Vacuum ultraviolet enhanced atomic layer etching of ruthenium films

Platinum group metals, such as Ru, are ubiquitous next generation applications not limited to catalysis^1,2 and nanoelectronics.^3,4 Ru is a potential interconnect material for integrated circuits as well.⁵ Ru metal (and RuO₂) can also be found in dynamic random access memory devices^6,7 and is employed for metal contacts and barrier liners.^8–12 However, as device dimensions continue to shrink and architectures become increasingly complex, atomic level fabrication allowing selective, low-temperature removal of metals is central to enabling next generation devices.

Ruthenium etching has been realized by oxidizing Ru and using a second chemical to remove the oxide. Thermal oxidation of Ru from O₂ is not observed until the temperature is 400–700 °C,¹³ which exceeds the thermal budget of many devices. Lower temperature oxidation methods typically involve an O₂ plasma,^14,15 or a neutral O beam,¹⁶ which oxidizes Ru to RuO₂. Oxidized Ru is removed chemically with the aid of a second reactive species, such as CH₃OH,¹⁴ CF₃CFH₂,¹⁵ or C₂H₅OH.¹⁶ Continuous etching of Ru without a plasma has also been reported, where Ru etching is accomplished using O₃, which is known to form volatile RuO₄ at temperatures above 100–150 °C.¹⁷ 

A self-limiting oxidation step is a likely prerequisite for Ru atomic layer etching (ALE). An electrochemical liquid ALE process has been reported for Ru, which proceeds by first forming Ru(OH)₂ on the surface by applying a positive voltage and completing the etch with Cl⁻ in the liquid phase.¹⁸ We have recently demonstrated a method of ALE, vacuum ultraviolet (VUV) enhanced ALE of Pd where oxidizing species are generated by co-exposing the substrate to VUV and O₂ at temperatures of 50–200 °C.^19,20 VUV-enhanced ALE proceeds by first oxidizing near-surface Pd layers through co-exposure of VUV/O₂ to form PdO_x. Following the oxidation half-cycle, samples are exposed to formic acid, HCOOH, vapor that completes an etch cycle. The extent of Pd oxidation is limited to the near-surface region through the VUV exposure time and sample temperature. Atomic O, formed in the gas phase, is the main species responsible for the oxidation of Pd under the ALE conditions. Herein, we report self-limiting oxidation half-cycles using a co-exposure of VUV/O₂ and etching half-cycles using HCOOH vapor results in the ALE of Ru.

Formic acid is known to etch transition metals (e.g., Co, Fe, Ni, Pt, and Pd), where the metal removed is limited to the amount of metal that is oxidized in an O₂ plasma as HCOOH does not etch the metal.^21,22 The exact etching mechanism of transition metal oxides with HCOOH is unknown. One report has suggested a bidentate Ni(COOH)₂ molecule etch product for etching of NiO with HCOOH.²² 

Density functional theory (DFT) complements experimental work and can aid in addressing issues of which oxidant(s) (atomic O, O₂, and O₃) may be responsible for oxidation, and nudged elastic band (NEB) calculations can aid in addressing how the oxide front propagates through a solid film. DFT has been combined with experiments to assess adsorption and bonding of O and O₂ onto Ru(1010),^23–26 Ru(0001),^27–29 and Ru(100),³⁰ where surface configuration, surface diffusion, and competitive adsorption all have been described. Experimental and DFT reports of O on Ru surfaces have found two equilibrium surface O coverages of 0.25 monolayer (ML) O–0.50 ML O.^23,31 The surface O loading increases to 1 ML O and 2 ML O after the temperature raised above 500 K.^31–33 

This work aims to explore VUV-enhanced ALE of Ru. We show that VUV/O₂ can oxidize Ru metal, that the oxidation is self-limiting under the conditions employed herein, and that HCOOH is capable of etching the oxidized Ru species that are formed. We find the amount removed per cycle decreases as the surface roughness decreases. This behavior is likely associated with the oxidation rate. DFT and NEB calculations are presented to provide insight into the self-limiting oxidation of Ru.

Ru metal films are sputtered with a >99.9% pure Ru target (Plasmaterials Inc.) at 30 mA emission current, and 4 mTorr Ar (99.999%, Matheson), yielding a deposition rate of 4 nm/min on large blanket wafers. The sputter system (EMS-Quorum, EMS300TD) is evacuated using a dual stage rotary vane mechanical pump. Substrates are single side polished p-Si(100) wafers (1–10 Ω cm, University Wafer Inc. and Nova Electronics Inc.), where the native oxide is present before Ru is sputtered. We have observed that other platinum group metals, such as Pd, delaminate from the Si substrate if an adhesion layer is not present. Because of this, 3 nm Ti (>99.995%, Plasmaterials Inc.) is sputtered onto the Si substrates before Ru deposition. A quartz crystal microbalance (Inficon) is used to evaluate sputtered thickness in situ. Film thickness is measured ex situ using x-ray reflectivity (XRR). All thicknesses reported are measured using XRR. XRR indicates that the Ru films are ∼8 nm thick, while the Ti adhesion layer cannot be detected.

