From Surface Reactions to Inhibitor Coverage: A Machine-Learning-Assisted Computational Study of Area-Selective Atomic Layer Deposition

Abstract

Area-selective atomic layer deposition (AS-ALD) enables selective material growth for the fabrication of nanoscale semiconductor devices. Predicting inhibitor coverage is essential for understanding and optimizing surface blocking in AS-ALD. Here, we present a computational workflow combining a pretrained universal machine-learning interatomic potential with off-lattice grand canonical Monte Carlo simulations and apply it to inhibitor adsorption and precursor blocking on amorphous SiO2. Realistic surface models reveal substantial site-dependent variations in reaction energies and activation barriers, with silanol groups serving as the primary adsorption sites for aminosilane and related inhibitors and metal precursors. Coverage simulations show that inhibitors producing identical adsorbed fragments can yield different coverages because the intact molecules impose distinct steric constraints, leading to different accessible site densities and blocking capabilities. Beyond coverage, we highlight the importance of the reactivity of adsorbed inhibitor fragments toward precursors in determining inhibitor stability and blocking performance. Together, these findings demonstrate the need to consider both steric effects and chemical reactivity on realistic amorphous surfaces. Our calculations connect local reaction chemistry with inhibitor coverage and provide a computational basis for understanding inhibition mechanisms and selecting suitable inhibitor-precursor combinations for AS-ALD.