Enhanced nucleation mechanism in ruthenium atomic layer deposition: Exploring surface termination and precursor ligand effects with RuCpEt(CO)₂ Free

The ongoing miniaturization of microelectronic devices, which is dictated by the drive for increased performance, requires manufacturing techniques capable of atomic precision and uniformity, particularly in the deposition of thin films.^1,2 Atomic layer deposition (ALD) has become an important technique for depositing thin films in various fields, including microelectronics, offering precise control over film thickness and composition, even on complex three-dimensional structures.^3–6 The uniqueness of ALD is due to the alternating self-limiting surface reactions, which ideally lead to a distinctive layer-by-layer growth mechanism of the thin film. Within the field of microelectronics, the use of ruthenium (Ru) thin films has garnered significant interest due to its desirable properties. Low electrical resistivity, thermal stability, and strong adhesion to underlying materials^7–10 make it a candidate of interest for advanced microelectronic applications, such as a barrier metal or seed layer for Cu electroplating,¹¹ capacitor electrodes for dynamic random access memory,^12,13 and a potential material for replacing Co, W, and Cu interconnects.¹⁴ However, ALD of high surface-free energy materials such as Ru on top of low surface-free energy surfaces, such as oxides, tend to have large nucleation delays. This, in turn, can lead to growth as dispersed nanoparticles at the beginning of deposition that eventually coalesce into a layer that is not always pinhole free or smooth at the desired thickness.^15–17 Poor nucleation is usually attributed to the limited chemisorption of the precursor on the surface and poor wetting of the deposited material.^18,19 Thus, in order to maximize the potential of ALD in these material systems, it is essential to improve the nucleation and growth processes and understand the mechanisms by which nucleation can be improved.

Several approaches have been studied to improve nucleation and growth, specifically for ALD of noble metals. Approaches include the use of radicals or plasma chemistry as pretreatment,²⁰ applying high surface-energy wetting layers,²¹ and treating the substrate with small organometallic molecules prior to deposition.^17,19,22 Recently, we studied the nucleation and growth of Ru layers deposited by ALD using Ru(Cp)₂ as the precursor and O₂ as the coreactant and showed that predosing one pulse trimethylaluminum (TMA) or diethylzinc (DEZ) on the Si substrate prior to Ru deposition results in nucleation enhancement and generates a higher density of Ru nanoparticles and a more continuous Ru layer, compared to a nonpretreated Si substrate.¹⁷ Supported by the Avrami model^23,24 for nucleation and growth, it was shown that the modified surface increases the number of chemisorption sites on the one hand, and enhances the surface diffusivity on the other hand. As a result, transport processes on the surface, such as surface diffusion and coalescence, are enhanced as well.

Those earlier findings contributed to a better understanding of the nucleation and growth mechanism of Ru layers but raised an important question about the generalizability of the enhancement mechanism and the role of the precursor ligands. It could be presumed that the enhancement of Ru growth using a homoleptic precursor, characterized by identical ligands linked to the central metal ion, may differ from that using a heteroleptic precursor, where the metal ion is surrounded by more than one type of ligand that might be more reactive. Reports on Ru ALD using the homoleptic cyclopentadienyl (Cp)-based precursor showed low reactivity of the ligands with different surface species (e.g., –OH, –NH₂), which led to long nucleation delays before a constant film growth was achieved.^25–27 Conversely, other studies showed that replacing one or two of the Cp ligands with other ligands that are more reactive toward the surface decreased the nucleation delay in Ru ALD. For instance, exchanging one ethyl (Et)-Cp ligand with the more reactive 2,4-dimethylpentadienyl (DMPD) to form EtCpRu(DMPD) resulted in better precursor adsorption during the initial ALD process.^28,29

