Growth and Structure of Cu and Au on the Nonpolar ZnO(101̅0

Aug 23, 2013 - Additionally, the Au clusters show a trend of increasing metallicity as the clusters grow and transition from single-layer islands to 3...
1 downloads 12 Views 2MB Size
Article pubs.acs.org/JPCC

Growth and Structure of Cu and Au on the Nonpolar ZnO(101̅0) Surface: STM, XPS, and DFT Studies Matthew C. Patterson,† Xiaowa Nie,‡ Fei Wang,† Richard L. Kurtz,† Susan B. Sinnott,§ Aravind Asthagiri,‡ and Phillip T. Sprunger*,† †

Department of Physics and Astronomy, Louisiana State University, Baton Rouge, Louisiana 70803, United States William G. Lowrie Department of Chemical and Biomolecular Engineering, The Ohio State University, Columbus, Ohio 43210, United States § Department of Materials Science and Engineering, University of Florida, Gainesville, Florida 32611, United States ‡

S Supporting Information *

ABSTRACT: The morphology and electronic structure of Cu and Au clusters deposited via thermal evaporation onto ZnO(101̅0) substrates have been studied via scanning tunneling microscopy (STM) and X-ray photoelectron spectroscopy (XPS). The initial stages of nucleation and growth (∼0.2 ML) of both Cu and Au are compared with density functional theory (DFT) calculations, which show an excellent agreement with the cluster morphologies observed by STM, with Cu nucleating three-dimensional (3D) islands even at small coverage while Au nucleates single-layer islands that grow layer by layer. DFT also gives insight into the diffusion behavior of Cu and Au adatoms on the ZnO substrate, showing strongly anisotropic diffusion barriers for Cu atoms which results in the experimentally observed preferential cluster nucleation along [0001] step edges, whereas Au shows no such anisotropy and Au clusters are observed to have no preferred nucleation sites. XPS results show a slight positive charging of the small Cu clusters at 0.2 ML coverage, which disappears at higher coverage. The single-layer Au islands formed at low coverage show some evidence of positive charging as well, which likewise disappears with increasing cluster size. Additionally, the Au clusters show a trend of increasing metallicity as the clusters grow and transition from single-layer islands to 3D structures, demonstrated by the increasing asymmetry in the Au 4f line shape as a function of Au coverage. In general, the observed charge transfer trends are supported by Bader charge analysis. ZnO(1010̅ ) systems that would explain the reported dependence of methanol yield on Au particle size. In this work, we use STM, X-ray photoelectron spectroscopy (XPS), and density functional theory (DFT) to study the morphology and corelevel electronic structure of Cu and Au clusters on the ZnO(101̅0) surface as model systems for practical Cu/ZnO and Au/ZnO catalysts. STM studies of metal cluster formation allow us to determine the morphology of deposited metal clusters and, in combination with DFT calculations, infer details about the diffusion behavior and nucleation sites for metal atoms at the surface. XPS gives details about the core-level electronic structure of the deposited metals, probing the cluster−substrate interactions well as any charge transfer to or from the metal clusters. It is found from these studies that Cu grows as metallic 3D clusters preferring to align at [0001] step edges from very low coverage, whereas Au initially grows as single-layer clusters with nonmetallic character below a certain critical coverage. This latter growth mode is similar to the Au/TiO2(110)

I. INTRODUCTION Catalysts consisting of Cu or Au supported on ZnO have been widely used in a variety of hydrogenation activities, especially methanol synthesis, for the past several decades.1−5 In various experiments involving the hydrogenation of CO2 to methanol with Au/ZnO catalysts, Au particle size has been shown to affect the methanol yield, with smaller Au particles giving higher methanol productivity per exposed Au surface area.1−5 Cu on ZnO(1010̅ ) has previously been studied with scanning tunneling microscopy (STM)6 and valence band photoemission,7 which indicate that Cu grows almost exclusively as three-dimensional (3D) clusters even at very low coverage and that the clusters are positively charged because of cluster− substrate interactions. Other work with Cu/ZnO catalysts suggests that Cu−Zn surface alloys may be a catalytically active species for methanol synthesis,2 with a Zn/Cu(111) model catalyst showing high activity for methanol synthesis from CO2 and model Cu/ZnO catalysts showing Cu−Zn alloy formation after high-temperature reduction of the ZnO substrate. To date, little core-level spectroscopy exists for Cu clusters on ZnO(101̅0) that might indicate Cu−Zn alloying in the asdeposited clusters, or that would explain the reported charging effects, and almost no surface science data exists for model Au/ © XXXX American Chemical Society

