Gas-phase etching mechanism of silicon oxide by a mixture of hydrogen fluoride and ammonium fluoride: A density functional theory study

In semiconductor manufacturing, removing native oxide on the silicon surface is one of the most critical processes in obtaining low contact resistance. The wet cleaning processes using dilute hydrogen fluoride (HF) solutions have been the most widely used method for removing oxide. However, they cannot completely prevent the formation of native oxides because the wafer is exposed to air before being loaded into the selective epitaxial growth chamber or metal deposition system. Thus, in situ gas-phase cleaning techniques, such as downstream plasma etching,^1–18 sputter etching,^19–22 or HF vapor etching,^23,24 were introduced to complement the wet cleaning process.

Nishino et al. reported a damage-free selective removal of native oxide using an NF₃/NH₃ downstream plasma and subsequent vacuum annealing at 100 °C.²⁵ A cleaning process based on NF₃/NH₃ remote plasma was commercialized under the trademark SiConi¹ and adopted in several applications, such as precleaning of silicon surface before the nickel deposition for silicidation^8–10 or before the selective epitaxial growth of silicon-germanium,^7,17 the profile modification of interlayer dielectric^12,16 or shallow trench oxide,⁵ and removal of dummy gate oxides.^2,4,6,15 The process represented a lower contact resistivity and a smoother surface than wet cleanings by HF.⁹ 

The removal of the native oxide by the SiConi^TM process comprises two steps. The first step is the formation of a fluorinated salt layer by a remote plasma of NF₃/NH₃ at near-room temperatures.^26–29 The second step is the sublimation of the salt layer by annealing with a temperature of 100 °C or higher.^1,10,26 The general reaction scheme has been established by several reports.^25,27,30,31 The NF₃/NH₃ plasma has been reported to generate active species such as HF and ammonium fluoride (NH₄F).³² The gas-phase etching by a mixture of HF and NH₃ without plasma also showed an etching mechanism similar to NF₃/NH₃ plasma.^30,31 Analysis by Fourier transform infrared spectroscopy and reflection absorption spectroscopy confirmed that the composition of the salt layer was ammonium hexafluorosilicate, (NH₄)₂SiF₆.^27,32 It was proposed that NH₃ and HF adsorbed on the surface react with the etching byproduct, silicon tetrafluoride (SiF₄), to form (NH₄)₂SiF₆. However, more detailed reaction mechanisms have not yet been elucidated. The etching mechanism of SiO₂ by HF to form SiF₄ has been studied by density functional theory (DFT) using the SiO₂ cluster model³³ or SiO₂ slab model³⁴ as a substrate surface. The etching of SiO₂ by HF is energetically favorable, with low activation energies of about 0.72–0.79 eV,³⁴ implying continuous etching even at near-room temperatures. However, the etching mechanism by NH₄F has not yet been reported. Also, the mechanism of the (NH₄)₂SiF₆ salt formation on silicon oxide surfaces was not elucidated. Understanding the detailed reaction mechanism underlying the cleaning process by a mixture of HF and NH₄F would lay a foundation for developing next-generation cleaning processes.

In the present study, we modeled and simulated the surface reaction mechanism during silicon oxide removal using a mixture of HF and NH₄F by DFT calculation. First, we simulated the successive fluorination of silicon oxide using slab models. Next, we simulated the mechanism of the (NH₄)₂SiF₆ salt formation on fluorinated SiO₂ surfaces. Finally, we compared silicon with SiO₂ to demonstrate the inherently selective etching of silicon oxide. We calculated the adsorption, reaction, and activation energies for the reactions to expect the probable route of the etching process.

