ALD 2026

Generalized Reaction Networks for Atomic Layer Deposition

Speakers:
Simon Elliott‚ Thomas Ludwig‚ Thomas Hughes‚ Chloe Luyet‚ Jacob Gavartin

Abstract:
ALD processes are defined in terms of their underlying chemistry, namely self-limiting gas-surface reactions. Much work has been done to determine the reaction mechanism in specific cases, but a general understanding is lacking regarding the criteria for ALD and the resulting limits on growth rates and sticking coefficients. In this work, we develop a generic reaction network for the deposition or etching of metal oxides by ALD or CVD and use microkinetic modelling (MKM) to compute these measurable process characteristics as a function of process parameters. We use ZnO‚ Al2O3 and HfO2 as illustrative examples that span a range of metal valences and restrict ourselves to water as the co-reagent, though the extension to sulfides or nitrides in similar Bronsted acid-base chemistry would be straightforward. Building on past mechanistic studies‚ we identify the elementary forward reactions that together comprise the reaction network as (i) precursor/co-reagent adsorption‚ (ii) ligand or proton exchange‚ (iii) elimination of protonated ligands as by-products and (iv) densification into solid film. The reverse of each reaction is also included in the reaction network. We streamline the network by omitting linearly dependent reactions. Since these elementary steps convert one surface intermediate into another, the size of the network scales in principle with the square of the number of intermediates. We therefore restrict the number of surface intermediates to the minimum for the number of ligands per precursor and number of protons per water molecule. Activation free energies for each elementary step would typically be computed with DFT, but here our interest is in how the pattern of relative activation energies across the network affects the overall process. Running MKM simulations of multiple pulse-purge cycles, we establish the bounds for ALD versus CVD behavior in terms of the reactivity of individual metals and ligands. The relative kinetics of ligand transfer and proton transfer are found to be the crucial factor.Having used MKM to identify the chemical spaces where ALD is viable‚ we then derive the corresponding ranges of growth/etch per cycle and sticking coefficients‚ which are measurable characteristics of the growth/etch process and can be used as inputs to higher-scale simulations. We present the dependence of these characteristics on precursor mass, metal valence, process temperature and pulse pressure.