Thermal adsorption-enhanced atomic layer etching of Si₃N₄

With the continued scaling demands of Si-based complementary metal-oxide-semiconductor devices in advanced chip technology such as smartphones and other connected devices, the introduction of novel patterning methods with fine control over extremely small dimensions is becoming increasingly important in order to enable a variety of emerging electronic devices beyond the 10 nm technology node.^1–3 Until now, reactive-ion plasma etching techniques have been widely utilized for conventional top-down patterning processes in the fabrication of semiconductor devices.^4–6 However, these conventional reactive ion processes can lead to physical damage of device surfaces and interfaces, in that surface defects are created due to the highly energetic, reactive ions used to achieve vertical etch profiles.^6–8 In addition, the high energy ion bombardment results in high rates of etching, leading to difficulties achieving the precise control needed to manipulate etching at the level of monolayers. Accordingly, a damage-free directional (anisotropic) etching method that can precisely remove angstrom-thick layers of material is the subject of intensive investigation.^1–3 

To this end, atomic layer etching (ALE), a reverse process inspired by atomic layer deposition,^9,10 has attracted increasing attention due to its ability to control etch rates at the atomic scale and to achieve wafer scale uniformity without physical damage.^1,2,8,11 More specifically, ALE is a layer-by-layer, thin film removal method enabled through the control of self-limiting surface reactions in conjunction with sequential exposures of chemical species. To date, many of the reported ALE processes have focused on Si,^11,12 Ge,^13,14 and compound semiconductors.^15,16 More recently, research efforts have shifted to include ALE of various oxides such as SiO₂,¹⁷ Al₂O₃,^18,19 HfO₂,^20,21 ZrO₂,²² and TiO₂.²³ 

Meanwhile, Si₃N₄ is one of the most widely used thin film materials in semiconductor technologies.^24–26 One of its key applications is as a spacer film, where the Si₃N₄ serves multiple roles such as a barrier layer, contact etch stop layer, and capping layer for a stress memorization technique.²⁷ For the application of Si₃N₄ as a spacer layer in extremely miniaturized transistor structures, the requirements of uniform material quality and conformal film thickness throughout the device, regardless of the transistor pitch, are becoming ever more rigorous. In spite of its versatile utilization and technical importance in modern semiconductor devices, there have been only a few previous publications associated with ALE of Si₃N₄ films.^28–30 First, in 1999, Matsuura et al. reported a role-sharing plasma etching method for the atomic-order, layer-by-layer removal of Si₃N₄ films, in which plasma-excited H₂ molecules removed a surface layer of nitrogen, followed by plasma-generated H⁺/Ar⁺ etching of silicon. More recently, Sherpa and Ranjan demonstrated quasiatomic layer removal of Si₃N₄ films using a sequential cyclic process of surface modification by hydrogen ions and the removal of modified layers by fluorine radicals.³⁰ However, the ALE etch rate reported in that process was over 60 Å/cycle, which may be too high for modern semiconductor manufacturing with sub-10 nm critical dimension features.³⁰ The lack of other layer-by-layer etch chemistries for Si₃N₄ films motivates investigation into different cyclic processes for self-limiting ALE reaction chemistries for this material.

In the present study, therefore, we report a combination process for ALE of Si₃N₄ films using alternating repetition of thermal CH₃F adsorption and Ar⁺ bombardment steps. Unlike in the case of plasma-assisted adsorption, restricting the etchant adsorption step to a thermal process is expected to avoid unwanted plasma-polymerization,^24,31,32 thereby providing well-controlled chemisorption during the first half cycle of etchant exposure to a gas such as CH₃F. Accordingly, our results suggest that the thermal chemisorption of etchant molecules helps to form a modified layer, which in turn can be etched away readily during energetic ion bombardment. On the other hand, including at least one ion bombardment-based step in the cyclic process allows the possibility for directional etching to be achieved, a benefit important to a majority of critical applications for fabrication of modern devices.³³ The experimental scheme used in the Si₃N₄ ALE process is illustrated schematically in Fig. 1, where this cyclic process is composed of four consecutive steps: (1) Exposure of CH₃F precursor molecules to the Si₃N₄ surface at elevated temperature; (2) evacuation of excess CH₃F molecules, so that only the chemically adsorbed layer remains on the surface, avoiding unwanted etching by gas-phase species; (3) exposure of Ar⁺ plasma to effectively remove the chemisorbed gas molecules; and (4) evacuation of the reaction products and achievement of the cyclic ALE process. It is worth noting that the CH₃F step proceeds through saturated adsorption onto the Si₃N₄ surface. Hence, we demonstrate the self-limiting etching characteristics of Si₃N₄ ALE with increasing CH₃F etchant flow rate and exposure time, respectively. We also show a good ALE synergy factor when both precursor steps are combined. Additionally, on the basis of chemical analyses by x-ray photoelectron spectroscopy (XPS) and optical emission spectroscopy (OES), potential etch steps during the ALE Si₃N₄ process are proposed.

