A Multiscale Framework for Atomic Layer Deposition/Atomic Layer Etching Process and Precursor Development Using Neural-Network-Potential Atomistic Simulations and COSMO-SAC Thermodynamic Modeling
Abstract: Atomic layer deposition (ALD) is essential for advanced semiconductor fabrication because it enables excellent conformality and precise thickness control in highly scaled 3D structures. To establish robust ALD/atomic layer etching (ALE) processes, both suitable precursor properties and detailed understanding of surface reactions are required.
We present a multiscale framework that combines COSMO-SAC thermodynamic modeling for precursor design and down-selection with neural-network-potential (NNP) atomistic simulations for analyzing ALD/ALE reaction mechanisms.
We improved COSMO-SAC for metal-organic precursors by tuning dispersion-related parameters, achieving good agreement between predicted and measured vapor pressures. For example, this method predicted high volatility for a Co precursor, and the synthesized compound showed ~1060 Pa at 85 degrees C. Such predictive capability is also useful for ALE, where volatility of reaction products is critical.
NNP-based simulations enabled practical analysis of surface reactions at near-density functional theory accuracy. For Co deposition, they revealed that controlling surface hydrogen coverage can suppress carbon incorporation, which was experimentally confirmed. For area-selective deposition (ASD), simulations showed that a Co precursor reacts readily on Cu but poorly on SiO2, consistent with selective-growth experiments.
To overcome selectivity loss caused by Co nuclei formed on SiO2, we designed an ALE pathway using SO2Cl2 and Hfac to generate volatile products. COSMO-SAC supported their volatility, and the integrated ALD + ALE scheme successfully removed unwanted nuclei, providing a route to robust ASD.
This framework demonstrates how precursor design, reaction analysis and integrated ALD/ALE process design can be co-optimized for advanced atomic layer processing.