Spatially Resolved Band Gap and Dielectric Function in Two-Dimensional Materials from Electron Energy Loss Spectroscopy

Abstract

The electronic properties of two-dimensional (2D) materials depend sensitively on the underlying atomic arrangement down to the monolayer level. Here we present a novel strategy for the determination of the band gap and complex dielectric function in 2D materials achieving a spatial resolution down to a few nanometers. This approach is based on machine learning techniques developed in particle physics and makes possible the automated processing and interpretation of spectral images from electron energy loss spectroscopy (EELS). Individual spectra are classified as a function of the thickness with K -means clustering, and then used to train a deep-learning model of the zero-loss peak background. As a proof of concept we assess the band gap and dielectric function of InSe flakes and polytypic WS 2 nanoflowers and correlate these electrical properties with the local thickness. Our flexible approach is generalizable to other nanostructured materials and to higher-dimensional spectroscopies and is made available as a new release of the open-source EELSfitter framework.

Heterostrain‐Driven Bandgap Increase in Twisted WS2: A Nanoscale Study

“…To scrutinise the local electronic properties of these twisted WS 2 specimens, we employed spatially‐resolved electron energy‐loss spectroscopy (EELS) and analyzed the acquired spectral images using the EELSfitter framework. [ 34,35 ] This approach involved parameterizing the dominant zero‐loss peak (ZLP) background using deep neural networks, [ 36 ] with the Monte Carlo replica method [ 37 ] ensuring robust error estimate and propagation. We then subtracted the ZLP pixel by pixel in the spectral image.…”

Localized Exciton Anatomy and BandGap Energy Modulation in 1D MoS2 Nanostructures

“…Our focus is primarily the low-loss region, with energy losses ΔE ⩽ 50 eV, providing direct insights into the sought-for excitonic and plasmonic behavior. This analysis builds upon the open-source EELSfitter framework [15,16] for the processing of spatiallyresolved EELS data, here extended with automated peak-tracking algorithms.…”

“…Complementary insights on the local electronic properties of the 1D-MoS 2 nanostructures are provided by the spatial distribution of their bandgap energy E bg . Here spatially-revolved maps of E bg are determined by analyzing individual EEL spectra after subtracting the ZLP background, following the procedure in [15,16] and summarized for completeness in Section S4 and S5 (Supporting Information). Figure 5 displays the E bg maps associated to the 1D-MoS 2 nanostructures studied in Figures 3 (individual) and 4 (connected).…”

“…In this study, we undertake a thorough investigation of localized excitons in 1D-MoS 2 nanostructures, complemented by the assessment of their bandgap energy modulation. This is achieved by leveraging and extending our previously developed approach [15,16] for the automated processing and interpretation of electron energy-loss spectroscopy (EELS) spectral images (SI) with nanometer spatial resolution implemented within the opensource EELSfitter framework. We identify several excitonic transitions and plasmonic resonances in 1D-MoS 2 nanostructures, echoing previous findings in the 2D-MoS 2 case.…”

Alignment-invariant signal reality reconstruction in hyperspectral imaging using a deep convolutional neural network architecture

“…Although ML methods, especially deep learning (DL), have been extensively used for image and signal deconvolution or feature detection and classification [36][37][38][39][40][41][42][43] , their capability in dealing with spectral features (broadened, low dose features) that may have scientific significance in an extremely distorted signal is less investigated particularly for near zero-loss EELS signals and in terms of validating physical reality of the signal. In this regard, publications towards low-loss EELS signal processing are mainly limited to study either denoising the signal or improving the background signal [44][45][46][47] .…”