Adsorption and Reaction Mechanisms of Direct Palladium Synthesis by ALD Using Pd(hfac) 2 and Ozone on Si (100) Surface

: Palladium nanoparticles made by atomic layer deposition (ALD) normally involve formaldehyde or H 2 as a reducing agent. Since formaldehyde is toxic and H 2 is explosive, it is advantageous to remove this reducing step during the fabrication of palladium metal by ALD. In this work we have successfully used Pd(hfac) 2 and ozone directly to prepare palladium nanoparticles, without the use of reducing or annealing agents. Density functional theory (DFT) was employed to explore the reaction mechanisms of palladium metal formation in this process. DFT results show that Pd(hfac) 2 dissociatively chemisorbed to form Pd(hfac)* and hfac* on the Si (100) surface. Subsequently, an O atom of the ozone could cleave the C–C bond of Pd(hfac)* to form Pd* with a low activation barrier of 0.46 eV. An O atom of the ozone could also be inserted into the hfac* to form Pd(hfac-O)* with a lower activation barrier of 0.29 eV. With more ozone, the C–C bond of Pd(hfac-O)* could be broken to produce Pd* with an activation barrier of 0.42 eV. The ozone could also chemisorb on the Pd atom of Pd(hfac-O)* to form O 3 -Pd(hfac-O)*, which could separate into O-Pd(hfac-O)* with a high activation barrier of 0.83 eV. Besides, the activation barrier was 0.64 eV for Pd* that was directly oxidized to PdOx by ozone. Based on activation barriers from DFT calculations, it was possible to prepare palladium without reducing steps when ALD conditions were carefully controlled, especially the ozone parameters, as shown by our experimental results. The mechanisms of this approach could be used to prepare other noble metals by ALD without reducing/annealing agents. planarity to improve the adsorption stability and was associated with the dissociative adsorption of Pd(hfac) 2 . Pd atom then bonded with two neighboring Si atoms to annihilate the dangling bonds. The lengths of the Pd–Si bonds were 2.22~2.28 Å, and the populations were estimated to be 0.20~0.23, which implied strong bonds. The Pd–O bonds had been weakened, as bond lengths increased to 2.23~2.45 Å.


Introduction
Atomic layer deposition (ALD) is widely used in many fields, owing to the fact that the large and complex 3D film surfaces can be used to grow nanoparticles uniformally and conformally over them. This can be done with precise control of the film thickness and nanoparticle size [1][2][3], where the films and nanoparticles chemisorb on the substrate surface with a strong binding. Many noble metals at the nanoscale have been synthesized successfully by ALD [4][5][6][7][8]. Among them, palladium has a wide range of applications at the nanoscale for catalysts [9,10], hydrogen storage [11][12][13], and sensors [14][15][16]. It is desirable in most of the applications to prepare uniform and conformal palladium particles on complex surfaces to improve palladium content per unit volume [17][18][19]. Elam et al. [20] have reported a method to nucleate and grow palladium directly on the surface of Al 2 O 3 using Pd(hfac) 2 (hfac = hexafluoroacetylacetone) and formalin (the aqueous solution of formaldehyde) by ALD. Weber et al. [21] have prepared palladium particles with high purity and low resistivity at 100~200 • C using Pd(hfac) 2 and H 2 plasma followed by an O 2 plasma step, where the O 2 plasma step was necessary to remove hydrocarbon fragments on the surface that remained after the H 2 plasma step. Since formaldehyde is toxic, and hydrogen gas is explosive, as well as plasma ALD equipment being expensive, an alternative method to prepare palladium particles by ALD is desirable.
