Thermal Atomic Layer Etching of Aluminum Oxide (Al2O3) Using Sequential Exposures of Niobium Pentafluoride (NbF5) and Carbon Tetrachloride (CCl4)

: Thermal atomic layer etching (ALEt) of amorphous Al 2 O 3 was performed by alternate exposures of niobium penta ﬂ uoride (NbF 5 ) and carbon tetrachloride (CCl 4 ). The ALEt of Al 2 O 3 is observed at temperatures from 380 to 460 ° C. The etched thickness and the etch rate were determined using spectroscopic ellipsometry and veri ﬁ ed by X-ray re ﬂ ectivity. The maximum etch rate of about 1.4 Å/cycle and a linear increase of the removed ﬁ lm thickness with the number of etch cycles were obtained at a temperature of 460 ° C. With the help of density functional theory calculations, an etch mechanism is proposed where NbF 5 converts part of the Al 2 O 3 surface into an AlF 3 or aluminum oxy ﬂ uoride layer, which upon reacting with CCl 4 is converted into volatile halide-containing byproducts, thus etching away the converted portion of the material. Consistent with this, a signi ﬁ cant surface ﬂ uorine content of about 55 at. % was revealed when the elemental depth pro ﬁ le analysis of a thick NbF 5 -treated Al 2 O 3 layer was performed by X-ray photoelectron spectroscopy. The surface morphology of the reference, pre-, and postetch Al 2 O 3 surfaces was analyzed using atomic force microscopy and bright-ﬁ eld transmission electron microscopy. Moreover, it is found that this process chemistry is able to etch Al 2 O 3 selectively over silicon dioxide (SiO 2 ) and silicon nitride (Si 3 N 4 ).


■ INTRODUCTION
In an era of emerging nano-technologies, the fabrication of sub-10 nm, complex, and 3-D device structures demands unprecedented thickness control in the atomic regime. Specific advances in semiconductor manufacturing techniques are therefore required, 1−4 and several techniques are being actively developed, such as thermal atomic layer deposition (ALD), 5 plasma enhanced ALD, plasma-based atomic layer etching (ALE), 1,2,6 and the recently explored thermal atomic layer etching (ALEt). 7−9 Complementary to ALD, ALEt is a technique that utilizes either partial or complete self-limiting sequential gas−solid reactions to allow the removal of material from the surface with atomic-level precision. 7,10,11 Due to the self-limiting nature of at least one of the ALEt half-reactions, it can be preferred over plasma-based ALE for isotropic etching of material from non-line-of-sight features.
Several thermal ALEt processes have been developed that employ a conversion-etch reaction mechanism. 12 During the conversion step, the surface is converted into a more reactive material, which during the etch step is exposed to a coreactant and forms volatile products. In ALEt, the conversion step can form an oxide, 13 fluoride, 8−10 or another type of reactive surface that differs from the actual etch target. In the etch step, ligands from the coreactant molecules can undergo surface reactions such as ligand-exchange transmetalation, 7,8,11 fluorination, 13 or chlorination 14,15 to form species that may leave the surface.
In most of the reported ALEt reactions, hydrogen fluoride (HF) is used as a fluorinating agent in either the conversion 7−11 or the etch step. 13 However, the use of HF can be restricted due to the safety concerns associated with its handling and storage. Moreover, its corrosive nature can pose some compatibility issues with gas-feed as well as exhaust lines, pumps, and other reactor parts, especially when kept at elevated temperatures. Therefore, alternatives to HF-based ALEt chemistries would prove to be beneficial. Several other highly reactive and toxic 16 fluorinating agents such as SF 4 , XeF 2 , F 2 , CCl 2 F 2 , and CHClF 2 have been proposed. 17−20 In this paper, we explore a novel ALEt chemistry for etching amorphous Al 2 O 3 . It is proposed that NbF 5 can be used as a fluorinating agent, replacing HF. CCl 4 is used to carry out the halide-exchange reactions with the fluorinated aluminum oxide surface in order to form volatile AlF X Cl Y species. Moreover, CCl 4 is relatively more stable 21 than the trimethyl aluminum (TMA) used in the ALE literature, 8,9,11 which decomposes below 400°C. 22−24 Despite NbF 5 being solid, it has sufficient vapor pressure (about 0.1 Torr at 45°C 25 ) and forms volatile species upon fluorinating the Al 2 O 3 surface (at least at 460°C) and thus leads to almost no surface contaminations. Moreover, it is relatively safer to handle and store and is compatible with the semiconductor vacuum processing equipment.