The ALE apparatus employed in this study (shown schematically in Fig. S1 in supplementary material)⁴⁴ has an upper gas inlet and lower vacuum pumping lines designed for efficient gas transport from the manifold to the substrate. It consists of an inductively coupled plasma (ICP) remote plasma system and an electrostatic chuck tuned with a radio frequency (13.56 MHz) power source. The ALE of Si₃N₄ films was carried out using alternating exposure of CH₃F gas and Ar⁺ plasma as the etchant precursor for formation of a surface modification layer and subsequent ion bombardment for removal of the previously reacted region, respectively, with computer-controlled solenoid valves. The flow of CH₃F gas molecules fed into the reaction chamber was controlled between 50 and 100 sccm by mass flow controller (MFC). Ar gas was delivered into the reaction chamber with a constant flow rate of 400 sccm to evacuate excess gas molecules and byproducts between each etchant and plasma etching step. For the Ar⁺ plasma step, the same amount of Ar gas controlled by the MFC was also flowed into the chamber upstream of the remote plasma generation system perpendicular to the center of the specimen holder. During the plasma step, 1000 W of the ICP source power and 150 W of the electrostatic chuck power were applied, respectively.

A single cycle of the ALE process consisted of four consecutive steps: (1) CH₃F exposure of 7 s, (2) Ar purging of 5 s, (3) Ar⁺ plasma exposure of 10 s, and (4) another Ar purging of 5 s. These process times were determined by carefully monitoring the process conditions to enable an optimized ALE mode. By monitoring chamber pressure, purge times above 3 s were found to be sufficient for residual species or byproducts to be evacuated from the reaction chamber; 5 s was chosen for both purge steps to safely avoid any undesired gas-phase reaction. An Ar⁺ plasma step of 10 s yields a nonzero sputtering rate, but it was selected as it yielded good wafer uniformity (see Sec. III). During the entire ALE process, a working pressure of 30 mTorr was maintained, and the substrate temperature was set to 80 °C to allow for stable chemisorption of CH₃F gas molecules without the thermal decomposition that occurs at higher temperatures.

The Si₃N₄ films were prepared on a Si(100) substrate by plasma enhanced chemical vapor deposition (PE-CVD) using SiH₄, NH₃, and N₂ at 400 °C. Samples used in the ALE studies were 2700 Å thick. For determination of etch rates, the film thickness was measured by spectroscopic ellipsometry together with cross-sectional field emission scanning electron microscopy (SEM). An atomic force microscope (AFM) was used to measure the surface roughness. An angle-resolved XPS (Al K_α monochromatic source of 1486.6 eV) with a take-off angle of 15° to the surface was utilized to analyze the chemical species adsorbed on the Si₃N₄ surface following the CH₃F dose, and OES was used to detect the etching products during the Ar⁺ plasma step.

To achieve highly uniform etching with monolayer accuracy, the ALE process must proceed through self-limiting steps that control the amount of material removed during each cycle.^1–3 In the case of CH₃F etchant, this requires that CH₃F undergo self-limiting, saturated adsorption at the surface. Figure 2 shows the etch rate per cycle (Å/cycle) and wafer nonuniformity (%) of Si₃N₄ films following ALE as a function of the CH₃F flow rate and CH₃F exposure time. The etch rate clearly saturates at CH₃F flow rates above 70 sccm, indicative of self-limiting behavior [Fig. 2(a)], which is a characteristic feature of the ALE method.^1,2 The saturated etch rate per cycle of the Si₃N₄ ALE is estimated to be near 6 Å/cycle, a value which is comparable to previous reports on plasma-assisted ALE methods of related materials.^1,16,34 This high etch rate is indicative of the presence of few Å-thick reactive layers upon the first half cycle of CH₃F etchant exposure (e.g., ∼7 ± 2 Å reactive layer thickness in simulations by Brichon et al. for Cl₂⁺-exposed Si, i.e., SiCl_x).³⁴ On an atomic scale, saturation of etching at a single unit step should lead to atomic level smoothness of the etching surface, hence measurement of film uniformity is an important metric of an ALE process.³ Accordingly, wafer nonuniformity values, defined by

were collected. Consistent with the saturated etch rate, the wafer nonuniformity is lowered with increasing CH₃F flow rate, saturating at a value of ∼3% at flow rates above 70 sccm. These nonuniformity values are close to meeting the standard for industry implementation.