Ozone has been used as a co-reactant to prepare noble metals by ALD. Dendooven et al. [22] grew thin platinum films using ALD at 100-300 • C by (methylcyclopentadienyl)trimethylplatinum and ozone. They found that the platinum film was uniform with low impurity levels and close-to-bulk resistivity. Hämäläinen et al. [23] reported that thin amorphous platinum oxide films were deposited at 120 and 130 • C using ALD by platinum(II) bis(acetylacetonate) and ozone, while metallic platinum films were obtained at 140 • C and above. By changing the oxygen flow, Park et al. [24] controlled the formation of Ru and RuO 2 using ALD by bis (cyclopentadienyl) ruthenium and oxygen. Therefore, ozone might be a good candidate to prepare palladium particles with Pd(hfac) 2 without reducing steps. In our previous work [25], palladium metal was prepared using Pd(hfac) 2 and ozone by ALD. However, the reaction mechanisms between Pd(hfac) 2 and ozone on the surface of the substrate were still not clear.
In this work, calculations based on density functional theory (DFT) were carried out to investigate the precursor adsorption and reaction mechanisms of direct palladium synthesis by Pd(hfac) 2 and ozone on an Si (100) surface; palladium was synthesized successfully on the Si (100) surface. Pd(hfac) 2 precursors first chemisorb on the Si (100) surface, which is accompanied by its dissociation to form Pd(hfac)* and hfac*. Pd* could be obtained through a low activation barrier of 0.46 eV by an O atom of the ozone by breaking the C-C bond of Pd(hfac)*. Pd(hfac-O)* could also be produced through a lower activation barrier of 0.29 eV, where an O atom of the ozone was substituted for the O atom of the Pd-O-C bonds in Pd(hfac)*. Pd(hfac-O)* could further react with an ozone molecule to form Pd* with an activation barrier of 0.42 eV, where an O atom of the ozone could cleave the C-C bond of Pd(hfac-O)*. An ozone molecule could also be directly absorbed on Pd(hfac)*, which could also occur to form O 3 -Pd(hfac-O)*, which was reversible and barrierless. The dissociation of O 3 -Pd(hfac-O)* into O-Pd(hfac-O)* and O 2 had a high activation barrier of 0.83 eV, where O-Pd(hfac-O)* could lead to PdO x eventually. Another possible pathway to form PdO x was the oxidation of the fabricated Pd* by ozone, which also had a high activation barrier of 0.64 eV. Therefore, the preparation of Pd metal through ALD could be realized by Pd(hfac) 2 and ozone with controlled conditions, which was also confirmed by our experiment.

Model
First-principal DFT calculations were used. The Vienna Ab Initio Simulation Package [26,27] (VASP) was used to execute the projector augmented wave (PAW) method [28,29] and generalized gradient approximation (GGA) [30]. Up to 400eV of energy cutoff was employed for the plane-wave basis set. The absolute value of force on each unconstrained atom <0.001 eV/Å was carried out for geometry optimization. The Brillouin zone was sampled using a Monkhorst-Pack 2 × 2 × 1 k-point mesh [31]. The nudged elastic band (NEB) method was used to obtain the transition-state structure to simulate the minimum-energy profile along the reaction pathway connecting the reactant and product [32,33].

Experiment
Si (100) substrates (10 mm × 10 mm × 0.2 mm) were washed sequentially under ultrasonication in ethanol, acetone, and iso-propyl alcohol (IPA), then dried with nitrogen flow. Subsequently, Si (100) substrates were placed in the ALD reactor. Nitrogen (99.999%) was used as a carrier and as a purging gas. The ALD deposition was carried out in a low-pressure ALD reactor under a nitrogen pressure of about 1 mbar. Pd(hfac)2 (Strem Chemicals) and ozone gas were added sequentially into the ALD reactor heated at 200 °C. The precursor was heated to 60 °C. The transfer line connecting the bubbler to the ALD reactor was heated to 100 °C to avoid condensation of the precursor gas. The ALD cycle

Experiment
Si (100) substrates (10 mm × 10 mm × 0.2 mm) were washed sequentially under ultrasonication in ethanol, acetone, and iso-propyl alcohol (IPA), then dried with nitrogen flow. Subsequently, Si (100) substrates were placed in the ALD reactor. Nitrogen (99.999%) was used as a carrier and as a purging gas. The ALD deposition was carried out in a low-pressure ALD reactor under a nitrogen pressure of about 1 mbar. Pd(hfac) 2 (Strem Chemicals) and ozone gas were added sequentially into the ALD reactor heated at 200 • C. The precursor was heated to 60 • C. The transfer line connecting the bubbler to the ALD reactor was heated to 100 • C to avoid condensation of the precursor gas. The ALD cycle consisted of two 0.4 s pulses of Pd(hfac) 2 and a 5 s purge, followed by one 0.2 s pulse of ozone and a 5 s purge. A total of 1000 cycles were used to prepare the palladium particles. Ozone with the concentration of 0.87 g m −3 was used, lower than that in our previous work [25]. The crystalline structure of the particles was estimated using grazing incidence X-ray diffraction (GIXRD). The data were collected with a MAR345 2D imaging detector using a 12.7 KeV X-ray source with an incident angle of 1 • . We also measured the GIXRD pattern of the substrates by the same way in order to subtract the diffraction from the substrates around the coated areas. The elemental composition and chemical bonding structure of the resulted particles were measured by X-ray photoelectron spectroscopy (XPS, Axis supra) using an Al Kα source. Ar ion sputtering was carried out to remove the topmost impurities. The morphology of the particles was analyzed by scanning electron microscopy (SEM, JEOL JSM-7600F).

