MACHINE LEARNING ASSISTED OPTIMIZATION OF OXIDE NANOSTRUCTURES FOR RAPID AND SELECTIVE MULTI-GAS SENSING LIKE SNO2, NIO, NB2O3 OR NB2O5, ETC.

Abstract

Tin (IV) oxide (SnO₂), Nickel (II) oxide (NiO), Niobium(V) oxide (Nb₂O5) and Niobium (III) oxide (Nb₂O₃) are metal oxide semiconductor (MOS) nanostructures, which have thus far been at the heart of chemiresistive gas sensing due to their adjustable surface chemistry, high sensitivity, and ability to be microfabricated. Although heavily material-engineered, conventional MOS sensors typically lack cross-sensitivity, have low levels of inherent selectivity, have sluggish response-recovery kinetics, and drift in multi-gas conditions. Latest progress has demonstrated that the combination of machine learning (ML) and oxide nanostructured sensor arrays can greatly improve the performance of gas discrimination on complex atmospheres. As an example, more than 99 percent accuracy in classification of volatile organic compounds based on MOS sensor arrays when analyzed with a Random Forest and k-Nearest Neighbor algorithm, and deep learning methods have allowed detection of dangerous gases in real-time with very high accuracy. This is a critical review of the role of ML-assisted strategies in optimization in the form of supervised learning, feature engineering, dimensionality reduction and neural network structures in enhancing selectivity, drift compensation, and morphology-performance relationships in MOS sensors. This discussion points at the two issues, which include the lack of data and the interpretation of models, and research and development perspectives of self-calibration edge-AI-powered sensor arrays and physics-informed ML models transforming future gas sensing technologies.