Ambient pressure x-ray photoelectron spectroscopy study on the initial atomic layer deposition process of platinum

Atomic layer deposition (ALD) of ultrathin noble metal films is an increasingly useful technique for several fields due to the films’ catalytic properties,^1,2 chemical stability, low propensity for oxidation, and numerous gas sensing applications.^3–8 Creating such films using ALD is very useful as the technique offers an unique combination of properties which are unattainable with other methods. With ALD, one can very accurately control the thickness of the deposited films and make highly conformal films over complex three-dimensional structures in a uniform manner.⁹ Several studies of the ALD of Pt films have been published previously.^10–18 The ALD of Pt films or nanoparticles is of exceptional interest due to the highly active catalytic nature of the deposited films. It has been suggested that the deposited film plays a role during the ALD process itself, since it can dissociate $O 2$ into highly reactive atomic oxygen. Such processes are further enhanced since Pt films are known to agglomerate heavily during the early stages of the deposition, creating a higher surface area with more active sites.^14,15

The execution of in situ and operando measurements during the deposition is the only way to obtain true mechanistic information of the process. We have recently developed instrumentation for x-ray photoelectron spectroscopy (XPS) to carry out in situ and operando monitoring of the film surface during the dosing of the ALD precursors.¹⁹ The measurements are carried out using ambient pressure XPS (APXPS), which is a technique allowing XPS measurements in a sample environment with elevated pressure (up to several tens of mbar) and temperature. The technique has been successfully used in several cases²⁰ to study, for example, the initial mechanisms of the ALD of high- $κ$ oxides such as HfO $2$ on InAs and SiO $2$⁠.^21,22 Recently published results indicate that the technique has the potential to provide new insights for, or challenge the established mechanisms of ALD, such as the ligand exchange process.^23,24 Here, we use the label operando to clearly indicate that in these measurements the XPS is measured simultaneously with precursor dosing, as opposed to measurements after the half-cycles, which is often the colloquial meaning of in situ XPS experiments within the ALD community.

The past decade has shown considerable interest in the growth mechanism of Pt, with many different aspects being studied. The initial growth displays a well-known nucleation delay effect, which can often delay the proper growth by several tens or even hundreds of ALD cycles²⁵ and many studies have attempted to provide a complete mechanistic understanding of the precursor chemistry. Rather counterintuitively, the reduction of the metal ion (e.g., Pt $4 +$ to Pt $0$⁠) that is needed in the process is often done using an oxidizing coreactant such as O $2$⁠, O $2$-plasma, or O $3$⁠. Most studies have arrived at similar conclusions about the combustion of the organometallic ligands by the oxygen coreactant.

The oxidation state of the Pt atoms during the initial half-cycles has remained somewhat unknown. Recent in vacuo XPS work by Li et al. showed that the Pt atoms remain metallic throughout the process, and accordingly surface O species exist only as adatoms which decay quickly on their own in vacuum.¹⁸ In another in situ x-ray absorption study by Setthapun et al., it was shown that there is a much higher ratio of Pt oxides on the surface of alumina, TiO $2$ and SrTiO $3$ supported Pt nanoparticles.¹⁰ Experiments by Nieminen et al. have also found oxygen within the films after reaching the steady state growth range which were attributed to adsorbed oxygen.¹⁷ 

In this work, we have studied the very first cycle of ALD of Pt on silicon using trimethyl(methylcyclopentadienyl)platinum(IV) [ $( C H 3 C 5 H 4 ) Pt ( C H 3 ) 3$ or MeCpPtMe $3$] and O $2$ gas as the precursors. The ALD with these precursors produces a metallic Pt film, as the O $2$ pulse which follows the dose of MeCpPtMe $3$ removes all of the ligands from the surface. In this work, the focus was on the surface processes during the first two half-cycles, in order to gain better understanding of the mechanism that produces the initial metallic film which is crucial for the reactions needed to get reactive atomic oxygen during the O $2$ half-cycles.