The DFT calculations were carried out using the Dmol³ package, material studio 7.0 (BIOVIA, USA).^35,36 The calculations used the generalized gradient approximation functional and the Perdew–Burke–Ernzerhof scheme correlation.³⁷ For dispersion correction, the DFT-D2 using Grimme's method was applied.³⁸ The basis set of the double numerical polarization (DNP) 4.4 was selected due to the relatively high accuracy and low computation cost.^39,40 The accuracy and cost of the simulation were compared in Fig. S1 in the supplementary material.⁴¹ The self consistent field tolerance was 10⁻⁶ Ha. The calculations employed the Monkhorst-pack grid of 2 × 2 × 1 k-point parameter and the global orbital cutoff of 4.6 Å. The spin-unrestricted calculation condition was used. All geometry optimization calculations used the tolerance convergence energy of 10⁻⁶ Ha (1 Hartree = 27.2114 eV), the atomic force of 2 × 10⁻⁴ Ha/Å, and the maximum displacement of 0.005 Å, as the same parameter setting in the previous works.^34,42,43

The hydroxylated and fluorinated SiO₂ slab models were used for the DFT simulation: –Si(OH)₂, –Si(OH)F, –SiF₂, and –SiF₃ terminated cristobalite SiO₂ (001) 2 × 2 surfaces. Also, the hydrogenated and fluorinated Si surface slab models were used: –SiH₂, –SiHF, –SiF₂, and –SiF₃ terminated Si (100) 2 × 2 surfaces. The details of the surface slab models can be found in our previous work.³⁴ 

For the fluorinations of SiO₂ by NH₄F molecules, we used four types of surface slab models with different degrees of fluorination for four fluorination steps by NH₄F. Therefore, a single energy diagram could not represent the four fluorination steps. Also, since we used the same method in our previous paper on the reaction of HF, a direct comparison of the HF and NH₄F reactions would be straightforward.

Figure 2 shows the energy diagram and the changes in atomistic structures during the first step in Eq. (9). At first, NH₄F physisorbed on the –OH terminated SiO₂ surface in R1, with an energy of −1.03 eV. It was more favorable than the adsorption of HF (−0.53 eV)³⁴ due to the more hydrogen bonding of NH₄F than HF. The length of the H–F bond of the physisorbed NH₄F was 1.09 Å, elongated from 0.98 Å of isolated NH₄F. Then, an F atom of NH₄F replaced an OH group on the surface, producing a –Si(OH)F surface group, an H₂O, and an NH₃ molecule in P1. The reaction energy was −0.44 eV, which is similar to that of the first fluorination step by HF (−0.37 eV).³⁴ The exothermicity can be described by the formation of Si–F (7.15 eV³⁴) and O–H (5.20 eV³⁴) bonds in P1 despite the dissociation of Si–O (6.15 eV³⁴), H–F (6.76 eV), and N–H (0.71 eV) bonds. We calculated the bond dissociation energies of H–F and N–H using the same methods in the literature,^48–50 as shown in Fig. S2 in the supplementary material.⁴¹ The desorption of the byproducts, H₂O and NH₃, was endothermic with an energy of 2.00 eV in UP1. However, H₂O would spontaneously desorb at near-room temperature because Gibb's free energy change was negative above 11 °C, as shown in Fig. S3 in the supplementary material.⁴¹ On the other hand, NH₃ would remain on the surface at near-room temperature because Gibb's free energy change was negative only above 106 °C, as shown in Fig. S4 in the supplementary material.⁴¹ However, in the mixture of HF and NH₄F, NH₃ would react with HF to form NH₄F.

In the transition state, TS1_, NH₄F was attached to –SiOH* to form an elongated Si–F bond with a length of 2.18 Å, as shown in Fig. 2(b). The H–F bond in NH₄F was elongated from 1.09 Å in R1 to 1.18 Å in TS1. The Si–O bond of –SiOH* was also slightly elongated from 1.64 to 1.69 Å. There was no four-membered ring structure of –Si–O–H–F–, observed in the previous work with HF.³⁴ Instead, an N–H bond was formed inside the NH₄F molecule. The activation energy was low at 0.67 eV, possibly due to the absence of bond dissociation in TS1_, and was similar to 0.77 eV of the HF case.³⁴ Therefore, the first step would be kinetically plausible.

Figure 3 shows the energy diagram and the changes in the atomistic structures of the second step in Eq. (10). First, in R2, NH₄F physisorbed on the –Si(OH)F terminated SiO₂ surface with an energy of −0.78 eV. It was less favorable than in R1 due to the fewer hydrogen bondings. We assumed that F of NH₄F was exchanged with OH of –Si(OH)F*, forming –SiF₂*, H₂O, and NH₃ in P2.

