From Bulk to Atoms: The Influence of Particle and Cluster Size on the Hydrogen Evolution Reaction

2.1 Sample preparation

As deposition support material titanium sheets (99.93%, Alfa Aesar) with a thickness of 0.5 mm and an area of 10 × 10 mm² were used. The experimental determination of the native oxide film thickness was done using X-ray photoelectron spectroscopy (XPS). A value of 2.5 ± 0.6 nm was found for the thickness of the native titanium dioxide layer. This is in accordance with several thin film studies reported in the literature [27], [28], [29], [30], [31], [32], [33]. As a reference material, a 0.25 mm thick platinum foil (99.995%, Alfa Aesar) was used. All samples were cleaned for 5 min each with acetone, isopropanol and ethanol in an ultrasonic bath. The solvents were all purchased as analytic grade (Carl Roth).

2.2 Characterisation

Atomic force microscopy (AFM) and transmission electron microscopy (TEM) were applied to investigate the roughness and topology of the surfaces, as well as the size of the deposited clusters and nanoparticles. For AFM measurements a commercial benchtop microscope Nanosurf EasyScan 2 operated under ambient conditions with a vibration-free mounted table was used. The images were analysed with the open source software Gwyddion v2.53. High-resolution TEM images were taken in cooperation with the Jooß group from Göttingen (Germany) on a research device. All images were taken using the secondary electron (SE) mode at a potential of 10 kV with an Everhart-Thornley detector for the overview images and a through-lens detector for the zoomed-in images.

Photoelectron spectroscopy measurements (XPS and UPS) took place within the integrated UHV laboratory DAISY-FUN within the Jaegermann group in Darmstadt (Germany). For XPS a high-resolution monochromatic Al Kα X-ray source at 1486.74 eV (XR 50, Specs Surface Nano Analysis GmbH) was used. The spectrometer was calibrated to the core levels of clean coinage metals (copper, silver, gold) as well as to the Fermi edge of silver. To estimate the energy resolution the full width at half maximum (FWHM) of the Au 4f_7/2 core line was investigated showing a FWHM of 0.5 eV. The analyser was a Specs Surface Nano Analysis GmbH Phoibos 150 hemispherical analyser.

The electrochemical measurements took place in an identical setup to the electrodeposition of platinum nanoparticles using the Gamry (Interface 1000) potentiostat (see Chapter 2.1.1). The setup was operated in a three-electrode configuration. An aqueous solution of 0.1 M sulphuric acid (analytical grade, Carl Roth) was the electrolyte for all electrochemical experiments. For cyclic voltammetry measurements, a scan rate of 10 mV s⁻¹ was applied.

3.1 Imaging

Surface roughness has a direct influence on the total number of platinum atoms reachable by the solvent in electrochemistry and hence on the overall catalytic performance. Thus, before a well-founded insight into the catalytic activity is possible, a picture of the topologies of the surfaces must first be obtained. AFM was implemented, to investigate the surface of the titanium foils and platinum reference. Since both metals were produced by a similar rolling process, the topologies differ insignificantly. In Figure 1 the topology of the platinum reference is shown in a 56 μm² large image section. For all samples, the mean roughness lies within the 1–2 nm range with a roughness factor of 1.04 ± 0.02. This implies a very planar surface with only small deviations. Therefore, the effective surface area deviates only a few percents from the geometric area.

Fig. 1:

AFM image of the bulk platinum reference sample Pt_bulk. The colour bar indicates changes in the height of the sample (z-direction).

