Presentation Information
[11a-B21-4]Kinetic Monte Carlo Study of Defect-Mediated Morphological Evolution during TMA/H2O Al2O3 Atomic Layer Deposition
〇(M2)Yichen ZOU1, Yuxuan Wu1, Noboru Sato1, Atsuhiro Tsukune1, Yukihiro Shimogaki1 (1.The Univ. of Tokyo)
Keywords:
Atomic Layer Deposition,Kinetic Monte Carlo,Neural Network Potential
Atomic layer deposition (ALD) is a key thin-film growth technique for advanced semiconductor applications because of its atomic-scale thickness control and conformal film formation. For the trimethylaluminum/water (TMA/H2O) process of Al2O3, morphological evolution is governed not only by growth rate but also by defect chemistry, including residual carbon and sub-stoichiometric voids on the Al and O sublattices. Although such defects are known experimentally to depend on deposition temperature, quantitatively linking surface reactions to in-film morphology remains challenging because demethylation, hydrolysis, and vacancy filling jointly convert local surface states into buried film structure.
In this study, neural-network-potential (NNP) atomistic analysis is coupled with an on-lattice kinetic Monte Carlo (KMC) framework to identify the residual-carbon formation mechanism and the distinct formation pathways of two void populations: Al voids and O voids. The TMA/H2O ALD process is simulated over 120–240 °C (393–513 K) under a fixed 1-5-5-5 s pulse/purge sequence. The model contains approximately 1400 elementary reaction events for TMA chemisorption, sequential demethylation, water adsorption/hydrolysis, and Al-vacancy filling.
The simulated film is divided into substrate, bulk, surface, and vacuum regions, yielding well-defined bulk plateaus (Fig. 1). Over 120–240 °C, Al voids decrease from 8.63% to 6.12% and O voids from 6.42% to 3.94%, while residual bulk methyl carbon decreases from 1.00 to 0.68 at%, in good agreement with the experimental XPS trend within 0.1–0.3 at%. These simultaneous reductions indicate that elevated temperature promotes both ligand removal and vacancy filling, leading to a denser film with fewer frozen-in voids. The model therefore provides a quantitatively validated description of defect-mediated morphological evolution during TMA/H2O Al2O3 ALD.
In this study, neural-network-potential (NNP) atomistic analysis is coupled with an on-lattice kinetic Monte Carlo (KMC) framework to identify the residual-carbon formation mechanism and the distinct formation pathways of two void populations: Al voids and O voids. The TMA/H2O ALD process is simulated over 120–240 °C (393–513 K) under a fixed 1-5-5-5 s pulse/purge sequence. The model contains approximately 1400 elementary reaction events for TMA chemisorption, sequential demethylation, water adsorption/hydrolysis, and Al-vacancy filling.
The simulated film is divided into substrate, bulk, surface, and vacuum regions, yielding well-defined bulk plateaus (Fig. 1). Over 120–240 °C, Al voids decrease from 8.63% to 6.12% and O voids from 6.42% to 3.94%, while residual bulk methyl carbon decreases from 1.00 to 0.68 at%, in good agreement with the experimental XPS trend within 0.1–0.3 at%. These simultaneous reductions indicate that elevated temperature promotes both ligand removal and vacancy filling, leading to a denser film with fewer frozen-in voids. The model therefore provides a quantitatively validated description of defect-mediated morphological evolution during TMA/H2O Al2O3 ALD.