Etching Mechanisms and Surface Conditions for SiO_xN_y Thin Films in CF₄ + CHF₃ + O₂ Inductively Coupled Plasma

Abstract

In this work, we investigated the etching characteristics of SiO_xN_y thin films in CF₄ + CHF₃ + O₂ inductively coupled radiofrequency (13.56 MHz) plasma. SiO_xN_y etching rates were measured as functions of the CF₄/CHF₃ mixing ratio at constant O₂ fraction, gas pressure (10 mTorr), input power (500 W) and bias power (100 W). The conditions of the etched surfaces were examined by X-ray photoelectron spectroscopy, atomic force microscopy and contact angle measurements. Data on internal plasma parameters and steady-state plasma composition were obtained by Langmuir probe diagnostics and zero-dimensional plasma modeling. It was found that the substitution of CF₄ for CHF₃ suppresses the SiO_xN_y etching rate as well as results in increasing both amount of residual fluorocarbon polymer and SiO_xN_y/Si etching selectivity. The SiO_xN_y etching mechanism was analyzed by considering the relationships between measured etching rates and model-predicted fluxes of active species (F atoms, CF_x radicals and positive ions). It was proposed that the SiO_xN_y etching process: (1) exhibits features of ion-assisted chemical reactions in the neutral-flux-limited mode, and (2) involves the contributions of by HF molecules. The effective probability of the SiO_xN_y + F reaction is correlated with the amount of deposited fluorocarbon polymer while the hydrophobic nature of the plasma-treated SiO_xN_y surface confirms the presence of a continuous fluorocarbon polymer film.

Introduction

Silicon dioxide (SiO₂) and silicon nitride (Si₃N₄) are silicon-based materials which have found many applications in various micro- and nano-electronic devices. These materials traditionally play the roles of final passivation and protective layers, hard masks, gate dielectrics in field-effect structures, spacer dielectrics and anti-reflective coatings for solar cells [1,2,3,4,5]. Accordingly, silicon oxynitride (SiO_xN_y) also has a dielectric nature and features a low density of surface states and high dielectric permittivity [6]. It was also found that SiO_xN_y is a very attractive material for optical devices [7]. Particularly, the advantageous optical properties of SiO_xN_y relate to its quite low optical loss (less than 0.2 dB/cm at 1550 nm [8]) and the wide range of refractive index values (between 1.45 for SiO₂ and 2.0 for Si₃N₄) which may be achieved by varying the O/N ratio. Since most of the aforementioned applications require the precise patterning of SiO_xN_y layers, the development and optimization of a dry etch process for SiO_xN_y thin films is an important task for achieving accurate pattern transfer and stable device parameters.

At present, there have been many studies reporting the plasma-assisted etching characteristics and mechanisms for SiO₂ [9,10,11,12,13,14,15,16,17,18] and Si₃N₄ [9,10,11,12,13,14,15,16,17,18,19,20] in fluorocarbon gas plasmas. Existing results can briefly be summarized as follows:

• 1. Under the typical reactive ion etching (RIE) conditions, the dominant role in the chemical etching pathway for both SiO₂ and Si₃N₄ belongs to F atoms. The contribution from CF_x radicals appears only through their ion-assisted dissociation on the surface [14] and is much smaller because of the much smaller ion flux compared with the F atom flux.

• 2. The spontaneous chemical reaction in the SiO₂ + F system is thermodynamically prohibited (since the Si–O bond strength of ~ 799 kJ/mol is greater than the Si–F bond strength of ~ 552 kJ/mol [21]) and, in fact, can be ignored at near-to-room temperatures [1, 2, 22]. That is why the SiO₂ dry etching process requires ion bombardment in order to break Si–O bonds as well as to sputter the low volatile fluorinated layer [12, 22, 23]. An ion bombardment energy of more than 150–200 eV results in the reaction-rate-limited etching regime [23, 24]. At least, the non-monotonic behavior (with a maximum at 30–40% O₂) of the etching rate for SiO₂ in the CF₄ + O₂ plasma is surely associated with the same non-monotonic behavior of the F atom density [2, 25].

• 3. The spontaneous reaction in the Si₃N₄ + F system is possible in principle, since the Si–N bond of ~ 470 kJ/mol is a bit weaker than the Si–F bond [21]. At the same time, it appears to be much less effective compared with the fluorination of silicon [26] while also undergoing an evident acceleration by ion bombardment [11]. Accordingly, under typical RIE conditions, the Si₃N₄ etching rate increases monotonically with increasing gas pressure and input power [11, 19] and shows a maximum at ~ 30–40% O₂ in CF₄ + O₂ plasma [10, 12, 13]. All these facts suggest the reaction-rate-limited etching regime of ion-assisted chemical reaction.

• 4. The highest etching rates for both SiO₂ and Si₃N₄ were obtained for the low-polymerizing fluorocarbon gases while the maximum etching selectivity over Si was found for high-polymerizing systems [10, 11, 27]. Mechanisms for obtaining high Si₃N₄/Si etching selectivity are same as those for obtaining high SiO₂/Si selectivity. Particularly, the thickness of the fluorocarbon (FC) polymer film on Si₃N₄ was found to be lower than on Si [9, 16] because of the lower sticking probability for polymerizing radicals [15].

Given this background, information on the etching characteristics for SiO_xN_y thin films is rather lacking. In fact, there have been only few purely experimental works [28,29,30] discussing etching and polymerization effects in fluorocarbon gas plasmas through relationships between measured SiO_xN_y etching rates and FC polymer film characteristics. Though the existing data provide reasonable information on the heterogeneous stages of the etching process and surface conditions, these do not match the changes in the gas-phase plasma characteristics. In such situations, one can speak only about the phenomenological description of the SiO_xN_y etching process, while additional investigation is needed to understand the whole process pathway; (in fact, the etching mechanism must be determined).