Received: April 16, 2013 Revised: August 16, 2013

A

dx.doi.org/10.1021/jp4037809 | J. Phys. Chem. C XXXX, XXX, XXX−XXX

The Journal of Physical Chemistry C

Article

correction. Zn 2p, Cu 2p, O 1s, and Zn 3p spectra were fitted in CasaXPS using a Gaussian/Lorentzian product line shape, while the Au 4f spectra were fitted using a hybrid Doniach− Sunjic line shape convoluted with a Gaussian/Lorentzian sum.

system, wherein a bilayer growth morphology has been shown to occur and is responsible, in part, for the enhanced catalytic activity of CO oxidation.8,9 Both Cu and Au show evidence of charge transfer from the clusters to the substrate, with lower coverages of both metals more positively charged than high coverages. In synergy with the experimental studies, DFT calculations help elucidate the different growth modes in terms of the different metal−substrate interactions (Cu preferring to bond to O, Au preferring to bond to Zn) and diffusion barriers for single atoms on the ZnO(101̅0) substrate (Cu shows very anisotropic diffusion barriers, whereas those for Au are much more uniform).

III. MODELING DETAILS The DFT calculations reported in this paper were performed using Vienna ab initio simulation package (VASP)14−17 with the ion cores represented by the projector augmented wave (PAW)18−20 potentials provided by VASP. The exchange and correlation energies were treated with the spin-polarized generalized gradient approximation (GGA) with the Perdew− Burke−Ernzerhof (PBE) functional.21−23 A plane wave expansion with a cutoff energy of 400 eV was used for all calculations. A Fermi-level smearing with a Gaussian width of 0.1 eV was adopted. The convergence criterion for maximum force was set to 0.03 eV/Å to allow for feasible geometrical relaxation. For bulk wurtzite ZnO we obtain optimized lattice parameters of 3.290 and 5.313 Å, which are comparable with the experimental values (3.250 and 5.207 Å24) and other DFT results (3.287 and 5.286 Å;25 3.266 and 5.247 Å26). In regard to Cu and Au bulk, the optimized lattice constants we obtained are 3.64 and 4.18 Å, which are 0.03 and 0.10 Å larger than the experimentally measured 3.61 and 4.08 Å, respectively. The DFT-PBE cohesive energies of Cu and Au are −3.48 and −2.98 eV/atom versus experimental values of −3.49 and −3.81 eV/ atom, respectively. Therefore, while DFT-PBE is accurate for Cu it does a poorer job of capturing Au lattice parameters and cohesive energy. Our values for the lattice parameter and cohesive energies match similar DFT studies in the literature.27,28 Au does not match expected trends of the coinage metals in the periodic table, but these deviations (i.e., smaller lattice parameter and larger cohesive energy than expected) are attributed to strong relativistic effects for Au.29,30 The VASP calculations contain fully relativistic pseudopotentials and scalar relativistic treatment for the valence electrons, and we have examined the effect of including spin orbit coupling (SOC). With SOC the PBE values for the bulk Au cohesive energy increases from −2.98 to −3.11 eV/atom and the Cu cohesive energy is unaffected. Using the local density approximation (LDA) functional gives values for the Au bulk cohesive energy of −4.28 (−4.43) eV/atom without (with) SOC and a lattice parameter of 4.06 Å, which agrees well with an earlier DFT-LDA study.31 For Cu the DFT-LDA cohesive energy and lattice constant is −4.57 eV/atom and 3.52 Å, respectively, again with the inclusion of SOC showing no changes. Therefore, while changing the functional can modify the cohesive energy substantially, neither the LDA or PBE functional nor the inclusion of SOC effects is able to reproduce the trend in cohesive energies between Cu and Au. This error in the trend for the cohesive energy for the noble metals has been noted by others, with the functional identified as a potential source of the error.32 We note that PBE-DFT with pseudopotentials has been used in several Au clusters on oxide surfaces, including comparisons with cluster behavior experimentally to explain observed cluster morphology changes.33 Bader charge analysis as implemented by Henkelman and coworkers34,35 was used to examine the charge distribution on the metal and ZnO surface. The top and side views of the ZnO(101̅0) surface are shown in Figure 1, with the (2 × 2) surface unit cell outlined by a dashed box in Figure 1a. The single metal and small metal cluster calculations were initially performed on a (2 × 2)