In this work, we investigate the mechanism behind the nucleation and growth enhancement observed in Ru ALD as a model system that employs the same small organometallic molecules previously utilized. We introduce a less-studied Ru precursor, RuCpEt(CO)₂, chosen for its promising industrial applicability owing to its liquid form and elevated vapor pressure (400 mTorr at 60 °C) compared to other Ru precursors (15–100 mTorr at the same temperature).³⁰ The RuCpEt(CO)₂ precursor is also of interest because of its heteroleptic ligands that may engender smaller intrinsic nucleation delays. A Si wafer with native oxide was chosen as the substrate to demonstrate the pretreatment's efficiency on a moderately hydroxylated surface. Scanning electron microscopy (SEM), spectroscopic ellipsometry, and x-ray photoelectron spectroscopy (XPS) were used to investigate the ALD growth mechanism on the treated and untreated substrates. Additionally, density functional theory (DFT) calculations were performed to simulate the reactions at the molecular level. The results show that the molecular surface pretreatment exerts a strong influence on Ru ALD nucleation, yielding up to a 3.2-fold increase in Ru surface coverage for the first 60 cycles of Ru ALD, as well as an increase in the thickness and nuclei density. We also observe a higher nucleation rate for this ruthenium precursor than for previously-studied Ru(Cp)₂. We postulate that two reaction mechanisms specific to this precursor's molecular structure enhance the nucleation and growth of the Ru film: first, the removal of CO ligands of the precursor during the deposition, which makes the precursor more reactive, and second, the ability of the ethyl ligand attached to the Ru precursor to undergo hydrogen abstraction with the newly formed surface termination. This enhancement mechanism was not observed with other Ru precursors such as Ru(Cp)₂.

Cyclopentadienylethyl(dicarbonyl)ruthenium [RuCpEt(CO)₂] (EMD Electronics, Merck KGaA, Darmstadt, Germany) and O₂ (99.6%, Linde Gas & Equipment Inc.) were used as the precursor and the coreactant, respectively. Si substrates were cut into ∼1 cm² coupons from Si(100) wafer (Pure Wafer, Inc.) with native oxide and were cleaned by sequentially rinsing in acetone, 2-propanol, and de-ionized water, followed by a 15 min cleaning with UV–ozone. Two organometallic molecules—trimethylaluminum (TMA) (Sigma-Aldrich) and diethylzinc—(DEZ) (Sigma-Aldrich) were used to pretreat the substrate surface.

A custom-made ALD reactor, controlled by LabView software, was used to grow the Ru layers. The reactor type is a showerhead inlet with vacuum pumping lines that are connected to the reactor's body. The Ru precursor was contained in a metal bubbler that was held at 80 °C to obtain a desired vapor pressure of ∼450 mTorr. The O₂ coreactant was pulsed using a mass flow controller at 50 SCCM. Precursor vapor was carried into the reaction chamber with N₂ carrier gas, controlled by a mass flow controller at 20 SCCM. Purging the excess gas molecules and byproducts during the ALD process was done by using the same N₂ flow rate. The optimal Ru ALD cycle consisted of a 2 s precursor pulse, 40 s of soaking while the vacuum pump valve was closed, a 45 s N₂ purging, a 6 s O₂ pulse, and an additional 45 s of N₂ purging. The substrate deposition temperature was 300 °C and deposition was done for 10–100 ALD cycles. Full growth curves are presented in Fig. S1 in the supplementary material.⁵² TMA and DEZ were contained in metal bubblers at room temperature. The half-cycle of one of these organometallic compounds consisted of a 3 s pulse followed by a 60 s N₂ purge and is expected to produce a submonolayer coverage at the surface. Further information on chamber configurations can be found in our previous report.¹⁹ 

The thicknesses of the resulting films were determined by spectroscopic ellipsometry (Alpha SE ellipsometer, J.A. Woollam Co., Inc.) at incident light angles of 65 and 70°. Software from Woollam was used to fit the ellipsometry data with a GenOsc model. The surface morphology of the grown film was determined using SEM (Magellan 400 XHR, FEI) equipped with a field emission gun (FEG) source. All SEM images were taken by using an accelerating voltage of 3−5 kV and an in-lens secondary electron detector. Particle size, particle distribution, and areal surface coverage were determined using imagej software,³¹ analyzing at least 1000 particles from at least five different SEM images at different magnifications. Chemical analysis of the grown films was done using XPS (PHI VersaProbe 3, PHI) with Al Kα (1486 eV) radiation, a 200 μm beam width, 50 W beam power, and 15 kV e-beam energy. The XPS data were fit using multipak software (Copyright 2019, Physical Electronics, Inc.).