The main idea of the current work is to combine experimental and model-based approaches to study the SiO_xN_y etching process in CF₄ + CHF₃ + O₂ inductively coupled plasma with matching of the gas-phase and heterogeneous chemistries. Both CF₄ and CHF₃ gases have been widely used in the semiconductor processing, even though it has a global worming potential and has been considered to be replace with a low potential gas, such as liquid Perfluorocarbons (L-PFCs). As a main variable parameter, we selected the CF₄/CHF₃ mixing ratio at constant O₂ fraction in the feed gas. In our opinion, the mixing of low- and high-polymerizing fluorocarbons allows one to better understand the role of the FC polymer film in the SiO_xN_y etching process. The interest in CHF₃ was also caused by the fact that it allows for the effective formation of HF in the gas phase [31, 32]. If this contributes to the SiO_xN_y etching rate, the CF₄/CHF₃ mixing ratio may be an effective tool to adjust the chemical etching mechanism and thus, the output process characteristics. Accordingly, the main goals are:

• 1. To study the influence of the CF₄/CHF₃ mixing ratio on SiO_xN_y etching kinetics, surface conditions and etching selectivities with respect to typical over- and under-layer materials in SiO_xN_y-containing structures. For this last purpose, the list of etched materials also includes Si and poly (methyl methacrylate) (PMMA).

• 2. To analyze the relationships between CF₄/CHF₃ mixing ratio, plasma chemistry, and gas-phase plasma characteristics (electron temperature, energy of ion bombardment, densities and fluxes of plasma active species). Since all these questions have not been studied yet for CF₄ + CHF₃ + O₂ plasma, the corresponding data are expected to be very useful for understanding the features of the SiO_xN_y etching process and thus, for the optimization of process regimes.

• 3. To perform analysis of SiO_xN_y etching kinetics with model-predicted fluxes of plasma active species, to establish the gas-phase-related parameters which adequately characterize the etching/polymerization balance in the SiO_xN_y etching process, and to formulate a reasonable approach to the etching mechanism in the given gas system.

Experimental and Modeling Details

Experimental Setup, Procedures and Conditions

The experimental part of this work (etching rate measurements and plasma diagnostics by Langmuir probes) was performed in the inductively coupled plasma (ICP) reactor described in our previous works [33, 34]. The reactor has a cylindrical ($r$ = 15 cm, $l$ = 12.8 cm) chamber made from anodized aluminum. The plasma was excited using a 13.56 MHz RF power supply connected to a planar upper-side coil. Another 12.56 MHz RF source powered the bottom electrode to control the negative DC bias voltage, $- \;U_{dc}$. The set of constant process parameters included gas pressure (p = 10 mTorr), input power ($W$ = 500 W) and bias power ($W_{dc}$ = 100 W). The variable parameter was the CF₄/CHF₃ mixing ratio. The latter was set by adjusting partial flow rates for CF₄ ($q_{{{\text{CF}}_{4} }}$) and CHF₃ ($q_{{{\text{CHF}}_{3} }}$) in the range of 0–40 sccm with both the fixed O₂ flow rate $q_{{{\text{O}}_{2} }}$ = 4 sccm and total gas flow rate $q$ = 44 sccm. Accordingly, the fraction of O₂ in the feed gas $y_{{{\text{O}}_{2} }} = q_{{{\text{O}}_{2} }} /q$ was always 0.09, or 9% while the fraction of CHF₃$y_{{{\text{CHF}}_{3} }} = q_{{{\text{CHF}}_{3} }} /q$ was changed from 0–0.91, or 0–91%. The lower end of this range $y_{{{\text{CHF}}_{3} }}$ = 0, corresponds to 91% CF₄ + 9% O₂ gas system while the upper end is a 91% CHF₃ + 9% O₂ mixture. From preliminary experiments and model (see “Plasma Modeling” section) runs, it was concluded that $y_{{{\text{O}}_{2} }}$ ~ 10% in CF₄ + O₂ and CHF₃ + O₂ gas mixtures increases the F atom formation rate, but does not cause sufficient changes in both densities of polymerizing radicals and FC polymer film destruction rate. As such, the variation of CF₄/CHF₃ mixing ratio in the low-oxygenated CF₄ + CHF₃ + O₂ gas mixture allows one to trace simultaneously the effects of gas-phase chemistry, etching kinetics and FC polymer film deposition/destruction balance.

The etched samples of SiON, Si, and PMMA (in fact, pieces of Si wafer covered by a PMMA layer with a thickness of 850 nm) had the dimensions of about 2 × 2 cm². All three samples were simultaneously placed in the middle part of the bottom electrode. The built-in water-flow cooling system allowed one to maintain a constant sample temperature at ~ 17 °C. The etching rates ($R$) for all three materials were determined from the corresponding etched depths ($\Delta h$) measured by the Alpha-step 500 surface profiler (Tencor) after a processing time of $\tau$ = 60 s. In preliminary experiments, it was found that the condition $\tau$ < 300 s surely provided a quasi-linear shape for the $\Delta h = f(\tau )$ function and thus, the steady-state etching regime. As such, one can simply assume $R = \Delta h/\tau$. Both treated and reference (non-treated) SiO_xN_y surfaces were examined using X-ray photoelectron spectroscopy (XPS), atomic force microscopy (AFM) and contact angle measurements. The XPS system was a VG Scientific ESCALAB 200R with Mg K_α 1253.6 eV radiation operating at 260 W. The binding energies were calibrated using the C(1s) peak at 284.5 eV. The AFM measurements were carried out using the XE-7 (Park system) tool. The mean square root (MSR) roughness was extracted from the AFM images using the software supplied by the equipment manufacturer. The contact angles were measured at room temperature using a drop shape analysis system (DSA-100, KRUSS) with polar (de-ionized water) and non-polar (CH₂I₂) liquids. Then, the free surface energy was calculated through the Owens–Wendt equation [35].