II. EXPERIMENTAL METHODS STM and XPS experiments were performed in an ultrahigh vacuum chamber with a base pressure of 1 × 10−10 Torr. This chamber is equipped with sputter gun, LEED (SPECS ErLEED 150), thermal Au evaporator, e-beam Cu evaporator, and an Omicron VT STM XA. XPS experiments are performed with an Omicron XM1000 X-ray source (monochromatic Al Kα1 radiation) in combination with a SPECS PHOIBOS 150 hemispherical analyzer. ZnO(101̅0) single crystals with one side polished were purchased from MTI Corporation. Sample surfaces were cleaned by cycles of Ne ion sputtering (30 min) and vacuum annealing to 1000 °C (30 min). This far exceeds the temperature necessary to desorb both surface and bulk hydrogen10 and ensures that any effects of hydrogen in the selvage region is minimized. Sample cleanliness was checked with low-energy electron diffraction (LEED) and XPS; clear and sharp (1 × 1) LEED patterns, and no contamination from carbon or other species visible in XPS, were observed after sputtering and vacuum annealing. In addition to STM and XPS, high-resolution electron energy loss spectroscopy (EELS) data (using an LK2000 EELS spectrometer) shows an absence of adsorbed water and/or hydroxyls (vibrational peak at ∼450 meV loss energy) on the freshly cleaned surface (see the Supporting Information). Before deposition, all evaporators were fully outgassed, and filaments were kept at a lower temperature than the evaporation temperature in order to ensure the Cu/Au evaporated was clean and without contamination. Deposition rates were calibrated by evaporating Cu/Au onto a clean Ru(0001) crystal (both Cu and Au are known to grow as single-layer islands at submonolayer coverages on Ru(0001)11,12) and determining the percentage of the surface covered using STM. The experimental STM and XPS results reported are for Cu deposited under the same conditions (temperature, flux, angle, background pressure) on the same and/or identically prepared ZnO substrates. STM measurements were performed at room temperature (RT) with electrochemically etched tungsten tips. Tunneling voltages and currents were set in the range of 0.5−3 V and 0.3− 1.5 nA, respectively, depending on the surface conditions. XPS measurements were performed at room temperature using monochromatic Al KαI radiation (hν = 1486.6 eV). Highresolution spectra of Zn 2p, Cu 2p, O 1s, Zn 3p, and Au 4f peaks were recorded using a pass energy of 20 eV. To eliminate any effects due to sample charging, the energy scale for XPS spectra of metals deposited on ZnO was corrected by setting the Zn 2p3/2 line to a binding energy of 1021.7 eV and the O 1s line to a binding energy of 530.4 eV (following the procedure described by Wöll13), whereas reference spectra taken from Cu(111) and Au(100) single crystals did not need such B

dx.doi.org/10.1021/jp4037809 | J. Phys. Chem. C XXXX, XXX, XXX−XXX

The Journal of Physical Chemistry C

Article

atoms, but the adsorption energy changes by less than 0.03 eV/ atom with the inclusion of spin polarization and no net magnetic moment is found for metal dimers and larger clusters. The primary measure of the strength of the metal−oxide bonding obtained from the DFT calculations is the adsorption energy. The adsorption energy of n metal atoms on ZnO(101̅0) surface, Eads, is defined by eq 1, where Eslab is the energy of the relaxed ZnO slab without any adsorbed metal atoms, EMetal,iso represents the energy of an isolated single metal atom, Eads‑slab is the total energy of the metal cluster/ZnO system, and n is the total number of metal atoms in the system. Eads = {Eads‐slab − (Eslab + nEMetal,iso)}/n

(1)

By our definition of Eads a more negative value indicates stronger binding of the metal atom to the ZnO surface.