■ EXPERIMENTAL SECTION
Substrates and Process Gases. The target films deposited on ptype 200 mm silicon wafers that were used in the etch experiments are thermally grown SiO 2 (TOx), high-quality low-pressure chemical vapor deposited Si 3 N 4 , and ALD-grown Al 2 O 3 . The SiO 2 and Si 3 N 4 films were about 20 and 30 nm thick, respectively. The Al 2 O 3 films were deposited at 300°C by the TMA and water (H 2 O) ALD process. 26 The fluorinating agent was 99.5% pure NbF 5 purchased from abcr GmbH, (Germany). The CCl 4 (anhydrous) was bought from Merck KGaA (Germany) and had a purity ≥99.5%. The NbF 5 vessel was kept at 45°C, and the CCl 4 source vessel was operated at room temperature. CCl 4 has a vapor pressure of about 90 Torr at room temperature 27 and was therefore used in a vapor-draw mode, whereas NbF 5 was used in a vapor-push mode. Nitrogen (99.999%) was used as an inert purge gas. Both the purge gas flows as well as CCl 4 dose were controlled by needle valves.
Experimental Setup. The etch experiments were conducted on a Pulsar 2000 (P2000) cross-flow reactor manufactured by ASM International N.V. designed to process 200 mm silicon wafers. The P2000 chamber was operated at isothermal conditions to a maximum T etch of 460°C. Prior to any etching, all target substrates were kept in a vacuum chamber for 5 min in order to ensure stabilization of the wafer surface temperature. The chamber pressure was between 2 and 4 Torr.
Characterization Techniques. The thickness and optical constants of the films were evaluated using SE-800 spectroscopic ellipsometer from SENTECH Instruments GmbH, (Germany). X-ray reflectometry (XRR) was performed for confirming the removed thickness values, measuring the postetched surface roughness, and film densities using Xpert PRO MRD X-ray diffractometer from Malvern Panalytical Ltd. (United Kingdom). For consistency, the exact center points of the wafers were chosen as measurement spots.
Elementary surface composition investigations were performed by X-ray photoelectron spectroscopy (XPS) and conducted on PHI Quantera SXM. Monochromatic Al Kα (1486.6 eV/15 kV) X-rays with a take-off angle of 45°were used. The analysis chamber pressure was 3 × 10 −6 Pa. In all XPS depth profile measurements, the argonion energy was set to 1 keV and the X-ray spot size was 100 μm. The pass energy of 140 eV was used. The detection limit of this particular tool is about 0.1−0.5 at. %.
Transmission electron microscopy (TEM) imaging was performed by Evans Analytical Group (California, USA) on FEI Tecnai TF-20 FEG/TEM operated at 200 kV in a bright-field imaging mode. The TEM lamellas for imaging were prepared by using the focused ion beam lift-out technique. The thicknesses of the TEM lamellas were around 100 nm. For preserving the contrast and structural integrity of the sample, a carbon coating was performed prior to the ion-milling.
Computational Techniques. The chemical mechanism of the ALEt process was investigated using first-principles density functional theory (DFT) as implemented in the Schrodinger Materials Science Suite. 28 Atomic-scale models of bulk materials, surfaces, and gas-phase molecules were optimized under periodic boundary conditions with the Perdew−Burke−Ernzerhof (PBE) functional, 29 a plane wave basis to a wavefunction cut-off of 40 Ry, PBE ultrasoft pseudopotentials, 30 and Monkhorst−Pack k-point sets 31 using the Quantum ESPRESSO code. 32,33 Total energies from DFT are used to calculate reaction energies ΔE for possible chemical transformations of these surfaces by the etchants during ALEt.