Plasma parameters were examined using a DLP2000 double Langmuir probe tool (Plasmart Inc.). In order to minimize inaccuracies in the measured current–voltage (I–V) curves due to FC polymer deposition on the probe tips, these were exposed in 50% Ar + 50% O₂ plasma for 1 min before each measurement. As a result, the difference between the data points recorded under the same experimental conditions within the time period $t > \tau$ did not exceed the standard experimental error. The data on electron temperature ($T_{e}$) and ion current density ($j_{ + }$) were extracted from the I–V curves using the well-known equations of double Langmuir probe theory [36, 37]. The total density of positive ions was calculated as $n_{ + } \approx j_{ + } /0.61\;{\text{ev}}$ [35], where the expression for ion Bohm velocity $v \approx \sqrt {eT_{e} /m_{i} }$ did not take into account the presence of negative ions. This simplification has been demonstrated to be reasonable for low-pressure electronegative plasmas in our previous studies [38, 39]. The effective ion mass ($m_{i}$) was estimated through the densities of dominant neutral species while accounting for the differences in their ionization rate coefficients.

Plasma Modeling

In order to obtain the volume-averaged densities and fluxes of plasma active species, we applied the 0-dimensional kinetic model with using the experimental data of $T_{e}$ and $n_{ + }$ as input parameters. The kinetic scheme (the set of chemical reactions and corresponding rate coefficients) was taken from our previous works which dealt with the modeling of CF₄ + O₂ + Ar [18, 40, 41] and CHF₃ + O₂ + Ar [41, 42] plasmas. The formation of C_xH_yF_z species with x > 1 and y > 1 was ignored due to their low densities and negligible influence on the overall reaction balance [43, 44]. Similar to Refs. [40, 41], the model accounted for the following assumptions:

• The electron energy distribution function (EEDF) is close to a Maxwellian one [44,45,46]. This allows one to obtain the rate coefficients for the electron-impact processes in the form of $k = f(T_{e} )$ using the well-known fitting expressions from Refs. [40,41,42,43,44].

• The heterogeneous loss of atoms and radicals can be described in terms of conventional first-order recombination kinetics [40, 41]. The corresponding rate coefficients were evaluated as $k \approx \gamma v_{T} /\varLambda$, where $\varLambda^{ - 2} = (2.405/r)^{2} + (\pi /l)^{2}$ is the diffusion length, $v_{T} = (8k_{B} T/\pi m)^{1/2}$, and $\gamma$ is the recombination probability. The recombination probabilities for F atoms and CF_x (x = 1, 2, 3) radicals were taken from the modeling works where these values were adjusted in order to obtain agreement between the measured and model-predicted densities of F and CF₂ in CF₄- based plasmas. The recombination probabilities for CHF_x (x = 1, 2) radicals were assumed to be the same as those for CF_x.

• The electronegativity of the CF₄ + CHF₃ + O₂ plasma under the given set of process condition is low enough to assume $n_{ - } /n_{e} < <$ 1 and $n_{e} \approx n_{ + }$ [44, 45], where $n_{ - }$ and $n_{e}$ are the densities of negative ions and electrons, respectively. This allows one to neglect the influence of dissociative attachment processes on both the plasma parameters and active species kinetics.

The adequacy of the given modeling algorithm, kinetic schemes and general approaches was demonstrated in previous works [44,45,46] by an acceptable agreement between measured and model-predicted plasma parameters and species densities.

Results and Discussion

Etching Rates and Selectivities

Figure 1 shows the etching rates for SiO_xN_y, Si and PMMA as functions of CHF₃ fraction in the CF₄ + CHF₃ + O₂ gas mixture. An increase in $y_{{{\text{CHF}}_{3} }}$ (in other words, the substitution of CF₄ for CHF₃) results in a monotonically decreasing SiO_xN_y etching rate which changes from 159–102 nm/min, or by ~ 1.6 times, for 0–91% CHF₃. Earlier, Ueno et al. [29] also mentioned the slower SiO_xN_y etching process in CHF₃ plasma compared with the CF₄ plasma under the RIE conditions. The etching rate of PMMA exhibited similar behavior with a quite close relative change (266–167 nm/min at 0–91% CHF₃), so that the SiO_xN_y/PMMA etching selectivity retained a nearly constant value of about 5.8 ± 0.5. At the same time, the etching rate of silicon decreases more rapidly compared with both SiO_xN_y and PMMA (167–71 nm/min, or by ~ 2.3 times, for 0–91% CHF₃). As a result, the SiO_xN_y/Si etching selectivity increases when using CHF₃-rich plasmas and occupies the range of 0.9–1.4.

Fig. 1

Measured etching rates (solid lines + open symbols) and etching selectivities (dashed lines + filled symbols) as functions of CHF₃ fraction in CF₄ + CHF₃ + O₂ gas mixture: 1—SiO_xN_y; 2—Si; 3—PMMA; 4—SiO_xN_y/Si; 5—SiO_xN_y/PMMA. Constant process conditions are p = 10 mTorr, W = 500 W and W_dc = 100 W

Similar changes in the SiO_xN_y, Si and PMMA etching rates versus CF₄/CHF₃ mixing ratio may be attributed to similar etching mechanisms for all three materials. In fact, this indicates that the etching processes of SiO_xN_y, Si and PMMA appear in the same regime, are limited by the same stages, and are driven by the same types of active species. Considering previous experience with the etching of SiO₂ and Si₃N₄, one can surely suggest the fluorine atoms as the chemically active species driving the SiO_xN_y etching process. Accordingly, the decrease in the SiO_xN_y etching rate shown in Fig. 1 may be caused by two principal factors, namely: (1) by the same behavior of both the F atom density and flux due to the changes in plasma parameters and gas-phase reaction kinetics; and/or (2) by a decrease in the effective probability of the SiO_xN_y + F reaction. Obviously, the latter may be sensitive to both ion bombardment intensity (through the breaking of Si–O bonds and sputtering of reaction products) and FC polymer thickness (through the change in the polymer deposition/destruction kinetics). Therefore, in order to provide a more precise understanding of the SiO_xN_y etching mechanism, data describing the plasma parameters as well as the densities and fluxes of F atoms, polymerizing radicals and positive ions are strongly required. For this purpose, we performed plasma diagnostics using Langmuir probes and plasma modeling.