In addition to the topology of the blank support material, it is also necessary to gain information about the deposited particles. In order to achieve relevant data for the catalytic activity, a homogeneous coverage of isolated particles is needed. To verify this, additional samples were prepared on TEM grids for both deposition methods (magnetron sputtering and electrodeposition). A Matlab (R2016a) script was used to analyse the images obtained. To determine the size distribution and coverage, a floating background was subtracted first before converting the images from grey scale to binary, which allows a more simplified identification and analysis of individual particles. In order to obtain a statistically meaningful quantity of particles for both the cluster/particle size and the total coverage, two different image sections were examined per sample. A larger section for determining the total coverage and one section with greater magnification for determining the size distribution based on the measurement of individual particles. The corresponding TEM images can be seen in Figure 2. Figure 2a illustrates Pt₁₆-clusters deposited on a carbon TEM-grid, which was prepared by sputtering in analogy to the samples Pt_atom, Pt₁₀ and Pt₁₃. The slightly larger Pt₁₆-clusters were used to achieve better contrast, since larger particles are easier to identify within the TEM images. The obtained atomic coverage is in agreement with the coverage received from ion-current measurement during deposition within 20–40%. Figure 2b illustrates platinum nanoparticles deposited by electrodeposition on a TiO₂ TEM-grid synthesised under the same conditions like the Pt_nano samples. Whereas the Pt₁₆-clusters have a diameter of (0.9 ± 0.3) nm, the nanoparticles have an average diameter of (232 ± 90) nm. By TEM imaging of a sample cross-section, it became visible that the nanoparticles possess a half-sphere shape. Applying the spherical cluster approximation (SCA) [36] to the Pt atoms within the nanoparticles as hard spheres with a Wigner-Seitz radius of 136 pm (calculated from the density of bulk platinum) [34], an average number of 3 ⋅ 10⁸ atoms per nanoparticle is obtained. The number of Pt atoms on the surface adds up to 0.6% of the total atom number per nanoparticle.

Fig. 2:

TEM images of differently deposited Pt particles, overview and zoomed in section. (a) Magnetron sputtered Pt clusters with a diameter (0.9 ± 0.3) nm on a carbon TEM-grid. (b) Electrodeposited Pt nanoparticles with an average diameter of (232 ± 90) nm on a TiO₂ TEM-grid.

3.2 Photoelectron spectroscopy

After preparation, all samples were introduced directly into the integrated UHV system connected to the analysis chamber in which XPS and UPS measurements were performed in sequence. Firstly all samples were checked for unwanted contaminations other than residual carbon which could not be washed completely of previously or which readsorbed under ambient conditions during transfer into the vacuum system. In addition, the intensity of the Ti- and Pt-core lines were utilised to validate the Pt coverage obtained by the ion-current measurements during cluster deposition via magnetron sputtering. In order to calculate the coverage the dampening of the Ti2p intensity assuming a homogeneous layer of platinum was used. Two inelastic mean free paths (IMFPs) are needed for the calculation: Ti2p electrons in platinum from the NIST IMFP Database (λ = 13.34 Å) [37] and Ti2p electrons in TiO₂. Due to a missing experimental value for the latter, the TPP-2M algorithm within the NIST IMFP Database for inorganic compounds was used with the following data for titanium dioxide: the number of valence electrons per molecule = 16, band-gap for native titanium oxide = 3.05 eV and a density of 4.22 g cm⁻³ [38] resulting in an IMFP of λ = 27.7 Å. This approach has a high amount of uncertainty leading to an error margin of approximately a factor of two. But this is still a feasible tool to re-validate the overall particle coverages. All values of the coverages determined from the current measurements are within this margin.