IV. RESULTS AND DISCUSSION IV.A. STM Results. Figure 2a shows an STM image of 0.2 ML Cu deposited on ZnO(101̅0) at room temperature. It is clearly seen that a 3D cluster growth mode (Volmer−Weber growth mode) is adopted for very low Cu coverage at room temperature. Cu clusters preferentially nucleate at step edges, but not all the step edges are equivalent for Cu growth. Most of the Cu clusters are located on the step edges perpendicular to the [121̅ 0] direction, while almost no Cu clusters can be seen on the step edges along the [12̅10] direction. This can be explained in terms of the strongly anisotropic diffusion of Cu adatoms, which is discussed more fully in the DFT Results section below. In addition to the preferential nucleation at step edges, there are also some small clusters formed on the terraces. The ZnO(101̅0) surface as prepared is near perfect (see the Supporting Information), and surface defects are probably not the reason for cluster nucleation on terraces. Dulub et al. observed near-identical morphology for very low Cu coverage on ZnO(101̅0) in a previous study, and observed drastic changes in cluster dimensions and nucleation sites on the same surface deliberately contaminated with CO and/or CO2, which provides additional confirmation that our sample preparation described above results in a pristine surface. The nucleation mechanism for clusters on terraces rather than step edges was explained as the Cu atoms meeting during diffusion and then nucleating, forming nucleation sites on the terrace.6 Most of the

Figure 1. Various perspectives of the ZnO(101̅0) surface with the dashed box indicating the 2 × 2 surface unit cell. The four unique adsorption minima for Cu and Au identified from DFT calculations are represented by yellow circles.

surface unit cell. The diffusion of metal atoms and larger metal clusters was examined using a (4 × 4) surface unit cell, which is sufficiently large to avoid spurious lateral metal−metal interactions between the periodic cells. These surface cells are larger than previous DFT studies,26 which has consequences to the minima and diffusion barriers as will be discussed in more detail in section IV. The slab model consists of six Zn−O substrate layers with the bottom two layers fixed. A 4 × 4 × 1 and 2 × 2 × 1 k-point mesh were used for calculations with the (2 × 2) and (4 × 4) ZnO surface cells, respectively. More dense k-meshes resulted in changes in adsorption energy of less than 0.03 eV/metal atom. An 18 Å vacuum region and dipole corrections were included to avoid spurious periodic interactions in the surface normal direction. All calculations were done with spin polarization and there are nonzero magnetic moments observed at certain binding sites for single metal

Figure 2. STM image of (a) 0.2 ML Cu growth on the ZnO(1010̅ ) surface at RT and (b) 2 ML Cu growth at RT. C

dx.doi.org/10.1021/jp4037809 | J. Phys. Chem. C XXXX, XXX, XXX−XXX

The Journal of Physical Chemistry C

Article

Figure 3. STM images of (a) 0.05 ML Au growth on ZnO(101̅0) at room temperature, (b) 0.4 ML Au on ZnO(1010̅ ) at room temperature, (c) 3 ML Au on ZnO(1010̅ ) at room temperature; (d) height profiles of clusters along the indicated lines in panel a (top) and panel b (bottom), with inset schematics showing Au cluster morphology.

Cu clusters have a size of ∼50 Å in diameter and ∼7 Å in height. The 7 Å height indicates that the clusters are mostly formed with three layers of Cu atoms. Figure 2b shows the results of 2 ML Cu deposited at room temperature. The biggest change is in the number density of clusters; the whole surface is now covered with Cu. The average cluster size did not change appreciablythe largest clusters are around ∼50 Å in diameter and ∼12 Å in heightindicating that the clusters did not grow laterally but added one or two more layers of Cu atoms in height. In contrast to the 3D growth and preferred step edge nucleation displayed by Cu at low coverages, STM studies of Au on ZnO(101̅0) show a distinctly different morphology. Figure 3a shows an STM image of 0.05 ML Au grown on ZnO(101̅0) at room temperature. It is clearly seen that many small, 2D-shaped Au islands are well-dispersed randomly over the surface. Unlike low coverages of Cu on ZnO(101̅0) as in Figure 2a, we find almost no islands at step edges, and autocorrelation of the STM images of 0.05 ML Au coverage shows no preferential nucleation behavior or anisotropic diffusion. The small Au islands have an average size of ∼30 Å across and ∼3 Å high; a sample line profile across one island is shown in Figure 3d. We inspected the island sizes at most

