The growing demand for solid-state memory devices has generated significant interest in process technology development for 3D-NAND fabrication along with challenges that need to be overcome. Particularly, plasma etch engineers are faced with the task of etching high-aspect-ratio contact (HARC) holes through up to ~100 layers of stacked ONON layers (i.e. silicon oxide, SiO₂, and silicon nitride, Si₃N₄) as well as dummy SiO₂ of varied thickness within the same structure, landing all holes simultaneously. To this end, the development of a plasma etch process that etches both SiO₂ and Si₃N₄ with a high etch rate and a near unity selectivity is highly desirable. One strategy is to make SiO₂ and Si₃N₄ etch fronts closer in composition via surface modification, illustrated by our previous simulation work on the feasibility of Si₃N₄ oxidation-etch¹ , where the Si₃N₄ surface oxidation to silicon oxynitride (SiO_xN_y) was explored. The indicated role of oxygen (O₂) addition to the feed gas in attaining closer etch rates for Si₃N₄ was also experimentally reported^{2 ,3,4}.

Another approach of surface modification is to convert the SiO₂ surface to SiO_xN_y via plasma nitridation. For the SiO₂ plasma nitridation per se, there have been extensive studies. N₂ feed was the most commonly used^{5 ,6,7,8,9}, while N₂O⁵ and NH₃¹⁰ were also reported as well as He/N₂ mixed gas¹¹. Multiple plasma sources, including inductively-coupled plasma (ICP)⁹, helical resonator plasma⁵, and electron-cyclotron resonance (ECR) plasma⁷, were investigated. In terms of the nitrogen penetration in the oxynitride product with remote He/N₂ plasma, Hattangady et al. found via XPS and SIMS characterizations that the implanted nitrogen was confined to the immediate vicinity of the surface ¹¹. Moreover, the extent of nitridation also depends on the substrate temperature based on the findings by Kobayashi et al.: a more complete nitridation to Si₃N₄ was observed at 25 ^oC than at 700 ^oC in their experiment using N₂ plasma generated by electron impact⁶. Plasma nitridation can also alter the surface stress of the silicon oxide film, and particularly Itakura et al. demonstrated that the compressive surface stress formed by thin oxide on Si(100) was relaxed⁸. Simulation studies of SiO₂ nitridation in pulsed inductively coupled N₂ plasma using a coupled plasma equipment-surface model show that N atoms and N₂⁺ ions are the primary species that contribute to the nitridation⁹.

The etching behavior of silicon oxynitride was also investigated, where the oxynitride was mostly prepared via plasma-enhanced chemical vapor deposition (PECVD). Cavallari and Gualandris found that the oxynitride shows a significantly higher etch rate than does the thermal oxide in a CF₄/O₂-based plasma, as well as a correlation between the etch rate to the nitrogen inside the layer¹². Focusing on plasmas based on CF₄/CHF₃/CO, Ueno et al.¹³ established a negative correlation between the SiO_xN_y etch rate and the amount of carbon deposition, with the highest SiO₂/SiO_xN_y selectivity afforded by the CHF₃ + CO plasma generating a more carbon-rich deposition. Moreover, they also observed that the etch rate of the low-N-percantage SiO_1.4N_0.4 to be higher than SiO₂ for CF₄ plasma and CHF₃ plasma. In addition to the nitrogen implantation effects, studies using downstream plasma reactors reveal the nitric oxide (NO) molecule as an additional contributor to the etching reaction. For instance, Kastenmeier et al. investigated the Si₃N₄ etch rate enhancement by N₂ addition to CF₄/O₂ plasma and identified the NO species formed in the plasma to be a key contributor through assisting in the Si-N cleavage¹⁴. The same enhancement by N₂ addition to CF₄/O₂ plasma for polycrystalline silicon etch was also observed, occurring via NO assisting in Si-O cleavage and exposing the open Si bond to be easily attacked by incoming fluorine species¹⁵.