All ALE experiments take place in a custom ultrahigh vacuum (UHV) stainless steel experimental and analysis facility described elsewhere.^19,20 The UHV facility is comprised of an ALE chamber, x-ray photoelectron (XP) spectrometer, load-lock chamber, and vacuum furnace, all connected to one another via a UHV transfer line (1 × 10⁻⁷ Torr). This allows etching and analysis to be completed without exposing samples to atmosphere.

ALE is performed in a custom stainless steel reactor as well, which is essentially a six-way cross, with a D₂ lamp (H2D2 Hamamatsu, L11798) mounted at the top of the reactor. The D₂ lamp is 110 W, with a broad emission range from 110 < λ < 400 nm, and is equipped with an MgF₂ window. This gives the strongest lines at 115 and 160 nm (6.5 < hν < 11.3 eV). Exposures are performed by moving the sample along the z axis of the reactor using a linear motion device with a stroke length that can span the entire length of the reactor. This allows the sample to be moved within 0.50 cm of the MgF₂ window during oxidation half-cycles, and greater than 15 cm away from the MgF₂ window during HCOOH exposure half-cycles, which is far enough away to fully attenuate VUV photons. The background pressure of the ALE reactor is maintained by introducing Ar (99.9999%, Matheson) at a total flow rate of 80 SCCM to yield a background pressure of 0.10 Torr. The sample is heated from below using a 24 V halogen light bulb (Osram), while the walls of the ALE reactor are maintained at 80 °C to minimize adsorption and pumping times.

Oxidation half-cycles are performed by first raising the sample to within 0.50 cm of the VUV light source and then introducing O₂ (99.9999%, Matheson) into the chamber using a mass flow controller at 150 SCCM. This yields an oxygen partial pressure of 1 Torr. Following the oxidation exposure, the gas is cycled off, the sample is lowered 15 cm for the etching half-cycle, and the chamber is purged with Ar to the starting background pressure. The etching half-cycle consists of exposure to HCOOH vapor (Sigma-Aldrich, >99%) isothermally. HCOOH is introduced into the ALE chamber by opening a pneumatic valve and dosing the head space of a saturator containing liquid HCOOH at room temperature. The vapor is metered using a needle valve. This yields a HCOOH partial pressure of 0.50 Torr. Following HCOOH exposure, the chamber is purged with Ar again. ALE cycles are indicated throughout the text with the following shorthand: VUV/O₂ co-exposure time/Ar purge/HCOOH dose time/Ar purge, where all units are in seconds (i.e., 20/10/50/30 corresponds to a 20 s VUV/O₂ exposure, 10 s Ar purge, 50 s HCOOH exposure, and 30 s Ar purge). Unless noted in the text, a new substrate is used for each experiment.

In situ x-ray photoelectron spectroscopy (XPS) is performed by transferring samples into the XP spectrometer (PHI 5600) without exposing them to the atmosphere. XPS is collected using an Mg Kα anode at 15 kV and 250 W. This yields a sample current of 40–50 nA. Analysis of XP spectra is performed using casaxps (v2.3.16).

Ex situ analysis of the film thickness and crystallinity is performed using XRR and x-ray diffraction (XRD). XRR and XRD are collected using a Rigaku Ultima IV Diffractometer with Cu Kα radiation. XRR patterns are fit using genx (2.4.10)³⁴ to determine film thickness. Film morphology is analyzed using atomic force microscopy (AFM, Asylum Research 3DMFP), collected with Si cantilevers (HQ:NSC15/Al BS, μmasch) in the tapping mode. Analysis of atomic force (AF) micrographs is completed using gwyddion (2.47).

DFT is performed using the plane wave basis code quantumespresso (v6.4.1)^35,36 using the Perdew–Burke–Ernzerhof functional for approximating the electronic exchange and correlation contributions.^37,38 Ultrasoft pseudopotentials describe the valency of atoms, where the wave function and charge density cutoff are 60 and 600 Ry, respectively. The valency of Ru and O is 4s²5s²4p⁶4d⁶ and 2s²2p⁶, respectively. van der Waals forces are included with the Grimme-D3 correction,³⁹ and Becke–Johnson damping,⁴⁰ which are needed for the accurate prediction of O₃ structures. Marzari–Vanderbilt⁴¹ “cold-smearing” is also employed, with a smearing width of 0.002 Ry. Spin-polarized calculations are employed for free molecular oxidants.