Chemical Adsorption of Pd(hfac) 2 on the Clean Si (100) Surface
In a typical ALD process, the precursors are first chemisorbed on the reactive sites, resulting in the release of by-products. Figure 1c shows the optimized structure of Pd(hfac) 2 . Pd coordinated four oxygens of two hfac ligands. The length of the covalent Pd-O bonds was 2.00 Å and the mean angle of O-Pd-O was 92.0 • . Bond lengths for C1-C2 and C2-C3 were 1.54 and 1.39 Å respectively, these are shown in Figure 1c. The results were consistent with experimental and DFT results that have been reported [37,38].
DFT calculation was used to explain the adsorption and reaction mechanisms of Pd(hfac) 2 on the clean Si (100) surface. Mulliken bond populations with 3.0 Å distance cut-offs were then carried out to assess the relative strength of the bonds; higher positive bond population value implied a stronger bond strength. The B2 site had the most stable configuration for the adsorption of Pd(hfac) 2 on the Si (100) surface, with an adsorption energy of −2.79 eV, as shown in Figure 2a. The Pd(hfac) 2 * was significantly distorted from planarity to improve the adsorption stability and was associated with the dissociative adsorption of Pd(hfac) 2 . Pd atom then bonded with two neighboring Si atoms to annihilate the dangling bonds. The lengths of the Pd-Si bonds were 2.22~2.28 Å, and the populations were estimated to be 0.20~0.23, which implied strong bonds. The Pd-O bonds had been weakened, as bond lengths increased to 2.23~2.45 Å.
ozone and a 5 s purge. A total of 1000 cycles were used to prepare the palladium particles. Ozone with the concentration of 0.87 g m −3 was used, lower than that in our previous work [25].
The crystalline structure of the particles was estimated using grazing incidence X-ray diffraction (GIXRD). The data were collected with a MAR345 2D imaging detector using a 12.7 KeV X-ray source with an incident angle of 1°. We also measured the GIXRD pattern of the substrates by the same way in order to subtract the diffraction from the substrates around the coated areas. The elemental composition and chemical bonding structure of the resulted particles were measured by X-ray photoelectron spectroscopy (XPS, Axis supra) using an Al Kα source. Ar ion sputtering was carried out to remove the topmost impurities. The morphology of the particles was analyzed by scanning electron microscopy (SEM, JEOL JSM-7600F).

Chemical Adsorption of Pd(hfac)2 on the Clean Si (100) Surface
In a typical ALD process, the precursors are first chemisorbed on the reactive sites, resulting in the release of by-products. Figure 1c shows the optimized structure of Pd(hfac)2. Pd coordinated four oxygens of two hfac ligands. The length of the covalent Pd-O bonds was 2.00 Å and the mean angle of O-Pd-O was 92.0°. Bond lengths for C1-C2 and C2-C3 were 1.54 and 1.39 Å respectively, these are shown in Figure 1c. The results were consistent with experimental and DFT results that have been reported [37,38].