The measurements were done on the APXPS branch of the SPECIES beamline at the MAX IV Laboratory in Lund, Sweden.^26,27 More specifically, the ALD ambient pressure cell of the SPECIES beamline was used, which is an APXPS sample environment specifically designed for studying ALD processes in real time.¹⁹ 

The Pt films were deposited onto a native oxide-covered Si(100) crystal which was initially cleaned in vacuum by Ar ion sputtering (1.5 keV at 10 mA current and at about $1× 10 − 5$ mbar chamber pressure for 10 min). This was mostly done in order to remove some of the organic contaminants, but the process also removed some of the native oxide from the Si surface. Prior to the deposition, the substrate was reoxidized by exposing it to O $2$ gas at a total pressure of 0.8 mbar (the gas mixture also contained about 10 % Ar) at a temperature of 300  $°C$⁠. The deposition was carried out in the ALD cell, into which the sample was transferred without breaking the vacuum after the preparation steps. To mitigate any x-ray beam damage to the sample and effects to the spectral features resulting from it, the sample was continuously rastered and automatically scanned into new spots for every XPS measurement. This movement causes slight intensity variations in the total XPS signal due to small drift of the sample in and out of the optimum focus position of the electron energy analyzer.

The XPS measurements were done either by using the snapshot mode of the electron analyzer (whereby only a small range of the electron spectrum is acquired very quickly) to achieve a time-resolved XPS dataset or by using the swept mode to obtain x-ray photoelectron (XP) spectra with better quality and slightly higher energy resolution. The time resolution in the snapshot mode data is approximately 2.5 s. The combined beamline and spectrometer energy resolution in the swept spectra is about 330 meV. During the half-cycles, the Pt 4f and O 2s core levels were recorded in a time-resolved manner at a photon energy of 400 eV. After both of the half-cycles, the higher resolution XP spectra were recorded for the O 1s, C 1s, Pt 4f, and Si 2p core levels and the valence band (VB). Each core level was measured using a photon energy which yields an electron kinetic energy of $∼$100 eV, thereby achieving highest surface sensitivity. The VB was measured at a photon energy of 50 eV.

The Pt deposition was done on a Si substrate at 300  $°C$⁠. During the first half-cycle, the total pressure inside the ALD cell was about 0.8 mbar. In this case, the total pressure is the sum of the precursor vapor pressure (⁠ $∼$0.07 mbar in room temperature²⁸) and the Ar carrier gas. The O $2$ (purity 99.999%) half-cycle was done at the same temperature and pressure without Ar carrier gas. After the Pt and O $2$ half-cycles, the swept mode XP spectra were measured with the sample in constant Ar flow, in a pressure of about 0.4 mbar. The binding energy scale in all cases was calibrated using the Si 2p $3 / 2$ peak of the bulk silicon located at 99.3 eV,²⁹ which is in good agreement with literature values.^30–32 

The first ALD half-cycle was carried out on the oxidized Si substrate after the pretreatments specified above. The substrate was exposed to the MeCpPtMe $3$ precursor for about 100 min. The time-resolved XPS data from this half-cycle are shown in Fig. 1, which contains the Pt 4f and O 2s core-level spectra. The O 2s also contains the Ar 3s line, which has a similar binding energy but a considerably smaller linewidth. The Pt core-level spectra clearly show that the surface was initially free from any Pt, which then slowly started to appear on the surface during the deposition with an exponential intensity increase until the surface would no longer adsorb any more molecules. The small oscillations in the Pt 4f signal intensity are due to the continuous sample movement, as described above. The first half-cycle timescale is very long compared to typical ALD cycle timing, but common to those used when studying ALD with APXPS.^19,24,33 Moreover, noble metal ALD processes are known to exhibit long nucleation delays of tens of cycles that are attributed to the difficulty of getting the very first metal atoms deposited in the absence of the catalytic effect of the metal.³⁴ It is assumed that the very first metal atoms form by incidental decomposition of the metal precursor, and only then the growth can start by the actual ALD mechanism. Here, however, metallic Pt was observed already during the first pulse because of its exceptional length which gives time for the incidental decomposition to occur.