Thick (23 nm) Ru films are deposited to identify crystalline facets that may be present in the 8 nm films used in the etching studies and to inform the surfaces to model with DFT and NEB calculations. XRD patterns for 23 and 8.7 nm Ru films are shown in Fig. S1 in the supplementary material.⁵⁹ Diffraction patterns are weak for 8.7 nm Ru. Three features are present in the diffraction pattern of 23 nm Ru at 2θ of 38.54°, 42.23°, and 44.01°, which are consistent with (100), (002), and (101) reflections, respectively. These peak locations correspond to interplanar distances of 2.334, 2.135, and 2.056 ± 0.005 Å for the (100), (002), and (101) facets, respectively. While three facets are detected, the intensity of reflections at (002) and (101) are the most prominent and, therefore, inform the simulations herein. The choice of (002) and (101) facets is also to align with other Ru DFT reports.⁴² 

The Ru unit cell is known to be hcp with lattice constant a = b = 2.705 Å and c/a = 1.582.⁴³ This structure is used to define the Ru slabs in this study, where stresses within the unit cell are minimized <2 meV/Å before slab geometries are defined. Adsorption simulations are done using (2 × 2) Ru(002) and Ru(101) surfaces, which correspond to surface coverages of 0–0.50 ML O. The surface coverage 0.25–0.50 ML O is the saturation coverage observed under most conditions and is used in this study as well. The irreducible volume of the Brillouin zone is discretized using a 5 × 5 × 1 Monkhorst–Pack grid. Periodic images normal to the slab are separated by 20 Å of vacuum to minimize periodic image interactions.

Four atomic Ru layers are used to represent the bare Ru slab, where the top two layers are allowed to relax freely and the bottom two layers are held at the bulk Ru position. Six atomic Ru layers are used to explore simulations with increasing O incorporation so at least two Ru layers are allowed to relax freely in between O-modified Ru layers and Ru layers fixed at bulk positions. O incorporation into Ru is performed by relaxing an Ru surface with 0.50 ML O coverage. The positions of O atoms corresponding to 0.50 ML O on the surface are then cumulatively propagated beneath the surface Ru layer (n = 1), between the first and second Ru layer (n = 1 and 2), and between the second and third Ru layer (n = 2 and 3). Incorporation of O on the surface layer only; on the surface and in between the first and second atomic layers; and on the surface, in between the first and second atomic layers, and in between the second and third atomic layers are denoted 1, 2, and 3 O layer structures, respectively. Convergence of atomic planes, k-points, charge density, and wave function cutoff are verified by changing from 4 to 6 atomic planes, 5 × 5 × 1 points to 8 × 8 × 1, 450–600 Ry and 45–60 Ry, respectively, which results in a change of <2 meV/atom. All simulations are optimized until the error in energy and the force are less than 1 × 10⁻⁸ Ry and 1 × 10⁻⁶ Ry/Å, respectively.

The NEB method⁴⁴ is used to approximate the minimum energy pathway (MEP) of O diffusion through 1 and 2, O layered Ru(002) structures, where 14 images are used to discretize the diffusion pathway. Structures are first optimized, where a “probe” O atom is placed at the nominal starting and ending positions. Optimized structures for the starting and ending geometries are then used as the input for the NEB calculation. For the NEB calculations, the wave function and charge density cutoff values are 45 and 450 Ry, and structures are optimized until the error in energy and the force are less than 1 × 10⁻⁶ Ry and 1 × 10⁻⁵ Ry/A, respectively. Ru structure visualization and analysis is performed using xcrysden (Version 1.6.2).⁴⁵ 

Following the procedure used in Ref. 20, adsorption energy is calculated according to Eq. (1),

where E_surf is the energy of the surface; E_Ox is the energy of the oxidant O, O₂, or O₃; and E_Ox/Surf is the energy of the Ru surface with the oxidant adsorbed. More details on the calculation of E_Ox is found in Ref. 20. Briefly, E_Ox, references all oxidants to the calculated energy of O₂, E_O2. The energy of atomic O, E_O, is calculated by destabilizing the energy of E_O2 by ΔE_O–O, the bond energy of an O₂ molecule (498 kJ/mol). The energy of O₃, E_O3, is calculated by adding in the stabilizing energy for the formation of O₃ per the reaction O₃→ O₂ + O, which has a change in internal energy of 102.4 kJ/mol.⁴⁶ In this convention, a positive adsorption energy indicates exoergic adsorption, while a negative value is endoergic.