A comparison of the Pt4f_7/2 core-levels reveals distinct electronic differences between the samples. For Pt_atom the core level is 72.9 eV, 72.3 eV for Pt₁₀ and decreasing further to 72.1 eV for Pt₁₃, 71.3 eV for Pt_nano and 71.2 eV for Pt_bulk as seen in Figure 3. In the case of the latter two samples, the usual line-shape for metallic core-lines can also be detected, which was not observed for the deposited clusters. The result for the Pt_nano sample is in agreement with platinum nanoparticles of comparable size on amorphous titan dioxide films [39], [40]. Compared to results from references [41], [42], the Pt4f_7/2 core-levels of the small cluster species (Pt_atom, Pt₁₀, Pt₁₃) are increased by 0.5–0.8 eV. This discrepancy can be explained by different TiO₂ support effects in the case of crystalline oxide. The core-level shifts are due to modified final state screening precipitated by titania’s electrophilicity and by particle size effects. Whereas electrons in metallic platinum easily screen a core hole created by the photoelectric effect and therefore stabilising the final state, this does not hold for very small particles. The final state energy is shifting due to a localised charge which is proportional to the inverse particle radius. Also, a partial polarisation of the valence electrons occurs since there are no delocalised conduction band electrons available [43], [44], [45], [46]. Complementary the electronic structure of the support material is relevant for both the initial and the final state effect. For titania-supported platinum clusters, this leads to a small positive charging of the cluster due to the high electronegativity of the oxygen in titania in comparison to platinum [26], [47]. Even though titanium is less electronegative, the resulting total charge will still be positive over the whole cluster (e.g. +0.97 e for Pt₁₃ according to Wang et al. [26]). A more conductive support can compensate this effect up to a specific margin. Also, the charge carrier density, acidity and defect concentration have an impotent impact on the initial state effect and therefore on the core shift in comparison to the particle bulk material [48], [49]. Overall a strong influence of cluster size onto the electronic structure can be seen.

Fig. 3:

Pt4f_7/2 binding energies for different platinum species on native TiO₂ before and after electrochemical measurements. The dashed line indicates the core-level for bulk platinum as known from literature [48].

A change in geometric and electronic structure for clusters and nanoparticles, in particular, precipitated though electrochemical measurements are well known from literature [50], [51], [52], [53], [54]. Which mechanism (agglomeration, migration, Oswald ripening, dissolving, detachment) is dominant depends strongly on the particle substrate combination [50], [51], [52], [53], [54]. As stated in Figure 3 for the majority of samples only small shifts in binding energy for the Pt4f_7/2 core-levels to lower values can be observed after electrochemical treatment indicating only minor modifications of the overall electronic structure of the platinum particles. This is not the case for platinum atoms supported on titania (Pt_atom), in as much as a clear shift of the binding energy to lower values can be recorded. The initial value for the Pt4f core-line of 72.9 eV shifts to 72.6 eV implying the formation of small clusters through sintering during the CV measurements. In contrast to larger particles, the platinum atom has much lower adsorption energy and a fewer number of possible anchoring sites, resulting in higher mobility on the surface. The platinum atom seems to agglomerate to dimers and small clusters. This matches with the obtained Pt4f binding energy of 72.6 eV which fits within the gap of Pt_atom and Pt₁₀. In addition, the total intensity of the platinum core-lines for all samples is altered through the electrochemical treatment resulting in a loss of 30–60% after 5 h of measurement within the electrolyte. The samples Pt₁₀, Pt₁₃ and Pt_nano lose material from the surface rather than to change their structure due to ripening mechanisms. Other core-lines were not subject to major change apart from expected small quantities of remaining –OH groups.

Valence band spectra reveal additional information of the electronic structure of the catalytic systems near the Fermi-edge. Relating to Figure 4 between 4 eV and 8 eV a superposition of metal (Ti, Pt) and oxide valence states can be found. In the UPS analysis of the plain native TiO₂ surface the valence band edge appears at 3.1–3.5 eV, which is in good agreement with values from the literature [55], [56]. The small Pt-species Pt_atom, Pt₁₀ and Pt₁₃ did not reach a metallic state, regarding the lack of density of states (DOS) near the Fermi-level. In contrast, Pt_nano and Pt_bulk show a high intensity of DOS at the Fermi-level indicating metallic behaviour matching with the value of Pt4f core-lines for very large particles and bulk platinum. Density functional theory calculations from Wang et al. showed that the DOS in the valence gap of TiO₂ is filled by deposited Pt_n-clusters with n = 5, 8, 13, 19 [26]. The resulting valence band gap narrows with an increase in cluster size analogous to the experimental data obtained here. In the UPS spectra of the platinum deposited surfaces the band edge can be found at 1.5–1.9 eV for Pt_atom, 0.7–0.8 eV for Pt₁₀ and 0.25–0.4 eV for Pt₁₃. This is in good agreement with calculations from Wang et al. who received 3.5 eV for pure titania, 0.87 eV for Pt₈ and 0.45 eV for Pt₁₃ supported on TiO₂(101) [26].