The reaction energy from R2 to I2 in Eq. (10a) was −0.14 eV. In TS2(1) between R2 and I2, NH₄F was attached to –Si(OH)F* with an elongated Si–F bond (2.17 Å), as shown in Fig. 3(b), which is similar to TS1. The activation energy was low at 0.29 eV, also due to the absence of bond dissociation. Therefore, the formation of I2 in Eq. (10a) would be energetically and kinetically favorable. The reaction from I2 to P2 in Eq. (10b) was endothermic, with an energy of 0.35 eV, mainly due to the strong ionic interaction between the NH₄⁺ cation and the –Si(OH)F₂⁻* surface anion. A transition state, TS2(2), was discovered between I2 and P2 with the formation of Si–OH₂*, as shown in Fig. 3(b). The activation energy was 0.53 eV due to the N–H bond dissociation, O–H bond formation, and Si–O bond elongation. As a result, the energy diagram of Fig. 3(a) suggested that the reaction with NH₄F would stop in I2.

Since the second fluorination step by NH₄F would not be plausible, the –SiF₂* terminated surface would not be generated by NH₄F. However, we considered the further fluorination of the –SiF₂* terminated surface by NH₄F because the –SiF₂* terminated surface is generated by HF. Figure 4 shows the energy diagram and the changes in the atomistic structures for the third-step reaction in Eq. (11). Initially, NH₄F is physisorbed on the –SiF₂ terminated surface, with an energy of −0.58 eV. It was less favorable than R1 or R2 due to the weaker interaction between NH₄F and the –SiF₂ terminated surface. We assumed that the Si–O bonds of the surface were cleavaged by NH₄F, forming –SiF₃*, –SiOH*, and NH₃ in P3.

The reaction energy from R3 to I3 in Eq. (11a) was −0.21 eV. The transition state between R3 and I3, TS3(1)_, shows the attachment of NH₄F to –SiF₂* to form an elongated Si–F bond (2.10 Å), as shown in Fig. 4(b), similar to TS1 and TS2(1). The activation energy was low at 0.10 eV due to the absence of bond dissociation. Therefore, the formation of I3 in Eq. (11a) would be energetically and kinetically favorable. The reaction from I3 to P3 in Eq. (11b) was endothermic, with a reaction energy of 0.57 eV. The transition state, TS3(2), was discovered between I3 and P3 with the formation of Si₂O–H*, as shown in Fig. 4(b). The activation energy was 0.80 eV due to N–H dissociation, O–H formation, and Si–O elongation. As a result, the energy diagram of Fig. 4(a) suggested that the reaction would stop in I3.

The formation of the intermediate state in Eq. (12a) was exothermic, with a reaction energy of −1.35 eV. However, the formation of P4 from I4 in Eq. (12b) was endothermic, with a reaction energy of 0.69 eV. The transition state, TS4, was discovered between I4 and P4 with the formation of Si₂O–H*, as shown in Fig. 5(b). The activation energy was 0.85 eV due to the N–H bond dissociation, O–H bond formation, and Si–O bond elongation, similar to the TS2(2) and TS3(2) cases. Therefore, the energy diagram of Fig. 5(a) suggested that the reaction in Eq. (12) would also stop in I4.

Figure 6 shows the energy diagrams and the changes in the atomistic structures during the reaction in Eq. (13). UR3' is I2 with the addition of an HF molecule at an infinite distance, and P3' is I3 with the addition of an H₂O molecule as the byproduct. The apostrophe symbol indicates the fluorination of a surface anion by HF. The adsorption of HF was favorable, with an energy of −0.64 eV. The OH of –SiF₂(OH)⁻* in R3' was exchanged with F of HF to form a –SiF₃⁻* surface anion and H₂O in P3', with a reaction energy of −0.59 eV. The transition state, TS3', was found between R3' and P3', as shown in Fig. 6(b). The H–F bond was dissociated, the O–H bond was formed, and the Si–O bond was elongated in TS3'. Since the activation energy was low at 0.37 eV, we concluded that HF would further fluorinate I2 to form I3, as shown in Eq. (13).