The timescale of the Pt growth can be further investigated from the Pt 4f $7 / 2$ signal by fitting a peak shape to the individual scans to extract the peak position and XPS intensity during each step of the growth. To increase statistical quality in the fitting, ten snapshots were averaged for each fit. A purely Gaussian peak shape was used for fitting together with a linear background. The extracted Gaussian peak positions (the binding energies of the peaks) and intensities are plotted in Fig. 1(c). The data follow a clear trend in the peak positions as well as in their intensities. To quantify these two trends, an exponential function was fitted to these changes. The function was in the form $A+B⋅exp(−t/τ)$⁠, where $A$ and $B$ are constants and $τ$ is the time constant characteristic of the change of intensity or binding energy, respectively. The resulting trend line and the $τ$ values are shown in Fig. 1(c). Interestingly, it appears that during the initial dose, the Pt 4f $7 / 2$ peak position starts at a binding energy of $∼71.55$ eV and rather rapidly shifts toward lower binding energy with a $τ$ of $1563±210$ s (error given as three standard deviations). The final binding energy is about 71.1 eV, which is consistent with metallic Pt.^35,36 The peak intensity, however, continues to increase even after the initial binding energy shift of the peak has slowed down. The $τ$ value for the peak intensity growth is significantly higher at $2685±272$ s. The binding energy shift of the Pt 4f could be correlated with a clusterization effect. It is known that the binding energy of a supported metal cluster on a poorly conducting surface depends on the size as small clusters exhibit higher binding energy.³⁷ This effect is also consistent with the known agglomeration in the early stages of Pt ALD.³⁸ 

The Si 2p region shows an inverse tendency to that seen in the Pt 4f region, with a steady decrease in intensity which corresponds to the surface being covered by the Pt-containing molecules. The corresponding map from the Si 2p region is shown in the supplementary material in Fig. S1.

The photon energy for the time-resolved measurements was chosen to be 400 eV in order to enhance the surface sensitivity of the Pt 4f signal, while still allowing to measure C 1s spectra. This implied that it was not possible to measure the O 1s core-level spectra simultaneously with the same photon energy. Instead, we recorded the shallow O 2s level located at approximately 25 eV binding energy. As expected, we see a constant decrease in the O 2s signal, corresponding to the slow coverage of the surface by the Pt containing precursor (which does not contain any O atoms). The decrease of the O 2s intensity follows a time constant of $3078±872$ s. However, the use of Ar as a carrier gas provided an opportunity to observe the change in the gas phase position of the Ar 3s peak (located also at about 25 eV), which, with its width of about 0.3 eV, is much narrower than the overlapping O 2s peak and therefore easily distinguishable. The position of the gas phase peak has a direct correlation to the work function of the sample surface.³⁹ The data in Fig. 1(b) very clearly show the changes in the work function as the gas phase peak first shifts to higher binding energy. The steady increase of the binding energy (or decrease of the work function) can be fitted in a similar way as was done for the Pt 4f $7 / 2$ peak. For the O 2s/Ar 3s case, each spectrum was fitted with two Gaussians: one wider representing the O 2s peak and one narrow for the Ar 3s core level. Figure S4 in the supplementary material shows an example from the fitting results for the O 2s and Ar 3s taken at about 1236 s. The binding energy evolution of the Ar 3s peak is plotted in Fig. 1(d) together with an exponential trend function, as specified earlier. The corresponding $τ$ for the exponential function representing the position of the Ar 3s peak is $1060±78$ s. The total shift in the Ar 3s peak during the deposition is about 0.9 eV. The position of the O 2s peak (shown in Fig. S3 in the supplementary material) did not seem to change considerably during deposition, but it is expected to be less sensitive to chemical changes. The signal quality of the O 2s peak is also considerably noisier especially toward the end of the deposition. The decrease of the O 2s signal intensity has approximately the same time constant as the increase of the Pt 4f signal.