Oxidation of 8 nm Ru thin films is explored by co-exposing substrates to VUV/O₂ at 1 Torr O₂ and 100 °C for 5, 10, and 15 min. The Ru 3d XP feature is shown in Fig 1(a). There are two clear features in the Ru 3d XP spectra at 280.7 and 284.8 eV, which are attributed to Ru 3d_5/2 and Ru 3d_3/2, respectively.⁴⁷ The 4.17 eV shift in binding energy between the 5/2 and 3/2 features as well as the asymmetry of the XP features is characteristic of Ru metal. Evidence for Ru oxidation can be found in the broadening of the 3d_5/2 and 3d_3/2 XP features. This is evident in the tail toward higher binding energies, which is identified as RuO₂.^2,48,49 However, despite this increase in the peak area to higher binding energies, Ru metal character remains after 5, 10, and 15 min of VUV/O₂ co-exposure. It is also worthwhile to note that the C 1s XP feature overlaps directly with the 3d_3/2 XP feature, which makes the oxidation state identification from the Ru 3d feature difficult. The Ru 3p XP feature is known to exhibit a shift toward higher binding energy as oxidation proceeds.³¹ There is no detectable shift for Ru 3p XP features in Fig. 1(b), which indicates there is limited bulk oxidation occurring during the VUV/O₂ co-exposure.

The O 1s XP spectrum similarly exhibits two features at 531.5 and 530.1 eV for the as-deposited film [Fig. 1(c)] that are indicative of an adsorbed molecular oxygen species^32,50 and RuO₂,^31,32,50,51 which we attribute to a surface oxide. The feature at 531.5 eV is nearly completely removed after Ar⁺ sputtering of the as-deposited film for 25 s confirming it arises from some form of adsorbed oxygen, while the feature at 530.1 eV remains (Fig. S2).⁵⁹ Adsorbed molecular oxygen, evidenced by the feature at 531.5 eV, is present following all VUV/O₂ exposures [Fig. 1(c)], as gas phase O₂ is present above the surface throughout each exposure. After 5 min of VUV/O₂ co-exposure, the RuO₂ feature at a binding energy of 530.1 eV increases. XP features of RuO₂ persist after 5 min of VUV/O₂ co-exposure; however, the peak area does not appear to increase with the exposure time. The fact that no further increase in RuO₂ can be detected after 10 and 15 min of VUV/O₂ co-exposure suggests self-limiting oxidation.

The effect of higher temperature on the amount of RuO₂ formed from VUV/O₂ co-exposure is explored with XPS and is shown in Fig. 2. We note that in separate experiments (not shown), an increase in the RuO₂ feature at 530.1 eV cannot be detected when Ru was co-exposed at 50 °C. 8 nm Ru films are co-exposed to VUV and O₂ for 5, 10, and 15 min at 150 °C. The Ru 3d, Ru 3p, and O 1s spectral regions are shown in (a), (b), and (c), respectively. Two features are clear in the Ru 3d XP spectrum at 280.7 and 284.8 eV. Relative to the XP spectra of the as-deposited Ru, there is an increase in the peak area toward higher binding energies indicative of RuO₂; however, there is negligible difference between the 100 and 150 °C exposures. Similar to oxidation at 100 °C, there is minimal bulk RuO₂ formed as indicated by the asymmetric metallic character in the 3d_5/2 and 3d_3/2 XP features, as well as the Ru 3p XP region in Fig. 2(b).

Deconvolution of the O 1s feature indicates an adsorbed oxygen species is present in all samples investigated. The area of the RuO₂ oxygen XP feature at 530.1 eV is approximately constant between 100 and 150 °C. RuO₂ formation saturates after 15 min at 100 °C but does not saturate after 15 min at 150 °C. We note that it appears that less RuO₂ forms at 150 °C. The ratio of the areas of the RuO₂ features at 100 and 150 °C are 1.08, 1.05, and 1.14 at 5, 10, and 15 min, respectively. This is within the 5%–10% error that we have similarly observed on Pd.^19,20

Atomic O, O₂, and O₃ are present in the gas phase when O₂ is subjected to VUV photons.^52,53 We have demonstrated that atomic O is responsible for oxidation when Pd metal is co-exposed to VUV/O₂ using two different sample configurations.²⁰ This point is similarly explored on 8 nm Ru with two different exposure configurations. In the first configuration, Ru is illuminated by the VUV lamp during O₂ exposure, and the second configuration Ru is shadowed from the VUV lamp by mounting the Ru sample face-down. In the first (illuminated) configuration, it is likely that atomic O, O₂, and O₃ oxidant species are incident on the Ru surface, while O₂ and O₃ are likely incident on the Ru surface in the second (shadowed) configuration. The ratio of the lifetime of atomic O to O₃ is approximately 2 × 10⁻⁶,^54,55 which suggests that atomic O is consumed in gas-gas and gas-surface interactions in the shadowed configuration before diffusing to and contacting the Ru surface, while atomic O is likely not consumed in the illuminated configuration.