Plasma Parameters and Densities of Active Species

The general mechanisms which determine the gas-phase characteristics in both CF₄ + O₂ + Ar and CHF₃ + O₂ + Ar plasmas with variable O₂ + Ar mixing ratios were the subjects of detailed discussion in our previous works [40, 41]. Unfortunately, these data do not provide a correct understanding of the situation when CF₄ and CHF₃ are combined together in one gas mixture. First, the mixing of two different gases normally causes changes in both $T_{e}$ (through the change in the collisional energy loss channels for electrons) and $n_{e}$ (through the change in their formation-decay balance). These phenomena suggest that the CF₄/CHF₃ mixing ratio will have an effect on the electron impact dissociation kinetics for all multi-atomic species. And secondly, CHF₃-containing plasmas are characterized by both effective decay of CH_xF_y radicals and generation of HF molecules via the gas-phase reactions CH_xF_y + H → CH_xF_y-1 + HF and CH_xF_y + F → CH_x−1F_y + HF [41]. Since the source species in these reactions are provided by both fluorocarbon components, the kinetics of atoms and radicals in a CF₄ + CHF₃ gas system are expected to be different compared with pure CF₄ and CHF₃ gases. All these facts suggest that the change of the CF₄/CHF₃ mixing ratio in a CF₄ + CHF₃ + O₂ plasma may result in non-linear and non-additive changes in the densities and fluxes of plasma active species responsible for the SiO_xN_y etching process. Below, we will summarize the existing data on reaction kinetics in individual CF₄ and CHF₃ gases as well as focus attention on the most principal issues related to CF₄ + CHF₃ mixing effects.

Figure 2 presents measured and model-predicted plasma parameters as functions of CHF₃ fraction in the CF₄ + CHF₃ + O₂ gas mixture. The substitution of CF₄ for CHF₃ results in decreases in both $T_{e}$ (4.3–3.5 eV at 0–91% CHF₃) and $n_{ + } \approx n_{e}$ (1.5 × 10¹⁰–1.3 × 10¹⁰ cm⁻³ at 0–91% CHF₃). A decrease in electron temperature toward CHF₃-rich plasmas probably results from the higher electron energy losses for low-threshold excitations (vibrational, electronic) of CHF₃ and HF compared with CF₄. The same behavior of both $n_{ + }$ and $n_{e}$ is caused by two phenomena which result in decreasing formation rates for both electrons and positive ions. First, an increase in $y_{{{\text{CHF}}_{3} }}$ enriches the plasma with harder ionizing species. This conclusion directly follows from the comparison of ionization rate coefficients for R1: CF₄ + e → CF₃⁺ + F + 2e ($k_{1}$ = 5.7 × 10⁻¹⁰ cm³/c at $T_{e}$ = 4 eV), R2: CHF₃ + e → CF₃⁺ + H + 2e ($k_{2}$ = 2.4 × 10⁻¹⁰ cm³/c at $T_{e}$ = 4 eV) and R3: HF + e → HF⁺ + 2e ($k_{3}$ = 1.9 × 10⁻¹⁰ cm³/c at $T_{e}$ = 4 eV) [43, 44]. As such, the substitution of CF₄ for CHF₃ lowers the total ionization frequency. And secondly, the decreasing $T_{e}$ suppresses ionization rate coefficients for all types of neutral species. This is because $\varepsilon_{iz}$ ≈ 12–15 eV > (3/2)$T_{e}$, where $\varepsilon_{iz}$ is the threshold energy for ionization [43], and (3/2) $T_{e}$ is the mean electron energy. Opposite behaviors of $n_{ + }$ and $j_{ + }$ (0.36–0.40 mA/cm² at 0–91% CHF₃, see Fig. 1b) result from the decrease in the effective ion mass according to the change in the dominant neutral species.

Fig. 2

Measured plasma parameters as functions of CHF₃ fraction in CF₄ + CHF₃ + O₂ gas mixture: 1—electron temperature (T_e), 2—total positive ion density (n₊) and electron density ($n_{e}$), 3—ion current density ($j_{ + }$), 4—negative dc bias (− U_dc)

