IP

MKa

b

c

a

ARRAA

KONIXXA

1

rfAatssTtdrg

0h

Catalysis Today 229 (2014) 2–13

Contents lists available at ScienceDirect

Catalysis Today

j o ur na l ho me page: www.elsev ier .com/ locate /ca t tod

n situ XAS and XRF study of nanoparticle nucleation during O3-basedt deposition

atthias Fileza, Hilde Poelmana,∗, Ranjith K. Ramachandranb, Jolien Dendoovenb,ilian Devloo-Casierb, Emiliano Fondac, Christophe Detavernierb, Guy B. Marina

Laboratory for Chemical Technology, Ghent University, Technologiepark 914, B-9052 Zwijnaarde, BelgiumDepartment of Solid State Sciences, COCOON, Ghent University, Krijgslaan 281/S1, B-9000 Ghent, BelgiumSynchrotron SOLEIL, L’Orme des Merisiers, Saint-Aubin, BP48, F-91192 Gif-sur-Yvette Cedex, France

r t i c l e i n f o

rticle history:eceived 30 June 2013eceived in revised form 9 January 2014ccepted 10 January 2014vailable online 12 February 2014

eywords:zone-based Pt depositionucleation

n situ characterization-ray absorption spectroscopy-ray fluorescence

a b s t r a c t

X-ray absorption spectroscopy (XAS) and X-ray fluorescence (XRF) were combined in situ to study theALD-based synthesis of Pt catalysts. This first time combination of synchrotron-based techniques wasapplied during the (methylcyclopentadienyl)trimethylplatinum/ozone deposition process executed at150 ◦C on a silica support. A nucleation delay indicative for nanoparticle formation was observed for Ptloadings below 1 equivalent monolayer. XAS and XRF were recorded simultaneously at different catalystloadings in this nucleation regime. Analysis of the combined in situ data yielded a quantitative picture ofthe evolution of the diameter, shape, lattice packing and density of the deposited Pt clusters. Additionally,the degree of oxidation at the cluster surface after the ozone pulse could be monitored. At the early start ofthe deposition process, Pt adatoms cluster together to form stable nuclei. A strong increase in the densityof nuclei is seen below 0.16 equivalent monolayers, after which coalescence gradually occurs. From0.04 to 0.71 equivalent monolayers, Pt clusters are fcc packed and correspond best to a hemispherical

tomic layer deposition (1 1 1)-truncated cuboctahedral shape. By crosslinking the XRF and XAS data, a linear increase in clusterdiameter with Pt loading is observed within this range. The surface of the Pt clusters is shown to beoxidized immediately after the ozone exposure. The degree of surface oxidation remains approximatelyconstant for clusters with a 1–3 nm diameter. This surface oxygen is shown to be crucial for furthergrowth during deposition. The combined application of in situ XRF and XAS thus allowed for an advancedidentification of the ALD-deposited Pt nanoparticles.

. Introduction

Atomic layer deposition (ALD) uses sequential self-saturatingeactions between gaseous precursor molecules and a solid sur-ace to grow thin films in a layer-by-layer fashion [1–3]. The cyclicB-type exposures make it possible to deposit material with antomic layer precision in a conformal way [4]. However, owingo nucleation difficulties, the start of metal deposition on oxideurfaces using the ALD procedure is often characterized by verymall growth rates and Volmer–Weber type island growth [3,5–9].his initial formation of metal nanoparticles opens the opportunityo use this deposition technique for the controlled atomic-scaleesign of catalysts [10–13]. Promising catalytic properties have

ecently been reported for Pd, Pt and Ru (bi)metallic nanoparticlesrown on different supports including carbon aerogels [13], SrTiO3

∗ Corresponding author. Tel.: +00 32 9 331 17 22; fax: +0032 9 264 49 99.E-mail address: hilde.poelman@ugent.be (H. Poelman).

920-5861/$ – see front matter © 2014 Elsevier B.V. All rights reserved.ttp://dx.doi.org/10.1016/j.cattod.2014.01.011

© 2014 Elsevier B.V. All rights reserved.

nanocubes [10], alumina spheres [11], and (surface-modified) silicagels [8,12].

The most common Pt deposition process utilizes alter-nating exposures to (methylcyclopentadienyl)trimethylplatinum(MeCpPtMe3) vapour and O2 gas [14,15]. During a precursor pulsein the nucleation regime, the Pt precursor may adsorb on boththe oxide support surface and the existing Pt clusters. In thenext O2 pulse, oxygen burns off the remaining ligands of thenewly adsorbed precursor molecules and is commonly believedto further oxidize the Pt cluster surface [14–16]. The chemisorbedoxygen atoms form anchoring points for the next arriving precur-sor molecules inducing combustion of the ligands. Recently, it wasshown that the MeCpPtMe3 precursor can also be combined withO2 plasma [17], N2 plasma, NH3 plasma [18] and O3 as a reactant[19]. The O3-based process enables Pt deposition in a lower tem-perature range due to the stronger oxidizing power of O3 compared

to O2. It is clear that oxygen plays a crucial role in both the O2-and O3-based Pt processes. However, the degree of oxidation of thesupported Pt clusters under in situ conditions is not yet known in aquantitative way.

sis Tod

bttcAra

di(XearmXcoiitfXfttu

2

2

cTspbtctcrt1

sSXowiof6o

ds1tabh

M. Filez et al. / Cataly

Besides knowledge on the oxidation state of the Pt clusters, aetter understanding of the mechanisms occurring during nuclea-ion is necessary for a more rational catalyst design. In particular,he metal loading and the density and diameter of the nanoparti-les are needed to reconstruct an evolutional picture of the process.dditionally, detailed structural information is required since theeactivity of Pt nanoparticles changes with both their geometry [20]nd degree of oxidation [21].

In this work, information on the diameter, shape, lattice packing,ensity and oxidation state of the Pt clusters is acquired by apply-

ng X-ray absorption spectroscopy (XAS) and X-ray fluorescenceXRF) simultaneously under in situ conditions during deposition.AS yields detailed structural and electronic information throughxtended X-ray absorption fine structure (EXAFS spectroscopy)nd X-ray absorption near edge structure (XANES spectroscopy),espectively [22,23]. XRF provides information on the amount ofaterial deposited [24]. A novel setup, allowing in situ XAS andRF during deposition, is used to record the Pt LIII-edge fluores-ence signal between MeCpPtMe3/O3 cycles. An evolutional picturef the structure and oxidation state of the deposited Pt clusterss pursued during the nucleation stage. The particular method ofn situ XRF-XAS experiments during nanoparticle deposition is,o the best of our knowledge, a first time combination. There-ore, this work is a proof-of-principle of the simultaneous in situRF-XAS application and aims to provide a stepwise procedure

or the applied methodology. In addition, a structural study ofhe O3-based Pt process was not yet performed before and ques-ions concerning the degree of cluster oxidation have remainednanswered.

. Experimental

.1. Setup and materials

A cold-wall ALD setup was designed to study deposition pro-esses using in situ XRF and in situ XAS at a synchrotron facility.he solid MeCpPtMe3 precursor (99% Strem Chemicals) was pre-erved in a stainless steel container and heated above its meltingoint of 30 ◦C. This container was connected to the vacuum cham-er by a 60 ◦C heated delivery line. The precursor was pulsedo the chamber using argon as carrier gas. The sample chamberontained a heated substrate holder and evacuated double Kap-on windows for X-ray transmission. A gate valve connected thehamber with a turbo pump, which was in turn backed up by aotary pump. An OzoneLabTM OL100 ozone generator was usedo produce O3 from pure O2. The O3 concentration obtained was75 �g/mL.

In order to minimize particle–support interaction, SiO2 was cho-en as substrate. Rather than a flat surface, a mesoporous film ofiO2 was used to increase the exposed surface area and hence the-ray fluorescence signal. Thus, all Pt depositions were performedn a thin film of mesoporous silica which was supported on a flat Siafer, i.e. SiO2/Si. The silica film had a thickness of 188 nm, a poros-

ty of 62% and an average pore diameter of 5 nm. A surface area ratiof ca. 30 cm2 of SiO2 substrate surface per cm2 of flat Si wafer sur-ace was obtained. This corresponds to a specific surface area of5 m2/g SiO2. Further details on the characteristics and synthesisf the mesoporous SiO2 film were published elsewhere [25].

Pt was grown at a substrate temperature of 150 ◦C. The firsteposition half-cycle consisted of controlled exposure of the sub-trate to the MeCpPtMe3 precursor pulse at pressures up to.6.10−1 mbar, followed by a 50 s pump time for precursor evacua-

ion to a pressure of ca. 2.10−5 mbar. Within the second half-cycle,n O3 pulse was injected into the chamber and again followedy a 50 s pump down, both at similar pressures as for the firstalf-cycle. When X-ray measurements were performed in situ, the

ay 229 (2014) 2–13 3

evacuation time after the O3 pulse was increased (Section 2.2). Nopurge gas was utilized. The pulse time for the MeCpPtMe3 precur-sor was 10 s. The gate valve connecting the chamber to the pumpingsystem was closed additionally for 5 s, yielding a total static expo-sure of 15 s. The O3 reactant gas pulse of 5 s was followed by another5 s of residence time in the chamber. Consequently, the static expo-sure acquired for O3 was 10 s.