Under saturation conditions with the CH₃F dose fixed at 100 sccm, the effect of CH₃F exposure time was also investigated. Very similar results to those seen in the CH₃F flow rate studies are obtained, yielding a saturated etch rate of 6 Å/cycle and nonuniformity of 3% when the CH₃F exposure time exceeds 7 s [Fig. 2(b)]. Hence, we have chosen a CH₃F exposure of 7 s for subsequent experiments. In addition, both SEM and AFM experiments were performed to confirm the etch rate and surface uniformity. Cross-sectional SEM images after 50 cycles of ALE (Fig. S2 in supplementary material) show removal of 300 Å of material, corresponding to an etch rate 6 Å/cycle, which is consistent with the value determined from ellipsometry. Root mean squared roughness values measured by AFM (Fig. S3 in supplementary material) before and after 50 cycles of ALE are nearly identical, confirming that uniformity is not degraded after the etch process. Therefore, we conclude that the current ALE process produces uniform layer-by-layer etching through a surface saturation mode.

In the ALE process developed in this study, the Ar⁺ plasma step is expected to effectively eliminate the chemical species generated from the adsorbed CH₃F layer. An ideal ALE system requires a cooperative etch process wherein both steps are necessary for film removal. For this study, however, we performed ALE under conditions at which there was a small Ar⁺ plasma sputter rate of ∼0.2 Å/s, slightly higher than the sputtering threshold condition; these conditions were chosen because they produced highly reliable, wafer scale uniformity for both ALE and sputtering processes. In order to interrogate the role of the Ar⁺ plasma step for the current process, etch yields with each step performed separately were compared to the etch yield achieved when both steps were performed sequentially. The results, shown in Fig. 3, clearly indicate that there is synergistic interaction between the self-limited adsorption of the CH₃F molecules and the Ar⁺ sputtering. Figure 3(a) compares the etch amounts under three different conditions: (1) CH₃F gas exposure only, (2) sequential exposures of CH₃F gas and Ar⁺ bombardment, and (3) Ar⁺ bombardment only, for which the same exposure times of the CH₃F (7 s) and Ar⁺ plasma (10 s) were used, respectively. The data show that combining both chemisorbed CH₃F molecules and the Ar⁺ bombardment on the reaction surface leads to an etch rate three times higher than that of Ar⁺ sputtering only.^4,5 On the other hand, for the CH₃F exposure only, negligible etch amounts were found in the absence of Ar⁺ sputtering, indicating that during the CH₃F dose, CH₃F molecules formed involatile chemical adsorption products, as we will describe later based on XPS analyses.

We can quantify the ALE synergy by defining it as a percentage relative to the total amount of material removed per cycle³³ 

where EPC is the “etch per cycle” rate and the values of “α” and “β” are contributions from each separate chemisorption and removal step. From this equation, the effect of ALE synergy can be calculated to be 67%. Although the ALE synergy may ideally approach 100% if there is no etching contribution from either step alone, synergy factors below 100% are typical and similar results can be found in the recent literature on ALE of Si₃N₄.³⁰ Nonzero contributions of α and β can originate from several factors such as physical sputtering, photon-induced etching, step contamination, and/or competing reactions of conventional etching in practice. We speculate that the synergy factor could be improved for this system by further tuning the plasma parameters beyond the range that we were limited to in the current experimental setup.