Since etching is likely to be driven by entropy (S), it is important to also determine reaction free energies (ΔG) that include the effect of temperature (T) and pressure of reactants (p react ) and products (p prod ), which for ideal gases at constant volume is given by ΔG = ΔE − TΔS − k B T ln(Πp prod μ/Πp react μ) where μ are stoichiometric coefficients. 34 For the etchants, we use p react = 0.1 Torr, approximating the experimental conditions, and assume equilibrium with product gases at p prod = sp react where the sticking coefficient s is an unknown precursor parameter that is typically in the range 10 −2 to 10 −4 for ALD precursors. 35 As explained in the Supporting Information, the entropy terms at 100°C < T < 600°C and p react or p prod were derived from molecular vibrational analysis of the gas-phase reactant and product molecules at the PBE/LAV3D** level with the Jaguar code. 36 The (2 0 1) surface of θ-Al 2 O 3 was used as the periodic slab model for the alumina surface during ALEt. θ-Al 2 O 3 was chosen because ions in this polymorph have a similar coordination environment to those in the amorphous as-deposited Al 2 O 3 . 37 The 2 × 1 expansion of a 3-layer (2 0 1) slab was selected as a representative model surface on the basis of its stability and level of corrugation when bare or fluorinated (see the Supporting Information).
As shown in Figure 1, half of the Al atoms on the bare surface are exposed (4-coordinate to O) and half are covered (6-coordinate).
Fluorinated and chlorinated surfaces were generated by progressively replacing O with F or Cl in the topmost layer of the slab, always adding two halide monoanions for each oxide dianion removed so as to preserve stoichiometry and charge neutrality.
The maximum level of chlorination was determined to be 3 Cl per 4-coordinate Al (12 Cl per 2 × 1 cell), which we designate as 100%-Cl coverage since higher coverages led to the spontaneous desorption of AlCl 3 molecules during optimization. By contrast, fluorination up to 166%-F (20 F per 2 × 1 cell) was energetically favorable, albeit with a substantial reorganization to give a surface resembling AlF 3 structural motifs. In this study, we focus on interconversions between the 166%- F, 100%-F, 100%-Cl, and 0% (i.e., bare alumina) surfaces depicted in Figure 1.
The aim of the DFT simulations is to map out the general form of the ALEt mechanism. Many reactions and byproducts can be possible, and we seek here to distinguish which surface transformations are thermodynamically favorable as a function of process conditions. This can reveal the nature of the saturating surface after each precursor exposure during the ALEt cycle and hence account for the etch rate. However, ALEt is actually a non-equilibrium process, driven by kinetics and the irreversible removal of byproducts from the surface into the gas flow. Computing the detailed sequence of surface adsorbates, intermediates, reaction pathways, and associated kinetics with DFT would be a possible future step but is beyond the current scope. It is worth remembering, therefore, that a reaction computed here to be thermodynamically feasible (with ΔG < 0 at a given T) may in fact face a high kinetic barrier and so not be observed on the experimental time scale. Nevertheless, in the absence of kinetic data, we assume that for a family of competing reactions, the kinetic barriers scale with reaction energies and therefore the most exoergic reaction is the most kinetically likely one.

■ RESULTS AND DISCUSSION
Etch-Process Characterization. Al 2 O 3 is etched by utilizing sequential exposures of NbF 5 and CCl 4 precursors at elevated temperatures (380−460°C). An inert gas purging was introduced after each exposure to remove the volatile byproducts and excess of precursor molecules from the reaction chamber. The etch per cycle (EPC) was estimated by subtracting the post-etch thickness from the initial thickness and dividing by the number of etch cycles. The Al 2 O 3 film thickness values were reliably determined by spectroscopic ellipsometry (SE) and in some cases confirmed with the help of XRR and TEM images.