Fig. 4:

UPS valence band spectra for different platinum catalysts on native TiO₂. The energy of excitation was hν = 40.81 eV (Helium II). For reasons of clarity the spectra are stacked, all lines below −2 eV are at zero intensity. The inserted graph shows the difference-spectra of the platinum catalysts (as-is spectra subtracted by the TiO₂ spectrum). Because the total amount of platinum is varying for different samples, the intensities of the measured photoelectrons of the various spectra should not be compared to each other.

Valence electrons near the Fermi level as well as the highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) positions are of decisive importance for the activity in catalysis [57]. According to Figure 4 at least the position of the HOMO can be tuned by optimising the cluster size. Further information regarding the LUMO could be gathered employing i.e. inverse photoemission spectroscopy. According to size-dependent shifts of the unoccupied states for small Pt-clusters can be found in literature [58]. To reduce hydrogen, an electron has to go through a two-step process (if not concerted) in which it passes from the substrate (electrode) to the cluster and finally to the hydrogen. Both processes require different optimal LUMO positions. The LUMO of the cluster must be located above the reduction potential of hydrogen, otherwise hydrogen can not be reduced. But to maximise the lifetime of an excited state with the electron located on the cluster, the electron needs to be trapped there. This can be realised with a low position of the LUMO beneath the conduction band of the supporting material. In the future, it might be interesting to investigate also the DOS of the unoccupied states for supported clusters. But the here presented data clearly suggest that the valence orbitals could be tuned by changing the particle size of the catalytic active cluster [59].

3.3 Electrochemistry

To ensure proper functionality of the used electrochemical setup and to gain a benchmark, two reference systems are studied, i.e. a platinum foil Pt_bulk and clean TiO₂. They are also important for comparison with the deposited samples, as well as with data from the literature. All cyclic voltammetry (CV) measurements are taken after a set of measurements to make sure, that quasi-stationary conditions were reached. The set consisted of impedance measurements between 0.1 Hz and 100 kHz ranging from 0.0 to −0.1 V vs RHE (reference hydrogen electrode) with 10 points per decade, cyclic voltammetry from +0.5 to −0.6 V vs RHE with 50 mV/s (15×) and with 10 mV/s (10×). In between the individual measurements, the open-circuit potential was held for 300 s each. In total each sample was in contact with the electrolyte for about 5 h. The measurements are done in 0.1 M H₂SO₄ with a scan rate of 10 mV s⁻¹. Each cycle starts at an electric potential of +0.5 V vs the value of the reference hydrogen electrode moving on to the turning point of −0.6 V vs RHE and back to +0.5 V vs RHE. As shown in Figure 5, the blank TiO₂ reaches a maximum of −1.6 mA cm⁻² at −0.6 V vs RHE and is therefore nearly inactive for the HER, which is in good agreement with the data from Proch et al. (under 2 mA cm⁻² in 0.1 M HClO₄, identical scan rate) [33]. The feature within the current densities slightly above 0.1 V vs RHE indicates the desorption of weekly bound hydrogen onto the electrode surface [60]. The negative branch below 0 V vs RHE is appertaining to the hydrogen evolution reaction. The hysterical course in this region is typical evidence for hydrogen bonding onto the platinum surface [61], [62]. Those features are more distinct for the Pt_bulk and Pt_nano samples rather than for the cluster deposited surfaces. The sample Pt_atom shows only a slight increase in catalytic performance compared with the TiO₂-reference. If the used catalyst lowers the overpotential far enough as seen for the Pt_bulk reference, significant current densities of over 40 mA cm⁻² can be reached. This is also the case for the deposited platinum particles Pt_nano, where nearly identical values to the bulk material are obtained.