Ru is co-exposed to VUV/O₂ in illuminated and shadowed configurations at 1 Torr O₂ for 5 min at 100 and 150 °C. XPS results are shown in Fig. 3. The Ru 3d XP spectra in Fig. 3(a) indicate that more RuO₂ is formed in illuminated configurations than shadowed configurations at 100 and 150 °C, which is revealed by the tail toward higher binding energy and reduction in feature intensity at 280.7 and 284.8 eV. Negligible difference is seen in the Ru 3p features at 100 or 150 °C, as seen in Fig. 3(b). RuO₂ formation is evidenced by a larger O feature in the O 1s XP spectra, which is shown in Fig. 3(c). RuO₂ features (i.e., the peak at 530.1 eV in O 1s) are present for the illuminated and shadowed configurations, where more RuO₂ forms when the sample is illuminated. The results in Fig. 3(c) indicate that atomic O dominates the oxidation of Ru for conditions herein. This result is similar to oxidation of Pd where atomic O also dominates the oxidation reaction.²⁰ 

ALE of Ru is explored by exposing 8 nm Ru thin films to an oxidation half-cycle consisting of 2 min VUV/O₂ co-exposure, followed by a 1.5 min Ar purge; and an etching half-cycle consisting of a 30 s HCOOH exposure at 0.50 Torr, followed by a 2.5 min Ar purge. One oxidation half-cycle followed by one etching half-cycle constitutes one ALE cycle, and cycles are written in the text with the time in seconds (i.e., 120/90/30/150 for the cycle just described). XPS measurements show Ru metal and no RuO₂ following the HCOOH half-cycles (not shown). XRR of the films is shown before and after ALE cycles at 100 °C in Fig. 4(a). XRR measurements of the Ru film before and after etching indicate a thickness change of 6, 13, and 14 ± 2 Å for 5, 10, and 15 ALE cycles, respectively. These results are presented in Fig. 5. The slope of a line of best fit to the data at 100 °C is 0.85 ± 0.15 Å/cycle.

The XRR patterns in Fig. 4(a) indicate all Ru surfaces are rougher as deposited compared to the surfaces after ALE is performed, which is evidenced by the lack of reflections in the range 3° < 2θ < 6° before etching and clear reflections emerging after ALE is performed. This observation is consistent across all XRR patterns in Fig. 4(a). 20 × 20 μm² AF micrographs of an as-deposited Ru film and Ru after 5 and 15 ALE cycles are also shown beneath the XRR curves. The roughness, R_a, is 2.4, 0.4, and 1.0 Å, for samples as-deposited, after 5 ALE cycles, and 15 ALE cycles, respectively. We note that R_a of the as-deposited surfaces varies by 2 Å; however, the clarity in reflections in XRR curves improves for all samples. This implies that as ALE is performed, the surface roughness is reduced. This could possibly affect the oxidation rate during the oxidation half-cycle.

ALE is also performed at 150 °C with the same oxidation and etching half-cycles as described for 100 °C and the XRR and AF micrographs are shown in Fig. 4(b). The change in thickness is 6, 12 and 16 ± 2 Å for 5, 10, and 15 ALE cycles, respectively. This results in an overall etch removal amount at 150 °C of 0.96 ± 0.15 Å/cycle (Fig. 5). Again, the surface roughness is reduced, as evidenced by the quality of reflections in the range 3° < 2θ < 6°, which improve after ALE is performed. AF micrographs reveal R_a of Ru as-deposited, after 5 ALE cycles, and after 15 ALE cycles is 2.4, 1.6, and 1.2 Å, respectively, which also suggests a reduction in roughness with etching.

The thickness change between 10 cycles and 15 cycles at 100 °C is modest at 1 Å in comparison to 6 Å removed in the first 5 ALE cycles. Figure 4(a) also presents a set of XRR curves labeled 20 cycles. The 20 cycles indicated are cumulative, where the starting substrate is the sample treated with 10 cycles exposed to 10 additional cycles (i.e., 10 cycles + 10 additional cycles of 120/90/30/150). The cumulative amount of material removed is 14 Å, and the incremental amount after the additional 10 ALE cycles is 1 Å. This cumulative thickness change for this sample is plotted using a diamond in Fig. 5. The amount of Ru etched in the HCOOH half-cycle depends on the amount of RuO₂ formed in the VUV/O₂ half-cycle. As previously noted, the roughness decreases with increasing numbers of ALE cycles, and we associate the reduced removal amount to a decreased rate of oxidation.

The VUV/O₂ exposure time was increased to test the hypothesis that the oxidation rate slows as the roughness is reduced. The film presented in Fig. 4(a) that was effectively not etched after 20 cumulative ALE cycles (120/90/30/150) was exposed to 5 additional cycles (i.e., 25 cumulative cycles) using a longer oxidation half-cycle. The 5 additional cycles used 5 min VUV/O₂ co-exposures (300/90/30/150). XRR of the film after 5 additional cycles (25 cumulative ALE cycles) is shown in Fig. 6. XRR of the samples with 10 and 20 cumulative cycles are reproduced for comparison. The cumulative amounts of Ru removed after 10, 20, and 25 cycles are 13, 14, and 19 ± 2 Å, respectively. Approximately 5 Å is removed after 5 additional 300/90/30/150 cycles, indicating the etch rate, and the extent of oxidation is restored, consistent with the hypothesis that oxidation rate decreases as the film is subjected to ALE cycles.