2.2. Instrumentation and methods

In situ XRF and in situ XAS measurements were performed inbetween Pt deposition cycles at the SAMBA beam line [26] of the2.75 GeV SOLEIL synchrotron (Saint-Aubin, France) using a top-upelectron beam of 430 mA. All experiments were recorded in fluo-rescence mode at the Pt LIII edge (E = 11564 eV) using a 35-elementGe detector. XRF measurements at 12 keV were performed aftereach deposition cycle. XANES measurements were executed every5th cycle (ca. 5 min/scan), whereas several EXAFS scans were takenafter each 10th cycle (ca. 30 min/scan). All measurements wererecorded during the pump down period situated after the pulseof O3 reactant gas. Background scans were obtained before thedepositions started and all raw data signals measured during thedeposition process were corrected for this background contribu-tion.

The XRF counts measured during in situ experiments were recal-culated to units of equivalent monolayers (EML) of Pt. This numberof EML equals the number of atoms deposited on 1 nm2 of SiO2support surface divided by the number of atoms necessary to form1 monolayer of Pt on 1 nm2 of Pt bulk. The latter quantity wasderived theoretically by multiplying the expected thickness of 1monolayer of Pt deposited on Pt bulk, i.e. 0.247 nm, and the num-ber of Pt atoms residing within the volume of 1 nm2 (base) × 1 nm(height), i.e. 66.2 atom/(nm3). This results in 16.4 atoms/nm2. Thenumber of Pt atoms/nm2 of SiO2 surface deposited during the syn-chrotron experiments was determined a posteriori by performingex situ XRF measurements (Bruker Artax, Mo X-ray source, XFlash5010 silicon drift detector) on the resulting sample. The number ofXRF counts recorded for the final sample was linked with an aver-age Pt film thickness by using an established relation between thePt film thickness and the number of XRF counts. The average filmthickness obtained for the final sample was then transformed to thenumber of Pt atoms deposited/nm2. By dividing the latter quantitythrough the number of atoms necessary to form 1 monolayer of Pton 1 nm2 of Pt bulk, the number of EML was obtained. The latterwas corrected for its specific surface area. All ex situ XRF measure-ments were recalculated to number of EML by applying the samemethodology.

XAS data reduction and analysis were executed with Athena andArtemis, part of the Demeter 0.9.13 software package [27]. The pre-edge background was subtracted from the data signal. The edgeenergy E0 was chosen at the maximum of the first derivative ofthe spectrum. The isolated atom absorption background �0(E) wascalculated by a cubic spline fitting procedure, using the AUTOBKroutine. Edge step normalization to the edge jump was performedon pre-edge background subtracted spectra, yielding the XANESsignal [22]. The k2-weighted EXAFS signal was Fourier transformedto R-space to obtain the radial distribution function.

The EXAFS signal was modelled using the extended formu-lation of the EXAFS function in which the third cumulant �(3)

is incorporated in the argument of the sine function [28]. Themetallic Pt fcc structure with a lattice parameter of 3.92 A wasintroduced as starting structural model for EXAFS fitting. In addi-

tion, Pt-O neighbours at 1.92 A distance from the Pt absorber wereimplemented. The FEFF 6.0 code [29] was applied to calculate thephase shifts and backscattering amplitude functions of Pt-O andPt-Pt contributions to the EXAFS signal. For the Pt metal structure,

4 sis Today 229 (2014) 2–13

spsferrqlTfuommrffts

smtiFbwsta2(d

dDdaGc[

3

3

tdsTaccadvEfcdldE

0

1

2

3

4

5

6

7

8

9

0 20 40 60 80 100 120

Equ

ival

ent m

onol

ayer

s (E

ML

) [-

]

Cycles [-]

in situ XRFex si tu XRF

0.0

0.2

0.4

0.6

0.8

0 10 20 30 40 50

EM

L [-

]

Cycles [-]

in sit u XRF

Fig. 1. The number of equivalent monolayers (EML) of Pt deposited as a functionof the number of MeCpPtMe3/O3 cycles is shown. Both in situ and ex situ XRF dataare depicted. The dashed line proves the linearly of the linear growth regime above60 cycles. In the inset, the in situ (synchrotron) XRF data are plotted more clearly

M. Filez et al. / Cataly

pherical clusters with radius 6.0 A and maximum photoelectronath length of 6.5 A were installed. The photoelectron paths withignificant contributions to the EXAFS signal were selected andurther used for modelling. The experimental spectra were mod-lled from R = 1.5–6 A (k = 2.7–12.2 A−1). However, for the spectrumecorded at the lowest Pt loading, i.e. 0.007 EML, a restricted R-ange (R = 1.5–3 A, k = 2.7–12.2 A−1) is used due to its poorer datauality. The Levenberg–Marquardt algorithm was used for non-

inear least-squares minimization of the objective function ��2.

his minimization, carried out by IFEFFIT [27], yielded estimatesor the structural parameters by multiple shell fitting in R-spacesing multiple k-weightings. The R-factor, defined as the closenessf fit to the data [30], is used to evaluate the agreement betweenodel and experiment. All binary correlations between the esti-ated parameters were below 0.85. From the fit to the Pt foil

eference signal with bulk Pt fcc structure, an amplitude reductionactor S0

2 of 0.82 ± 0.04 was obtained. The latter value was used forurther Pt LIII edge EXAFS modelling of the signals measured duringhe MeCpPtMe3/O3 deposition process onto the mesoporous SiO2upport.

Additional ex situ structural analysis was carried out on theample in the end state of the in situ deposition by means of trans-ission electron microscopy (TEM). A specimen slice was cut out of

he sample in a dual beam scanning electron microscopy – focusedon beam (SEM-FIB) apparatus (FEI Nova 600 Nanolab with SirionEGSEM and Magnum Ion Column). After deposition of a carbonuffer and Pt protector layer for sample strength, a FIB treatmentith Ga ions was applied to obtain a 100 nm slice. The thinned

ample was mounted on a copper support grid and transferredo the TEM. A microscope JEOL JEM2200FS-Cs-corrected, operatedt 200 kV and equipped with Schottky-type FEG, EDX JEOL JED-300D and JEOL in-column omega filter, was used. High resolutionHR) TEM, scanning transmission bright field (STEM-BF) and energyispersive X-ray (EDX) analysis were applied.

Crystallographic analysis was performed by means of X-rayiffraction (XRD) measurements in �–2� mode using a Siemensiffractometer Kristalloflex D5000, with CuK� radiation. The pow-er patterns were collected in a 2� range from 15◦ to 60◦ with

step of 0.02◦ and 30 s counting time per angle. By fitting aaussian function to a diffraction peak, the crystallite diameteran be determined from the peak width via the Scherrer equation31].

. Results and discussion

.1. X-ray fluorescence (XRF) measurements

Firstly, knowledge is obtained on the evolution in Pt loading ofhe mesoporous SiO2 support with the number of MeCpPtMe3/O3eposition cycles. This is done by performing ex situ XRF mea-urements on samples with 10, 30, 50, 75, 100 and 125 cycles.he XRF intensities are plotted in terms of the number of equiv-lent monolayers (EML) of Pt (Fig. 1, ex situ XRF). Three regionsan be distinguished in the XRF growth curve. During the first 30ycles, it appears that very little material is deposited. Thereafter,

monotonic increase in growth rate is seen, implying an increasedeposition rate of Pt. This growth rate evolves towards a constantalue resulting in a linear growth regime starting from ±60 cycles (1ML). The linear growth is indicative for nanoparticle coalescenceollowed by film closure [32]. Because the formation of nanoparti-les is of particular interest from a catalysis point of view, the focus

uring the in situ synchrotron experiments will be on the range of

oadings below the onset of the linear growth regime, i.e. below 60eposition cycles. This region corresponds to Pt loadings below 1ML.

for the first 50 cycles. An additional zoom around 40 cycles is presented within thisinset.

Although the ex situ XRF measurements are able to indicate thestarting point of the linear growth regime, the XRF signal at low Ptloadings is less conclusive about the actual growth behaviour due toits low sensitivity. This sensitivity can be improved by performingin situ XRF measurements at a high brilliance synchrotron facilityusing the same process conditions as before (Fig. 1, in situ XRF, seealso the inset). A subtle increase in XRF intensity is now observedfrom the first cycles. The latter implies that there is already growthfrom the beginning of the O3-based Pt process. As the deposi-tion evolves and the number of cycles increases, the growth rateincreases monotonically during the nucleation regime in line withthe ex situ measurements.

An additional zoom is made from the in situ XRF data around40 cycles in the inset of Fig. 1. The number of EML of Pt after40 and 41 cycles are quasi-identical. This implies that no addi-tional Pt is deposited during the 41st MeCpPtMe3/O3 cycle. Thelatter is observed for each cycle following a XAS measurement.A beam influence as the underlying cause can be excluded sincethe same effect is observed at locations on the support that arenot irradiated during the XAS measurements. In fact, the reducedgrowth is in agreement with an earlier study showing a decreasinggrowth rate of the O3-based Pt process as a function of the pumptime following the O3 step in the deposition process [19]. Duringa XAS measurement the chamber is evacuated for ca. 90 min, i.e.the duration of three EXAFS scans, which is much longer than thestandard pump time of 50 s between consecutive ALD cycles. Giventhe previous study, this difference in pump time between ALDcycles with and without intermediate XAS measurement seems tobe the underlying cause for the reduced growth rate following aXAS measurement.