As shown in Figure 2, an increase in EPC is observed with increasing NbF 5 pulse time at an etch temperature (T etch ) of 460°C. In the figure, NbF 5 pulse time is varied from 0 to 3 s, while the CCl 4 pulse time is fixed to 3 s. For each data point plotted in Figure 2, a total of 150 etch cycles were performed. In the absence of NbF 5 , no Al 2 O 3 etching was observed as confirmed by XRR, TEM, and XPS. When the NbF 5 pulse time is increased to 0.5 s, an etch rate of about 0.5 Å per cycle is noticed. Further increase in the NbF 5 pulse time to 3 s provides an etch rate of about 1.4 Å/cycle. Longer NbF 5 pulse times were not tested. The figure shows that at a T etch of 460°C , the EPC did not show strong self-limiting characteristics but may still indicate a soft saturation behavior. A nonsaturative fluorination behavior may affect the etch uniformity across the wafer especially in a cross-flow reactor. This nonself-limiting behavior can be explained by enhanced diffusion of fluoride ions at 460°C. The thickness of the converted layer may depend on factors such as surface temperature, partial pressure, as well as flux of the precursor molecules, 9,11,38 reaction time, adsorbate concentration, interaction between surface and diffusing adspecies, presence of defects, being accompanied by phase transitions, and so forth. 39 Thus, in thermal-based ALE processes there may not exist an ALE window.
For studying the temperature-dependent kinetics of the fluorination step, excess fluorination was carried out by exposing the aluminum oxide surface to NbF 5 at 400 and 460°C (Figure 3). At 400°C, excess fluorination was performed for 150 and 600 cycles of 4.5 s NbF 5 pulse times, while 150 and 800 cycles of 3 s NbF 5 pulse times were used at 460°C. After performing the excess fluorination step, the thickness of the formed AlF X [O] layer was measured by ex situ SE. About 11 and 14 nm thick fluorinated layers were obtained after 450 and 2400 s of total NbF 5 exposure time, respectively, at 460°C. The formation of thick (>10 nm) AlF X [O] layers must be due to the prolonged fluorination step and the fluoride ions diffusing deeper into the aluminum oxide layer. The diffusion of fluoride is evidently dependent on the temperature and thus may hint at the existence of no ALE window. According to Morelock et al., 40 aluminum trifluoride adopts a cubic form above 440°C, which upon cooling changes to a rhombohedral phase. The X-ray diffraction (XRD) analysis of about 11 nm AlF X [O] layer revealed the rhombohedral phase (see the Supporting Information S4) and is thus consistent with the literature. 40−44 On the contrary, at 400°C, the thickness of the fluorinated layer stayed around 3 nm even after prolonged total NbF 5 exposure of 2700 s. Chen et al. showed the fluoride ions penetrating up to 2 nm thick aluminum oxide after treating fluorinated alumina at 400°C, 45 and neutral/ion beams-assisted fluorination lead to a thicker AlF X [O] layer. 46 Moreover, up to 0.5 nm thick AlF 3 layer was obtained at a lower temperature of 300°C with HF as a fluorinating agent. 38 This temperature-dependent thickness of  the converted fluorinated layer suggests that the conversion step is diffusion-limited.
The fluorination step produced a very hydrophilic surface with water contact angles ≤15°, which is consistent with earlier reports from Roodenko et al. 19 The low water contact angle of the AlF X [O] surface is attributed to the high Lewis acidity of Al−F species, resulting in strong adsorption of H 2 O molecules at under-coordinated aluminum sites. 