Fig. 5:

CVs of different platinum catalysts on TiO₂ support compared to bulk platinum in 0.1 M H₂SO₄ acquired with a scan rate of 10 mV s⁻¹.

Considering the high current densities achieved with TiO₂ as support material, problems with electric resistance due to the isolating properties of the native oxide on the titanium foil can be ruled out. In addition to electron tunnelling through the thin oxide layer, electron transport can also occur through the bulk titania which is mainly determined by a strong electron-phonon coupling [63], [64]. Due to the thickness of the native oxide layer of about 2.5 nm and an overall electric conductivity of the TiO₂ in the range of 10⁻³ (Ω m)⁻¹ − 10⁰ (Ω m)⁻¹ [32], [65], the resulting resistance of the native oxide layer is in the range of the detection limit of about 1 mΩ. Hence the electron transport can be disregarded and only the electrocatalytic activity is investigated for our experiments. When discussing the catalytic performance the current density value for the titanium dioxide as a reference system was subtracted for further current densities, so only the additional platinum-induced catalytic activity is taken into account.

The wide range of catalytic species, as well as the large gap of overall used catalytic material, leads to difficulties in the comparability of catalytic activity. The usual comparison of the considered catalysts by the selection of a current density (e.g. −1.0 mA cm⁻¹) to be reached and the consideration of the corresponding electric potential is not suitable in this case. The so gained potentials could only be analysed for the as-is samples but cannot be put in relation to the used catalyst amount which is necessary here. The catalytic activity needs to be evaluated as a current density at a potential of choice. The used CVs are not IR-corrected in addition due to the variation in the onset potentials for the HER. With an IR-correction the lowest potentials for the bulk platinum would be about −0.15 V vs RHE which is still below the onset potentials for the small cluster species at this low cluster coverage. The catalytic activity will therefore depend on the potential and coverage choosen. The different CVs can still be compared since the sum of resistances (electrolyte, cables, support) is equal for all investigated samples (22 ± 0.5 Ω, determined by impedance measurement). This resistance is also the reason why a linear increase can be found for high current densities rather than an exponential one.

The current densities at −0.6 V vs RHE for the supported monodisperse platinum species (TiO₂ reference subtracted) with −1.18 mA cm⁻¹ for Pt_atom, −5.56 mA cm⁻¹ for Pt₁₀ and −6.86 mA cm⁻¹ for Pt₁₃ are still low compared to the value of the large nanoparticles or even to the bulk material. However, these values are obtained using a different amount of platinum and can hardly be compared with each other by considering the current densities alone. To gain a better inside, the total mass of platinum was determined for all samples using the coverages (see Table 1) and set in relation to the current densities. The results can be seen in Figure 6 where the current densities at −0.6 V vs RHE are compared with the mass-specific current densities for all samples. In summary, much higher current densities per mass of −229 mA μg⁻¹ for Pt₁₃ and −172 mA μg⁻¹ for Pt₁₀ are achieved for the clusters, than for the other deposited samples (Pt_atom: −36 mA μg⁻¹, Pt_nano: −1.16 mA μg⁻¹).

Fig. 6:

Current densities and mass-specific current densities at −0.6 V vs RHE in 0.1 M H₂SO₄ for different platinum catalysts on TiO₂-support compared to bulk platinum. For Pt_nano and Pt_bulk the specific current densities for the surface atoms only are shown in addition.