We note that etching arising from O₃ and O₂ must be considered in co-exposures of VUV/O₂. Ozone has been shown to form RuO₄ at 100–150 °C, and this forms the basis of an Ru etching process via RuO₄ sublimation.¹⁷ The possibility that RuO₄ could have formed and contributed to the etching reported herein is explored by co-exposing Ru to VUV/O₂ at 1 Torr O₂, and 100 °C, for 30 min. Thirty minutes correspond to the equivalent VUV/O₂ exposure associated with 15 ALE cycles with 2 min VUV/O₂ oxidation half-cycles. XRR before and after exposure (not shown) indicate no change in Ru film thickness. RuO₄ formation could similarly occur with O₂, as others have reported.³² We hypothesize that, as our DFT calculations indicate, O₂ and O₃ can form RuO₄; however, etching via RuO₄ formation is unlikely for the conditions herein. The only detectable change, however, is a roughening of the surface, that is indicated with lower quality reflections in the range 3° < 2θ < 6°. Roughening of the surface with co-exposure of VUV/O₂ is consistent with previous observations on Pd.¹⁹ Additionally, ALE cycles without VUV photons (i.e., a thermal O₂ exposure at 100 °C, with 120/90/30/150 ALE cycles) result in no change in thickness (XRR not shown). This indicates ALE of Ru requires VUV/O₂ to proceed.

The experiments presented above illustrate that atomic O and O₃ that are incident on the Ru and oxidized Ru surfaces contribute to the formation of RuO₂ in the near-surface region under VUV/O₂ exposure. There are numerous first principles studies of Ru, RuO₂, and oxygen interacting with Ru and RuO₂.^23–30 We employ DFT as well to investigate VUV-enhanced oxidation. The detailed DFT results are presented in the supplementary material⁵⁹ and are in agreement with the structural models and the energetics of atomic and molecular oxygen on Ru(101) and Ru(002). Table S1 presents the structural results.⁵⁹ Adsorption geometries are presented in Fig. S3 for atomic O and O₃ on (2 × 2) Ru(002) and Ru(101).⁵⁹ Table S2 presents adsorption energies predicted by DFT for O, O₂, and O₃ on (2 × 2) Ru(002) and Ru(101).⁵⁹ The adsorption energy, E_ads, for atomic O adsorbing onto bare (101) and (002) surfaces is 5.63 and is 6.07 eV, respectively, which are in agreement with previous reports. Other DFT reports estimate that the adsorption of atomic O on Ru has an adsorption energy of 5.37–5.32 and 5.15–5.26 eV on Ru(101) with 0.50 and 1.0 ML O coverage,^23,56 and 5.64–6.28, 5.52–5.43, and 4.78–5.07 eV on Ru(002) with 0.25, 0.50, and 1.0 ML O coverage, respectively.^23,48,56,57

Adsorption energetics and structures are also calculated for Ru slabs with O incorporated in between Ru atomic layers. Top and side views of the optimized O layered Ru slabs are shown in Fig. 7. Atomic O incorporation into Ru is performed by propagating an equivalent surface of 0.50 ML surface coverage to subsurface layers. This is done by optimizing O atom positions on the surface, noted to be one hcp site with an occupied fcc site adjacent for (002), and a “zig-zag” series of surface rows for (101) [denoted 1 O layer in Figs. 7(a) and 7(b) for (002) and (101), respectively], which is consistent with other reports.^24,26,31 Atomic O is incorporated into the substrate Ru by keeping the same surface coverage, and propagating O atom positions in between the first and second Ru atomic layers (denoted 2 O layers in Figs. 7(c) and 7(d) for (002) and (101), respectively), and between the first, second, and third atomic Ru layers [denoted 3 O layers in Figs. 7(e) and 7(f) for (002) and (101), respectively]. Slabs with O layers are optimized, and following optimization, oxidants are adsorbed onto the surfaces.

E_ads for O adsorbing onto the O layered structures is given in Table I. E_ads for O₂ and O₃ adsorbing onto O layered structures is given in Table S3 in the supplementary material.⁵⁹ E_ads for atomic O on Ru(002) and Ru(101) with one O layer is 5.32 and 5.04 eV, respectively, in line with previous reports.^23,48,56,57 This is a change of −0.75 and −0.59 eV relative to bare Ru(002) and Ru(101). While adsorption is made less favorable by the presence of a 2D surface oxide, it is still exoergic. E_ads for atomic O does not monotonically decrease, however, as the trend in E_ads has a local minimum when two O layers are incorporated on Ru(002) and when three O layers are incorporated on Ru(101). As Ru surfaces are known to saturate at 0.50–0.75 ML O, we hypothesize that adsorption would be made endoergic if the surface coverage was raised >0.75 ML O.