3.2. X-ray absorption near edge structure (XANES): spontaneousreduction.

In the previous decades, numerous XANES studies haveappeared on the effect of particle size, support and process condi-tions upon the resonance shape at the Pt edge [33–36]. It is widelyknown that at the Pt LIII edge an electronic transition occurs fromthe 2p3/2 core level to the 5d3/2 or 5d5/2 valence state [37]. The

white line (WL) intensity, i.e. the area of the first maximum atthe edge, reflects the number of transitions occurring. More elec-tronic transitions will take place when the density of unoccupied

M. Filez et al. / Catalysis Tod

0

1

2

3

4

0 5 10 15 20 25 30

Ave

rage

Pt o

xida

tion

stat

e [-

]

Residence ti me [min]

0.52 EML

0.37 EM L: residen ce time = 60 m in

0.71 EM L: re sidence ti me = 60 min

0

0.5

1

1.5

11550 11560 11570 11580 11590

Nor

mal

ized

µx

[-]

Energy [eV]

Fig. 2. Pt oxidation state evolution as a function of the residence time of the sam-ple in the deposition chamber following the O3 step in the deposition process. Twoadditional oxidation state values at higher/lower Pt loadings are represented as hori-zontal lines. These two values are measured after a 60 min evacuation in the vacuumchamber. Additionally, in the inset, the normalized XANES spectra after 0.52 EMLaa

dt

PtdcbtAaXc

psPtigtB(attcsqlabdip

ltie

utions to the 0.04 EML k-space EXAFS signal, especially in the high

re shown from which the corresponding values of the average Pt oxidation statere obtained.

-electron states increases. Hence, the WL intensity is proportionalo the density of unoccupied d-electron states [38].

As indicated in the introduction, the degree of oxidation of thet phase is a key feature determining growth during the deposi-ion process. However, this factor is not fully known during theeposition process and remains a question of great interest. Inonventional metal oxides, ionic bonds are expected to be formedetween the oxygen atoms and the metal. In the course of Pt oxida-ion, d-electrons could therefore participate in chemical reactions.s a result, the density of unoccupied d-electron states changesccordingly during these reactions. Therefore, probing the Pt LIIIANES WL intensity in situ can yield a deeper understanding of thehemical state of Pt during deposition.

In particular, the WL intensity at the Pt LIII edge is linearly pro-ortional to the oxidation state of the Pt phase within a given clusterize range [39,40]. In order to determine this oxidation state of thet phase, the WL intensity needs to be quantified and connectedo the Pt0 and Pt+4 states. The quantification of the WL intensitys obtained by simultaneously fitting a Gaussian and an arctan-ent function to the XANES WL intensity area. The WL intensityhen corresponds to the area intensity of the Gaussian function.y applying this fitting procedure to both metallic Pt (Pt0) and PtO2Pt+4), the WL intensities for this lowest and highest oxidation statere obtained. By connecting these values to their respective oxida-ion state, the linear relationship between the WL intensity andhe Pt oxidation state is established. In the following, the quantifi-ation of the WL intensity of all spectra is obtained through theimultaneous Gaussian-arctangent fitting. Once the WL intensityuantified, the oxidation state is determined by using the estab-

ished WL intensity-oxidation state relationship. This WL intensitynalysis is preferred to a linear combination fitting (LCF) analysisecause of the direct link between the WL intensity and the Pt oxi-ation state. In addition, in LCF two or more separate phases are

mplicitly assumed to be present, which is not evidenced at thisoint.

Using the WL intensity methodology, the oxidation state evo-ution of the Pt phase is obtained as a function of the residence

ime of the sample in the vacuum chamber, following the O3 stepn the deposition process (Fig. 2). Based on Yoshida et al. [39], therror bars shown for these oxidation state data are estimated to be

ay 229 (2014) 2–13 5

around ±0.1. The normalized XANES spectra utilized to calculatethis oxidation state evolution are represented in the inset of Fig. 2.After the first XANES measurement obtained immediately after theO3 pulse, each fourth minute another scan is recorded. The arrowin the inset indicates the time evolution of the WL peak. A gradualdecrease of the XANES WL intensity peak and average Pt oxida-tion state with time is observed for the sample with 0.52 EML. Thisindicates a spontaneous reduction, caused by oxygen desorptionfrom the sample at 150 ◦C in vacuum [41]. As explained in Section3.1, an X-ray beam effect could be excluded, since the spontaneousreduction also occurs without X-ray measurements.

In Fig. 2, two additional oxidation states measured 60 min afterthe ozone pulse are shown as horizontal lines. The two measure-ments are done at higher/lower Pt loadings, i.e. 0.37 EML and 0.71EML, and represent their respective stabilized reduced oxidationstates. As explained in Section 3.10, the stabilized oxidation stateof the sample with a 0.52 EML loading will be intermediate to theformer two. Hence, the major part of the spontaneous reductiontakes place within the first 30 min after the O3 pulse. This transientregime covers the same timescale as the one required to record anEXAFS scan. In view of having EXAFS measured on a close to stablestructure, it was chosen to perform the EXAFS measurements aftera 30 min evacuation time following the O3 pulse in the depositionprocess.

Some fundamental conclusions can be made concerning theevent of spontaneous reduction and the reduced growth rateobserved following a XAS measurement. During spontaneousreduction, originally chemisorbed oxygen desorbs from the Ptphase. After this spontaneous reduction, as illustrated in the insetof Fig. 1, Pt deposition is disabled. Consequently, chemisorbed oxy-gen is essential for reaction with new precursor molecules enteringthe chamber during the next pulse. Equivalently, the oxygen atomsfunction as anchoring points for new Pt precursors. Furthermore,the spontaneous reduction occurs to such extent that no or a neg-ligible amount of oxygen remains at the cluster surface.

3.3. Extended X-ray absorption fine structure (EXAFS): modelling.

In Fig. 3a, the k2-weighted EXAFS signals are represented forfive different loadings below 1 EML. These spectra are recordedduring the nucleation regime of the deposition process at 0.007,0.04, 0.16, 0.37 and 0.71 EML of Pt. The data quality of the EXAFSsignal at the lowest Pt loading, i.e. 0.007 EML, is low due to the verylittle amount of Pt deposited at this stage of deposition. The signalquality significantly improves with increasing Pt loading. Similarfeatures, somewhat shifted to higher k-values for lower loadings,are observed from 0.04 EML on.

Fig. 3brepresents the magnitude of the Fourier transformed k2-weighted EXAFS signals shown in Fig. 3a. In addition, the metallicfcc Pt Fourier transformed signal is plotted. It is reported thatthe first nearest neighbour Pt-O and Pt-Pt peaks appear at about∼2 A and ∼2.8 A, respectively, in the phase corrected R-space [22].The latter distances correspond to the real interatomic distancesbetween the respective atomic pairs and are indicated accordinglyin Fig. 3b.

At the very beginning of the Pt deposition, i.e. 0.007 EML, a Pt-O neighbour peak is predominantly present compared to the Pt-Ptfirst nearest neighbour peak (Fig. 3b). However, with increasing Ptloading the Pt-O peak amplitude decreases significantly and thePt-Pt first nearest neighbour peak gains intensity. Also, from 0.04EML, a Pt fcc packing pattern can be observed, conformal to theone of the Pt fcc bulk structure. Despite the small noise contrib-

k-range (0.04 EML, Fig. 3a), the fcc pattern is clearly discerned inR-space. This fcc packing appears to be a stable configuration fromthe beginning of the deposition process. This is not evident, since it

6 M. Filez et al. / Catalysis Tod

-1.5

0.0

1.5

3.0

4.5

6.0

7.5

2 4 6 8 10 12

k².χ

(k)

[Å-2

]

k [Å-1]

0.007 EML 0.04 EML 0.16 EML 0.37 EML 0.71 EML

a

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

0 1 2 3 4 5 6

|χ(R

)| [Å

-3]

R [Å]

0.007 EML (10 cycles)0.04 EML (20 cycles)0.16 EML (30 cycles)0.37 EML (40 cycles)0.71 EML (50 cycles)Pt foil

Pt-Pt

Pt-O Pt-Pthigher shells

b

Fig. 3. (a) The k2-weighted EXAFS signals are shown for five different loadings below1 EML, i.e. at 0.007, 0.04, 0.16, 0.37 and 0.71 EML. (b) Magnitudes of the Fourier trans-formed k2-weighted EXAFS signals (k = 2–10 A−1, phase corrected) are representedfor five different loadings below 1 EML. A Pt foil with metallic fcc Pt bulk structurein

ititai

of the first Pt-Pt and Pt-O shell are constrained to the same value.