Mechanistic insight into the ALE process can be gained by looking at the transition from the process combining CH₃F gas and Ar⁺ to the sputter-only process. We clearly observed a decrease in the etch rate (corresponding to the linear slope in a plot of Si₃N₄ etch amount per cycle) by changing the etch mode from ALE (slope: ∼6 Å/cycle) to Ar⁺ sputtering alone (slope: ∼2 Å/cycle), as shown in Fig. 3(b). In the second part of the experiment, the sputtering mode was performed in the absence of the CH₃F dose; i.e., the single cyclic process consisted of Ar⁺ plasma exposure followed by evacuation. After removing the CH₃F exposure steps in the second part of the experiment, the etch rate immediately dropped to that of Ar⁺ sputtering only: i.e., no extension of the higher ALE etch rate was observed at the transition from ALE to sputtering mode (i.e., cycle 4 → cycle 5). This behavior indicates that the surface modification layer generated by the CH₃F exposure is limited to a single layer directly at the surface. If the CH₃F molecules had penetrated more deeply into the Si₃N₄ surface, a higher etch rate would be expected to persist beyond the transition point due to the residual presence of the modified layer. Therefore, the result in Fig. 3(b) implies that the CH₃F molecules contribute to thermal chemisorption primarily on top of the Si₃N₄ surface during the first half reaction cycle, followed by well-controlled removal of the modified layer during the Ar⁺ plasma half cycle.

In addition to a higher etch rate for ALE compared to pure Ar⁺ sputtering, the film uniformity was also significantly better during the ALE process. Film thickness contour maps (∼200 mm wafer scale) with increasing number of etching cycles are depicted in Figs. 3(c)–3(j) for both the ALE and pure sputtering modes. The thickness variations are smaller for the films treated by ALE [Figs. 3(c)–3(f)] compared to those for the films treated by Ar⁺ sputtering alone [Figs. 3(g)–3(j)]. The estimated wafer nonuniformity during the ALE mode was determined to be ∼3%, which is two times lower than that from the sputtering mode (∼6%). We propose that in the current ALE Si₃N₄ process, the atomically smooth surfaces originate from the layer-by-layer etching mechanism by well-controlled elimination of the adsorbed CH₃F molecules.^1,2

To identify the chemical species at the Si₃N₄ surface resulting from the CH₃F exposure, angle-resolved XPS analyses were carried out before (as-deposited Si₃N₄ films) and after the CH₃F exposure as shown in Fig. 4. The Si 2p high resolution spectrum of as-deposited Si₃N₄ is deconvoluted into two peaks, which are assigned to the Si–O (103.4 eV) and Si–N (102 eV) bonding components, respectively [upper data in Fig. 4(a)]. These peaks correspond well with literature values for PE-CVD Si₃N₄ films.^26,35 The presence of oxygen in the XPS spectrum is expected for PE-CVD Si₃N₄ samples, as bulk oxygen is known to be incorporated, likely due to the highly porous structure created by the large amount of hydrogen trapped in the PE-CVD Si₃N₄.^26,35,36 In addition, some surface oxygen may arise during the ex situ XPS measurements. Following the CH₃F exposure, notable changes in the Si 2p spectrum are observed, with the appearance of a new broad feature that can be deconvoluted into two peaks [lower data in Fig. 4(a)]. These high binding energy chemical states are tentatively assigned to SiF_x (109.2 eV) and Si(CH₃)_x (108 eV) species present in a reaction layer formed during the CH₃F chemisorption.³⁷ 

In the N 1s high resolution spectrum of as-deposited Si₃N₄, two deconvoluted peaks are seen [upper data in Fig. 4(b)]. The lower binding energy peak centered at 397.6 eV is assigned to Si–N bonding.³¹ The other peak located at 400 eV can be possibly attributed to chemical species containing N–H, N–O, and N–C bonds, based on literature reports.^35,38,39 As discussed earlier, PECVD Si₃N₄ generally contains a large amount of hydrogen incorporated in the form of appreciable N–H bonds, followed by the probable ambient contamination of N-C and N-O content, consistent with the presence of the 400 eV XPS peak.^26,35 However, after the CH₃F dose, as with the case of Si 2p, a new peak centered at 401.4 eV appears [lower data in Fig. 4(b)]. This high binding energy peak can be attributed primarily to nitrogen species bonded to more electronegative elements such as carbon with fluorine neighbors, indicating the existence of a fluorocarbon-modified Si₃N₄ reaction layer.³¹ High resolution scans over the C 1s and F 1s regions of the spectrum (Fig. S4 in supplementary material) show the appearance of CF_x species after the CH₃F dose. Therefore, all the high binding energy peaks in the Si 2p and N 1s XPS spectra are clearly indicative of the presence of a modified surface layer caused by various forms of chemisorbed species from the CH₃F exposure.

