Wet Hydrofluoric Acid Etching Reaction Mechanism Analysis of Silicon Oxide Using GRRM with Universal Neural Network Potential

In semiconductor manufacturing technology that is advancing towards miniaturization, the improvement of wet etching process reactions at the atomic level, which utilizes dilute hydrofluoric (HF) acid solution to remove oxides from substrate surfaces, has become increasingly important. Although there has been active research aimed at understanding at the atomic level, it has been difficult to handle the formation and cleavage of bonds in molecular dynamics simulations, and due to the high computational cost, first-principle calculations have struggled with the analysis of extensive phenomena. Despite these difficulties, several studies have been conducted on the formation process of SiF4.<br/>In the case of MD simulation, studies have been conducted on slab models using ReaxFF[1]. In the case of DFT simulations, cluster models[2] and slab models[3][4] have been investigated. However, these studies have been conducted under anhydrous conditions or in the presence of NH4, and the elementary reactions in the presence of surrounding H2O molecules, which are closer to the wet etching environment, have not yet been reported.<br/>Recently, the development of neural network potentials(NNP) has been actively pursued, enabling high-speed examination of reaction pathways for systems with hundreds of atoms at the DFT level. The Preferred Potential (PFP)[5] is a one of the NNP with the unique feature of universality[1]. PFP is trained on large DFT data sets, including not only stable crystals and molecules, but also surfaces, adsorption and disordered structures. As a result, it is applicable to predict reaction mechanisms such as surface reactions in multi-component systems.<br/>While PFP allows for a rapid examination of reaction pathways, manual examination is difficult due to the need to consider processes important to the reaction but not fluorination, such as molecule movement and proton relay.<br/>Therefore, we have conducted a comprehensive study of reactions using the SC-AFIR method[6][7], which enables the automatic search for changes in the relative position of molecules and intermolecular reaction pathways.<br/>In this study, we report the results of our attempt at a comprehensive reaction analysis of wet etching, using PFP implemented in Matlantis^TM and the SC-AFIR method implemented in GRRM20. As an example, we report the comparison results of reaction pathways considering the stabilization of intermediates by surrounding HF and H₂O in the four-step fluorination reaction of Si.<br/>[1] D. H. Kim, <i>et al</i>., <i>ACS Omega</i>, <b>6</b>, 16009 (2021).<br/>[2] J. K. Kang and C. B. Musgrave, <i>J. Chem. Phys</i>., <b>116</b>, 275 (2002).<br/>[3] R. Hidatat, <i>et al.</i>, <i>Phys. Chem. Chem. Phys.</i>, <b>25</b>, 3890 (2023).<br/>[4] R. Hidayat, <i>et al.</i>, <i>J. Vac. Sci. Technol. A</i>, <b>41</b> 032604 (2023).<br/>[5] S. Takamoto, <i>et al.</i>, <i>Nat Commun.</i>, <b>13</b>, 2991 (2022).<br/>[6] S. Maeda, <i>et al.</i>, <i>J. Comput. Chem.</i>, <b>35</b>, 166 (2014).<br/>[7] S. Maeda, and Y. Harabuchi, <i>WIREs Comput. Mol. Sci.</i>, <b>11</b>, e1538 (2021).