We adopted DFT calculations to elucidate the proposed adsorption mechanism of RuCpEt(CO)₂, on a TMA-treated surface. All DFT calculations were performed using the Gaussian 16 suite of programs.³² Becke's three-parameter hybrid functional (B3LYP)^33,34 was used for the calculations. In addition to the B3LYP functional, we used Grimme's dispersion with Becke–Johnson damping (GD3BJ) for dispersion correction.^35,36 The effective core potential (ECP) basis set with LANL2DZ for Ru, the Def2-TZVP^37,38 basis set for Al, C, H, and O atoms, and the Def2SVP basis set for Si atom were used. An AlCH₃-terminated Si(100)-like cluster [Si₁₅H₁₆(O–AlCH₃–O)(OH)₂] was used to represent the TMA pretreated SiO₂ substrate. To calculate thermochemical properties at 300 °C [thermal energy (E), enthalpy (H), entropy (S), and Gibbs free energy(G)], we conducted vibrational frequency calculation for all geometries. For energy states with molecules adsorbed on the surface, translational and rotational vibrations of the complex were assumed to be nonexistent, and translational and rotational corrections were excluded from the calculations.

Metallic Ru layers were deposited by ALD on nonpretreated Si substrates containing native oxide using RuCpEt(CO)₂ and O₂ for 10, 20, 30, 50, and 100 cycles, as described in the Methods section. This process resulted in a noncontinuous film, as shown in the SEM image of Fig. 1(a). All the deposited samples were chemically characterized by XPS. The results showed metallic Ru deposited on the surface, with the characteristic doublet Ru3d peak at 280.2 eV, and no evidence for RuO₂ (Fig. S2 in the supplementary material).⁵² The growth on nonpretreated Si presents a nucleation delay of ∼20 cycles, as seen in the black data points for thickness versus ALD cycle number in Fig. 1(d). This delay is similar to the nucleation delay previously observed by Park et al.³⁹ (∼22 cycles), but much shorter than that was reported by Leick et al.³⁰ (∼100 cycles), using the same RuCpEt(CO)₂ precursor. The growth per cycle (GPC) can be estimated from the slope of the thickness versus ALD cycle curve in Fig. 1(d) and equals 0.11 ± 0.01 nm/cycle. This value is comparable to the previously reported GPCs using the same Ru precursor^30,39 but higher than GPC values obtained for Ru ALD using other Ru precursors.^10,40,41 These differences in the GPC values may be attributed to several factors related to the physical and chemical properties of the precursor, such as precursor volatility and vapor pressure, reactivity, stability, surface adsorption, ligands, and the metal center oxidation state.⁴² 

In order to enhance the nucleation and growth of Ru toward achieving a pinhole-free film, a second set of experiments was performed by predosing one pulse of TMA or DEZ on the Si substrate prior to Ru deposition, as suggested previously.^17,19 Nucleation and growth enhancement were achieved using pretreatment, as reflected in the uniformity of the Ru layer by SEM [Figs. 1(b) and 1(c)]. In addition, consistent with the SEM results, the pretreated Ru layers showed higher thicknesses and a shorter nucleation delay, according to the growth curves in Fig. 1(d). Despite the lower nucleation delay, the calculated GPC values for TMA-pretreated and DEZ-pretreated samples, once growth begins to occur, are 0.10 ± 0.01 and 0.09 ± 0.01 nm/cycle, respectively, similar to the GPC value observed for Ru growth on a nonpretreated substrate. The nucleation enhancement can be quantified by calculating the enhancement factor (EF), which was previously introduced by de Paula et al.¹⁹ and is defined as the ratio of Ru areal coverage on a treated substrate to Ru areal coverage on an untreated substrate. The areal coverages used to determine EF are obtained from the SEM images, as described in Methods and are plotted in Fig. 2. The calculated EF values are presented in Fig. 1(e) as a function of the ALD cycle number. The values range between 1.1 and 3.2 for both TMA- and DEZ-pretreated substrates, meaning that pretreatments enhance the growth of the Ru films up to 3.2 times, for 40 ALD cycles, compared to the nonpretreated substrates. The EF value decreases as the number of ALD cycles increases, eventually converging toward a value of 1. This trend is expected because of the way EF is calculated: over time, even substrates that have not undergone pretreatment achieve complete areal coverage of Ru.