An interesting effect is also seen toward the end of the first half-cycle at about 6000 s. According to the intensity increase of the Pt 4f peak [Fig. 1(a)] and the intensity decrease of the Si 2p peak [supplementary material, Fig. S1(b)], probably the surface saturation has nearly been reached at this point. At 6000 s, however, the binding energy of the Ar 3s peak, and therefore the work function, starts shifting back toward lower values. Therefore, there must be some other effect on the surface, which begins to occur toward the end of the half-cycle. It should be emphasized that the experimental conditions were stable during the whole deposition: the pressure was about 0.8 mbar and temperature about 300  $°C$ during the measurements shown in Fig. 1. The valves to the Pt precursor bottle were opened at 0 s and closed at approximately 3000 s, where a small increase in carrier gas pressure caused a temporary decrease in XPS signal. Otherwise, no other changes were made. During the measurement that followed after closing the Pt precursor lines, it is possible that the precursor concentration in the gas mixture at the sample position decreased. The decrease on precursor concentration might also manifest itself as a decrease in the Pt 4f area growth, but we believe most of the effect originates from the fact that the surface saturation is being reached, which is especially notable as the binding energy of the Pt 4f peak has, to a large degree, stopped changing before the valves were closed at the 3000 s point. However, it should be pointed out that the precursor concentration and the saturation of the surface could be correlated.

The C 1s changes were also monitored during the deposition and this is shown in Fig. S2 in the supplementary material. The C 1s region shows a similar steady increase of intensity. If a similar exponential function is fitted to the increase, a time constant of $3051±226$ s can be extracted. This value is similar to the O 2s decrease as well as the Pt 4f intensity increase. This indicates that the increase of the C 1s region is correlated with the carbon containing ligands in the MeCpPtMe $3$ molecule.

Figure 2 shows higher resolution O 1s, C 1s, Pt 4f, Si 2p, and valence-band XP spectra after the initial oxygen treatment, after the MeCpPtMe $3$ half-cycle, and after the O $2$ half-cycle, respectively. In each measurement, the photon energy was chosen as to render the measurement surface sensitive by proper choice of the electron kinetic energy. The photon energies are indicated in the figure. The core-level fitting results are shown in detail in the supplementary material (Figs. S5, S6, S7, and S8 and Tables S1, S2, S3, and S4), and the overview spectra are shown in Fig. S9 in the supplementary material.

The Pt 4f peaks were curve fitted with two different line shapes: the metallic peaks at 71.3 eV (Pt 4f $7 / 2$⁠) and 74.6 eV (Pt 4f $5 / 2$⁠) were fitted with Doniach–Šunjić (DS) functions, and Voigt lines were used to model the shoulders at 1.4–1.2 eV higher binding energy. The binding energy and the spin–orbit splitting of the DS peaks are in good agreement with the expected values for bulk metallic Pt.^35,36,40,41 It is notable that no surface peak is observed on the low-binding energy side. The binding energies of the shoulders match well with those expected for PtO, i.e., Pt in the +2 oxidation state.⁴⁰ The relative area of the PtO peak with respect to the metallic Pt changes after the Pt pulse and the O $2$ pulse, with it contributing to about 8% and 10% of the total Pt 4f peak areas, respectively. The total area of the Pt 4f region increases by approximately 4.9 times after the second half-cycle, which, according to the postulated reaction mechanism, should clean off the residual ligands from the surface. In addition, the PtO contribution is slightly more intense after the O $2$ pulse in the second half-cycle.

