Seminar Dr Murilo Moreira: "Quantitative EDS Analysis supported by unsupervised machine learning. Chemical gradients in small nanoalloy and a taste in EELS nanoplasmonics"

"Quantitative EDS Analysis supported by unsupervised machine learning. Chemical gradients in small nanoalloy and a taste in EELS nanoplasmonics"
Seminar by Dr Murilo Moreira, Institut Lumière Matière, Villeurbannes

Abstract

Advancements in Energy-dispersive X-ray Spectroscopy (EDS) combined with Scanning Transmission Electron Microscope (STEM) have significantly improved the analysis of chemical composition in nanoscale objects, particularly bimetallic nanoparticles (BNPs) smaller than 10 nm. This breakthrough enables precise quantitative analysis of individual particles. BNPs exhibit diverse physico-chemical properties based on their chemical structure (e.g., random, gradient, or core-shell). Achieving high signal-to-noise ratios (SNR) at such small scales is challenging, requiring robust data acquisition, treatment methods, and rigorous error analysis for a complete comprehension of the BNPs chemical structure.

In this presentation, the study of the chemical composition of AgAu benchmark BNPs will be discussed. The results reveal chemical gradients featuring Ag enrichment towards the particle surface, due to differences in surface energy between the two elements. To further validate these findings, unsupervised machine learning methods (Principal Component Analysis-PCA and Non-negative Matrix Factorization-NMF) were applied, providing statistically reliable verification of radial composition changes observed in EDS spectra. This study demonstrates the ability to quantify chemical composition within 3-9 nm BNPs, offering opportunities for further research on segregation effects and chemical reactivity. The particles were coated with a thin carbon film (~20 nm), annealed, and analyzed after structural relaxation to the ground state. Notably, these findings show that the two elements are well mixed in the BNP volume, with Ag segregation occurring mostly in the last atomic layers (~1 nm) of the BNP surface. Finally, using similar methodologies of unsupervised machine learning, as it was done with EDS, spectral and spatial correlations can also be found in low loss Electron Energy Loss Spectroscopy (EELS) experiments. These preliminary results might help to improve the understanding of the different plasmon modes existent in small metallic nanoparticles.

This seminar will be held in English