Self-limiting behavior in the oxidation of Ru from co-exposure to VUV/O₂ is examined in more depth with the NEB method, where oxygen diffusion into an Ru surface with O incorporation is explored. NEB structures are first optimized and a “probe” O atom is placed on the surface for diffusion. The O atom is then propagated through the structures between the n = 1 and 2 layers, followed by the n = 2 and 3 layers, and finally between the n = 3 and 4 layers as indicated in Fig. 7. All NEB diffusion studies, like O layered incorporation studies, are performed using Ru slabs consisting of six atomic layers. This is done to allow at least one atomic layer to relax in between the layers containing the O “probe” atom and the layers representing the bulk Ru film. Ru(002) with one and two layers of O incorporation, as shown in Figs. 7(a) and 7(c), are explored in the NEB calculation, as clean Ru is unlikely to be a relevant surface kinetically, as experiments and simulations indicate.

The approximate minimum energy pathway (MEP) for Ru(002) with one and two O layers incorporated is shown in Fig. 8 in the top and bottom, respectively. For O diffusion through Ru(002) with one O layer on the surface, the activation energy, E_a, of O to move to the subsurface is 2.11 eV, and going from a subsurface location to the surface, E_a is 0.17 eV. This disparity indicates that diffusion subsurface is difficult and requires a high O concentration on the surface to proceed. Diffusion through the n = 3 and 4 layers must overcome an overall barrier of 2.77 eV. As the trend in E_a indicates, continued O diffusion into Ru is challenging, becoming increasingly difficult for O moving past the second atomic layer of Ru. We note that for a stoichiometric RuO₂ to form from any oxidation process, Ru:O stoichiometry must be 1:2, which would correspond to a two O layered system in this illustration. It is also worthwhile to mention that atomic O has a calculated E_ads of 5.32 eV on Ru(002) with one O layer, which is slightly less than E_max − E_min, which is 5.46 eV. Overall, the barrier to begin forming RuO₂ on the surface is the E_a of O to move the subsurface. As E_a increases moving deeper into the substrate, further oxidation is not energetically favorable.

Atomic O diffusion through a two O layered structure is shown in Fig. 8 as well. There is a large barrier to diffusion subsurface, E_a of 1.98 eV, which is slightly reduced compared to the one O layer diffusion diagram (−0.13 eV). O is at an energetic minimum when it resides in between the first and second atomic Ru layers. The calculated E_a to move subsurface is consistent with other reports of O diffusion through Ru(0001) of 1.80 ± 0.15 eV (Ref. 58) and 1.99 eV for O to move subsurface for 0.75 ML O coverage on an hcp Ru surface.²⁷ Continued diffusion past the second atomic Ru layer has an E_a of 5.10 eV, which we note is larger than E_ads for O adsorbing onto a two O layered Ru(002) system (4.23 eV). Therefore, it is energetically favorable for O to remain between the n = 1 and 2 atomic Ru layers, which is consistent with the observation of self-limiting behavior from XPS. McCoy et al.³¹ report a self-limiting oxidation of Ru from O generated by thermally cracking O₂ as well.

Chang and co-workers report^21,22 etching of oxidized Co, Fe, Ni, Pd, and Pt metals with exposure to HCOOH vapor. Our results show that HCOOH also works to etch oxidized Ru and RuO₂. We find the RuO₂ XP feature at 530.1 eV that results from VUV/O₂ co-exposure is removed upon isothermal exposure to HCOOH vapor (not shown). The etching half-cycle in a metal ALE process is relatively straightforward; however, the challenge in realizing metal ALE is to find a self-limiting oxidation process. Through the application of VUV/O₂ at 100 to 150 °C, we show that gas phase atomic O generated with a D₂ lamp produces a sufficient amount of adsorbed atomic O to drive the diffusion of this atomic O into Ru, which is in agreement with the findings of McCoy et al. over a similar temperature range.³¹ We observe that atomic O is responsible for oxidation at the conditions explored herein, as shown in Fig. 3. We also note that there is likely a competition between oxidation by atomic O and O₃ for Ru, as regimes of etch selectivity have been reported for controllable ratios of atomic O to O₃,¹⁷ which suggests that selectivity in oxidation may be achievable.