Figure 2 illustrates the influence of CHF₃ content in the in CF₄ + CHF₃ + O₂ gas mixture on the densities of neutral species. In pure CF₄ plasma, the main sources of F atoms (~ 85% of total F atom formation rate) are R1, R4: CF₄ + e → CF₃ + F + e and R5: CF₃ + e → CF₂ + F + e. The total contribution from the CF₂ and CF radicals through R6: CF₂ + e → CF + F + e and R7: CF + e → C + F + e does not exceed 5% due to the low densities of these species which is a result of their multi-step formation mechanism. The remaining part comes from R8: F₂ + e → 2F + e, which is supported by the high F → F₂ recombination rate on the reactor walls. Accordingly, the last process represents the dominant loss channel for F atoms. The addition of even 9% of O₂ to CF₄ noticeably lowers the rates of R1, R4 and R5 due to the simultaneous decrease in $n_{e}$, $n_{{{\text{CF}}_{4} }}$ and $n_{{{\text{CF}}_{ 3} }}$. The density of CF₃ radicals decreases because of their decomposition via R9: CF₃ + O → CF₂O + F, R10: CF₃ + O(¹D) → CF₂O + F, R11: CF₃ + CFO → CF₄ + CO and R12: CF₃ + CFO → CF₂O + CF₂. The behavior of $n_{{{\text{CF}}_{4} }}$ follows that of $n_{{{\text{CF}}_{ 3} }}$ because the latter represents the main source of CF₄ molecules in the plasma chemical reactions. At the same time, the addition of O₂ introduces new effective channels for the formation of F atoms through R13: CF₂O + e → CFO + F + e and it accelerates R8. The high formation rate and density of CF₂O are provided by R9, R10, R14: 2CFO → CF₂O + CO and R15: CFO + F → CF₂O. The acceleration of R8 is due to the increase in $n_{{{\text{F}}_{2} }}$ because of the formation of these species in R16: CF₂O + O(¹D) → F₂ + CO₂. Therefore, the condition $n_{\text{F}} \approx n_{{{\text{F}}_{ 2} }} \approx n_{{{\text{CF}}_{ 2} {\text{O}}}} > n_{{{\text{CF}}_{x} }}$ (x = 1 − 3) is valid for the 91% CF₄ + 9% O₂ plasma. The substitution of CF₄ for CHF₃ at $y_{{{\text{O}}_{2} }}$ = const rapidly reduces the total F atom formation rate due to two basic phenomena. First, the CHF₃-containing plasma introduces effective mechanisms for the conversion of F into HF via gas-phase reactions R17: CHF_x + F → CF_x + HF ($k_{17}$ ~ 3.0 × 10⁻¹¹ cm³/s for x = 1, 2), R18: CHF_x + H → CHF_x−1 + HF ($k_{18}$ ~ 3.0 × 10⁻¹⁰ cm³/s for x = 1, 2) and R19: CF_x + H → CF_x−1 + HF ($k_{19}$ ~ 8 × 10⁻¹¹ cm³/s for x = 3 and ~ 4 × 10⁻¹¹ cm³/s for x = 2). Particularly, the high density of HF molecules (and even the domination of HF over the other neutral species) in CHF₃-based plasmas has been mentioned in Refs. [32, 47, 48]. And secondly, the formation of F atoms from both CHF₃ (R20: CHF₃ + e → CHF₂ + F + e, $k_{20}$ = 1.6 × 10⁻¹¹ cm³/c at $T_{e}$ = 4 eV) and HF (R21: HF + e → H + F + e, $k_{21}$ = 8.1 × 10⁻¹⁰ cm³/c at $T_{e}$ = 4 eV) seems to be less effective compared with CF₄ ($k_{1} + k_{4}$ = 8.3 × 10⁻¹⁰ cm³/c at $T_{e}$ = 4 eV). As such, the combination of decreasing formation rate and increasing loss rate for F atoms produces the deep decrease in $n_{\text{F}}$ (3.1 × 10¹³–4.5 × 10¹¹ cm⁻³ at 0–91% CHF₃, see Fig. 3). The main fluorocarbon radicals at $y_{{{\text{CHF}}_{ 3} }}$ < 60% are CF_x (x = 2, 3) while the further addition of CHF₃ leads to the domination of CHF. The reasons are the increase in both the formation rate for CHF in R18 and the decay rates for CF_x (x = 2, 3) in R19. The main sources of H atoms for the last processes are R21 and R22: CHF₃ + e → CF₃ + H + e. It is also important to note that the decay of O atoms in CHF_x + O → CF_xO + H ($k$ ~ 1.0 × 10⁻¹¹ cm⁻³) is less effective compared with CF_x + O → CF_x−1O + F ($k$ ~ 3.2 × 10⁻¹¹ cm⁻³). That is why, despite the increasing total density of CH_xF_y radicals (1.1 × 10¹³–2.3 × 10¹³ cm⁻³ at 0–91% CHF₃), the total loss rate for O atoms decreases toward CHF₃-rich plasmas. Accordingly, this causes an increase in O atom density (1.9 × 10⁹–3.3 × 10¹⁰ cm⁻³ at 0–91% CHF₃).

Fig. 3

Model-predicted densities of neutral species as functions of CHF₃ fraction in CF₄ + CHF₃ + O₂ gas mixture

Summarizing the above results, one can conclude that the substitution of CF₄ for CHF₃ in the CF₄ + CHF₃ + O₂ gas mixture: (1) lowers the efficiency of electron-impact processes due to the decrease in both $T_{e}$ and $n_{e}$, (2) results in decreasing F atom density, and (3) causes an increase in the total density of fluorocarbon radicals and oxygen atoms.

Etching Mechanism and Surface Conditions

The data on plasma parameters and densities of plasma active species allow one to trace the relationships between the gas-phase characteristics and the kinetics of the heterogeneous stages of the SiO_xN_y etching process. According to Refs. [49,50,51], the basic features of ion-assisted chemical reaction in polymerizing plasmas may be summarized as follows:

• 1. Any chemical etching pathway (the interaction of both the target material and FC polymer film with chemically active neutral species) has the rate of $\gamma_{R} \varGamma_{N}$, where $\varGamma_{N}$ is the flux of neutral species, and $\gamma_{R}$ is the effective reaction probability [18, 49]. Since a thicker FC film reduces the flux of chemically active neutrals to the FC film/etched surface interface, $\gamma_{R}$ for the target reaction depends on the polymer deposition/destruction kinetics [15, 16].

• 2. Any physical etching pathway (the breaking of chemical bonds between surface atoms, the ion-stimulated desorption of reaction products and the destruction of the FC polymer film) has the rate of $Y_{S} \varGamma_{ + }$, where $Y_{S}$ is the ion-type-averaged sputtering yield [49, 50], and $\varGamma_{ + } \approx j_{ + } /e$ is the ion flux. Since $Y_{S}$ is determined by the momentum which the incident ion transfers to the surface atoms in a single collision [22], one can assume $Y_{S} \sim\sqrt {\varepsilon_{i} }$, where $\varepsilon_{i} \approx e\left| { - U_{f} - U_{dc} } \right|$ is the ion bombardment energy, and $- \;U_{f} \approx 0.5T_{e} \ln (m_{e} /2.3m_{i} )$ is the floating potential.