Figure 7 shows the energy diagrams and the changes in the atomistic structures during the reaction in Eq. (14). UR4' is I3 with the addition of an HF molecule at an infinite distance, and P4' is I4 with the addition of –SiOH*. The adsorption of HF was favorable, with an energy of −0.62 eV. The Si–O bond beneath the –SiF₃⁻* in R4' reacted with HF to form –SiF₄⁻* and –SiOH* in P4'. The reaction was exothermic, with an energy of −0.25 eV. The transition state, TS4', was obtained between R4' and P4', as shown in Fig. 7(b). One Si–O bond was dissociated, the O–H bond was formed, the H–F bond was elongated, and the other Si–O bond was shortened in TS4'. A low activation energy of 0.61 eV indicates that the reaction in Eq. (14) would be energetically and kinetically favorable at near-room temperatures.

As explained so far, the intermediate states, I2 and I3, can be further fluorinated by HF to form I4 eventually. As a result, two product states are formed by a mixture of HF and NH₄F: (1) a surface with SiF₄ molecules and new –OH* groups produced by HF only and (2) a surface with –SiF₄⁻* surface anions and NH₄⁺ cations formed by a mixture of HF and NH₄F. In Subsection III B, we will examine the (NH₄)₂SiF₆ formation from these two products.

Figure 8(a) shows the energy diagram for the first step in Eq. (15a). P5 was spontaneously formed from UR5 by geometry optimization, with an energy difference of −1.49 eV, and the physisorption of the first NH₄F molecule could not be obtained. The F atom of NH₄F was attached to the Si atom of SiF₄, as shown in Fig. 8(b).

Next, we simulated the second step of the salt formation of Eq. (15b), and the energy diagram is shown in Fig. 9(a). The second NH₄F molecule in UR6 reacted with SiF₅⁻ and NH₄⁺ to form (NH₄)₂SiF₆ in P6. The adsorption of the second NH₄F molecule was favorable in R6, with an energy of −1.24 eV. The reaction from R6 to P6 was also exothermic, with a reaction energy of −0.08 eV, due to the ionic interaction between NH₄⁺ cations and SiF₆⁻ anion. The activation energy was low at 0.27 eV, which attributes to the interaction between H and F atoms, despite the dissociation of the weak NH₄–F bond, which was already elongated in R6. Therefore, the salt formation of Eq. (15) would be expected to be favorable and fast because both steps were exothermic with low activation energies.

Figure 10 shows the energy diagrams and the changes in the atomistic structures during the reactions in Eqs. (16a) and (16b). The adsorption of HF was favorable, with an energy of −0.71 eV. However, the reaction was endothermic, with an energy of 0.12 eV, indicating that the chemisorption was unfavorable. The adsorption of NH₄F was more exothermic than HF, with an energy of −1.12 eV. The reaction was endothermic, with an energy of 0.90 eV. As a result, both HF and NH₄F only physisorb on the surface.

Since both Eqs. (16a) and (16b) were endothermic, we modeled the (NH₄)₂SiF₆ formation reaction from the physisorption states in Fig. 10. Resulting reaction equation is the same as Eq. (16). Figure 11 shows the energy diagram and the change in the atomistic structure. The adsorption energy of HF on the surface where NH₄F was already physisorbed was −0.67 eV. The reaction of HF with the surface to form (NH₄)₂SiF₆ was exothermic, with a reaction energy of −0.16 eV. The activation energy was low at 0.30 eV, which is similar to that of 0.27 eV for Eq. (15b). Therefore, the salt formation reactions of both Eqs. (15) and (16) would be favorable even for near-room temperature processing.