19 Figure 4 reveals the effect of variation in the CCl 4 pulse time on EPC at 460°C. The NbF 5 pulse and N 2 purge lengths were set to 3 and 6 s, respectively, and the CCl 4 pulse time was changed from 0.5 to 3 s. The figure shows that an EPC around 1.1 Å is observed for CCl 4 pulse times between 0.5 and 1.75 s. However, when 3 s CCl 4 pulse time is used, an etch rate of 1.4 Å/cycle is noticed. This could indicate a partial self-limiting nature of the etching step, which could be due to limited AlF X material available at the surface for CCl 4 to react with and volatilize. Moreover, Figures 3 and 4 together indicate a twostep ALEt mechanism of conversion followed by etching. Figure 5 plots EPC values at various etch temperatures for the Al 2 O 3 ALEt process by NbF 5 + CCl 4 (red curve). The figure also demonstrates that CCl 4 alone (blue curve) does not etch Al 2 O 3 within the tested temperature range of 380−460°C . The incapability of CCl 4 alone in etching Al 2 O 3 is also confirmed by XRR and TEM images as seen below. However, at 380°C, the etch rate of Al 2 O 3 by the NbF 5 + CCl 4 ALEt process is comparatively low, at about 0.08 Å/cycle, and no significant etching was observed at temperatures <380°C. From Figure 5, it can be seen that the EPC increases with the etch temperature and reaches an etch rate of 1.4 Å/cycle at 460°C . The Figures 2−4 also corroborate the proposed ALEt mechanism, whereby NbF 5 is necessary to fluorinate the Al 2 O 3 and the converted layer is etched by a subsequent pulse of CCl 4 . In most of the thermal ALE reports, the ALE window is not observed, 9,11,12,38,47−50 and the EPC is dependent on the partial pressure 9,11,38 and temperature. 47,49,51 Similarly, it is evident from our findings that there exists no ALE window. However, a few ALE processes exhibit an ALE window such as ALE of HfO 2 , 47 TiO 2 , 13 and TiN. 51 Selective Etching. In thin-film technology, many applications may benefit from thermal ALEt processes that provide selective etching of Al 2 O 3 over SiO 2 and Si 3 N 4 . Figure  6 shows that the NbF 5 + CCl 4 ALEt process is able to etch Surface Characterization. XPS studies were conducted in order to understand the etch mechanism. An Al 2 O 3 surface exposed to NbF 5 was the test subject. Figure 7 shows the XPS elemental depth profile analysis of the surface that has been fluorinated. The aluminum oxide surface was subjected to a total of 150 cycles of 3 s NbF 5 pulses with intermittent 6 s N 2 purges at 460°C. Reasons for conducting the fluorination step under such extreme conditions were to ensure a significant as well as meaningful fluorine signal, to understand the fluorination step, and to deduce a reliable reaction mechanism. From the XPS depth profiling and SE measurements, the thickness of the AlF X [O] layer formed under these conditions was estimated to be around 11 nm.
After 15 s of sputtering, a very high fluorine content of about 55 at. % (similar F content was found by Kim et al. 52      were performed with CCl 4 being the last pulse. The zoomed part of the figure depicts the fluorine content around 5 at. % and very low amounts of niobium (<0.3 at. %) and chlorine (<0.2 at. %). The presence of surface fluorine may indicate that a CCl 4 pulse time longer than 0.5 s is necessary to either fully remove the converted layer or complete the halidetransmetalation reaction or it does not go to completion. From SE and XPS depth profiling, the fluorine-containing top layer is estimated to be around 1.2 nm thick.
In Figure 9, cross-sectional bright-field transmission electron micrographs of Al 2 O 3 -coated silicon substrate are displayed. The thickness of an original Al 2 O 3 film was 56 nm as revealed by SE and confirmed by XRR. Figure 9a shows the Al 2 O 3 film that remained unetched after being exposing to 150 cycles of 3 s CCl 4 pulses. The figure also shows very smooth films devoid of significant surface defects. In Figure 9b, it is shown that after exposing the Al 2 O 3 film to 150 etch cycles of the NbF 5 + CCl 4 ALEt process, about 18 nm of the film is removed and the remaining Al 2 O 3 film appears to be rougher than the unetched film. In addition, atomic force microscopy (AFM) measurements (see the Supporting Information) revealed a slight increase in the surface root-mean-squared (or R a ) roughness. This increase in surface roughness can be associated with the fluorination step and therefore can be due to either the variation in the diffusion depths or the reorganization of the near-surface region during diffusion-based fluorination. The etched surface does not appear to be crystalline.