TT3

s shown too. Around ∼2 A the Pt-O peak is expected, whereas the Pt-Pt first nearesteighbour peak appears at ∼2.8 A. Higher Pt-Pt fcc shells are indicated as well.

s known that during the Pt(acac)2/O2-plasma deposition process atemperatures below 130 ◦C, a Pt oxide phase is formed [42]. Evolv-ng to higher Pt loadings, i.e. from 0.04 up to 0.71 EML, the Fourier

ransformed EXAFS signals undergo a peak amplitude increase inll near neighbour Pt-Pt fcc peaks. The latter implies that the orig-nal fcc packing perseveres to be the most stable packing within

able 1he estimates of the structural parameters, obtained by EXAFS modelling. The tabulated q.3 and 3.6. The values in italic should be looked at as to give the reader a qualitative feel

0.007 EML 0.04 EML

R-factor 0.095 0.032

NO,NN1 [–] 1.93 ± 0.34 0.80 ± 0.18

NPt,NN1 [–] 4.08 ± 0.84 5.88 ± 0.60

N Pt,NN2 [–] – 0.72 ± 1.20

N Pt,NN3 [–] – 3.36 ± 5.52

N Pt,NN4 [–] – 0.84 ± 1.20

�2Pt,NN1 [10−3Å2]a 7.2 ± 1.5 9.1 ± 0.8

�2Pt,NN2-4 [10−3Å2] – 11.9 ± 9.8

�(3)Pt,NN1 [10−4Å3] 18.8 ± 2.0 5.9 ± 2.5

RO,NN1 [Å] 1.995 ± 0.019 1.984 ± 0.022

RPt,NN1 [Å] 2.742 ± 0.014 2.719 ± 0.019

RPt,NN2 [Å] – 3.846 ± 0.027

RPt,NN3 [Å] – 4.710 ± 0.034

RPt,NN4 [Å] – 5.439 ± 0.039

�E0 [eV] 2.78 ± 1.93 3.75 ± 1.34

NNN1 [–] 6.01 ± 1.18 6.68 ± 0.78

NO,NN1/NNN1 [–] 0.321 ± 0.057 0.120 ± 0.027

NPt,NN1/NNN1 [–] 0.679 ± 0.140 0.880 ± 0.090

a �2Pt,NN1 = �2

O,NN1 .

ay 229 (2014) 2–13

the cluster. The increased peak intensity is caused by the increasedcoordination indicating that Pt growth is occurring.

Based on the above findings, an EXAFS model can be constructedfor further refinement. For this model, the dominant photoelec-tron scattering paths of the Pt fcc lattice configuration are used upto R = 6 A. These photoelectron scattering paths are those permit-ting an adequate reconstruction of its bulk structure signal. In linewith other reports [43], nine scattering paths are used up to R = 6 A.These consist of four single scattering paths, i.e. the first four near-est neighbour shells, three triangular scattering paths, two collinearpaths and one triple scattering path. Additionally, a Pt-O shell isadded to the model to account for the predominant presence of theoxygen peak in the low R-range at 0.007 EML. This model (Pt-O + Ptfcc) is used as a starting point for modelling the five spectra withloadings below 1 EML from Fig. 3.

To let the EXAFS model converge to the experimental signals,path parameters are varied to minimized the objective function[27]. Therefore, the photoelectron energy origin correction �E0,the half path length R, the path degeneracy N and the Debye–Waller(DW) bond length disorder �2 are used as path parameters. In addi-tion, the third cumulant �(3) is also necessary to include in the fit[28,43,44]. The latter parameter is only implemented for the Pt-Ptfirst nearest neighbour and is a measure for the asymmetry in thepair distribution function.

The spectra from 0.04 EML up to 0.71 EML are modelledfrom R = 1.5–6 A (k = 2.7–12.2 A−1). For these spectra, the modeldescribed above (Pt-O + Pt fcc) is used. Since the in situ XAS mea-surements are performed at increased temperatures (150 ◦C) andlittle Pt is deposited (<1 EML), some approximations have beenmade concerning the path parameters. The latter are defined suchthat its number is maximally reduced, while conserving a physi-cally defensible model. It is assumed that �E0 is the same for allscattering paths. Also, an isotropic expansion coefficient is usedto estimate the Pt-Pt interatomic distances and the same DW fac-tors are used for the second through fourth Pt-Pt single scatteringpaths. The latter approximation is justified since the DW factorsare expected to vary little from the second Pt shell on. The Pt-Ptmultiple scattering DW factors are implemented as mathematicalcombinations of the single scattering ones. Finally, the DW factors

The oxygen contribution to the signal is very small, making a reli-able estimation of a separate DW factor for oxygen challenging.In addition, this approximation does not affect the estimation of

uantities, given for five loadings below 1 EML, are defined and discussed in Sectionsing. The remaining results are meant for quantitative use.

0.16 EML 0.37 EML 0.71 EML

0.017 0.012 0.0080.45 ± 0.12 0.35 ± 0.12 0.30 ± 0.127.44 ± 0.36 8.52 ± 0.36 9.12 ± 0.361.20 ± 0.72 2.58 ± 0.84 3.30 ± 0.785.76 ± 2.88 11.28 ± 3.60 12.96 ± 3.362.88 ± 0.96 4.20 ± 0.96 5.40 ± 0.967.8 ± 0.5 7.5 ± 0.4 7.1 ± 0.38.6 ± 2.5 9.1 ± 1.6 8.9 ± 1.21.9 ± 0.9 2.7 ± 0.7 2.4 ± 0.51.944 ± 0.026 1.918 ± 0.035 1.911 ± 0.0392.744 ± 0.006 2.761 ± 0.005 2.766 ± 0.0043.881 ± 0.008 3.904 ± 0.007 3.912 ± 0.0054.753 ± 0.010 4.782 ± 0.008 4.791 ± 0.0075.489 ± 0.011 5.522 ± 0.009 5.533 ± 0.0085.76 ± 0.51 6.38 ± 0.39 7.11 ± 0.34

7.89 ± 0.48 8.87 ± 0.48 9.42 ± 0.480.057 ± 0.016 0.039 ± 0.014 0.032 ± 0.0120.943 ± 0.046 0.961 ± 0.041 0.968 ± 0.038

sis Today 229 (2014) 2–13 7

tc

ouantiissb

sPNPt�eagspeqsw

trIim

omeaifmesnt

FwmE

Fig. 5. In the inset, an artifact is shown which schematically describes the procedureto compare a cluster shape model to the data (Section 3.4). The degree of overlap

M. Filez et al. / Cataly

he DW factor of the Pt-Pt first shell significantly since the oxygenontribution is very small compared to the Pt contribution.

For the experimental spectrum recorded at the lowest loadingf 0.007 EML, a restricted R-range (R = 1.5–3 A, k = 2.7–12.2 A−1) issed due to its poorer data quality. In this region, only the first Pt-Ond Pt-Pt shells are appropriate to model the restricted R-space sig-al. The DW factors of the Pt-O and Pt-Pt shells are constrained tohe same value to decorrelate the Pt-Pt/O coordination number andts DW factor. This time, the latter assumption and the data qual-ty do not allow quantitative interpretation of the estimates of thetructural parameters. However, the presence of a Pt-O and Pt-Pthell in the spectrum at 0.007 EML is evidenced since convergenceetween model and data is obtained.

Fitting the described models to the data yields estimates for thetructural parameters. The results are tabulated in Table 1: (i) thet-O and Pt-Pt nearest neighbour coordination numbers NO,NN1 andPt,NNi (NNi = i’th coordination shell; i = 1–4), respectively, (ii) thet-O and Pt-Pt Debye–Waller factors �2

O,NN1 and �2Pt,NNi, respec-

ively, (iii) the third cumulant for the first Pt-Pt nearest neighbour(3)

Pt,NN1, (iv) the Pt-O first and Pt-Pt first through fourth near-st neighbour interatomic distances RO,NN1 and RPt,NNi, respectively,nd (v) the numbers obtained for the photoelectron energy ori-in correction �E0. As mentioned before, the estimates for thetructural parameters of the spectrum at 0.007 EML should be inter-reted as indicative of the actual values rather than quantitativestimates. In contrast, the results for the other spectra are suited foruantitative use. These findings are also reflected in the R-factorshown in Table 1. A gradual decrease of the R-factor is observedith increasing Pt loading.

The comparison of the model fit with the k2-weighted Fourierransformed EXAFS signal at 0.71 EML is depicted in Fig. 4. Both theeal part and the magnitude of the transformed signal are depicted.n the inset of Fig. 4, the corresponding k2-weighted EXAFS signals shown in k-space. A clear correspondence between the experi-

ental signal and the model fit is observed (R-factor = 0.008).The previously made observations based on visual inspection

f the EXAFS signals in R-space (Fig. 3) are confirmed by theodelling results. At 0.007 EML, as mentioned before, the pres-

nce of Pt-Pt neighbours is shown, implying that Pt atoms meetlready at this early stage of deposition. A Pt fcc lattice pack-ng is not detected at this very low Pt loading. For the remainingour loadings above 0.007 EML, i.e. 0.04–0.71 EML, the imple-

ented model (Pt-O + Pt fcc) is adequate to accurately describe thexperimental signal. Therefore, the Pt fcc packing is the favourable

tacking configuration from 0.04 EML on. Also, the total coordi-ation in the first shell, i.e. NNN1 = NPt,NN1 + NO,NN1, as a function ofhe loading is obtained from and shown in Table 1. The degree of

-1.5

-1.0

-0.5

0.0

0.5

1.0

1.5

0 1 2 3 4 5 6

|χ(R

)| [Å

]

R [Å]

experimental signalmodel fit

-1

0

1

2 4 6 8 10 12

k².χ

(k)

[Å-2

]

k [Å-1]

ig. 4. Comparison of the model fit to the experimental Fourier transformed k2-eighted EXAFS signal at 0.71 EML (k = 2.7–12.2 A−1, R = 1.5–6 A, R-factor = 0.008,agnitude and real part, not phase corrected). The corresponding k2-weighted

XAFS signal k2�(k) at 0.71 EML is depicted in k-space in the inset.

for the three indicated models is shown for the sample with a Pt loading of 0.71EML. Additionally, the 3D shape of a hemispherical (1 1 1)-truncated cuboctahedralcluster is shown.

under-coordination decreases with increasing loading, implyingparticle growth. Lastly, the relative fraction of Pt-O neighbourscompared to the fraction of Pt-Pt first nearest neighbours, i.e.NO,NN1/NNN1 and NPt,NN1/NNN1 respectively, decreases by a factor of10 over the investigated 0.007–0.71 EML range (Table 1). Based onex situ XAS measurements, Christensen et al. [45] found a similarevolution in the Pt-O and Pt-Pt coordination numbers with pro-gressing O2-based Pt ALD growth on SrTiO3 nanocubes.