The C 1s intensity before deposition is rather small, cf. Fig. 2(b). It can be deconvoluted into three different contributions at binding energies of approximately 283.2, 284.6, and 286.3 eV. The higher-binding energy peaks are most likely due to adventitious carbon, while the first one could be some form of carbide, as it is quite low in binding energy. After the first half-cycle, the intensity of the C 1s line increases significantly. The new spectrum comprises two peaks at 284.2 and 284.8 eV. These energies are in line with values expected for double- and single-bonded carbon. Here, the C 1s peak assignment is made based on typical values found for graphitic materials, where the single- (C–C) and double-bonded (C=C) C 1s binding energy differs by about 0.5 eV.⁴² The overall central binding energy of the peak is more toward the lower-binding energy side, which indicates that the peak has components from double-bonded carbon (i.e., unsaturated hydrocarbon). After the second half-cycle, the entire carbon intensity decreases again to a level which is slightly higher than in the initial spectrum, measured on the bare Si oxide before any deposition. While the O $2$ half-cycle should clear the surface of any carbon containing MeCpPtMe $3$ precursor ligands, in this case it appears that some intensity was left on the peak at 284.5 eV, perhaps indicating incomplete removal of the ligands. A remote likelihood also exists for the addition of different contamination carbon during the two half-cycles, which is seen as an increase of the adventitious carbon peak. However, for the O atoms to clean the surface more effectively, O $2$ needs to dissociate to a more reactive form, which in this case is catalyzed by the created (metallic) Pt film. After the O $2$ pulse, the C 1s line consists of just two components at 284.5 and 286.0 eV. The overall shape of the C 1s line is similar in the initial Si oxide and after the second half-cycle, with the exception of slightly higher intensity of the peak at 284.5 eV. This indicates that the initial contamination carbon is still present on the Si atoms and was not removed during the first or second half-cycles, which only affected the Pt-bound carbon. Thus, any carbon species related to MeCpPtMe $3$ that were on the surface after the MeCpPtMe $3$ half-cycle were mostly removed by the reactive oxygen.

The O 1s spectrum on the initial Si oxide of substrate (before the first half-cycle) shows a single broad peak centered at about 532.4 eV, due to Si oxide. After the first MeCpPtMe $3$ pulse, the peak intensity has decreased drastically, similar to that of the Si 2p spectrum. After the subsequent O $2$ pulse, the largest O 1s peak has shifted toward lower binding energies by about 0.1 eV, but appears roughly equally broad. A new component has appeared at a binding energy of about 529.7 eV, in agreement with Pt oxide formation. The same feature also exists in a very small form in the spectrum take after the MeCpPtMe $3$ pulse on the higher binding energy side. The feature is too small to be able to fit accurately due to statistical noise, but it appears at a binding energy of approximately 530.9 eV, which would indicate it is, for example, surface oxygen atoms. The feature is tentatively attributed to oxygen migrated from the native Si oxide.

The Si 2p core level initially shows the bulk and oxide Si contributions. After the first half-cycle, the Si signal intensity is significantly reduced, but it appears again at rather high intensity after the O $2$ half-cycle.

The valence-band region is shown in Fig. 2(e). The spectra are normalized to integrated intensities in order to show relative peak changes. A characteristic SiO $2$ surface VB structure is initially seen, which then clearly evolves into broader features after the first half-cycle. The states near the Fermi edge grow significantly after the MeCpPtMe $3$ half-cycle due to the more metallic nature (instead of a semiconductor Si-like) of the deposited Pt. The Pt 5d peak should appear at about 3.9 eV, but overlaps heavily with contributions from oxygen and carbon valence states.⁴³ The combined effect of these various valence contributions makes analysis of the resulting shape difficult. After the O $2$ half-cycle, the valence-band shape changes significantly: The density of states near the Fermi edge and at binding energy region 8–5 eV increases, most likely due to the removal some of the C species. The Ar 3p peak from the carrier gas (which was present during the acquisition of the non-time-resolved spectra) is seen with different binding energies in the region 12–10 eV due to the changing work function.

The basic reaction mechanisms of the initial stage of noble metal deposition propose an intriguing question about the exact role of the precursor and its attachment mechanisms. The noble metal film created in the Pt process catalyzes the dissociation of $O 2$ into atomic oxygen, which is highly reactive toward breaking the ligands. However, $O 2$-based noble metal film growth commonly experiences some nucleation delay in the first cycles. This delay could be caused by a reaction where the first MeCpPtMe $3$ molecules adsorb on the surface but need to undergo some minor decomposition to a more metallic state, thereby allowing further cycles to continue the growth in a manner described by the common mechanisms. Indeed, the results here from the time-resolved measurement on the very first half-cycle may point to a slow reduction of the Pt atoms, which could be evidence for the decomposition of the MeCpPtMe $3$ molecules on the surface of the Si oxide.^44–49 

On the other hand, as pointed out earlier, the lowering of the Pt binding energy also happens as a result of clusterization of the atoms. Whether the processes of clusterization and ligand decomposition are linked or not remains unclear, but it is evident that this process of Pt reduction is on-going while the MeCpPtMe $3$ ligands attach to the surface over the long first pulse. Either of these mechanisms could contribute to the well-known nucleation delay effect.