Computation of Mechanism. Although the gaseous reactants undergo individual reaction steps when they adsorb to the surface during ALEt, along with bulk-surface diffusion, the net effect of each cycle is the removal of bulk Al 2 O 3 and regeneration of surface functionality. The most straightforward way to assess the viability of ALEt is therefore to compute the thermodynamics of bulk etching, which is presented next. Afterward, DFT results for the thermodynamics of surface reactions are presented (more details in the Supporting Information) We first investigate whether it is thermodynamically favorable for either precursor alone to etch bulk alumina, transforming it into gas-phase products. NbF 5 can in principle remove both Al and O as follows and for this reaction, periodic DFT yields ΔE = +140 kJ/mol-NbF 5 and ΔG 300°C = −100 kJ/mol-NbF 5 . The positive value of ΔE indicates a net cost in bond energies, while the negative value of ΔG reflects the entropy gain in volatilizing the solid. Bulk alumina may be more readily etched by the Cl source alone according to the reaction which has DFT-computed energetics of ΔE = −18 kJ/mol-Al 2 O 3 and ΔG 300°C = −248 kJ/mol-Al 2 O 3 . (If the reaction would proceed to completion, CO 2 could be produced, but it seems more likely that COCl 2 would desorb first). However, no etching is observed experimentally with CCl 4 alone ( Figure  4), which suggests that kinetic barriers prevent this reaction from taking place. Consistent with this, periodic DFT simulations show that the CCl 4 molecule resists adsorption to a bare alumina surface probably because the molecule is nonpolar. We therefore exclude these single-etchant reactions from the reaction equations below and restrict our analysis to reactions that etch away either Al or O as volatile products but not both simultaneously. Such reactions are the prerequisite for a successful, self-limiting ALEt process. The conversion of bulk alumina into solid aluminum fluoride according to is computed to show that ΔE = −24 kJ/mol-NbF 5 and ΔG 300°C = −58 kJ/mol-NbF 5 (Table 1). Although labeled as "conversion" in the ALEt literature, this reaction necessarily involves the etching of oxygen, as one oxide dianion is exchanged with two fluoride monoanions. The thermodynamics of conversion of the bare alumina surface into a 100% Figure 7. XPS depth profile of Al 2 O 3 exposed to NbF 5 for a total of 450 s at 460°C. The atomic percentage of elements found in the converted film is plotted with respect to the sputter time.  fluorinated surface is found to be similar in terms of energetics to that of the bulk (R3: ΔE = −27 kJ/mol-NbF 5 and ΔG 300°C = −60 kJ/mol-NbF 5 ). Further fluorination of the surface is slightly less favorable (e.g., R4: ΔE = +0.8 kJ/mol-NbF 5 and ΔG 300°C = −33 kJ/mol-NbF 5 for 100% F to 166% F) but still exoergic, indicating that surface conversion does not self-limit. As shown in Figure 10, there is little variation with temperature in these free energies of fluorination of the bulk and surface by NbF 5 . We conclude that continuous conversion of alumina into aluminum fluoride is moderately exoergic under the experimental conditions, but again, the extent to which such conversion actually takes place depends on the kinetics of individual steps that bring NbF 5 and alumina into contact. We suggest that this is likely to be dictated by the kinetics of diffusion across the Al 2 O 3 −AlF 3 interface. The NbF 5 pulse therefore produces a surface layer and possibly also subsurface layers that are fluorinated and is experimentally verified in Figures 3 and 8.
We now examine the potential reactions of such fluorinated surfaces (considering the sample surfaces 100% F and 166% F) with gas-phase CCl 4 in the next phase of the ALEt cycle. The computations show that the most energetically favorable products are CFCl 3 and AlCl 3 , and the resulting thermodynamics are given in Table 1 and Figure 11. CFCl 3 is the likely product of halide-exchange between the CCl 4 reagent and the fluorinated surface. The simplest of such reactions produces a chlorinated surface with the same coverage which is computed to be endoergic across a wide temperature range (ΔG 300°C = +41 kJ/mol-CCl 4 ). The reaction becomes more favorable if Al is also etched away as AlCl 3  and ΔG 300°C = −58 kJ/mol-CCl 4 for complete etching to a bare surface: Similar results for the 166% F surface are displayed in Figure  11. As expected, these etch reactions are driven by entropy and thus become more favored at high temperatures. We can not confidently state the exact temperature above which a given reaction becomes exoergic since this is affected by the pressure correction and thus by the arbitrary choice of the sticking coefficient s. Nevertheless, Figure 11 allows us to conclude that it is thermodynamically favored for CCl 4 to entirely etch away the aluminum fluoride layer and reveal bare alumina (0% surface).