• 3. The growth of the FC polymer film is driven by fluorocarbon radicals with two or more free bonds [22, 52], and the polymerization probability increases in fluorine-poor plasmas. This is because the polymer surface contains less saturated fluorocarbon groups and thus, reacts more easily with the CF_x species from the gas phase [52].

Based on these features, one can suggest a set of gas-phase-related parameters to analyze the SiO_xN_y etching kinetics in CF₄ + CHF₃ + O₂ plasma. Particularly, the effective probability of the SiO_xN_y + F reaction may be characterized by the $R_{{{\text{SiO}}_{x} {\text{N}}_{y} }} /\varGamma_{\text{F}}$ ratio, where $R_{{{\text{SiO}}_{x} {\text{N}}_{y} }}$ is the measured etching rate, and $\varGamma_{\text{F}}$ is the flux of F atoms. Accordingly, the FC polymer deposition rate is traced by the $\varGamma_{pol} /\varGamma_{\text{F}}$ ratio (where $\varGamma_{pol} = \varGamma_{{{\text{CF}}_{2} }} + \varGamma_{\text{CF}} + \varGamma_{{{\text{CHF}}_{3} }}$ is the total flux of polymerizing radicals) while the parameters $\varGamma_{pol} /\sqrt {\varepsilon_{i} } \varGamma_{ + } \varGamma_{\text{F}}$ and $\varGamma_{pol} /\varGamma_{\text{O}} \varGamma_{\text{F}}$ reflect the changes in polymer deposition/destruction balance in respect to physical (sputtering by ion bombardment) and chemical (etching by oxygen atoms) destruction pathways, respectively.

Figures 3 and 4 show it can be seen that the behaviors of $\varGamma_{\text{F}}$ versus CF₄/CHF₃ mixing ratio completely follows the change in $n_{\text{F}}$. Although a comparison of Figs. 1 and 4 shows the general agreement between the changes in the model-predicted $\varGamma_{\text{F}}$ and the measured SiO_xN_y etching rate, the situation seems to be not clear enough. The reason is that the much deeper decrease in $\varGamma_{F}$ produces an increased $R_{{{\text{SiO}}_{x} {\text{N}}_{y} }} /\varGamma_{\text{F}}$ ratio (2.5 × 10⁻¹⁶–1.1 × 10⁻¹⁴, or by ~ 45 times at 0–91% CHF₃, see Fig. 4b) that assumes the same change in $\gamma_{R}$. At the same time, this phenomenon cannot be explained within the conventional ion-assisted reaction mechanism which takes into account only the chemical effect from F atoms:

$${\text{F}} \to {\text{F}}_{(s)}$$

(R23)

$${\text{SiON}}_{(s)}{\mathop{\longrightarrow}\limits^{ions}} {\text{SiN}}_{(s)} + {\text{O}},{\text{SiON}}_{(s)}{\mathop{\longrightarrow}\limits^{ions}}{\text{SiO}}_{(s)} + {\text{N}}$$

(R24)

$${\text{SiN}}_{(s)} + {\text{F}}_{(s)} \to {\text{SiNF}}_{(s)} ,{\text{SiO}}_{(s)} + {\text{F}}_{(s)} \to {\text{SiOF}}_{(s)}$$

(R25)

$${\text{SiNF}}_{(s)}{\mathop{\longrightarrow}\limits^{ions}}{\text{SiF}} + {\text{N}},{\text{SiOF}}_{(s)} {\mathop{\longrightarrow}\limits^{ions}}{\text{SiF}} + {\text{O}}$$

(R26)

First, the parameter $\sqrt {\varepsilon_{i} } \varGamma_{ + }$ maintains a nearly constant value of ~ 4.5 × 10¹⁶ eV^1/2 cm⁻² s⁻¹ because the weak increase in ion flux ($\varGamma_{ + }$ = 2.3 × 10¹⁵–2.5 × 10¹⁵ cm⁻² s⁻¹ at 0–91% CHF₃) is balanced by the decreasing ion bombardment energy ($\varepsilon_{i}$ = 358–319 eV at 0–91% CHF₃). Therefore, it is hard to expect the intensification of both Si–O and Si–N bonds breaking (R24) and/or ion-stimulated desorption of reaction products (R26) with increasing CHF₃ fraction in the feed gas. And secondly, Fig. 4a shows that an increase in $y_{{{\text{CHF}}_{ 3} }}$ results in increases in both $\varGamma_{pol} /\sqrt {\varepsilon_{i} } \varGamma_{ + } \varGamma_{\text{F}}$ (by ~ 1400 times at 0–91% CHF₃) and $\varGamma_{pol} /\varGamma_{\text{O}} \varGamma_{\text{F}}$ (by ~ 85 times at 0–91% CHF₃) ratios. The reason is the rapidly increasing polymer deposition rate ($\varGamma_{pol}$ = 1.3 × 10¹⁶–2.8 × 10¹⁷ cm⁻² s⁻¹ and $\varGamma_{pol} /\varGamma_{\text{F}}$ = 0.02–30 at 0–91% CHF₃) which overwhelms the changes in both $\sqrt {\varepsilon_{i} } \varGamma_{ + }$ and $\varGamma_{\text{O}}$. Obviously, such a situation corresponds to increasing thickness of the FC polymer film as well as retarded the transport of F atoms to the etched surface (R23).