The SEM images suggest that organometallic pretreatment enhances nucleation and growth, and analyzing these images can help to clarify the mechanisms behind this enhancement. Figure 2(a) displays the areal coverage of Ru layers as a function of the ALD cycle number for surfaces pretreated with no-, TMA-, and DEZ- pretreatments, as determined from the SEM images. Similar to the EF data, the areal coverage data demonstrate enhancement when using TMA or DEZ pretreatments. The graph depicts a rapid initial increase in areal coverage that subsequently slows as it approaches saturation, forming an S-shaped curve that illustrates the relationship between coverage and ALD cycle number. The trend in areal coverage observed in this study is consistent with previous findings on Ru ALD enhancement using similar organometallic molecules and Ru(Cp)₂ as a precursor. However, a significant difference between the earlier study¹⁷ and the present work lies in the number of ALD cycles required to achieve more than 90% coverage. While the nucleation-enhanced Ru layer in this case reaches over 90% coverage after just 30 ALD cycles, it took 200 ALD cycles to reach similar coverage when using Ru(Cp)₂ as the precursor. This shorter time period for full coverage may be attributed to the unique chemisorption of the RuCpEt(CO)₂ precursor on the pretreated surface, as will be discussed later.

Further analysis of the SEM images and comparisons among the various samples provide important morphological insights, such as information on nuclei density. Figure 2(b) shows the nuclei density, calculated for each cycle number by dividing the total number of nuclei by the uncovered surface area, as extracted from SEM images of the non-, TMA-, and DEZ-pretreated substrates. The nuclei density data show a similar trend of constant increase for the nonpretreated and pretreated surfaces. This constant slope of nuclei density versus cycle number observed in Fig. 2(b) suggests a constant nucleation rate for each of the substrates. On the other hand, the values of nuclei density for the different substrates differ significantly, with 5−7 × 10³ nuclei per μm² for the nonpretreated surface and much higher values of 1.5−2.75 × 10⁴ nuclei per μm² for the pretreated ones at the early stages of the ALD process.

To elucidate the mechanism behind the enhancement, it is beneficial to compare the current data with that from the previous study that employed the same organometallic molecules for pretreatment but utilized a different Ru precursor.¹⁷ While the samples with and without organometallic molecules using the current RuCpEt(CO)₂ precursor showed a constant increase in the nuclei density, as described above and shown in Fig. 2(b), the data for Ru enhancement using Ru(Cp)₂ as a precursor are different (Fig. S3 in the supplementary material).⁵² Interestingly, the nuclei density for the pretreated substrates using Ru(Cp)₂ as a precursor rapidly increases until 50 ALD cycles and then nearly plateaus, suggesting a high initial nucleation rate that then approaches zero as the ALD process continues. This differing behavior is highlighted by comparing the calculated nucleation rates, as plotted in Fig. 3. The nucleation rates are the changes in the nuclei densities per cycle and were calculated as the difference between the nuclei densities divided by the difference of the ALD cycle numbers for two adjacent points. Because the nuclei densities were already determined per uncovered surface area in Fig. 2(b), the nucleation rate is the change in nuclei density per cycle per uncovered surface area, with units of nuclei/μm²/cycle as shown in Fig. 3. A true linear dependence in Fig. 2(b) should translate into a constant nucleation rate in Fig. 3. However, as Fig. 3 indicates, the actual nucleation rate follows a more complex trend rather than remaining constant, as described further below.

Figure 3 displays data ranging from 10 ALD cycles up to the point where the Ru film achieves full areal coverage, occurring at 50 cycles for RuCpEt(CO)₂ and 150 cycles for Ru(Cp)₂, respectively. The data illuminate two main differences. The first is the values of the nucleation rates, which are ∼7 times higher when using RuCpEt(CO)₂ compared to Ru ALD using Ru(Cp)₂ as a precursor_. Since the pretreated substrate surfaces are identical between the two systems (both employing saturation coverages of either TMA or DEZ), the difference in the magnitudes of the nucleation rate must be attributed to the differences in reactivity between the two Ru precursors. We expect that the larger nucleation rate for RuCpEt(CO)₂ is the result of its high reactivity and potential interaction between the RuCpEt(CO)₂ ligands and surface termination. Previous studies that used carbonyl-based precursors for ALD showed thermal CO dissociation at around 200 °C.^43–46 

The metal–CO bond is a coordinate covalent bond, in which the CO ligand donates both electrons to form the bond. When heated, the vibrational energy excites the metal–CO bond, causing the CO ligands to detach from the metal center.^47,48 Therefore, the loss of CO ligands from the precursor either in the gas phase or the adsorbed state leaves the metal center with unfilled coordination sites, which makes the precursor more reactive. An additional potential explanation for the higher reactivity of RuCpEt(CO)₂ is its ethyl ligand, which can interact with the metal-alkyl groups on the surface, as discussed later.