DFT is used to explore the adsorption characteristics of oxidants onto Ru surfaces with increasing O content. Structures and energetics of oxidants adsorbing onto bare Ru(101) and Ru(002) are shown in Fig. S3 and Tables S1 and S2.⁵⁹ Adsorption energetics for atomic O adsorbing onto O layered Ru are presented in Table I. Briefly, we find that adsorption, while exoergic, is less favorable as more O is incorporated into the Ru slab evidenced by a decrease in E_ads as more O is incorporated into the surface. This suggests that there is little to limit the amount of O that adsorbs onto the Ru surface, so that the rate of subsurface O diffusion is likely not limited by the amount of atomic O on the surface. The NEB calculations (Fig. 8) illustrate the energetic barrier for atomic oxygen to diffuse into Ru(002) is close to E_ads of atomic O on the Ru surfaces studied in DFT (Table I and Tables S2 and S3). Through a combination of substrate temperature to overcome this energy barrier to diffusion and VUV/O₂ exposure time, it is possible to realize Ru etch removal amounts of ∼1 Å/cycle (Fig. 5). It is this removal amount per cycle, and the self-limiting oxidation under the exposure conditions shown in Figs. 1 and 2, that leads us to describe the work reported herein as an ALE process.

The AFM and XRR results herein illustrate that the Ru films are rougher at the outset and become smoother after experiencing only five ALE cycles (Fig. 4). It is important to note that, while atomically smooth facets are used to understand adsorption characteristics in DFT and NEB calculations, it is not expected that this reflects the true mechanism of oxidation on Ru films owing to the possibility of additional surface facets, and the existence of higher order surface facets at the outset. Higher order facets undoubtedly accompany the roughness of as-deposited Ru thin films. As etching proceeds, the roughness, along with the prevalence of higher order facets, is reduced. This has the effect of reducing the exposed surface area and the number of surface configurations available during oxidant adsorption. The different facets present different diffusion paths into the substrate than modeled in Fig. 8 with likely different energy barriers.

We explored this by increasing the VUV/O₂ co-exposure time during oxidation half-cycles for a ten ALE cycle-treated substrate that was observed to experience minimal (i.e., 1 Å) etching with ten additional ALE cycles (120/90/30/150). The film thickness was reduced after increasing the VUV/O₂ co-exposure time from 2 to 5 min (i.e., ALE cycles of 300/90/30/150), suggesting that the oxidation rate is a limiting factor in allowing ALE to proceed. The existence of higher order facets likely is responsible for the higher oxidation rates on Ru. The effect of roughness/facets on Ru ALE is consistent with ALE of Pd where we showed the extent of Pd oxidation, and in turn the amount of Pd etched using VUV/O₂ is greater for discontinuous Pd films than it is for void-free films.¹⁹ Further work is needed to understand the effects of microstructure on the rate and extent of metal oxidation as that defines the amount of metal removed during the HCOOH etching half-cycle.

VUV-enhanced ALE of Ru metal is demonstrated. ALE is accomplished with a half-cycle that oxidizes near-surface Ru in a self-limiting manner under the conditions of exposure time and temperature employed herein, followed by a half-cycle that etches oxidized Ru metal. Oxidation half-cycles consist of co-exposure of VUV/O₂ for 2 min at 1 Torr O₂, while etching half-cycles consist of exposure to HCOOH vapor at 0.50 Torr for 30 s. Ar is used to purge the reactor between oxidation and etching half-cycles for 1.5 and 2.5 min, respectively. This yields controllable oxidation of the Ru and an etch rate that decreases with ALE cycles if co-exposure time is held constant. The differential removal amount for five cycles (i.e., material removed between 0 and 5 cycles) at 100 and 150 °C is 1.10 and 1.20 Å/cycle, respectively, indicating a slight increase in the etch rate with the temperature. DFT results suggest that all oxidants (O, O₂, and O₃) adsorb strongly to the bare Ru surface and equilibrate at a surface coverage between 0.50 and 0.75 ML O. Subsurface O diffusion is probed using the NEB method, which demonstrates a large activation energy, E_a, barrier to diffusion for an O atom to move beneath the subsurface layers (i.e., beyond n = 2 atomic layers of Ru). These findings suggest that self-limiting behavior is due to the inability of the RuO₂ front to propagate through bulk Ru. We use this reasoning to suggest that the decrease in etch rate as ALE is performed is likely due to a slowing oxidation rate, which tends toward self-limiting behavior with longer VUV/O₂ co-exposure times. This hypothesis is substantiated using an Ru sample exposed to ten additional ALE cycles with 2 min VUV/O₂ oxidation half-cycles, which is observed to be only minimally etched. The same film is then exposed to five ALE cycles with 5 min VUV/O₂ oxidation half-cycles. This increased time of oxidation half-cycles restores the etch removal amount per cycle that is observed for the initial ALE cycles.

The authors acknowledge the Texas Advanced Computing Center (TACC) at The University of Texas at Austin for providing high performance computing that has contributed to the research results reported herein. This work was funded by the National Science Foundation (NSF) (Grant Nos. 1610403 and EEC-1160494).

The data that support the findings of this study are available within the article and its supplementary material.⁵⁹