In terms of activity per mass, Pt_nano and Pt_bulk show a low specific catalytic activity due to the unfavourable surface-to-volume ratio. To allow a correlation with these samples as well, only the surface atoms can be used. Specifically, this corresponds to the first atomic layer for Pt_bulk and the outer most layer for the Pt-nanoparticles. Here, the bulk platinum reaches a specific activity of −85 mA μg⁻¹ and Pt_nano possesses a value of −193 mA μg⁻¹. This demonstrates that the catalytic activity varies only by about 20% considering Pt clusters with 10 atoms and nanoparticles with 10⁸ atoms when taking the number of surface atoms into account. Noticeable is the reduced value of −36 mA μg⁻¹ for the sample with the least efficient electrocatalyst Pt_atom. If this is caused by an unideal LUMO position or by a missing second-layer of platinum, which is needed to achieve a proper catalytic activity for titania-supported platinum species, as [26] is still an open question.

If the catalytic activities are directly compared with the Pt4f core levels, no clear trend can be found. This is also the case for the valence band spectra and HOMO positions. However, Weber et al. were able to achieve a direct correlation of catalytic activity with core level position for small platinum clusters for ethanol electro-oxidation on indium tin oxide (ITO) [66] or the oxygen evolution reaction on ITO and glassy carbon [67]. Nevertheless, the same clusters on ITO used for oxygen reduction do not follow a trend regarding the Pt4f core-levels [13]. Effects induced due to changes in the valence electronic structure might be of minor importance compared to the influence of reachability and the overall amount of active sites. Still, there might be an optimum for both in the case of Pt₁₀ and Pt₁₃, but to clarify this more experiments with a wider range of cluster sizes are required. Here, the best activity can be seen for the subnanometer-clusters. Taking the total amount of platinum into account this is even much more pronounced. If only the surface atoms are considered the catalytic activity for the platinum clusters and the larger nanoparticles lies within 20%.

A high reactivity with ambient oxygen can also have altered the activity since for all samples the measurements took place under application-related ambient conditions. The particles may have an even higher catalytic activity without the presence of residual gases, as already known from the literature [66], [68]. The active site tends to be the (partially) oxidised platinum on the oxide to cluster boundary [69], creating an increase in activity for the H-H coupling on the surface, especially for small Pt-clusters. An activation with elemental oxygen forming PtOx is not possible and would even result in a loss of catalytic performance [66], [68]. The activation effect drops off quickly for additional layers of platinum atoms and hence diminishing the activity for large clusters [70]. This predicts the highest activity per atom to be found for clusters without core atoms [71] which is also observed here for Pt nanoclusters having a size of about 10 atoms.

In agreement with the results from Proch et al. [68] after the electrochemical measurements, some degradation of the platinum catalysts, especially for the small clusters were found. These could be seen as a Pt4f core line intensity loss as well as shifts to lower binding energies in the case of Pt_atom after hours within the electrochemical setup, mostly induced through detachment and in the case of Pt_atom additional agglomeration took place. Since the positions of Pt4f core lines are strongly dependent on the cluster size, they give a good insight into the change of the size during the electrochemical measurements. For all investigated platinum catalysts, besides of the supported platinum atoms, no Pt4f shift could be observed. This is in good agreement with results from the literature [72] were a dissolution of platinum and therefore a reduction in particle size was found at potentials below −0.9 V vs RHE and could not be observed at −0.6 V vs RHE as used in this work. As described before, the increase in cluster size, which causes the core level shifting towards lower binding energies for the platinum atoms, can be explained due to the high mobility of single atoms resulting in agglomeration to slightly larger species on the titania surface [73]. After the set of electrochemical measurements, about 5 h in total, up to 60% of Pt4f intensity loss was observed for all deposited particles (Pt_atom, Pt₁₀, Pt₁₃ and Pt_nano) independent of size and deposition technique. The soft-landing regime ensures the integrity of cluster geometry, but results in a much smaller particle-support interaction and binding energy than with embedded or even implanted particles [74]. Since the particles are only soft-landed on the TiO₂ substrate and not further embedded into the surface they can be detached easily if in contact with a liquid electrolyte.