3.4. Cluster shape determination

It has been shown in literature that the deposition of Pt byapplying the ALD procedure leads to the formation of islands[5]. However, the precise morphological and compositional evo-lution of these island clusters during the O3-based Pt process hasremained unanswered. Applying in situ XAS can help in answer-ing these questions. In particular, EXAFS literature provides studies[43,46–48] which can aid to reconstruct the particle shape duringdeposition. Simulations have shown that the nanoparticle shapecan be determined most reliably in the small cluster diameterrange, i.e. smaller than 3–5 nm [43]. In addition, the more dis-tant shells are more adequate for sampling the particle exteriorand discriminate between differently shaped clusters. EXAFS mod-elling can provide the experimental coordination numbers of theneighbouring shells in (sub)nanometer particles (Table 1). By find-ing the best match between the latter and simulated coordinationnumbers for different cluster shapes, insight into the particle shapeevolution during the initial stages of Pt deposition can be obtained.

A first morphological screening applies the methodologyprovided by Jentys et al. [48]. Such screening is possible bycomparing the experimentally obtained estimates for NPt,NN1 andNPt,NN3/NPt,NN1 with the theoretical values simulated for differentcluster shape models. Based on this analysis, spherically shapedclusters can be excluded, whereas cubes and distorted cubes cor-respond best to the obtained experimental coordination numbers.Despite this acceptable result, additional refinement of the clustermorphology towards more realistic shapes is desired.

Assuming a given cluster shape model, Frenkel et al. simu-lated the average coordination numbers of the first few shellsaround Pt as a function of the cluster diameter [43]. If thefour nearest neighbouring shells are considered, four curves aregenerated as represented artificially in the inset of Fig. 5. Sub-sequently, the set of experimental coordination numbers with

uncertainties {NPt,NNi ± �NPt,NNi} (NNi = i’th coordination shell withi = 1–4, Table 1) are implemented on these simulated curves.The possible range in which NPt,NNi occurs, i.e. [NPt,NNi–�NPt,NNi,NPt,NNi + �NPt,NNi], and its corresponding diameter range can be

8 sis Today 229 (2014) 2–13

vyFttCaomait

tdtdbaiTt(flh

riaifsabenctttse

3

diidTbtoadmcteaafiF

0.0

0.5

1.0

1.5

2.0

2.5

0.0 0.2 0.4 0.6 0.8

Clu

ster

dia

met

er [

nm]

Equivalent monolayers [-]

Fig. 6. Cluster diameter evolution during the deposition process, shown as a func-

although with scatter (Fig. 7). This increasing trend with decreasingsize is additionally supported by the large (qualitative) value of�(3)

Pt,NN1 at 0.007 EML (Table 1). The increased third cumulant is

2.62

2.64

2.66

2.68

2.70

2.72

2.74

2.76

2.78

0.000

0.003

0.006

0.009

0.012

0.015

0.018

0.021

0.5 1.0 1.5 2.0 2.5

Firs

t nei

ghbo

ur P

t-P

t dis

tanc

e RPt,NN1

[Å]

Deb

ye-W

alle

r (D

W)

fact

or σ² Pt,NN1

[Ų]

T

hird

cum

ulan

t σ(3) Pt,NN1

x 10

[10

.ų]

Clus ter diameter [nm]

third cumulant

DW facto r

First neighb our Pt -Pt di stance

M. Filez et al. / Cataly

isualized by a rectangular shaped area for each i’th shell. Thisields four rectangular shaped areas, as depicted in the inset ofig. 5. Within this plot, all true coordination numbers would lie onhe same vertical line at a given cluster diameter if the actual clus-er shape fully corresponds the considered cluster shape model.onsequently, there should be a large overlap between the rect-ngular shaped areas in the diameter dimension. The degree ofverlap between these areas is therefore a measure for the agree-ent between the cluster shape model and the actual shape. The

rea of overlap is indicated by the area enclosed by the dotted linesn the inset of Fig. 5. The degree of overlap corresponds to width ofhe diameter range occupied by this area of overlap.

In this study, the aim is to obtain a decent model matching closeo the actual cluster shape, rather than a thorough cluster shapeetermination. Therefore, the described procedure is repeated forhree cluster shape models [43], being (i) a spherical cuboctahe-ron, (ii) a hemispherical cuboctahedron truncated with a (0 0 1)asal plane and (iii) a hemispherical cuboctahedron truncated with

(1 1 1) basal plane. The degree of overlap for these three modelss shown in Fig. 5 for the sample with a Pt loading of 0.71 EML.he cluster shape model best matching our data, i.e. presentinghe largest region of overlap, is unambiguously the hemispherical1 1 1)-truncated cuboctahedron. Similar results are obtained for allour spectra investigated, i.e. 0.04, 0.16, 0.37 and 0.71 EML. For theowest loading of 0.007 EML this procedure is not applicable sinceigher coordination numbers are not obtained.

Although using Frenkel’s method leads to a more detailed andealistic cluster shape, some remarks should be made. As statedn Section 3.2, the EXAFS spectra are measured in a stabilized statefter spontaneous reduction. Clusters which are reduced possess anncreased order owing to the absence of adsorbed oxygen [20]. Thisacilitates the determination of the cluster structure owing to theharpened peaks in R-space. Despite this advantageous effect, thislso implies that the structure immediately after the O3 pulse coulde different from the one obtained in the stabilized state. The pres-nce of more adsorbed oxygen could then cause rounding of theseanoparticles [49,50]. Hence, the hemispherical (1 1 1)-truncateduboctahedron can be considered a valid cluster shape model forhe reduced state. Concerning the cluster shape after the O3 pulse,his model is considered as a good starting point for further calcula-ions. However, due to possible nanoparticle rounding effects morepherical Pt cluster shapes, e.g. spherical cuboctahedron, cannot bexcluded.

.5. Cluster size and evolution

Since the cluster shape is determined, it is straightforward toetermine the corresponding average particle diameters [43]. This

s done for all recorded spectra for loadings from 0.04 EML, yield-ng the particle size evolution during the deposition process. This isepicted in Fig. 6 as a function of the number of EML of Pt deposited.he error bars shown correspond to the width of the area of overlapetween the coordination numbers (inset Fig. 5). Once the clus-er shape determined, this width is a good estimate of the errorn the cluster diameter. Some interesting trends can be noticed:t the very beginning of Pt deposition a strong increase in clusteriameter is seen for very low loadings, i.e. below 0.04 EML. Thiseans that although little Pt is deposited at this stage, the atoms

luster together to form stable nuclei. Implicitly, this signifies thathe combustion of the precursor ligands by O3 takes place in anfficient way. The combustion yields PtOx species which possessn increased diffusion coefficient compared to non-oxidized Pt

datoms [51]. This makes it possible to migrate over the surface toorm stable nuclei, thus enhancing the nucleation process. Arriv-ng at 0.04 EML, an average particle diameter of 0.75 nm is found.or loadings ranging from 0.04 up to 0.71 EML, within error bars, a

tion of the number of EML of Pt deposited. The linear curve, obtained by regressionanalysis from 0.04 to 0.71 EML, is shown as a dashed line.

linear increase in particle diameter as a function of the Pt loadingis observed. At a loading of 0.71 EML, an average particle diameterof 2.3 nm is obtained.

3.6. Detailed structural evolution with cluster size

More in-depth insight into the structural behaviour of the Ptclusters during the nucleation regime is obtained by correlat-ing the structural EXAFS parameters with cluster diameter. Forthis purpose, the evolution of the Pt-Pt first nearest neighbourDebye–Waller factor, third cumulant and interatomic distance, i.e.�2

Pt,NN1, �(3)Pt,NN1 and RPt,NN1 respectively, with cluster size is dis-

cussed from 0.04 EML up to 0.71 EML. The estimates for the latterthree quantities at 0.007 EML (Table 1) are not considered sincethese are not well-suited for quantitative use, as mentioned before.

In Fig. 7, �2Pt,NN1, �(3)

Pt,NN1 and RPt,NN1 are shown as a func-tion of the cluster diameter during the Pt deposition process. Asubtle but significant increase of the Pt-Pt first nearest neighbourDW factor �2

Pt,NN1 is seen with decreasing cluster diameter. Thisis caused by an increased fluctuation in the bond-length betweentwo neighbouring Pt atoms, leading to a broader bond-length distri-bution. A similar trend is detected for the third cumulant �(3)

Pt,NN1,

Fig. 7. Pt-Pt first nearest neighbour Debye–Waller (DW) factor �2Pt,NN1, third cumu-

lant �(3)Pt,NN1 and distance RPt,NN1 (Table 1) as a function of the cluster diameter

during the Pt deposition process. The error bars depicted correspond to the fittingerrors obtained by EXAFS modelling in Artemis (IFEFFIT).