DFT calculation was used to explain the adsorption and reaction mechanisms of Pd(hfac)2 on the clean Si (100) surface. Mulliken bond populations with 3.0 Å distance cutoffs were then carried out to assess the relative strength of the bonds; higher positive bond population value implied a stronger bond strength. The B2 site had the most stable configuration for the adsorption of Pd(hfac)2 on the Si (100) surface, with an adsorption energy of −2.79 eV, as shown in Figure 2a. The Pd(hfac)2* was significantly distorted from planarity to improve the adsorption stability and was associated with the dissociative adsorption of Pd(hfac)2. Pd atom then bonded with two neighboring Si atoms to annihilate the dangling bonds. The lengths of the Pd-Si bonds were 2.22~2.28 Å, and the populations were estimated to be 0.20~0.23, which implied strong bonds. The Pd-O bonds had been weakened, as bond lengths increased to 2.23~2.45 Å.  The reaction energy for Reaction (1) was −1.87 eV. DFT calculations performed showed that Pd(hfac) 2 * dissociated to Pd(hfac)* and hfac*. The optimized structures are shown in Figure 2b,c. The lowest adsorption energy of Pd(hfac) on the clean Si (100) surface was −4.08 eV. After Pd(hfac) was adsorped at the B2 site, two Pd-Si bonds were formed. The coordination configuration of Pd was preserved with a coordination number of 3. Figure 2c shows the most stable structure of hfac adsorption on the Si. Hfac preferentially adsorbed with the perpendicular species axis at the B1 site with an adsorption energy of −4.21 eV. The two carbonyl groups of hfac* adsorbed on two neighboring Si atoms.

Surface Reaction between Pd(hfac) and Ozone
Ligands are expected to remain intact upon adsorption so they can subsequently volatilize the metal atoms to be removed by the co-reactant. The reaction between Pd(hfac)* and ozone was studied since Pd(hfac) 2 * dissociated to produce Pd(hfac)* and hfac*. Figures 3 and 4 show the potential energy profiles and optimized reactant, transition state, and product structures. There were two reaction pathways as the ozone approached (step 1, Figure 3). One activation barrier of the reaction between the ozone and Pd(hfac)* was 0.46 eV (green lines). The relaxed adsorption structure with adsorption energy −3.01 eV is shown in Figure 4. The O atom of the ozone adsorbed above the bridging C2-C3 site (as marked in Figure 1c) to form an epoxide (the inset in Figure 4) and then underwent a dissociative chemisorption process. The O-C3 bond and oxygen molecule were formed on the surface (the transition state 1 inset in Figure 4). This was consistent with DFT calculations and infrared spectroscopy of the reaction between trimethylaluminum and ozone [39,40] The distance between the O1 (as marked in Figure 1c) and Pd atoms of Pd(hfac)* increased from 2.12 Å to 4.09 Å as the C2-C3 bond was broken. This indicated that the O1-Pd bond was cleaved. As a result, the formed complex dissociated into the adsorbed Pd, CF 3 -OC, and CF 3 -CO-CHO (referred to as "Pd*", "CF 3 -OC*", and "CF 3 -CO-CHO*", respectively). Gaseous fragments of the ligand could be formed. C atoms (except for the C atom of -CF 3  preferentially adsorbed with the perpendicular species axis at the B1 site with an adsorption energy of −4.21 eV. The two carbonyl groups of hfac* adsorbed on two neighboring Si atoms.