We summarized all possible reaction routes for the (NH₄)₂SiF₆ formation by a mixture of HF and NH₄F in Fig. 12 with activation energy values. Our previous work shows that HF without NH₄F forms volatile SiF₄ with low activation energies, resulting in the continuous etching of SiO₂. This work shows that the etching of SiO₂ by NH₄F is energetically unfavorable at near-room temperatures without HF. Instead, we found that NH₄F contributes to surface modification by forming a salt layer, (NH₄)₂SiF₆.

Next, we investigated the mechanism by which the silicon oxide film is selectively removed over silicon by a mixture of HF and NH₄F. We summarized the reaction routes for the fluorination of Si by HF or NH₄F in Fig. 13 with activation energy values. In the previous work, we studied the fluorinations of silicon by HF molecules in four consecutive steps to form the SiF₄ byproduct.³⁴ The etching of silicon by HF is kinetically unfavorable, with relatively high activation energies of about 1.22–1.56 eV, indicating that continuous etching of silicon is impossible at near-room temperatures. However, since NH₄F is also present in the mixture, we investigated whether NH₄F could etch silicon at near-room temperature in the present work.

In the first two steps, Si–H, H–F, and N–H bonds are dissociated while Si–F and H–H bonds are formed. Then, in the last two steps, Si–Si, H–F, and N–H bonds are cleavaged, whereas Si–F and Si–H bonds are generated. The –SiH* surface groups are replaced by –SiF* by Eqs. (17) and (18), and then regenerated by Eqs. (19) and (20). The reactions were simulated using the –SiH₂, –SiHF, –SiF₂, and –SiF₃ terminated Si surfaces.

Figure 14 shows the energy diagram and the change in the atomistic structure for the first step in Eq. (17). In the beginning, NH₄F is physisorbed on the –SiH₂ terminated Si surface in R1, with an energy of −0.42 eV. It was less favorable than the physisorption of NH₄F on SiO₂ in Eq. (9) due to the absence of a hydrogen bond. Then, an F atom of NH₄F was replaced with an H atom on the surface, forming a –SiHF* surface group, an H₂ molecule, and an NH₃ molecule in P1. Unlike the first fluorination by NH₄F on the SiO₂ surface or by HF on the Si surface, the reaction was endothermic, with an energy of +0.15 eV. The endothermicity of the reaction can be described by the dissociation of Si–H (3.23 eV³⁴), H–F (6.76 eV), and N–H (0.71 eV) bonds in R1 despite the formation of Si–F (5.97 eV³⁴) and H–H (4.25 eV³⁴) bonds in P1. Unlike HF showing an exothermic reaction with the Si surface, NH₄F requires the dissociation of the N–H bond. The desorption of the byproducts, H₂ and NH₃, was endothermic with an energy of 0.18 eV in UP1.

In TS1_, HF, dissociated from NH₄F, attached to –SiH* to form a four-membered ring of –Si–H–H–F–, as shown in Fig. 14(b), which is similar to the previous work with HF.³⁴ The elongated Si–F and H–H bonds were formed with a length of 2.02 and 0.96 Å. The Si–H bond of –SiH* and the H–F bond in HF were elongated from 1.49 and 1.00 Å in R1 to 1.69 and 1.25 Å in TS1, respectively. The activation energy was very high at 1.94 eV, which is even higher than the HF case, due to the dissociation of N–H and the elongation of two bonds despite the formation of elongated Si–F and H–H bonds. Therefore, the first step with NH₄F would be energetically and kinetically unfavorable.

Figures S5–S7 in the supplementary material⁴¹ show the energy diagrams and the changes in the atomistic structures of the second, third, and fourth steps in Eqs. (18)–(20). Unlike the first step, the second to the fourth steps were exothermic, with a reaction energy of −0.51, −0.76, and −2.01 eV. The transition state geometries for the second to the fourth steps, TS2, TS3, and TS4, were very similar to those by HF in previous work.³⁴ The only difference from the HF cases was the dissociation of NH₄F into HF and NH₃ at the surface. As a result, the activation energies of the four steps by NH₄F (1.94–2.46 eV) were higher than HF cases (1.22–1.56 eV) due to the dissociation of N–H (0.71 eV) of NH₄F. Therefore, the fluorination of Si by NH₄F would be kinetically unfavorable at near-room temperatures.

