Multidimensional analysis of excitonic spectra of monolayers of tungsten disulphide: toward computer-aided identification of structural and environmental perturbations of 2D materials

Abstract

Despite 2D materials holding great promise for a broad range of applications, the proliferation of devices and their fulfillment of real-life demands are still far from being realized. Experimentally obtainable samples commonly experience a wide range of perturbations (ripples and wrinkles, point and line defects, grain boundaries, strain field, doping, water intercalation, oxidation, edge reconstructions) significantly deviating the properties from idealistic models. These perturbations, in general, can be entangled or occur in groups with each group forming a complex perturbation making the interpretations of observable physical properties and the disentanglement of simultaneously acting effects a highly non-trivial task even for an experienced researcher. Here we generalise statistical correlation analysis of excitonic spectra of monolayer WS2, acquired by hyperspectral absorption and photoluminescence imaging, to a multidimensional case, and examine multidimensional correlations via unsupervised machine learning algorithms. Using principal component analysis we are able to identify four dominant components that are correlated with tensile strain, disorder induced by adsorption or intercalation of environmental molecules, multi-layer regions and charge doping, respectively. This approach has the potential to determine the local environment of WS2 monolayers or other 2D materials from simple optical measurements, and paves the way toward advanced, machine-aided, characterization of monolayer matter.

Multidimensional Imaging Reveals Mechanisms Controlling Multimodal Label-Free Biosensing in Vertical 2DM-Heterostructures

“…In our sample with only two substances, the intensity of the fingerprint lines of DOX already allows us to confirm the analyte structure and determine the presence of analyte (we note that it cannot be extracted from PL data channel alone, which could respond to other components of sample). While in general, for an agnostic biosensor, the whole Raman spectrum should be analyzed by machine learning correlation analysis of the data. ,, Here, GERS, the second data channel, complements the PL detection. PL channel provides information on the concentration of the drug (while the intensity of the GERS signal depends on enhancement factors and cannot be used to measure the amount of analyte).…”

“…Structural characterization with a high spatial resolution, capable to resolve such defects, such as electron microscopy, often does not detect materials optical properties, while optical microscopy lacks the required resolution. Thus, in order to reveal mechanisms that control sensing, multiple characterization tools should be combined and correlated . In this work, correlated multidimensional imaging, including Raman and near-field microscopies, scanning probe, and electron microscopies, was applied to unveil physical processes behind label-free multimodal detection of doxorubicin (DOX), an anthracycline cancer drug, by 2DM vertical heterostructures.…”

“…Thus, in order to reveal mechanisms that control sensing, multiple characterization tools should be combined and correlated. 19 In this work, correlated multidimensional imaging, including Raman and near-field microscopies, scanning probe, and electron microscopies, was applied to unveil physical processes behind labelfree multimodal detection of doxorubicin (DOX), an anthracycline cancer drug, by 2DM vertical heterostructures. Doxorubicin is one of the most common drugs against different types of cancer (hematological, thyroid, breast, ovarian, lung, and liver cancer).…”

Molecular excited states through a machine learning lens

“…[26][27][28][127][128][129] For example, unsupervised learning can be used to analyze how the spectrum depends on the structure and environment, enabling the formulation of rules for structure determination. 68,69,135,136 Such insight may also be useful for designing new materials, as we discuss in the section Design of optoelectronic materials.…”

Multidimensional imaging reveals mechanisms controlling label-free biosensing in vertical 2DM-heterostructures

“…Structural characterization with a high spatial resolution, such as electron microscopy, often does not detect materials optical properties, while optical microscopy lacks the required resolution. Thus, in order to reveal such mechanisms, multiple characterization tools should be combined and correlated 18 . In this work, correlated multidimensional imaging, including Raman and near-field microscopies, scanning probe and electron microscopies, was applied to unveil physical processes behind label-free multimodal detection of doxorubicin (DOX), an anthracycline cancer drug, by 2DM vertical heterostructures.…”