Surface Reaction between Pd(hfac) and Ozone
Ligands are expected to remain intact upon adsorption so they can subsequently volatilize the metal atoms to be removed by the co-reactant. The reaction between Pd(hfac)* and ozone was studied since Pd(hfac)2* dissociated to produce Pd(hfac)* and hfac*. Figure 3 and 4 shows the potential energy profiles and optimized reactant, transition state, and product structures. There were two reaction pathways as the ozone approached (step 1, Figure 3). One activation barrier of the reaction between the ozone and Pd(hfac)* was 0.46 eV (green lines). The relaxed adsorption structure with adsorption energy −3.01 eV is shown in Figure 4. The O atom of the ozone adsorbed above the bridging C2-C3 site (as marked in Figure 1c) to form an epoxide (the inset in Figure 4) and then underwent a dissociative chemisorption process. The O-C3 bond and oxygen molecule were formed on the surface (the transition state 1 inset in Figure 4). This was consistent with DFT calculations and infrared spectroscopy of the reaction between trimethylaluminum and ozone [39,40] The distance between the O1 (as marked in Figure  1c) and Pd atoms of Pd(hfac)* increased from 2.12 Å to 4.09 Å as the C2-C3 bond was broken. This indicated that the O1-Pd bond was cleaved. As a result, the formed complex dissociated into the adsorbed Pd, CF3-OC, and CF3-CO-CHO (referred to as "Pd*", "CF3-OC*", and "CF3-CO-CHO*", respectively). Gaseous fragments of the ligand could be formed. C atoms (except for the C atom of -CF3) might have been oxidized into CO2 and H atoms might have been oxidized to form H2O [21]. The other reaction between the ozone and Pd(hfac)* at the toping Pd site hadw a 0.29 eV activation barrier, thus was kinetically favorable (black lines). The O in the ozone was reactive towards the Pd in Pd(hfac)*. The distance between the O1 and the Pd atom then increased from 2.12 Å to 4.50 Å, suggesting that the O1-Pd bond was split. The O1-Si bond was formed from the transfer of the residual electrons on the O to the topmost Si The other reaction between the ozone and Pd(hfac)* at the toping Pd site hadw a 0.29 eV activation barrier, thus was kinetically favorable (black lines). The O in the ozone was reactive towards the Pd in Pd(hfac)*. The distance between the O1 and the Pd atom then increased from 2.12 Å to 4.50 Å, suggesting that the O1-Pd bond was split. The O1-Si bond was formed from the transfer of the residual electrons on the O to the topmost Si atom. The adsorbed O of the ozone bonds with the C2 atom also caused the chemically adsorbed Pd(hfac-O) (referred to as "Pd(hfac-O)*") and oxygen formation, through an adsorption energy of −3.46 eV.  There were two reaction pathways as another ozone approached the Pd(hfac-O) (step 2, Figure 3). Figure 5 shows the optimized reactant, transition state, and produ structures. The reaction between ozone and Pd(hfac)-O* at the C-C bridging positio could have formed a stable structure with an activation barrier of 0.42 eV and adsorptio energy of −3.14 eV, shown in Figure 3 (orange lines) and Figure 5, which was similar the reaction described in Figure 3 (green lines) and Figure 4. The final products were Pd adsorbed CF3-CO2, and CF3-CO-CHO*. The latter two might then have been oxidized form CO2, H2O, and gaseous fluoride. There were two reaction pathways as another ozone approached the Pd(hfac-O)*, (step 2, Figure 3). Figure 5 shows the optimized reactant, transition state, and product structures. The reaction between ozone and Pd(hfac)-O* at the C-C bridging position could have formed a stable structure with an activation barrier of 0.42 eV and adsorption energy of −3.14 eV, shown in Figure 3 (orange lines) and Figure 5, which was similar to the reaction described in Figure 3 (green lines) and Figure 4. The final products were Pd*, adsorbed CF 3 -CO 2 , and CF 3 -CO-CHO*. The latter two might then have been oxidized to form CO 2 , H 2 O, and gaseous fluoride. implying that further decomposition was favorable. The activation barrier of decomposition reaction was 0.83 eV (gray lines), almost twice as high as that of reaction between ozone and Pd(hfac-O)* displayed in