It is interesting that Pt reduction has a timescale which is much faster than the surface saturation. The reduction of the Pt on the surface also has an effect on the surface work function, which we see from the shifting of the gas-phase Ar peak. As this shifting is also much quicker than the saturation of the surface (or at least the growth of the Pt on the surface) and has roughly the same timescale (⁠ $τ$ in the order of 1100–1500 s), it could indicate that the surface work function change and Pt reduction are correlated.⁵⁰ 

It should also be explicitly pointed out that the timescale in this experiment is orders of magnitude longer than in any typical ALD process. Usually, precursor doses are much smaller and do not necessarily allow for enough time for potential thermal decomposition reactions of this timescale to take place. This could be a reason why, in the typical ALD cases, they then affect several subsequent cycles during the early stages of the deposition. Here, we purposefully made a very long first pulse and observed the signal from the Pt 4f core level until we were sufficiently sure that the surface saturation was finished based on the intensity of the Pt 4f lineshape with respect to the underlying Si signal. Due to the very long first pulse and the fact that the surface changes during the pulse, the effects could be more accurately described as chemical vapor deposition (CVD) reactions instead. The long timescale also resulted in a rather thick surface layer. Assuming an average thickness of MeCpPtMe $3$ molecule, the surface is estimated to be covered by approximately 0.75 molecular layers. This assumption was calculated based on the size of a full MeCpPtMe $3$ molecule, and therefore a more realistic coverage is likely to be slightly higher. More details on the coverage calculations are given in the supplementary material.

Nevertheless, the metallic nature of the film is evident also from the VB measurements, in addition to the Pt 4f core level. A clear Fermi level is formed already after the MeCpPtMe $3$ dose, and it becomes even clearer after the subsequent O $2$ dose. The fact that the surface appears to have bulk-like Pt very early on the deposition is evident from the results. As the amount is also significant after the first half-cycle has finished, and it being metallic, Pt must, in principle, have a catalytic nature.

The results show that the resulting film after the long MeCpPtMe $3$ and O $2$ doses is mostly metallic. However, some small amounts of surface oxide is present. Comparison to Pt surface science studies⁴⁰ does reveal that the film is mostly metallic according to the Pt 4f binding energies and the relatively strong intensity of the Fermi edge. There is some amount of Pt oxide even after the MeCpPtMe $3$ half-cycle, according to the Pt 4f region fit. The O 1s spectrum measured at this condition has much more noise in it, and it has been fitted with only one component corresponding to the Si oxide component. It is possible to add another component to the lower binding energy side representing the Pt oxide. However, due to the moderate quality of this spectrum, such fit is not presented in this analysis.

The absence of any Pt-C or Pt-Si components in any spectra is also notable. Any of those components would typically appear at higher binding energies than the metallic lineshape⁵¹ in Pt 4f, but through fitting procedure no such features can be identified, except Pt oxide. Surface components are also curiously not seen on the Pt 4f spectra. This could be notable as the surface appears more bulk-like even from the very start. In earlier works, the form of the carbon on the surface has remained unclear. Several studies^44,49 have found evidence of dehydrogenation reactions forming long-chained amorphous carbon. Recently, temperature programed desorption experiments¹⁷ found that the ligands mostly compose of C $x$H $y$ species, which are released from the surface subsequent to the $O 2$ pulse.