M. Filez et al. / Catalysis Today 229 (2014) 2–13 9

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0 0.2 0.4 0.6 0.8

Equi valent monol ayer s [-]

Den

sity

of

nucl

ei [

nm-2

]

Fs

ct

olaaasrat

adoWiEdfm

3

pzetlt

tyAcbognpiicfi

20 30 40 50 60

XR

D in

tens

ity

(a.u

.)

Diffra ction angle 2θ (º )

Pt (111)

Pt (200)

Fig. 9. XRD scan at atmospheric conditions. Pt(1 1 1) and Pt(2 0 0) peaks are indi-

XRD proves that the lattice packing is metallic fcc, even at

ig. 8. Density of nuclei as a function of the number of EML of Pt during the depo-ition process.

aused by an increased skewness in the bond-length pair distribu-ion function with decreasing diameter.

These two trends point towards anisotropic bond-length dis-rder. The latter is the consequence of distortions in the idealattice positions at specific locations of the cluster [43,47]. Suchnisotropic disorder effects are triggered by phenomena occurringt the cluster exterior, e.g. surface relaxations, support inter-ction, surface adsorbates. Therefore, these will influence thetructural parameters more strongly when the surface-to-bulkatio increases, i.e. for small clusters. Consequently, the DW factornd the third cumulant have an increased value at the beginning ofhe deposition process.

These findings are supported by the evolution of the Pt-Pt inter-tomic distance RPt,NN1 in the first shell as a function of clusteriameter (Fig. 7). At 0.04 EML, clusters have an average diameterf 0.75 nm with an interatomic Pt-Pt distance of 2.719 ± 0.019 A.hen the Pt loading increases and clusters grow, the Pt-Pt distance

ncreases and evolves towards the Pt bulk value of 2.77 A. At 0.71ML, i.e. average cluster diameter of 2.3 nm, the Pt-Pt interatomicistance is 2.766 ± 0.004 A. The bond-length contraction originatesrom the under-coordination of the surface atoms, leading to inho-

ogeneous strain at the cluster surface [43,52].

.7. Density of nuclei

Up to now, the lattice packing, shape, diameter and structuralarameters of the clusters yielded a detailed structural characteri-ation of the deposited Pt phase. However, in order to obtain a fullvolutional picture of the deposition process, the average nanopar-icle spacing on the support surface should be known as well. Theatter can be quantified by the density of nuclei which is defined ashe number of Pt clusters present per nm2 of SiO2 surface.

Combining EXAFS, i.e. Pt atoms per cluster, with XRF, i.e. theotal number of Pt atoms deposited, and the specific surface area,ields this density of nuclei during the deposition process (Fig. 8).n equidistant 2D spacing grid is assumed between the depositedlusters. The data points depicted in Fig. 8 possess significant errorars due to error bar propagation originating from the uncertaintyn the cluster size (Fig. 6). However, two mechanisms can be distin-uished. Below around 0.16 EML a sharp increase in the density ofuclei is seen. This evidences that a substantial nucleation is takinglace. Focusing on the region from 0.16 EML up to 0.71 EML, the

ncreasing trend stabilizes and slightly decreases. This behaviour

ndicates that existing nuclei are merging together through coales-ence, though this does not mean that new nuclei can’t still beormed as well. The mentioned two mechanisms could aid to clar-fy the linear relationship between cluster diameter and Pt loading

cated. The scan originates from the resulting sample of the in situ synchrotronexperiments, with Pt loading 0.93 EML.

seen in Fig. 6 (0.04–0.71 EML). Apparently, the balance betweennanoparticle growth, new nuclei formation and coalescence leadsto such linear diameter-loading relationship seen in Fig. 6.

In Fig. 8, the maximal density of nuclei is situated around 0.07clusters/nm2 of SiO2 surface. At 0.71 EML, on average 5.4 Pt clusterswith cluster diameter of 2.3 nm are present on a SiO2 surface area of10 nm × 10 nm. This corresponds to a ca. 25% surface coverage. Thislooks reasonable since it is known that the linear growth regimestarts at about 1 EML, as evidenced by ex situ XRF measurements(Section 3.1., Fig. 1).

3.8. X-ray diffraction (XRD) measurements

PtOx/SiO2 at 0.71 EML is not recovered as a separate sampleduring the synchrotron measurements, but rather measured asan intermediate state of Pt deposition on this sample. Therefore,the resulting sample of the in situ synchrotron experiments, witha 0.93 EML end loading, is used to check the crystallinity of thedeposited Pt phase with XRD. The diffraction peaks found corre-spond to metallic fcc Pt (Fig. 9). From the width of the peaks, theaverage crystallite diameter is determined to be 4 nm ± 0.2 nm. Thisvalue is larger than expected from the EXAFS analysis. Extrapo-lating the linear relationship in Fig. 6 up to 0.93 EML yields anestimated average particle diameter of 2.8 nm. However, this dis-crepancy is not unexpected since the Scherrer equation is knownto overestimate the diameter of particles below ca. 10 nm [53].

3.9. (Scanning) transmission electron microscopy ((S)TEM)measurements

The same sample with 0.93 EML loading is also investigated withTEM. The left part of Fig. 10 shows a HRTEM image taken at therim of a pore hole in the mesoporous SiO2 film. The SiO2 matrixis amorphous, as expected [25]. The Pt clusters are visible as sup-ported crystals packed with a high degree of order, supporting boththe XRD and EXAFS results. The right part of Fig. 10 is a STEM imageof a collection of supported Pt nanoparticles, again located at therim of a pore hole. The particle diameter distribution obtained fromthis image is shown in the inset. The corresponding volume aver-aged particle diameter of 2.8 nm is in excellent agreement with theextrapolated particle diameter from the EXAFS analysis, provingthe correctness of the latter. Finally, a qualitative comparison ofthe density of nuclei based on Fig. 10 and Fig. 8 shows agreementat first glance.

atmospheric oxygen pressures. TEM confirms that the diametersobtained by EXAFS are correct and shows the ordered crystallinepacking of the supported nanoparticles.

10 M. Filez et al. / Catalysis Today 229 (2014) 2–13

F f a porB ents, w

3c

dfaatoacttitt

wbaTcFqFaWoStsraSiitai

dWi

ig. 10. (left) HRTEM image. (right) BF STEM scan. Both images are taken at the rim ooth images originate from the resulting sample of the in situ synchrotron experim

.10. X-ray absorption near edge structure (XANES): clusteromposition

The following intends to improve our understanding about theegree of cluster oxidation during the deposition process. There-ore, the cluster shape model derived in Section 3.4 will be useds a starting point for further calculations. By appointing oxygentoms to (a group of) Pt atoms of the cluster, e.g. interface atoms,he degree of oxidation of the cluster can be altered. Calculationf the oxidation state for each – possibly oxidized – Pt atom andveraging over all atoms present in the cluster, yields a theoreti-al average oxidation state. This theoretical value can be comparedo the experimental one. By adapting the degree of oxidation ofhe model cluster, the correspondence between theory and exper-ment can be optimized. This opens up the possibility to elucidatehe degree of oxidation of the cluster. To the best of our knowledge,his approach has never been applied before.

Before presenting the modelling results, the experimental dataill be introduced first. Fig. 11 consists of three parts. Fig. 11a and

shows the evolution of the XANES spectra measured immedi-tely after the O3 pulse at different stages of the deposition process.hese two figures aim to give the reader a feeling of the signifi-ance of the WL intensity variations during deposition. In addition,ig. 11c shows two different experimental data series, recorded fre-uently at different loadings during the same deposition process.irst, the WL intensity obtained from the XANES spectra immedi-tely after the ozone pulse (Fig. 11a and b) is depicted. Second, theL intensity acquired from XANES spectra recorded 1 h after the

zone pulse is presented. The EXAFS measurements discussed inection 3.3 were performed in the same stabilized catalyst state ashe latter XANES spectra. The calculation method of the WL inten-ity is described in Section 3.2. This WL intensity is plotted in aescaled unit (vide infra). All data points mentioned are plotteds a function of the average cluster diameter during deposition.mall (large) clusters therefore correspond to low (high) Pt load-ngs during this deposition process (Fig. 6). The vertical error barsn Fig. 11c are estimated to be around ±0.1 [39]. Based on Fig. 6 andhe determination of the errors on the particle sizes using Frenkel’spproach (Section 3.5), the horizontal error bars are obtained andmplemented in Fig. 11c.

At the initial stages of nucleation, the WL intensity of theeposited Pt phase corresponds to the one of metallic Pt (Fig. 11c).hen the material deposited increases and clusters grow, an

ncrease in WL intensity is observed. Unravelling the observed

e hole after FIB slicing. The inset of the STEM image shows a cluster size distribution.ith Pt loading 0.93 EML.

trends in the XANES WL intensity below ±1 nm is not straightfor-ward. General rules do not exist to predict the WL intensity [32]within this region. A combination of support effects, lattice con-traction, cluster diameter and the nature of the Pt O bond couldexplain the observed trends [33,34].

