learning algorithms, their main difference being whether or not a manually labeled training set is needed.…”

“…Supervised and unsupervised learning are the two main classes of machine learning algorithms, their main difference being whether or not a manually labeled training set is needed. Due to limited labor costs and the potential advantage of avoiding predefined bias in solving specific problems, many previous studies prefer unsupervised learning methods, [10][11][12][13][14][15][16][17][18][19][20][21][22] such as multi-parameter vector-based classification (MPVC), [10] deep auto-encoder K-means (DAK), [11] K-means + + , [12] principal component analysis (PCA), [13] Alexnet-enhanced autoencoder [14] and spectral clustering. [19] In general, the conductance traces collected for a certain molecule or different molecules are grouped by their mutual similarities to uncover the underlying features.…”

Grid-Based Correlation Analysis to Identify Rare Quantum Transport Behaviors

“…One strategy for separating qualitatively different behaviors that may overlap in 1D and 2D histograms is to employ clustering. Indeed, over the past five years several clustering approaches have been designed specifically for breaking traces 34,[43][44][45][46][47][48][49][50][51][52][53] and related data, 54,55 and these approaches have had varied success in extracting known and potential features, including "hidden" features, from real and simulated datasets. However, clustering algorithms by definition look for groupings of similar data, and so they will always struggle with truly rare behaviors that may not be common enough to form a clear grouping.…”

Advancing Physical Chemistry with Machine Learning

“…The efficiency of calculations of excited-state properties and nonadiabatic dynamics of large systems is enhanced by combining ML with fragmentation methods and by direct deep learning of relevant properties with an automatically determined representation of molecular structures . ML analyses have been applied to various spectroscopies, including prediction of surface-enhanced Raman spectra (SERS) and analysis of complex electron paramagnetic resonance (EPR) signals and molecular conductance traces …”

“…6 ML analyses have been applied to various spectroscopies, including prediction of surface-enhanced Raman spectra (SERS) 7 and analysis of complex electron paramagnetic resonance (EPR) signals 8 and molecular conductance traces. 9 Both supervised and unsupervised ML has demonstrated great successes in thermodynamics and statistical mechanics. Recent examples include the deep NN study of the inverse problem of the liquid-state theory aimed at obtaining interaction potentials from distribution functions, 10 the prediction of distribution functions, 11 and the extraction of the order parameter that leads to universality of supercritical fluid properties.…”