Since CCl 4 may in principle etch bulk alumina (reaction R7), we need to consider whether the bare alumina surface is reactive with this precursor. Converting the bare surface to a 100%-chlorinated surface is found to be exoergic, with ΔE = −40 kJ/mol-CCl 4 and ΔG 300°C = −51 kJ/mol-CCl 4 (reaction R9) following a similar trend to the conversion of bulk alumina to bulk aluminum chloride (reaction R8). However, as noted above, the experimental result of no etching by CCl 4 alone indicates that this reaction pathway is not accessible. A bare alumina surface is therefore the thermodynamically favored outcome of the CCl 4 pulse.
In case fluorinated or chlorinated portions of the surface persist into the NbF 5 pulse of the next ALEt cycle, we include exemplary reactions of 100% F and 100% Cl surfaces with NbF 5 in Figure 10 (reactions R4−R6). In this case, halideexchange to NbClF 4 is the most favorable reaction, but the addition of F and removal of O is also exoergic across the entire temperature range. Therefore, these reactions also lead to a fluorinated surface as the outcome of the NbF 5 pulse.
Having identified the saturated surfaces at the end of each precursor pulse, we are now able to combine half-reactions (R3 + R13 or R4 + R14) into the overall ALEt reaction The overall energies per ALEt cycle computed with DFT are ΔE = +421 kJ/mol-Al 2 O 3 and ΔG 300°C = −329 kJ/mol-Al 2 O 3 .  a Bulk solid is labeled (s), gas-phase molecules (g), and coverage is indicated as a percentage for surface slabs.
Chemistry of Materials pubs.acs.org/cm Article Consistent with the data in Figure 2, the EPC in this mechanism depends on the level of conversion that is achieved in the NbF 5 pulse. 100% F means the etching of 2 (Al 2 O 3 )/ cycle in the 2 × 1 slab. Dividing by the cross-sectional area of the slab (84.5 Å 2 ) and multiplying by the molar volume of θalumina (48.5 Å 3 /Al 2 O 3 ) yields a theoretical EPC of 1.2 Å/ cycle for this F-coverage. Converting a greater proportion of the surface zone to AlF 3 means a higher EPCfor example, 166% F means etching 3.3 (Al 2 O 3 )/cycle per slab and an EPC of 1.9 Å/cycle.
Proposed Reaction Mechanism. Many reported thermal ALEt reactions undergo a two-step, conversion-etch mechanism, 7−11,13 with the conversion (or fluorination) step followed by an etch step. The experimental findings in this report also suggest a two-step etch mechanism.
The first step of the proposed reaction mechanism shown in Figure 12 is the conversion step. Figure 12a depicts the initial Al 2 O 3 film on the silicon substrate, and Figure 12b shows the fluorinated Al 2 O 3 layer after exposing to the NbF 5 . As evident from Figures 3 and 7, the NbF 5 is believed to act like a fluorinating agent similar to HF 7 or WF 6 13 and converts the Al 2 O 3 surface into AlF X [O]. The experimental findings suggest that the fluorination is not self-limiting, and therefore, with excessive fluorination, a thicker AlF X [O] layer is formed, as depicted in Figure 12c.
From the DFT calculations for gas-bulk (R1 and R2) as well as gas−surface reactions R3−R5, it is proposed that niobium most likely forms the volatile NbOF 3 species. The existence of gaseous NbOF 3 has been previously reported, and its volatility has been studied. 53−56 The reaction NbOF 3(s) → NbOF 3(g) is favorable (ΔG < 0) above 140°C. 25 Moreover, thermogravimetric analysis of NbOF 3 showed mass loss above 80°C. 25 During this conversion step, the oxygen from the Al 2 O 3 layer is also removed at least partially if not completely and therefore the fluorinated layer is labeled as AlF X [O]. Moreover, it is also possible that the fluorinated layer is AlF 3 which partially oxidizes in the air to form AlF X [O]. Figure 12d depicts an outcome of the excess fluorination step and is supported by Figures 3 and 7.
In the second step, the AlF X [O] layer can be etched as shown in Figure 12 (d−f). The etch step involves halideexchange transmetalation to form volatile byproducts such as AlF a Cl 3−a and CCl 1+a F 3−a , where 0 ≤ a ≤ 2. The DFT computations reveal AlCl 3 and CFCl 3 as being the most favorable etch products from the various chlorination halfreactions R10−R14. In this fashion, an etching of Al 2 O 3 can be achieved by alternating exposures of NbF 5 and CCl 4 , which   may follow the described conversion-etch mechanism by the overall reaction R15.