From loadings of 0.03 EML, i.e. estimated particle diameter of0.6 nm, the WL intensity becomes dependent on the evacuationtime of the chamber after the ozone pulse. From this point, theevolution of the upper and lower boundary of the WL intensity isshown in Fig. 11c. The upper one corresponds to the WL inten-sity measured immediately after the ozone pulse. In contrast, thelower one is the WL intensity of the stabilized reduced state after 1 hof spontaneous reduction in vacuum. Obviously, the WL intensityevolves gradually with time from the upper to the lower bound-ary (Fig. 2). Also, the considerable discrepancy between the twolimits proves that the XANES WL intensity is sensitive to oxidationstate changes of the Pt phase. Therefore, it should be possible toobtain more quantitative information on the degree of oxidation ofan average cluster.

In both curves in Fig. 11c, the increase in WL intensity stabilizesand starts to decrease gradually between 0.5 and 1 nm. As indi-cated in Section 3.2, a linear relationship is established betweenthe WL intensity and the x = O/Pt ratio for supported PtOx clusterswithin the 1–5 nm range [39,40]. The Pt oxidation state (oxidationstate = 2x) can therefore be monitored through the WL intensity. Allvalues of the WL intensity in Fig. 11c are rescaled to correspond tothe Pt oxidation state within the 1–3 nm range. Within this rangethe WL intensity equals the Pt oxidation state. However, for therange below 1 nm, the effects mentioned earlier this section (sup-port, size, . . .) lead to the breakdown of the linear relation betweenthe WL intensity and the O/Pt ratio. Therefore, the latter region willnot be considered in further calculations concerning the degree ofcluster oxidation.

In an attempt to describe the previous trends, first, the oxida-tion state evolution after 1 h in vacuum is modelled. This is donefor the 1–3 nm range shown in Fig. 11c. The hemispherical (1 1 1)-truncated cuboctahedron can be used as a realistic cluster shapemodel for this reduced state. First, it can be assumed that no oxy-gen atoms are adsorbed on the external surface of the Pt cluster,as proven by XANES analysis (Sections 3.1 and 3.2, Figs. 1 and 2).

Second, EXAFS and XRD prove that the Pt structure at the interiorof the cluster is metallic fcc Pt, implying Pt0 atoms. Since the clus-ter shape is known, the average oxidation state of the Pt atomsat the cluster–support interface can now be estimated. During the

M. Filez et al. / Catalysis Today 229 (2014) 2–13 11

Fig. 11. (a) Normalized XANES spectra immediately after the O3 pulse during the deposition process for sizes ranging from 0.2 nm to 0.9 nm and (b) 0.9 nm to 2.9 nm. (c)WL intensity as a function of particle diameter during deposition. Two datasets are shown: one immediately after the ozone pulse and one after 1 h in vacuum. These trendsare modelled by different cluster models (Section 3.10). The resulting fits to the data series are plotted too. Finally, two additional curves are shown which possess differentc presenm

oaeacm

dispaagttsts

toc

luster surface oxidation states for the hemispherical cluster model. These are reethodology to the degree of cluster surface oxidation (Section 3.10).

ptimization procedure, the latter parameter is kept identical forll sizes. It is found that in order to describe the oxidation statevolution in an optimal way, a Pt interface atom should have anverage oxidation state of Pt+4/3. Therefore, a Pt3O2 interface stoi-hiometry is obtained. As can be seen in Fig. 11c, the implementedodel describes the data in a satisfactory way (dotted curve).In an attempt to simulate the oxidation state evolution imme-

iately after the ozone pulse, first, the same cluster shape models used again. At the cluster-support interface, the Pt3O2 interfacetoichiometry is implemented since no change is expected at thisoint. Additionally, the Pt atoms residing at the cluster interiorre set to a Pt0 oxidation state. This time, the Pt atoms residingt the external cluster surface have the possibility to adsorb oxy-en atoms. Therefore, the average oxidation state of Pt atoms athe cluster surface can be estimated. Again, during this optimiza-ion procedure, the latter parameter is kept identical for all sizes. Aurface atom with oxidation state of Pt+2 yields the optimal descrip-ion of the oxidation state evolution, corresponding to a PtO surfacetoichiometry (solid dark grey curve, Fig. 11c).

As mentioned in Section 3.4, the cluster shape could alterhrough oxidation. A rounding of the original cluster shape has beenbserved [49,50]. Considering a more spherical cluster shape modelould therefore be useful. A spherical cuboctahedron is ideally

ted in order to give the reader a feeling on the sensitivity of the applied XANES

suited for this. Thus, the spherical clusters are reconstructed withthe same number of atoms as their hemispherical counterparts. Thesame interface stoichiometry is installed and an identical parame-ter is optimized using the same procedure as for the hemisphericalshape. Now, a PtO1.15 surface stoichiometry yields the optimaldescription of the oxidation state evolution immediately after theozone pulse (dashed curve, Fig. 11c). Therefore, an increased oxy-gen adsorption at the cluster surface is found compared to thehemispherical case. This can be clarified by the increased numberof Pt atoms residing in the interior of the cluster. These internal Ptatoms possess a Pt0 oxidation state and do not contribute anymoreto the increase of the average oxidation state. This effect is correctedby the observed increase of the number of adsorbed oxygen atomsat the Pt cluster surface. Thereby, the same average oxidation statecan be preserved.

Finally, some comments are worth mentioning. First, the XANESmethodology applied in this section is sensitive to changes in thedegree of cluster surface oxidation within the 1–3 nm range. Toillustrate this, the Pt surface oxidation state is varied, i.e. Pt+1.5,

Pt+2.0 and Pt+2.5, using the hemispherical cuboctahedral clustershape. In Fig. 11c, the result of this variation is shown in thesolid lines with different grey shades. An increase (decrease) in thedegree of cluster surface oxidation results in a considerable increase

1 sis Tod

(titrttdsdddpt±rPHt

4

tOsipdp

ddwt1

tbswmttadddn

sodr

A

u(sbtGs

[

[

[

[

[

[

[[

[

[

[

[[

[

[

[

[

[[

[

[

[[[[[

[

[

[

2 M. Filez et al. / Cataly

decrease) in the WL intensity in the 1–3 nm range. The change ofhe oxidation state of the Pt surface atoms by ±0.5 with respect tots optimal value results in a model curve located outside the uncer-ainty region of the data points, especially in the higher diameterange. The clear variation of these modelled curves, in response tohe relatively small surface oxidation state changes (±0.5), proveshe sensitivity of the applied XANES methodology. Second, theegree of surface oxidation remains quasi-constant during depo-ition within the investigated 1–3 nm size range. Therefore, theecrease in oxidation state with size (Fig. 11c) is caused by aecreased surface-to-bulk ratio rather than a decreased surface oxi-ation. Third, based on considering the uncertainty area of the dataoints shown in Fig. 11c, the error bars on the modelled values ofhe cluster surface oxidation states are determined to be around0.25. Therefore – fourth – the oxidation states of the Pt atoms

esiding at the surface of hemispherical and spherical clusters, i.e.t+2(±0.25) and Pt+2.3(±0.25), have overlapping uncertainty regions.ence, the employed cluster shape models do not significantly alter

he resulting degree of surface oxidation.

. Conclusions

Combining XAS and XRF measurements under in situ condi-ions provides first-hand insights into the nucleation stage of the3-based Pt deposition process. By EXAFS modelling, in-depth

tructural information on the Pt clusters is gained. This informations directly used to obtain the cluster shape, diameter and latticeacking. XANES spectra yield the degree of oxidation during theeposition. XRF quantifies the amount of Pt deposited on the sup-ort, which can be expressed in equivalent monolayers (EML).

Crosslinking all these reveals interesting elements about theeposition process such as the evolution of cluster diameter andensity with Pt loading and the degree of cluster surface oxidationith diameter. These refined details complete our picture about

he design of Pt catalysts using sequential MeCpPtMe3/O3 pulses at50 ◦C.

During the nucleation regime of the deposition process, Pt clus-ers are deposited on the mesoporous SiO2 support. At the veryeginning of deposition, Pt adatoms cluster together to form moretable nuclei. From 0.04 up to 0.71 EML, these clusters grow linearlyith Pt loading. They have an fcc packing and their cluster shapeatches most closely to a hemispherical (1 1 1)-truncated cuboc-

ahedron in the reduced state. Immediately after the ozone pulse,he degree of oxidation of the cluster at the surface corresponds ton average oxidation state around Pt+2 within the 1–3 nm clusteriameter range. This surface oxygen is necessary for further growthuring the Pt deposition process. The nucleation of new clusters isominant up to 0.16 EML, after which the coalescence of existinguclei occurs.

These conclusions prove the synergy of performing XAS and XRFimultaneously under in situ conditions. The presented method-logy exploits a detailed evolutional picture of the catalyst stateuring deposition and can be applied systematically for a diverseange of future studies.

cknowledgements

Experiments were performed on the SAMBA beam line [26],sing the chemistry support lab, at SOLEIL Synchrotron, France,proposal number 20120293). We are grateful to the SOLEILtaff for smoothly running the facility. This work was supported

y the Fund for Scientific Research Flanders (FWO: G.0209.11),he ‘Long Term Structural Methusalem Funding by the Flemishovernment’, the ‘Special Research Fund BOF of Ghent Univer-ity’, the IAP7/05 Interuniversity Attraction Poles Programme

[[[[

ay 229 (2014) 2–13

– Belgian State–Belgian Science Policy, and the CALIPSO TransNational Access Program funded by the European Commission insupplying financing of travel costs. C. Detavernier acknowledgesthe ‘European Research Council for a Starting Grant’ (No. 239865).The authors acknowledge V. Bliznuk for support with the TEMequipment and Sreeprasanth Pulinthanathu Sree for supplying themesoporous silica samples.