While the studies above mentioned provide valuable insights into the feasibility of an enhanced nitridation-etch of SiO₂, such evidence is mostly indirect: firstly because of the difference between the PECVD SiO_xN_y and the plasma nitridation SiO_xN_y, and secondly because of the nitrogen implantation effect not explicitly accounted for by downstream studies. Thus, in this paper we provide a theoretical understanding of the SiO₂ nitridation-etch via molecular dynamics and quantum chemistry simulations, in order to acquire a better knowledge of the atomistic mechanisms. We evaluate the effectiveness of the nitrogen implantation into the SiO₂ substrate as well as its mechanism. We also investigate the energetics of the key reactions in the etching reactions of SiO₂ by fluorine and nitrogen, the most important etchants produced by a fluorocarbon/nitrogen-based plasma. Furthermore, we also highlight the synergy between nitrogen and fluorine in delivering a high SiO₂ etch rate. This simulation work will offer insight for future experimental and theoretical studies aimed at developing high-performance non-selective SiO₂/Si₃N₄ etch processes in 3D-NAND all-in-one etching applications and beyond.

Molecular dynamics simulations were performed with LAMMPS¹⁶ incorporated in MedeA®¹⁷ with the ReaxFF reactive force field for C/H/O/N/S/F/Si parametrized in Ref¹⁸ . The subsrate slab was prepared with O-terminated {001} surface of α-quartz, and consists of 108 Si atoms and 240 O atoms, around 16 Å in thickness (along Z). The supercell lattice constant along Z is 50 Å to create sufficient vacuum area. The bottom Si and O layers were fixed during the simulation, and the next 3 layers of Si and O were thermalized with a Langevin thermostat at 300 K. The rest of the system was allowed to freely propagate with the microcanonical ensemble (NVE). The time step was chosen as 0.05 fs. The substrate was equilibrated with the canonical ensemble (NVT) at 300 K for 5 ps before the ion bombardment. Two sets of bombardment simulations, N only bombardments and alternating N/F bombardments, were performed. For N only bombardments, one N atom is launched every 5 ps at 25 eV after randomly placing it above the surfacewith random ion angles at most ± 10^o from the -Z direction (40 N atoms launched in total, 220 ps total simulation time). For the alternating N/F bombardment, the same set-up was used, except that the launching sequence is N-F-N-F… rather than N-N-N-N … (40 N atoms plus 40 F atoms launched in total, 420 ps total simulation time). All volatlized products were regularly removed from the system. Statistics were performed over 10 repeated simulation runs. The Pearson correlation coefficient, defined as \({\rho }_{XY}=\frac{\mathrm{c}\mathrm{o}\mathrm{v}\left(X,Y\right)}{{\sigma }_{X}{\sigma }_{Y}}\), was calculated to determine the correlation between the volatility of N and that of O. Here X and Y are the amount of N and O removals, respectively, while, \(\mathrm{c}\mathrm{o}\mathrm{v}\left(X,Y\right)\) is the sample covariance, and \({\sigma }_{X}\) and \({\sigma }_{Y}\) are the sample standard deviations; positive (negative) \({\rho }_{XY}\) values indicate positive (negative) linearities.

The quantum chemsitry calculations were performed with the Gaussian 16¹⁹ package using B3LYP/6-31++G(d,p)^{20 ,21,22,23,24}. A molecular cluster containing 4 Si atoms and 10 O atoms was used to simulate the reaction energetics. The structures were verified as local minima and transition states from harmonic frequency calculations²⁵. The transition states were located using the STQN method^{26 ,27}.