The second difference between the two Ru precursors in the data of Fig. 3 relates to the nucleation rate trends. The data indicate that when using the Ru(Cp)₂ precursor, nucleation rates increase rapidly between 0 and 50 ALD cycles before decreasing to nearly zero by 100 cycles. In contrast, nucleation rates remain constant when using the RuCpEt(CO)₂ precursor. This disparity points to a different nucleation enhancement mechanism when using these organometallic precursors. For ALD using Ru(Cp)₂, it has been suggested that an excess of hydroxyl groups on the Si substrate, combined with the cleaning effect of the pretreatment that removes contaminants inhibiting nucleation, leads to a relatively high number of potential nucleation sites.¹⁷ Nevertheless, there may be an incubation time for nucleation, as revealed in the initial increase in the rate. However, as the ALD process progresses, the surface groups may be reacted, diminishing the availability of active sites for further nucleation. Conversely, the constant nucleation rate observed when using the RuCpEt(CO)₂ precursor indicates a continuous formation of nuclei over the measured range of ALD cycles for this precursor.

Given the significantly higher nucleation rate for RuCpEt(CO)₂, an additional mechanism must be at play, likely involving interactions between the precursor ligands and the surface termination groups. Such a mechanism was previously suggested for Pt ALD.¹⁹ We hypothesize that the enhancement mechanism is related to the abundance of alkyl ligands present after the organometallic pretreatment, which increases the total number of active sites available specifically for RuCpEt(CO)₂ chemisorption. Considering that Al and Zn are strong Lewis acids, TMA or DEZ are expected to react vigorously with the Brønsted acidic surface hydroxyl groups, effectively consuming those sites and leaving adsorbed Al–CH₃ or Zn–C₂H₅ species at the surface.⁴⁹ The lower electronegativity of Al and Zn compared to Ru results in more polar Al–CH₃ and Zn–C₂H₅ bonds. Consequently, their ligands exhibit stronger Brønsted basicity, capable of accepting protons, compared to the organic ligands of RuCpEt(CO)₂. We postulate that the Brønsted basic –CH₃ and –C₂H₅ groups can abstract hydrogen from the ethyl ligand bonded to the Ru precursor, potentially facilitating the attachment of the precursor molecule to the surface. This suggested mechanism is not available when using the Ru(Cp)₂ precursor due to the absence of an alkyl ligand such as ethyl in the precursor.

Interestingly, previous studies reported the Al site density after a TMA pulse on an SiO₂ surface as 3.5 × 10¹⁸ per m² (Ref. 50) and 3.9 × 10¹⁸ per m².⁵¹ The Ru nuclei density in our study after the TMA pulse was calculated as 1.5−2.5 × 10¹⁶ per m² between 10 and 50 ALD cycles. Comparing the results suggests differences between surface Al–CH₃ site density reported in the literature and Ru nuclei density measured here. We posit several possible explanations: (1) Not every Al–CH₃ species may serve as a nucleation site for Ru ALD. (2) The TMA pulse in previous studies was done at 200 °C, whereas in our study, we dosed TMA at 300 °C. At higher temperatures, the surface mobility of the adsorbed species increases. This can lead to a decrease in the nuclei density as species can migrate and form fewer, larger nuclei rather than many small ones. (3) Higher temperatures can also increase the desorption rates of weakly adsorbed species. If the TMA or its reaction intermediates desorb more readily at higher temperatures, the effective coverage of TMA on the surface may decrease, leading to a lower Al site density.

The DFT results support the proposed enhancement mechanism. Initial calculations explore the detachment of the carbonyl ligands that is expected to lead to enhanced reactivity for the RuCpEt(CO)₂ precursor. Figure 4(a) presents the change in Gibbs free energy for the dissociations of the different ligands at a temperature of 300 °C. As seen in Fig. 4(a), the calculations reveal only a small cost in energy for the dissociation of the carbonyl ligands (∼25 kcal/mol per CO) compared to the dissociation of the cyclopentadienyl and ethyl ligands (177.0 and 194.3 kcal/mol, respectively). The detachment of the CO groups at a relatively early stage of the process is expected to render the precursor molecules more reactive, thereby increasing their probability of adsorption on the surface.

