Machine learning-based exploration of molecular design descriptors for area-selective atomic layer deposition (AS-ALD) precursors.

Abstract

Area-selective atomic layer deposition (AS-ALD) is a thin film deposition technique developed using conventional ALD by considering the surface chemical nature of the substrate. Selecting appropriate precursors is a critical step in developing an efficient AS-ALD process with high deposition selectivity. However, the current efficiency of research on viable AS-ALD precursors is limited because of the absence of theoretical design rules for precursor chemical structures. In this study, our objective is to propose molecular design principle for precursors for AS-ALD, particularly focusing on achieving high deposition selectivity of oxides on diverse substrates. Current preliminary results suggest that ML-based prediction model may provide a fundamental molecular-level understanding of the reactivity of metal oxide precursors, that can be useful for efficient selection of suitable precursors for AS-ALD. We employ density functional theory (DFT) calculations and machine learning (ML) techniques to analyze the relationship between the structure and the surface reactivity of the precursor. Considering DFT calculation data (M06L/def2-tzvp, Gaussian 09 and Orca 4.0) and information on precursor structures, artificial neural networks (ANN, neuralnet, R) are applied to identify critical descriptors of the AS-ALD process. Furthermore, we utilize this ANN model to predict precursor reactivity according to surface terminations.