places on the surface and found that all the clusters have about the same height. This 3 Å height represents just one layer of Au, which means a truly 2D growth mode is adopted for Au on ZnO(101̅0) at initial coverage. An STM image of 0.4 ML Au on ZnO(101̅0) surface is shown in Figure 3b. At 0.4 ML coverage, there is an obvious increase of island density on the surface, and step sites are still not preferred for island growth. We found the height of islands ranges from 3 to 9 Å, and most of them are less than 7 Å high, as shown in the second line profile in Figure 3d. The existing 2D Au islands already present from the prior 0.05 ML Au deposition do not act as nucleation sites for new Au atoms arriving on the surface. With the deposition of more Au, the old Au islands remain one or two atomic layers in height. The new Au atoms tend to find unoccupied sites as new nucleation sites. After an island grows to a critical size of 30−40 Å across by 7 Å high, more Au atoms will not stick to the island, and instead new islands will be formed. After the small Au islands cover a large portion of the surface, no new nucleation sites will be created, and the existing Au islands increase in size. As Figure 3c shows, at a coverage of 2 ML, 3D Au clusters of ∼70 Å across and ∼15 Å high are well-distributed all over the surface. The diameter is much bigger than height at this size, meaning D

dx.doi.org/10.1021/jp4037809 | J. Phys. Chem. C XXXX, XXX, XXX−XXX

The Journal of Physical Chemistry C

Article

Figure 4. (a) Cu 2p3/2 spectra for 0.2 ML Cu on ZnO(1010̅ ), 2 ML Cu on ZnO(1010̅ ), and a Cu(111) single crystal for reference. The Cu(111) spectrum was recorded using a pass energy of 10 eV; all others used a pass energy of 20 eV. Shirley backgrounds have been subtracted from all data. Open symbols show the experimental data, and solid lines show the fit envelope. (b) O 1s core-level spectra for clean ZnO(1010̅ ), 0.2 ML Cu, and 2 ML Cu on ZnO(101̅0). Shirley backgrounds have been subtracted from all spectra. Open symbols show experimental data, dotted lines show the individual fit components, and solid lines show the fit envelope.

Figure 5. (a) Zn 3p and Au 4f core-level spectra for clean ZnO(101̅0) and 0.25, 1, 1.25, and 2.5 ML Au on ZnO(101̅0), and from an Au(100) single crystal for reference. Linear backgrounds have been subtracted from all spectra. The indicated binding energies on the plot are the fitted Au 4f7/2 peak positions. (b) O 1s core-level spectra for clean ZnO(1010̅ ) and 0.25, 1, 1.25, and 2.5 ML Au on ZnO(1010̅ ). Shirley backgrounds have been subtracted from all spectra. Open symbols are experimental data, dotted lines are the individual fit components, and solid lines are the fit envelopes.

cannot be directly compared to the peak width from bulk Cu.) The shifted binding energies and broadened peaks might indicate that the Cu deposited on ZnO(1010̅ ) is oxidized, resembling CuO more than metallic Cu at low coverages; however, the high-intensity shakeup peaks at approximately 941 and 944 eV expected of CuO are absent in the 0.2 ML sample.36 Also, the energy difference ΔE between the 2p3/2 and 2p1/2 levels remains basically constant instead of increasing as it should if the Cu in the 0.2 ML sample is in the form of CuO and the Cu in the 2 ML sample is metallic Cu.37 (ΔEbulk = 19.85 eV, ΔE2 ML = 19.89 eV, ΔE0.2 ML = 19.90 eV). The rigid shift in the spectra from 0.2 ML Cu on ZnO(101̅0) is better explained as an initial-state effect. In qualitative agreement with the Bader charge analysis presented below, we see that the bilayer Cu clusters formed at 0.2 ML coverage are positively charged, thus showing higher binding energies than a neutral cluster. Both our XPS results and the DFT calculations are consistent with the angle-resolved photoemission study of Cu on ZnO(101̅0) by Ozawa et al., which showed through analysis of the Cu-induced Zn 3d band bending that evaporation of

the clusters are oblate disks and are still not spherical. The exact mechanism behind the growth of Au clusters and their critical size will require more extensive DFT studies of several atomiclevel mechanisms including Au adatom addition to clusters, Au diffusion on an existing Au cluster, and Au diffusion across the Au−oxide interface (the so-called Ehrlich−Schwoebel barrier). These studies are beyond the scope of this paper and will be addressed in the future. IV.B. XPS Results. Figure 4a shows XPS results for increasing Cu coverage on ZnO(101̅0), corresponding to the STM images of 0.2 and 2 ML Cu deposited on ZnO(101̅0) shown in Figure 2. Qualitatively, the spectra are similar to those of a bulk Cu(111) single crystal, shown at the top of the plot. In the 0.2 ML Cu sample, the 2p3/2 peak shows a shift of 0.5 eV toward higher binding energy with respect to the bulk Cu, whereas the 2 ML Cu sample agrees well (