Fig. 4

Model-predicted fluxes, and flux-to-flux and rate-to-flux ratios as functions of CHF₃ fraction in CF₄ + CHF₃ + O₂ gas mixture. The labels on the curves in a “F”—$\varGamma_{\text{F}}$; “HF”—$\varGamma_{\text{HF}}$, “F + HF”—$\varGamma_{\text{F}} + \varGamma_{\text{HF}}$, “pol”—$\varGamma_{pol}$, “pol/F”—$\varGamma_{pol} /\varGamma_{\text{F}}$, “pol/(Fi⁺)”—$\varGamma_{pol} /\sqrt {\varepsilon_{i} } \varGamma_{ + } \varGamma_{\text{F}}$ (× 10¹⁸), “pol/(FO)”—$\varGamma_{pol} /\varGamma_{\text{F}} \varGamma_{\text{o}}$ (× 10¹⁴); “F/+”—$\varGamma_{pol} /\sqrt {\varepsilon_{i} } \varGamma_{ + }$. The labels on the curves in b “R/F”—$R_{{{\text{SiO}}_{x} {\text{N}}_{y} }} /\varGamma_{\text{F}}$ (× 10¹⁶), “R/(F + HF)”—$R_{{{\text{SiO}}_{x} {\text{N}}_{y} }} /(\varGamma_{\text{F}} + \varGamma_{\text{HF}} )$ (× 10¹⁶)

Based on this, one can suggest that F atoms are not the only type of chemically active species participating in the SiO_xN_y etching process. In our opinion, additional chemical etching may occur due to HF molecules. From previous works [53,54,55,56], it can be understood that HF may be an effective etchant for Si, SiO₂, and Si₃N₄ in a dry plasma-less process combined with water or methanol. Though the reaction mechanism is quite complicated and passes through several intermediate stages, it can be surely associated with the dissociation and/or ionization of HF molecules due to their interaction with highly polar OH groups [53,54,55]. Clements et al. [54] suggested a simplified reaction scheme for the SiO₂ + HF system which assumes the hydrogenation of the etched surface and provides reasonable agreement between measured and model-predicted etching rates. Based on these data, an etching mechanism involving HF molecules, and O and H atoms may be proposed as follows:

$${\text{HF}} \to {\text{HF}}_{(s)} ,\;{\text{O}} \to {\text{O}}_{(s)} ,\;{\text{H}} \to {\text{H}}_{(s)}$$

(R27)

$${\text{SiN}}_{(s)} + {\text{O}}_{(s)} \to {\text{SiO}}_{(s)} + {\text{N}}$$

(R28)

$${\text{SiO}}_{(s)} + {\text{H}}_{(s)} \to {\text{SiOH}}_{(s)}$$

(R29)

$${\text{SiOH}}_{(s)} + {\text{HF}}_{(s)} \to {\text{SiF}}_{(s)} + {\text{H}}_{ 2} {\text{O}}$$

(R30)

$${\text{SiF}}_{(s)}{\mathop{\longrightarrow}\limits^{ions}}{\text{SiF}}$$

(R31)

The action of O atoms results in the oxidation of partially decomposed SiO_xN_y in the form of SiO_x (R28). Since the strength of the Si–O bond is higher than that of the Si–N bond, this reaction can occur spontaneously even at room temperature. The interaction of non-saturated SiO_x with hydrogen atoms produces Si(OH)_x (R29) while the latter reacts with HF to form SiF_x (R30). The increasing density of HF as well as the condition $n_{\text{HF}} > n_{\text{F}}$ obtained for $y_{{{\text{CHF}}_{ 3} }}$ > 10% cause the monotonic increase in $\varGamma_{\text{F}} + \varGamma_{\text{HF}}$ toward CHF₃-rich plasmas (6.3 × 10¹⁷–1.3 × 10¹⁸ cm⁻² s⁻¹ at 0–91% CHF₃, see Fig. 4a). Then, assuming similar partial reaction probabilities for F and HF with the SiO_xN_y surface, the effective reaction probability may be roughly characterized by the $R_{{{\text{SiO}}_{x} {\text{N}}_{y} }} /(\varGamma_{\text{F}} + \varGamma_{\text{HF}} )$ ratio. From Fig. 4b, it can be seen that this parameter decreases with increasing CHF₃ fraction in the feed gas (2.5 × 10⁻¹⁶–8.2 × 10⁻¹⁷, or by ~ 3 times at 0–91% CHF₃) and thus, shows good agreement with the model-predicted change in the FC polymer film thickness. Such a situation is absolutely typical for SiO₂ and Si₃N₄ etching processes in fluorocarbon gas plasmas [13,14,15,16].