We examine different possible chemisorption pathways resulting from hydrogen abstraction using the surface Al–CH₃ groups as the model system. Four different pathways are considered, and the DFT results are presented in Figs. 4(b) and 4(c). The four calculated reaction pathways for the adsorption of RuCpEt(CO)₂ molecule on the –CH₃ terminated surface are (1) hydrogen abstraction by the surface –CH₃ group from the ethyl ligands bonded to the intact Ru precursor (without detachment of carbonyl groups) resulting in the release of a CH₄ molecule, (2) hydrogen abstraction by the surface –CH₃ group from the ethyl ligands bonded to the Ru precursor resulting in the release of a CH₄ molecule, subsequent to the detachment of the carbonyl groups, (3) hydrogen abstraction by the surface –CH₃ group from the ethyl ligands bonded to the Ru precursor resulting in the release of a C₂H₅ molecule, subsequent to the detachment of the carbonyl groups, and (4) adsorption through a direct Ru–Al bond without releasing any other byproduct molecules, subsequent to the detachment of the carbonyl group. Note that based on the energetics in Fig. 4(a) for CO loss, the precursor can coexist in several forms at 300 °C, with two, one, or no CO ligands. Here, the energy values for pathways 2–4 do not include the additional energy cost of CO loss, based on the assumption that those pathways start with RuCpEt, the decarbonylated form of the precursor.

Comparing the four calculated reaction pathways for chemisorption on the –CH₃ terminated surface with the four corresponding pathways on a nonpretreated, –OH terminated surface (results presented in Fig. S4 in the supplementary material⁵²) reveals a clear thermodynamic favorability for chemisorption on the pretreated surface. Adsorption energetics range from −47.3 to −62.9 kcal/mol on the CH₃-terminated, pretreated surface, whereas they range from endergonic (+7 kcal/mol) to mildly exergonic (−30 kcal/mol) on the –OH terminated surface. Thus, pretreatment provides for more exergonic adsorption chemistry.

On the pretreated surface, the calculations show only minor differences in the overall Gibbs free energy change for the two most exergonic pathways, reactions (3) and (4), with very close values of −59.0 and −62.9 kcal/mol, respectively. The similarities in Gibbs free energy changes indicate that both pathways are similarly favorable thermodynamically. Since both pathways are nearly equally feasible from an energetic standpoint, it is possible that both reactions could occur simultaneously or under similar conditions. However, while the Gibbs free energy values indicate the thermodynamic feasibility of the pathways, they do not provide direct insights into the kinetics of the reactions. The two reactions can be thermodynamically favorable, but their rates may differ based on activation energies or reaction intermediates.

The consistent thermodynamic favorability of pathways (3) and (4) facilitates a stable chemisorption environment for the RuCpEt(CO)₂ molecule on the –CH₃ terminated surface, which might promote uniform growth rates across the surface, and account for the relatively large particle sizes, discussed in the following paragraph and presented in Fig. 5. In comparison, the previously studied precursor RuCp₂ lacks an ethyl group, the moiety which is central to all four reaction pathways calculated for the RuCpEt(CO)₂ molecule [Fig. 4(c)]. Thus, we speculate that RuCp₂ has fewer chemisorption pathways available, which could impact its nucleation despite the same TMA or DEZ pretreatment. This difference highlights the critical role of molecular structure in determining the efficiency of nucleation and growth processes in ALD.

As discussed above, the differences in nucleation and growth that result between the two Ru precursors, Ru(Cp)₂ and RuCpEt(CO)₂, provide insights into the latter's enhancement mechanism. However, examining not only the differences but also the similarities between the two cases can further clarify the overall ALD enhancement mechanism. It was shown that pretreating the Si surface with TMA or DEZ reduces the surface free energy from a value of 56.9 ± 1.4 mJ/m² to values of 46.4 ± 2.3 and 44.2 ± 3.3 mJ/m² for TMA- and DEZ-pretreated Si substrates, respectively.¹⁷ It was suggested that the change in the surface free energy increases Ru diffusivity on the pretreated substrate surfaces and, as a result, facilitates a greater influx of material toward the growing nanoparticles, leading to accelerated growth over similar ALD cycles and, hence, larger particles. In addition, the study with Ru(Cp)₂ observed increased nuclei densities for the pretreated surfaces, which were hypothesized to arise from an increase in the number of reactive sites. Since we employ the same pretreatments on the same substrates in the current ALD process using RuCpEt(CO)₂ as a precursor, it is reasonable to expect that similar effects will also occur in this case. Additionally, it can be expected that a variation of reaction pathway (4), which is the adsorption of the Ru precursor through a direct Ru–Al bond, could also proceed on existing Ru sites to form Ru–Ru bonds, potentially leading to larger Ru particles.