3.1. Molecular dynamics (MD) simulations of ion bombarding on the SiO₂ surface

We begin this section by examining the post-bombardment substrate structures (Figure 1), in order to identify the major chemical reactions and etch products. For both sets of bombardment simulations, i.e. the alternating N/F bombardments (40 N’s and 40 F’s) and the N only (40 N’s) bombardments, the formation of Si-N bonds is observed. Also common to both sets of simulations is the formation of N-O bonds, which leads to the volatilization of O from the substrate as NO_x species. Together, the implantation of N atoms via Si-N formation combined with the O volatilization as NO_x results in the substitution of O atoms in the SiO₂ substrate by N atoms, thus forming an oxynitride-like surface (Figure 1 (b)). Furthermore, the addition of F to the bombardment leads to Si-F bond formation, which is essential for the ultimate volatilization of Si as SiF_x (Figure 1 (a)). Finally, the impinging N atoms can also bond with N atoms that are already on the surface. This generates volatile N₂ molecules, creating a pathway for N self-scavenging. At the steady state, the outflux of N via self-scavenging as N₂ plus O abstraction as NO_x, must equal its influx as impinging N ions, consequently maintaining the surface concentration of N (as well as other elements) at a quasi-static level.

The steady-state concentrations of various elements, however, also depends on the composition of impinging species. As illustrated in Figure 2, the average numbers of N atoms in the final structure are different for the alternating N / F bombardments and the N-only bombardments. The addition of F results in a visibly lower (7.7 atom counts) average N content compared to that from N-only bombardments (12.1 atom counts). This is likely due to the passivating role of F for under-coordinated Si atoms on the surface, by forming Si-F bonds. With the under-coordinated surface Si terminated with F, its recombination with surface N atoms is inhibited, leaving N more volatile. This inhibition is further warranted by the higher Si-F bond energy than those of both Si-O and Si-N bonds, making the substitution of Si-F by Si-O / Si-N energetically unfavorable. The higher volatility of O given the addition of F, similarly to that of N, is evidenced by the increase in the average numbers of O removed from the surface during the bombardments (from ~13.5 atom counts to 18.0 atom counts), with the number of impinging N atoms being 40 for both the N / F bombardments and N bombardments.

To further demonstrate the correlation between the N and O volatility, Figure 3 shows the scatter plots for the number of N atoms in the final substrate structure and the number of O atoms volatilized from the surface. The data points for alternating N/F bombardments, N only bombardments, and their combination, are given in Figures (a), (b), and (c), respectively. Despite the relatively small sample size, the consistently negative Pearson correlation coefficients nonetheless reveal a positive correlation between N and O volatility. This is consistent with the observation of NO species as one of the O etch products (Figure 1), which also suggests the simultaneous O and N removal in the form of NO_x as an underlying mechanism.

To summarize this section, the MD simulations have revealed that the N treatment of the SiO₂ through ion bombarding partially replaces the O atoms by N atoms, forming an oxynitride-like surface. This is effected by the Si-N bond formation combined with O volatilization as NO_x molecules. Moreover, the F addition to the ion bombardments leads to a higher removal rate for O; however the F additionalso reduces the extent of N implantation via surface Si passivation as Si-F. In fact, the presence of F simultaneously makes both O and N more volatile, in addition to its crucial role of volatilizing Si, showing synergistic effects and trade-offs between impinging ions of different elements.

We have investigated the nitridation-etch of silicon oxide in a fluorocarbon/nitrogen-based plasma with atomistic simulations. Ion bombardment simulations with molecular dynamics (MD) and static energetics calculations with quantum chemistry (QC) have been employed. The simulations indicate a surface conversion from oxide to oxynitride enabled by the implantation of sheath-accelerated ions. During the coversion, Si-N bonds replace Si-O bonds and O atoms are volatilized as NO molecules, whereby generating a surface that is closer in composition to the nitride surface. We have also shown that the crucial reaction steps for oxygen and silicon abstraction are energetically favorable. In addition, our presented data reveal a synergy between fluorine and nitrogen etchants, where fluorine and nitrogen mutually enhances each other’s reactivity via creating additional surface active sites. These microscopic mechanistic insights indicate the potential value of utilizing a fluorocarbon/nitrogen-containing plasma to efficiently etch silicon oxide layers at a similar rate to the nitride layers in 3D-NAND applications and beyond.

The authors would like to thank to Aelan Mosden, Peter Biolsi and Alexander Oscilowski from TEL Technology Center, America, LLC for their continuous support and moral encouragement.