Discussion of Etch Rate. According to the overall etch reaction R15, the theoretical EPC of 1.2 Å/cycle for 100% Fslab coverage is evaluated. For prolonged fluorination as per reactions R4 and R5, where 166% F-slab is used, an EPC of 1.9 Å/cycle is obtained. An EPC of 1.4 Å/cycle is recorded for 3 s NbF 5 and CCl 4 pulse lengths (as shown in Figure 2), which is therefore consistent with a coverage of 120% F during etching by R15. Moreover, both the experiment and the DFT calculations indicate that the EPC is not self-limiting for NbF 5 pulse time variation and strongly depends on the fluorination step.
Another parameter that largely impacts the EPC is the etch temperature. Higher EPC values are observed at higher temperatures (≥380°C). At temperatures below 380°C, no etching was observed. However, DFT thermodynamic calculations did not show any evidence for this behavior. Therefore, this temperature dependence probably reflects the kinetics of diffusion and structural reorganization at the Al 2 O 3 −AlF 3 interface. Based on our examination of the DFT-computed structures, it may be that atoms near the interface have to reorganize so that aluminum takes a low coordination number (i.e., 4 rather than 6) as the prerequisite for the fluoride to form and hence for the etching to take place.

■ CONCLUSIONS
ALEt of amorphous Al 2 O 3 is demonstrated using cyclic exposures of NbF 5 and CCl 4 etchants. The Al 2 O 3 ALEt is found to follow a two-step etch mechanism. In the first step, NbF 5 converts part of the Al 2 O 3 layer into AlF X [O], or more evidently AlF 3 , which is described as a conversion or fluorination step. The most likely volatile byproduct is found to be NbOF 3 . In the second step, the chlorine atoms from CCl 4 undergo halide-exchange with the converted AlF X [O] surface. As a result of such an interaction, various volatile byproducts such as AlF 2 Cl, AlFCl 2 , AlCl 3 , and several chlorofluorocarbon molecules can be produced. Of these, AlCl 3 and CFCl 3 are found by DFT to be the most favorable, which leads to the overall etching reaction 2AlCl 3NbOF 6CFCl 2 3(s) 5(g) 4(g)

3(g) 3(g) 3(g)
Similar reactions can be written for the other possible byproducts.
In this manner, the converted AlF X [O] layer is volatilized from the surface by CCl 4 and hence the desired etching takes place. The feasibility of the proposed reaction mechanism was confirmed with thermodynamic calculations based on DFT. In addition, XPS and XRD analyses confirmed the formation as well as the removal of the fluorinated AlF X [O] layer.
The dependence of the EPC on NbF 5 pulse time shows soft self-limiting characteristics, which can be due to slow kinetics, diffusion-limited gas−solid reaction, high residence time of NbF 5 , low concentration or partial pressure of NbF 5 , or slow sublimation of volatile NbOF 3 surface species. The CCl 4 reaction is found to be more self-limiting than that of NbF 5 .
Aluminum oxide ALEt is observed between 380 and 460°C. From its onset (removing just 0.08 Å/cycle at 380°C), the etch rate tends to increase with etch temperature. We tentatively link this temperature-dependence to the kinetics of diffusion across the oxide−fluoride interface. The linear removal of Al 2 O 3 with cycles is observed with an etch rate of about 1.1 Å/cycle at 460°C. An etch rate of about 1.4 Å/cycle is measured when the exposure time for each precursor is set to 3 s, separated by 6 s N 2 purges. This etch rate is consistent with fluorination that extends partly into the subsurface layer. As revealed by TEM analysis, the surface of the post-etch Al 2 O 3 film was rougher than the unetched film, again pointing to diffusion at the interface.
It is proposed that the thermal NbF 5 + CCl 4 ALEt process reported here could etch other metal oxides in a similar twostep etch mechanism with selectivity against SiO 2 and Si 3 N 4 .