Measuring the average Ru nanoparticle diameter grown on nonpretreated and pretreated substrates supports this phenomenon. As an example, Fig. 5 presents the SEM images and diameter size distribution for 50 cycles of ALD Ru on the two substrates. While the average diameter of Ru nanoparticles for the nonpretreated Si substrate is 18.12 ± 5.47 nm, the average diameter of Ru nanoparticles for the TMA-pretreated Si substrate is 27 ± 2.66 nm, a larger size which we attribute to the enhanced surface diffusivity. A comparison of different average nanoparticle diameters for the entire ALD process is presented in Fig. S5 in the supplementary material.⁵² The measured values of the average diameters are similar to the ones measured when using Ru(Cp)₂ as a precursor. The main difference, however, is the number of ALD cycles to get these values. While for the Ru(Cp)₂ case, it took ∼150 ALD cycles to reach a maximum average diameter of ∼30 nm, a similar value is achieved only after ∼50 ALD cycles for the current RuCpEt(Cp)₂ system.

This paper investigates the mechanisms behind enhanced nucleation and growth in Ru ALD using a less-studied precursor, RuCpEt(CO)₂, with promising industrial applications, and O₂ as a coreactant. SEM images and ellipsometry measurements show that the as-deposited Ru forms a noncontinuous film on Si with dispersed Ru nanoparticles, a nucleation delay of ∼20 cycles, and a GPC of 0.11 nm/cycle. Predosing one pulse of TMA or DEZ on the Si substrate prior to the Ru ALD process results in nucleation enhancement, a higher density of nuclei, and a shorter nucleation delay. EF values up to 3.2 for 40 ALD cycles indicate enhanced growth on the pretreated surfaces compared to the nonpretreated ones. SEM analysis shows a higher areal coverage for the pretreated substrates with a rapid increase in coverage that subsequently slows as it approaches saturation.

A comparison of these results with previous studies using Ru(Cp)₂ as a precursor reveals higher nuclei densities and sustained nucleation rates for RuCpEt(CO)₂, versus a quick decline to the zero nucleation rate observed with Ru(Cp)₂, suggesting that an additional mechanism of nucleation enhancement is accessible for the RuCpEt(CO)₂ precursor. Supported by DFT calculations, we hypothesize that the enhanced nucleation behavior is attributed to (1) the removal of the CO ligands of the precursor during deposition, which leaves the metal center with unfilled coordination sites and makes the precursor more reactive, and (2) a hydrogen-abstraction reaction between the ethyl ligand of the Ru precursor and the metal-alkyl surface termination after the pretreatment. The former mechanism was also observed for Pt ALD using MeCpPtMe₃ as a precursor on a pretreated Si substrate with TMA.¹⁹ That Pt precursor has a free alkyl ligand, the same as the RuCpEt(CO)₂ precursor, unlike the RuCp₂ precursor, emphasizing the important role of molecular structure in determining the nucleation processes in ALD. Moreover, an increase in the average diameter of Ru nanoparticles during the ALD process on pretreated substrates suggests that the nuclei growth is further enhanced by a reduction in surface free energy, which promotes surface diffusion. These findings not only advance our understanding of Ru ALD but also underscore the critical roles of precursor chemistry and surface treatments in optimizing thin-film deposition processes for microelectronics applications.

This work was supported by the U.S. Department of Energy under Award No. DE-SC0004782. A part of this work was performed at the Stanford Nano Shared Facilities (SNSF), supported by the National Science Foundation (NSF) under Award No. ECCS-2026822.

The authors have no conflicts to disclose.

Amnon Rothman: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Writing – original draft (equal). Seunggi Seo: Formal analysis (supporting); Writing – review & editing (supporting). Jacob Woodruff: Resources (supporting); Writing – review & editing (supporting). Hyungjun Kim: Writing – review & editing (supporting). Stacey F. Bent: Conceptualization (equal); Funding acquisition (equal); Resources (equal); Supervision (equal); Writing – review & editing (equal).

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