We have studied the first cycle of the Pt ALD on silicon using MeCpPtMe $3$ precursor and $O 2$ as the coreactant. The surface reactions were investigated using time-resolved APXPS, making it possible to obtain surface XP spectra during the ongoing deposition. In addition, high resolution XP spectra were measured after each half-cycle where we have obtained detailed information on the chemical state of the surface atoms after each half-cycle. The ALD process was investigated over the first two half-cycles. The surface exhibits changes during the Pt precursor dose which indicate more CVD-like growth than what could be expected from typical self-limiting deposition.

The time-resolved data from the Pt 4f signal on the surface reveal a gradual shifting of the peak toward lower binding energy, which indicates a slight reduction of the Pt atoms during the initial stage of the MeCpPtMe $3$ pulse. This reduction is seen to occur at a different time scale than the growth of Pt on the surface, indicating dynamic reactions over the time span of the first half-cycle. At the same time, the work function of the surface increases as is seen from the shifted binding energy of the gas phase Ar 3s peak. These results might help explain the commonly observed delay in the initial stage of noble metal deposition.

The length of the first MeCpPtMe $3$ pulse was extremely long compared to values in typical ALD processes. This resulted in a surface which contained metallic Pt already after the first half-cycle. The metallicity is evident from the Pt 4f binding energy as well as from a clearly formed Fermi edge. The subsequent O $2$ half-cycle therefore has available Pt reaction sites which decompose O $2$ into reactive oxygen species that will further break off some of the ligands from the surface.

The high-resolution C 1s XP spectra after the first MeCpPtMe $3$ half-cycle indicates a surface which contains a rather high amount of unsaturated carbon compounds in addition to saturated carbon in C-C and C-H environments. The carbon is, however, efficiently removed during the $O 2$ pulse in the subsequent half-cycle.

The supplementary material contains further information about the time-resolved experiment including the results from the C 1s and Si 2p regions and more details on the fitting results for the Ar 3s position, O 2s area, and O 2s position changes during the MeCpPtMe 3 pulse. The supplementary material contains an example spectrum extracted from Fig. 1(b) at a time 1236 s which shows the fit used to obtain the positions and areas of the O 2s and Ar 3s peaks. In addition, the supplementary material contains the fitting results for all the spectra shown in Fig. 2 with the detailed fitting parameters explained in their respective tables. The supplementary material also contains further information about the surface molecular coverage calculations.

The ALD cell is a result of collaboration between the University of Helsinki (Finland) and the MAX IV Laboratory and funded by the Faculty of Science, University of Helsinki and Academy of Finland (Grant No. 295696) under the operations collaboration agreement between Finland (FIMAX consortium) and the MAX IV Laboratory. We acknowledge the MAX IV Laboratory for beamtime on the SPECIES beamline under Proposal No. 20200020. Research conducted at MAX IV is supported by the Vetenskapsrådet (Swedish Research Council, VR) under Contract No. 2018-07152, Vinnova (Swedish Governmental Agency for Innovation Systems) under Contract No. 2018-04969, and Formas under Contract No. 2019-02496. J.S. acknowledges project support from Vetenskapsrådet under Grant No. 2023-03492. M.P. acknowledges funding from the Jane and Aatos Erkko Foundation by the project “Novel materials for energy efficient microelectronics.” The authors wish to thank Mikko Kaipio for help during the experiments.

The authors have no conflicts to disclose.

E. Kokkonen: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Software (lead); Validation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (lead). H.-E. Nieminen: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Resources (equal); Validation (equal); Writing – review & editing (equal). F. Rehman: Investigation (equal). V. Miikkulainen: Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – review & editing (equal). M. Putkonen: Conceptualization (equal); Investigation (equal); Supervision (equal); Writing – review & editing (equal). M. Ritala: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Project administration (equal); Resources (equal); Supervision (equal); Writing – review & editing (equal). S. Huotari: Funding acquisition (lead); Project administration (equal); Supervision (equal); Writing – review & editing (equal). J. Schnadt: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Project administration (equal); Supervision (equal); Writing – review & editing (equal). S. Urpelainen: Funding acquisition (lead); Investigation (equal); Project administration (equal); Supervision (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.