OES analyses were carried out to monitor the volatile etch products formed by the Ar⁺ bombardment of the CH₃F-modified Si₃N₄ layer. Figure 5 shows the optical emission spectrum collected halfway through exposure of the CH₃F-treated Si₃N₄ film to the Ar⁺ plasma. For comparison, an OES spectrum collected during exposure of a CH₃F-exposed Si wafer is also shown. In spectra from both samples, the strong emission lines of Ar are present, indicating the generation of electronically excited Ar during this plasma step. For the CH₃F-exposed Si specimen, the only peaks detected other than those from the excited Ar are weak emission signals from N₂ (357.9 nm) and SiF (440.1 nm).⁴⁰ The presence of only these signals suggests that there is an absence of chemical reaction pathways to form H- and C-containing species on the Si surface by the incoming CH₃F molecules. In contrast, the CH₃F-modified Si₃N₄ film exhibits strong emission lines associated with NH (336 nm), N₂ (357.9 nm), CN (386.2 nm), and SiF species (440.1 nm), which suggests the existence of volatile etch products such as NH₃, N₂, CN-containing species of (CN)₂, hydrogen cyanide (HCN) and cyanogen fluoride (FCN), and SiF-containing species of SiF_x and SiF_x(CH₃)_y occurring during the Ar⁺ bombardment, respectively.⁴⁰ We note that no NF-related products, which would exhibit an OES peak at 528.8 eV, are observed, suggesting that species with direct N–F bonds are not formed during the process (Fig. S5 in supplementary material).⁴¹ These results are in good correspondence with the previous literature on plasma-assisted etching of Si₃N₄ films, in which these same types of etch by-products, i.e., those associated with the chemisorbed species implicated by XPS and consistent with species detected by OES in the present work, were found.^24,40,42,43

Based on the XPS and OES results, as well as comparison with the literature,^24,40,42 the following reaction steps for the ALE process are proposed:

• Step 1. Formation of the surface modification layer by the CH₃F chemisorption

  $Si 3 N 4 ( solid ) + CH 3 F ( gas ) → SiF x ( ads ) + Si ( CH 3 ) x ( ads ) + CH 3 N ( ads ) + FCH x N ( ads ) + HF ( gas ) .$

   

  (1)
• Step 2. Desorption of reaction products by the Ar⁺ plasma sputtering

  $SiF x ( ads ) + Si ( CH 3 ) x ( ads ) + CH 3 N ( ads ) + FCH x N ( ads ) + Ar + ( plasma ) → SiF x ( gas ) + SiF x ( CH 3 ) y ( gas ) + ( CN ) 2 ( gas ) + HCN ( gas ) + FCN ( gas ) + N 2 ( gas ) + NH 3 ( gas ) .$

   

  (2)

Overall, the current experimental results demonstrate an excellent approach for achieving self-limiting layer-by-layer etching of Si₃N₄ films. We foresee that this ALE method can potentially provide opportunities applicable to fine etch control of metals, semiconductors, and dielectrics for fabrication of emerging 3D nanodevices.

In this work, we demonstrated the atomic layer etching of Si₃N₄ films by sequential exposures of the CH₃F etchant and an Ar⁺ plasma. The self-limiting, layer-by-layer etching characteristics of an ALE process were experimentally confirmed in studies showing that the etch rate saturates with increasing CH₃F etchant flow rate and exposure time. For the Si₃N₄ ALE process, using alternating exposures of CH₃F and an Ar⁺ plasma yielded higher etch rates and better film uniformity than a process using Ar⁺ ion sputtering alone, and an ALE synergy factor of 67% was determined. On the basis of XPS analyses, it was found that a surface modification layer was generated after CH₃F exposure that contained various possible forms of SiF_x, Si(CH₃)_x, and nitrogen species bonded to carbon and more electronegative carbon with fluorine neighbors. Additionally, OES analysis revealed the existence of etch products that may include SiF₄, SiF_x(CH₃)_y, (CN)₂, HCN, FCN, N₂, and NH₃ during Ar⁺ bombardment. The current ALE scheme we present here is envisaged to extend opportunities for highly controllable and uniform etching technologies applicable to upcoming 3D nanoelectronic devices.

This work was supported by the Samsung Electronics Project (D.S., S.O., J.W., S.L., and H.L.) and by the Department of Energy under Award No. DE-SC0004782 (W.H.K. and S.F.B.).