References

[1] R.L. Puurunen, J. Appl. Phys. 97 (2005).[2] V. Miikkulainen, M. Leskela, M. Ritala, R.L. Puurunen, J. Appl. Phys. 113 (2013).[3] S.M. George, Chem. Rev. 110 (2010) 111.[4] C. Detavernier, J. Dendooven, S.P. Sree, K.F. Ludwig, J.A. Martens, Chem. Soc.

Rev. 40 (2011) 5242.[5] L. Baker, A.S. Cavanagh, D. Seghete, S.M. George, A.J.M. Mackus, W.M.M. Kessels,

Z.Y. Liu, F.T. Wagner, J. Appl. Phys. 109 (2011).[6] H.B.R. Lee, S.F. Bent, Chem. Mat. 24 (2012) 279.[7] S.P. Sree, J. Dendooven, K. Masschaele, H.M. Hamed, S.R. Deng, S. Bals, C. Detav-

ernier, J.A. Martens, Nanoscale 5 (2013) 5001.[8] Y. Lei, B. Liu, J.L. Lu, R.J. Lobo-Lapidus, T.P. Wu, H. Feng, X.X. Xia, A.U. Mane, J.A.

Libera, J.P. Greeley, J.T. Miller, J.W. Elam, Chem. Mat. 24 (2012) 3525.[9] H.B.R. Lee, M.N. Mullings, X.R. Jiang, B.M. Clemens, S.F. Bent, Chem. Mat. 24

(2012) 4051.10] J.A. Enterkin, W. Setthapun, J.W. Elam, S.T. Christensen, F.A. Rabuffetti, L.D.

Marks, P.C. Stair, K.R. Poeppelmeier, C.L. Marshall, ACS Catal. 1 (2011) 629.11] S.T. Christensen, H. Feng, J.L. Libera, N. Guo, J.T. Miller, P.C. Stair, J.W. Elam, Nano

Lett. 10 (2010) 3047.12] J.H. Li, X.H. Liang, D.M. King, Y.B. Jiang, A.W. Weimer, Appl. Catal. B-Environ. 97

(2010) 220.13] J.S. King, A. Wittstock, J. Biener, S.O. Kucheyev, Y.M. Wang, T.F. Baumann, S.K.

Giri, A.V. Hamza, M. Baeumer, S.F. Bent, Nano Lett. 8 (2008) 2405.14] T. Aaltonen, M. Ritala, T. Sajavaara, J. Keinonen, M. Leskela, Chem. Mat. 15 (2003)

1924.15] W.M.M. Kessels, H.C.M. Knoops, S.A.F. Dielissen, A.J.M. Mackus, M.C.M. van de

Sanden, Appl. Phys. Lett. 95 (2009).16] A.J.M. Mackus, N. Leick, L. Baker, W.M.M. Kessels, Chem. Mat. 24 (2012) 1752.17] H.C.M. Knoops, A.J.M. Mackus, M.E. Donders, M.C.M. van de Sanden, P.H.L. Not-

ten, W.M.M. Kessels, Electrochem. Sol. St. Lett. 12 (2009) G34.18] D. Longrie, D. Deduytsche, S. Van den Berghe, C. Driesen, C. Detavernier, Solid

State Sci. Technol. 1 (2012).19] J. Dendooven, R.K. Ramachandran, K. Devloo-Casier, G. Rampelberg, M. Filez,

H. Poelman, G.B. Marin, E. Fonda, C. Detavernier, J. Phys. Chem. C 117 (2013)20557.

20] S. Mostafa, F. Behafarid, J.R. Croy, L.K. Ono, L. Li, J.C. Yang, A.I. Frenkel, B.R.Cuenya, J. Am. Chem. Soc. 132 (2010) 15714.

21] H. Yoshida, S. Nonoyama, Y. Yazawa, T. Hattori, Cat. Tod. 153 (2010) 156.22] D.C. Koningsberger, B.L. Mojet, G.E. van Dorssen, D.E. Ramaker, Top. Catal. 10

(2000) 143.23] W. Setthapun, W.D. Williams, S.M. Kim, H. Feng, J.W. Elam, F.A. Rabuffetti, K.R.

Poeppelmeier, P.C. Stair, E.A. Stach, F.H. Ribeiro, J.T. Miller, C.L. Marshall, J. Phys.Chem. C 114 (2010) 9758.

24] K. Devloo-Casier, J. Dendooven, K.F. Ludwig, G. Lekens, J. D’Haen, C. Detavernier,Appl. Phys. Lett. 98 (2011).

25] S.P. Sree, J. Dendooven, D. Smeets, D. Deduytsche, A. Aerts, K. Vanstreels, M.R.Baklanov, J.W. Seo, K. Temst, A. Vantomme, C. Detavernier, J.A. Martens, J. Mater.Chem. 21 (2011) 7692.

26] V. Briois, E. Fonda, S. Belin, L. Barthe, C. La Fontaine, F. Langlois, M. Ribbens,F. Villain, UVX 2010; Applications et Développements Récents. EDP Sciences(2011), 41–47. http://dx.doi.org/10.1051/uvx/2011006

27] B. Ravel, M. Newville, J. Synchrot. Radiat. 12 (2005) 537.28] E. Bus, J.T. Miller, A.J. Kropf, R. Prins, J.A. van Bokhoven, PCCP Phys. Chem. Chem.

Phys. 8 (2006) 3248.29] J.J. Rehr, J.M. Deleon, S.I. Zabinsky, R.C. Albers, J. Am. Chem. Soc. 113 (1991)

5135.30] B. Ravel (2000), EXAFS Analysis with FEFF and FEFFIT: Part 2: Commentary.

http://cars9.uchicago.edu/∼ravel/course/notes.pdf31] P. Scherrer, Nachr Ges Wiss Göttingen 28 (1918).32] A.J.M. Mackus, N. Leick, A.A. Bol, W.M.M. Kessels, Chem. Mat. 25 (2013) 6.33] Y. Lei, J. Jelic, L.C. Nitsche, R. Meyer, J. Miller, Top. Catal. 54 (2011) 334.34] N. Ichikuni, Y. Iwasawa, Catal. Lett. 20 (1993) 87.35] K. Asakura, T. Kubota, W.J. Chun, Y. Iwasawa, K. Ohtani, T. Fujikawa, J. Synchrot.

Radiat. 6 (1999) 439.36] D.E. Ramaker, D.C. Koningsberger, PCCP Phys. Chem. Chem. Phys. 12 (2010)

5514.37] D.E. Ramaker, M.K. Oudenhuijzen, D.C. Koningsberger, J. Phys. Chem. B 109

(2005) 5608.38] A.N. Mansour, D.E. Sayers, J.W. Cook, D.R. Short, R.D. Shannon, J.R. Katzer, J.

Phys. Chem. 88 (1984) 1778.39] H. Yoshida, S. Nonoyama, Y. Yazawa, T. Hattori, Phys. Scr. T115 (2005) 813.40] Y. Yazawa, H. Yoshida, T. Hattori, Appl. Catal. A-Gen. 237 (2002) 139.41] L.K. Ono, J.R. Croy, H. Heinrich, B.R. Cuenya, J. Phys. Chem. C 115 (2011) 16856.42] J. Hamalainen, F. Munnik, M. Ritala, M. Leskela, Chem. Mat. 20 (2008) 6840.

sis Tod

[[[

[[

[[[50] P. Nolte, A. Stierle, N.Y. Jin-Phillipp, N. Kasper, T.U. Schulli, H. Dosch, Science

M. Filez et al. / Cataly

43] A.I. Frenkel, C.W. Hills, R.G. Nuzzo, J. Phys. Chem. B 105 (2001) 12689.44] B.S. Clausen, J.K. Norskov, Top. Catal. 10 (2000) 221.45] S.T. Christensen, J.W. Elam, F.A. Rabuffetti, Q. Ma, S.J. Weigand, B. Lee, S. Seifert,

P.C. Stair, K.R. Poeppelmeier, M.C. Hersam, M.J. Bedzyk, Small 5 (2009) 750.46] A.M. Beale, B.M. Weckhuysen, PCCP Phys. Chem. Chem. Phys. 12 (2010) 5562.47] A.I. Frenkel, A. Yevick, C. Cooper, R. Vasic, in: R.G. Cooks, E.S. Yeung (Eds.),

Annual Review of Analytical Chemistry, 4, Annual Reviews, Palo Alto, 2011,p. 23.

[[[

ay 229 (2014) 2–13 13

48] A. Jentys, PCCP Phys. Chem. Chem. Phys. 1 (1999) 4059.49] F. Mittendorfer, N. Seriani, O. Dubay, G. Kresse, Phys. Rev. B 76 (2007).

321 (2008) 1654.51] P. Wynblatt, Acta Metallurgica 24 (1976) 1175.52] E. Bus, J.T. Miller, J.A. van Bokhoven, J. Phys. Chem. B 109 (2005) 14581.53] J.B. Cohen, Ultramicroscopy 34 (1990) 41.