Application of machine learning on quantitative XPS analysis for heteroatoms (F, La and N) Alloyed HfO2 thin film
Journal
Surfaces and Interfaces
Journal Volume
72
ISSN
2468-0230
Date Issued
2025-09
Author(s)
Tseng, Yin-Bo
Abstract
The urgent need for new state-of-the-art materials has necessitated a rapid advancement of analytical tools for material science, first and foremost among them is X-ray Photoelectron Spectroscopy (XPS). With the advancement of materials science, we have found that XPS spectra can be influenced by lattice distortions caused by alloyed heteroatoms. Therefore, our aim is to identify the correlation between XPS spectral features and lattice distortion, specifically in terms of alloying element concentration. Alongside the boost in computational power in recent years, we have built an Artificial Neural Network (ANN) machine learning model to quantify the concentration of heteroatoms (F, La and N) with various sets of concentration alloyed in HfO2 samples. In this work, we discuss not only the accuracy of the prediction results, but also examine the prediction mechanism of the ANN model to provide a guide for proper model training. Our preliminary results show that most of the models are able to precisely predict the unknown sample as long as the element of interest is within its training process. Also, we characterize the model by the SHapley Additive exPlanation (SHAP) method, which further suggests that the characterization mechanism is varied from element to element. The success prediction of our work provides a novel approach in complex material characterization.
Subjects
Machine learning
Material characterization
Quantitative analysis
Surface science
X-ray photoelectron spectroscopy
Publisher
Elsevier BV
Type
journal article