Ab Initio Study of Adsorption and Decomposition of NH 3 on Si(100)-(2×1) Yuniarto Widjaja, Michael M. Mysinger, and Charles B. Musgrave* Department of Chemical Engineering, Stanford UniVersity, Stanford, California 94305 ReceiVed: October 18, 1999; In Final Form: December 9, 1999 We investigate the mechanism of NH 3 adsorption and initial decomposition on the (2×1) reconstructed Si- (100) surface using B3LYP density functional theory. The Si(100)-(2×1) surface is described using cluster approximation. Ammonia is found to adsorb on the “down” atom of the buckled silicon dimer with no activation barrier. We also find that only half of the surface silicon atoms are active sites for ammonia adsorption. Ammonia adsorption on the Si(100)-(2×1) is exothermic with an adsorption energy of 29 kcal/mol. Dissociation of the adsorbed ammonia to form NH 2 (a) and H(a) proceeds with a low activation energy of 5 kcal/mol below the NH 3 (g) and bare Si(100)-(2×1) energy. Our calculated recombination desorption energy of 51 kcal/mol is found to be in good agreement with the temperature-programmed desorption experimental result of 47 kcal/mol. Additionally, our calculated vibrational spectra of NH 2 (a) and H(a) agree within 2% of the experimental high-resolution electron energy loss spectra. Introduction Silicon nitride thin films have become pervasive in the microelectronics industry. Specifically, thin films of Si 3 N 4 are being used as insulators, oxidation masks, diffusion barriers, and gate dielectrics. 1-5 Proposed future applications include using it as a precursor to oxynitride dielectrics and as an interface between Si and high-K dielectrics. 6,7 These applications all involve growing a thin silicon nitride film, typically by chemical vapor deposition (CVD) of NH 3 or mixtures of silicon- containing precursors and NH 3 . NH 3 is an excellent nitriding agent and preferred over N 2 because of its high sticking coefficient and reactivity. 8-11 Plasma-enhanced (PE) CVD is more common than thermal CVD because PECVD allows deposition at lower temperatures. However, the resulting PECVD films are not reliable and contain a high concentration of electrically active defects, 12 resulting in high leakage current and other deleterious effects. Low- pressure thermal CVD has been investigated as an alternative growth method since it results in films with better electrical characteristics. 12 To optimize deposition processes to control the properties of the Si 3 N 4 grown by low-pressure thermal CVD, an under- standing of the chemistry of CVD of Si 3 N 4 film growth is necessary. In the absence of detailed knowledge of the chemistry, computational prototyping of reactors and processes must be based on empirical, lumped kinetics, which generally lacks true predictive ability. Currently, even the initial steps of the silicon nitride depositionsthe adsorption and decomposition of NH 3 on the Si(100) surfacesare not fully understood. Understanding the initial steps of deposition is crucial because these steps determine the properties of the interface, which in turn play an important role in device electrical characteristics. Here, we investigate the detailed chemical mechanism of the reaction of NH 3 with the Si(100)-(2×1) surface, which is the most important surface in microelectronics. We calculate first- principles geometries and energies of the stable points on the potential energy surface (PES) for the adsorption and decom- position reactions. These calculations provide the quantum mechanical input for chemical kinetics models. Consequently, this study will lead to kinetics for accurate CVD reactor modeling and design. Dresser et al. 13 investigated NH 3 on Si(100)-(2×1) using various techniques. Using low-energy electron diffraction (LEED), they confirmed that the (2×1) reconstruction is preserved after the adsorption and decomposition of NH 3 (g). Scanning tunneling microscopy (STM) images by Hamers et al. 14,15 also show that the (2×1) structure is retained after the reaction. This indicates that adsorption and decomposition only take place at the dangling bonds and that the underlying dimer structure remains intact. Dresser et al. used deuterium-ammonia coadsorption temperature-programmed desorption (TPD) to determine that NH 2 (a) and H(a) were the surface species at room temperature. High-resolution electron energy loss spectroscopy (HREELS) results by Fujisawa et al. 16 show peaks assigned to the stretching and bending modes of Si-NH 2 and the stretching mode of Si-H after NH 3 adsorption, further supporting NH 2 (a) and H(a) as the room temperature surface species. At temperatures below 120 K, molecularly adsorbed NH 3 (a) is observed along with NH 2 (a) and H(a). 10 Recent theoretical studies have also provided some insight into the detailed mechanism. Fattal et al. 17 have predicted the structures and detailed energetics for the adsorption and initial decomposition reactions by using ab initio quantum chemical simulations at the CASSCF/MRSDCI level of theory (complete active space self-consistent field geometry optimizations fol- lowed by multireference single and double excitation configu- ration interaction single-point energies) using a Si 9 H 12 cluster model and effective core potentials. Their results are able to qualitatively explain the high sticking probability of NH 3 , the observation of NH 3 (a) at low temperatures, and the observed stability of NH 2 (a) and H(a). However, the desorption energy found (75 kcal/mol) is not in good agreement with desorption measurements (47 kcal/mol). 13 On the other hand, this result is in better agreement than that of Miotto et al., 18 who used pseudopotential local density approximation (LDA) density functional theory (DFT) on a periodic slab to give a desorption energy of 100 kcal/mol. Lee et al. 19 used the PBE 20 generalized gradient approximation (GGA) DFT method on a slab to 2527 J. Phys. Chem. B 2000, 104, 2527-2533 10.1021/jp9936998 CCC: $19.00 © 2000 American Chemical Society Published on Web 02/24/2000