In order to obtain additional information on the SiO_xN_y etching process as well as to verify the above analysis of the etching mechanism, we examined the plasma-treated surfaces by XPS. Figures 5, 6 and 7 represent the XPS narrow scan spectra for the C 1s, F 1s, and Si 2p peaks of SiO_xN_y surfaces treated in 91% CF₄ + 9% O₂ ($y_{{{\text{CHF}}_{ 3} }}$ = 0) and 91% CHF₃ + 9% O₂ ($y_{{{\text{CF}}_{ 4} }}$ = 0) plasmas. The substitution of CF₄ for CHF₃ in the CF₄ + CHF₃ + O₂ gas mixture is accompanied by increasing peak intensities and areas for both C − F_x (x = 1–3) and F–C bonds. This directly points to an increasing amount of residual FC polymer and, in fact, confirms model-based conclusions concerning changes in both the polymer deposition/destruction balance and polymer film thickness. Another indirect proof for the increasing polymer deposition rate and growth of polymer film thickness is the increasing surface roughness (2.6–5.1 nm at 0–91% CHF₃, see Fig. 8a). Particularly, the direct correlation between these parameters has been mentioned in Ref. [57]. It is important to note that an increase in polymer deposition rate toward CHF₃—rich plasmas allows one to explain the changes in SiO_xN_y/PMMA and SiO_xN_y/Si etching selectivities shown in Fig. 1. Really, since SiO_xN_y and PMMA are oxygen-containing materials, these can be characterized by identical FC film chemical etching mechanisms at both the FC film/plasma and FC film/etched surface interfaces. As such, one can reasonably assume similar FC film thicknesses, similar effective reaction probabilities, and thus, similar slopes of the corresponding etching rates with increasing CHF₃ fraction in the feed gas. At the same time, the FC polymer film on a Si surface is thicker compared with both SiO_xN_y and PMMA due to the absence of the chemical etching effect at the FC film/etched surface interface. Accordingly, an increase in the polymer deposition rate with increasing $y_{{{\text{CHF}}_{ 3} }}$ results in faster film growth as well as a deeper decrease in both the effective reaction probability and Si etching rate. The decreases in peak intensities and areas for F–Si and Si–F bonds (see Figs. 6 and 7) are correlated with decreasing SiO_xN_y etching rate. The opposite changes in peak intensities and areas for F–Si and Si–F bonds on one side and F–O bonds on the other hand also look reasonable and may be attributed to the increasing fraction of SiF_xO_y compounds among the etching products. At least, this suggestion agrees with the change in O atom flux ($\varGamma_{\text{o}}$ = 4.2 × 10¹³–7.4 × 10¹⁴ cm⁻² s⁻¹ at 0–91% CHF₃). The weak change in peak intensity and area for Si–O bond (Fig. 7) does not contradict with the proposed reaction mechanism R27–R31. Since an increase in $y_{{{\text{CHF}}_{ 3} }}$ results in similar increasing trends for fluxes of O, H and HF species, an increase in the S–O formation rate in R28 is balanced by an increase in decomposition rates through R29 and R30. As a result, the steady-state amount of oxidized silicon on the etched surface remains at nearly constant level. The data of Fig. 8b also show that plasma-treated SiO_xN_y surfaces are characterized by a quite high contact angle for deionized water (53.6–71.5° at 0–91% CHF₃) and a relatively low free surface energy (47.2–45.0 mN/m at 0–91% CHF₃). Corresponding values for the untreated SiO_xN_y surface are 31.35° and 67.75 mN/m. Obviously, the treatment in fluorocarbon gas plasma leads to increased surface hydrophobicity. This phenomenon is in good agreement with published data [57, 58] and, in fact, confirms the presence of continuous FC polymer film. The slight decrease in free surface energy (and thus, increase in hydrophobicity) toward CHF₃-rich plasmas may be connected with the increasing hydrogen content in the FC film structure.

Fig. 5

XPS narrow scan spectra for C 1s of SiO_xN_y surfaces treated in 91% CF₄ + 9% O₂ (a) and 91% CHF₃ + 9% O₂ (b) plasmas

Fig. 6

XPS narrow scan spectra for F 1s of SiO_xN_y surfaces treated in 91% CF₄ + 9% O₂ (a) and 91% CHF₃ + 9% O₂ (b) plasmas

Fig. 7

XPS narrow scan spectra for Si 2p of SiO_xN_y surfaces treated in 91% CF₄ + 9% O₂ (a) and 91% CHF₃ + 9% O₂ (b) plasmas

Fig. 8

Physical surface characteristics as functions of CHF₃ fraction in CF₄ + CHF₃ + O₂ gas mixture. a Measured MRS roughness. b Measured contact angles (1—de-ionized water, 2—diiodomethane) and calculated free surface energy (3)

Finally, we would like to note that the proposed mechanisms provide mainly qualitative analysis of the SiO_xN_y etching process due to the evident simplifications made in the primary assumptions. At the same time, the given model-based analysis does not contradict with experimental results and allows one to trace the interconnections between operating conditions and the SiO_xN_y etching kinetics through the gas-phase plasma characteristics (internal plasma parameters, kinetics of plasma active species) and the FC film thickness-sensitive reaction probability. We also suggest that the proposed etching mechanism is valid for CF₄ + CHF₃ + O₂ gas mixtures with higher $y_{{{\text{O}}_{2} }}$ values. The reasons are that an increase in $y_{{{\text{O}}_{2} }}$, (1) provides similar qualitative changes in densities of F atoms and polymerizing radicals in both CF₄- and CHF₃-based plasmas [40, 41], (2) does not disturb the quantitative correlations between densities and fluxes of active species in these gas systems [40, 41], and (3) does not influence the gas-phase chemistry of HF molecules through the reactions with oxygen-containing species [41]. As such, one can surely expect no principal changes in both gas-phase and heterogeneous reaction kinetics, except the absolute species densities and process rates.

Conclusion

In this work, we investigated the etching characteristics, etching mechanisms and surface conditions for SiO_xN_y thin films in CF₄ + CHF₃ + O₂ inductively coupled plasma. Experiments showed that the substitution of CF₄ for CHF₃ in the feed gas at a constant gas pressure, input power and bias power causes the monotonic decrease in the SiO_xN_y etching rate and results in increasing SiO_xN_y/Si etching selectivity. Gas-phase plasma parameters, chemistry and the steady-state composition were analyzed using plasma diagnostics by Langmuir probes and 0-dimensional (global) plasma modeling. It was found that an increase in the CHF₃ fraction in the feed gas: (1) suppresses electron-impact kinetics through decreases in both electron temperature and density, (2) is accompanied by opposite changes in the densities of F atoms and polymerizing radicals, and (3) results in increasing O atom density. The SiO_xN_y etching mechanism was analyzed through the relationships between the measured etching rates and the model-predicted fluxes of active species together with the examination of etched surfaces by XPS, AFM and contact angle measurements. It was shown that an increase in the CHF₃ fraction in the feed gas shifts the polymer deposition/destruction balance toward deposition and results in increasing amounts of residual polymer on the SiO_xN_y surface. It was proposed that the SiO_xN_y etching process: (1) involves contributions from HF molecules; and (2) exhibits the features of the (F + HF)-flux-limited etching regime with a polymer-thickness-sensitive reaction probability. The hydrophobic nature of plasma-treated SiO_xN_y surfaces is connected with the presence of continuous fluorocarbon polymer films.