In the mixture HF and NH₄F, the etch selectivity of SiO₂ against Si originated from the low activation energy values for the fluorinations of silicon oxide and high values for silicon. For the reaction with SiO₂, HF plays a dominant role in fluorination. The fluorination of silicon oxide by HF showed lower activation energy values of 0.72–0.79 eV.³⁴ It is because all transition states showed the formation of stable four-membered ring structures of –Si–O–H–F– without bond dissociations. On the other hand, the fluorination of silicon showed relatively higher activation energy values than the silicon oxide case, 1.22–1.56 eV by HF (Ref. 34) and 1.94–2.46 eV by NH₄F. The relatively high activation energy for Si can be explained by less stable transition state structures. In addition, the stronger Si–F bond on SiO₂ than on Si in the transition states also contributed to the lower activation energies for the SiO₂ substrate. We discussed the origin of selectivity in the previous work.³⁴ 

The next-generation cleaning process requires self-limited surface modification and the removal of the modified layer. In this work, we modeled and simulated the selective cleaning process of SiO₂ over Si by DFT. The same strategy can be applied to designing the selective self-limited cleaning process of different materials. Generally, the fluorinations of dielectric materials, such as silicon oxide, silicon nitride, and metal oxides, are exothermic due to the formation of strong bonds between F and Si or metal atoms. Hence, reaction energy alone cannot expect if the material can be etched by fluorine-containing molecules. However, the activation energy can expect if the etching is possible at a specific process temperature. Therefore, the DFT study can provide the information to design the next-generation etching or cleaning process of different materials.

We studied the selective etching mechanism of silicon oxide using a mixture of HF and NH₄F gases by DFT calculation using surface slab models. The fluorination reactions of silicon oxide by the mixture produced a volatile SiF₄ molecule or a surface anion of –OSiF₄⁻* with an NH₄⁺ cation with low activation energies. Unlike HF, NH₄F produced surface salt species consisting of a surface anion and an ammonium cation. From the two reaction products, (NH₄)₂SiF₆ can be formed exothermally with low activation energies (0.27 or 0.30 eV). The fluorination reactions of silicon by the mixture were also exothermic, except for the first fluorination step by NH₄F. However, the activation energies were significantly higher than the silicon oxide cases, 1.22–1.56 eV by HF and 1.94–2.46 eV by NH₄F due to the less stable transition state geometries. Therefore, the selective salt formation on silicon oxide, not on silicon, is expected in near-room temperature processing, which enables selective etching of silicon oxide.

This work was supported by the Ministry of Trade, Industry & Energy (MOTIE) of the Republic of Korea (Project No. 20012588) and the Korea Semiconductor Research Consortium (KSRC) support program for the development of the future semiconductor device. This research was partly supported by the Korea Basic Science Institute (National Research Facilities and Equipment Center) grant funded by the Ministry of Education (No. 2022R1A6C101A774).

The authors have no conflicts to disclose.

Romel Hidayat: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Validation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (supporting). Khabib Khumaini: Formal analysis (equal); Investigation (equal); Methodology (equal); Validation (equal); Writing – review & editing (equal). Hye-Lee Kim: Conceptualization (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Validation (equal); Visualization (equal); Writing – review & editing (equal). Tanzia Chowdhury: Formal analysis (equal); Investigation (equal); Methodology (equal); Validation (equal); Writing – review & editing (equal). Tirta Rona Mayangsari: Investigation (equal); Methodology (equal); Validation (equal); Writing – review & editing (equal). Byungchul Cho: Conceptualization (equal); Methodology (equal); Validation (equal). Sangjoon Park: Conceptualization (equal); Methodology (equal); Validation (equal). Jongwan Jung: Conceptualization (equal); Methodology (equal); Validation (equal). Won-Jun Lee: Conceptualization (equal); Formal analysis (equal); Funding acquisition (lead); Methodology (equal); Project administration (lead); Resources (lead); Supervision (lead); Writing – review & editing (lead).

The data that support the findings of this study are available within the article and its supplementary material (Ref. 41).