Figure 3 (orange lines). Thus, it more difficult to produce PdO than Pd, resulting in less PdO in the Pd metal, which consistent with the previous report [25]. The relaxed adsorption structure is show Figure 5 (O-Pd(hfac)*+O2). An O atom of ozone adsorbed at the toping Pd site and t an oxygen molecule was produced (the transition state 3 inset in Figure 5), wh underwent a dissociative chemisorption process. The distance between the adsorbe and Pd atoms decreased from 2.14 Å to 2.00 Å as the oxygen molecule formed. Besi since Si was a tetravalent species and the Si atom bonding with the Pd atom had dangling bond left, the Si atom bonds with the adsorbed an O atom to improve structural stability. Hence, the formed complex dissociated into the adsorbed O-Pd(h Since ozone prefers to be broken down into an oxygen molecule after the atomic O adsorption, the decomposition of O 3 -Pd(hfac-O)* was assumed (reaction (4)). The reaction energy for Reaction (4) could be written as: The obtained reaction energy was −2.06 eV, implying that further decomposition was favorable. The activation barrier of the decomposition reaction was 0.83 eV (gray lines), almost twice as high as that of the reaction between ozone and Pd(hfac-O)* displayed in Figure 3 (orange lines). Thus, it was more difficult to produce PdO than Pd, resulting in less PdO in the Pd metal, which was consistent with the previous report [25]. The relaxed adsorption structure is shown in Figure 5 (O-Pd(hfac)* + O 2 ). An O atom of ozone adsorbed at the toping Pd site and then an oxygen molecule was produced (the transition state 3 inset in Figure 5), which underwent a dissociative chemisorption process. The distance between the adsorbed O and Pd atoms decreased from 2.14 Å to 2.00 Å as the oxygen molecule formed. Besides, since Si was a tetravalent species and the Si atom bonding with the Pd atom had one dangling bond left, the Si atom bonds with the adsorbed an O atom to improve its structural stability. Hence, the formed complex dissociated into the adsorbed O-Pd(hfac) (referred to as "O-Pd(hfac)*") and an oxygen molecule.
Atomic Pd moves might have occurred as the ligand was consumed. The most stable structure of atomic Pd adsorption on the Si surface is shown in Figure 6. Pd atom preferred to adsorb at the H1 site with an adsorption energy of −0.21 eV, as it was the most efficient annihilation of the dangling bonds.
Processes 2021, 9, x FOR PEER REVIEW 8 of 13 to adsorb at the H1 site with an adsorption energy of −0.21 eV, as it was the most efficient annihilation of the dangling bonds. Excess ozone might have reacted with Pd* directly with an activation barrier of 0.64 eV (Figure 7), lower than the activation barrier of the decomposition of O3-Pd(hfac-O)* (gray lines in Figure 3). Thus, the reaction between Pd* and the ozone was also an important reason for the generation of PdO. Figure 7 shows the optimized reactant, transition state, and product structures. The O atom of the ozone, located at the bridging Pd-Si site, preferentially bonded with the Pd and Si atoms to terminate two surfacedangling bonds, causing the formation of Pd-O-Si. Meanwhile, the distance between the O2 and O3 atoms (marked in Figure 7) increased from 1.28 Å to 2.55 Å, indicating that the O2-O3 bond was broken and an oxygen molecule was formed. The final adsorption energy was −2.64 eV. Hence, excess ozone caused the formation of PdO.  Figure 3). Thus, the reaction between Pd* and the ozone was also an important reason for the generation of PdO. Figure 7 shows the optimized reactant, transition state, and product structures. The O atom of the ozone, located at the bridging Pd-Si site, preferentially bonded with the Pd and Si atoms to terminate two surface-dangling bonds, causing the formation of Pd-O-Si. Meanwhile, the distance between the O 2 and O 3 atoms (marked in Figure 7 In conclusion, palladium metal can be synthesized when the ozone concentration is carefully controlled. Palladium metal should be prepared with less ozone. In conclusion, palladium metal can be synthesized when the ozone concentration is carefully controlled. Palladium metal should be prepared with less ozone.

Experiment Results
The adsorption and dissociation of Pd(hfac) 2 on the hydroxylated Si (100) surface was similar to those on clean Si (100) surfaces. This is because the OH* group is not the adsorption site of Pd(hfac) 2 according to a study on the effect of substrates' surface conditions on adsorption and reaction mechanisms of Pd(hfac) 2 on Si (100) surfaces, as described in the Supporting Information. Hence, palladium metal can be prepared on the Si (100) surface without specific surface treatment (e.g., heat treatment or H 2 SO 4 /H 2 O 2 mixture) by ALD. Thus, Si (100) substrates cleaned by ethanol, acetone, and IPA were used. Figure 8a presents a typical SEM surface image of the particles prepared on the Si (100) surface. The particles' diameters were 10~20 nm, indicating that palladium nucleated as discrete islands and grew laterally with increased ALD cycles. The GIXRD pattern (bottomleft corner in Figure 8a) shows that the particles had the face-centered cubic (fcc) crystal structure. Meanwhile, Pd (111), Pd (200), Pd (202), and Pd (311) planes were observed, which was consistent with the literature's report [21]. Hence, palladium was grown on the Si (100) surface, which was consistent with the simulation results. To further assess the chemical state of the palladium particles, XPS was carried out. Typical XPS spectra of the Pd 3d5/2, Pd 3d3/2, and Pd 3p3/2 levels for the particles on the Si (100) surface are shown in Figure 8b,c. The peak positions were referenced to carbon at 284.8 eV. Curve fitting analysis showed that the Pd 3d spectra resulted from two pairs of spin-orbit components. The Pd 3d5/2 peak at 335.6 eV had been attributed to metallic palladium (Pd(0)) [41] (Figure 8b). The second Pd 3d5/2 signal presenting at 336.4 eV was due to palladium atoms with lower charge densities. The origin of such palladium species might have been due to a perturbation occurring during the dispersion process of palladium clusters [42]. For comparison, the Pd 3d5/2 signal for Pd-O was at 337.0 eV, based on Zhou et al. [43]. Also, an O 1s signal at 529.3 eV was observed [44], which did not appear in the resulting particles (Figure 8c). Hence, it was reasonable to exclude the To further assess the chemical state of the palladium particles, XPS was carried out. Typical XPS spectra of the Pd 3d 5/2 , Pd 3d 3/2 , and Pd 3p 3/2 levels for the particles on the Si (100) surface are shown in Figure 8b,c. The peak positions were referenced to carbon at 284.8 eV. Curve fitting analysis showed that the Pd 3d spectra resulted from two pairs of spin-orbit components. The Pd 3d 5/2 peak at 335.6 eV had been attributed to metallic palladium (Pd(0)) [41] (Figure 8b). The second Pd 3d 5/2 signal presenting at 336.4 eV was due to palladium atoms with lower charge densities. The origin of such palladium species might have been due to a perturbation occurring during the dispersion process of palladium clusters [42]. For comparison, the Pd 3d 5/2 signal for Pd-O was at 337.0 eV, based on Zhou et al. [43]. Also, an O 1s signal at 529.3 eV was observed [44], which did not appear in the resulting particles (Figure 8c). Hence, it was reasonable to exclude the presence of palladium oxide, which agrees with the GIXRD spectrum.

Conclusions
DFT simulations were used to elucidate the adsorption and reaction mechanisms for the synthesis of palladium from Pd(hfac) 2 and ozone on Si. The dissociative chemical absorption of Pd(hfac) 2 produced Pd(hfac)* and hfac*. The ozone cleaves the C-C bond located at the bridging location of Pd(hfac) to produce Pd* (activation barrier 0.46 eV). With excess ozone molecules, the Pd* formation activation barrier became 0.42 eV. The ozone molecule at the toping Pd site could reversibly bond with the Pd in Pd(hfac-O)* to produce O 3 -Pd(hfac-O)* without a barrier. It could then decay to produce O-Pd(hfac-O)* and oxygen though it is kinetically less favorable (activation barrier 0.83 eV). The Pd* that was produced could further react with ozone at the Pd-Si site (activation barrier 0.64 eV). Therefore, palladium metal must be prepared by controlling the ozone condition carefully. Moreover, we prepared palladium metal successfully on the Si (100) surface without specific surface treatment (e.g., heat treatment or H 2 SO 4 /H 2 O 2 mixture) by ALD.