Activity and Evolution of Vapor Deposited Pt-Pd Oxygen Reduction Catalysts for Solid Acid Fuel Cells

Fuel cell polarization curves (corrected for ohmic losses) for electrodes with selected sample compositions are shown in Figure 1. Electrodes containing Pd-rich catalysts sustain a high cell potential with increasing current density, operating with much greater efficiency than Pt-rich electrodes.

Zoom In Zoom Out Reset image size

Two metrics were adopted as a quantitative basis for comparing the performance of the experimental cathodes. The first, the mass activity (i_m), is simply defined by the cell current per mg of catalyst at an iR_Ω-corrected cell voltage of 0.8 V. The second, the charge transfer resistance (R_ct) is the width of the PEIS arc at a cell voltage of 0.8 V, shown in the lower portion of Figure 1. Empirically, the charge-transfer resistance is the instantaneous slope of the iR_Ω-corrected polarization curve, in analogy with Ohm's Law, but instead comprising the non-ohmic loss channels within the electrode. Figure 2 summarizes these metrics for the tested compositions (here, we plot the inverse of R_ct, which has the form of a conductance); a full table of results is shown in Table I.

Zoom In Zoom Out Reset image size

An equivalent circuit was used to obtain the full charge transfer resistance of the MEA at 0.8 V in cases for which the impedance arc did not intersect the real axis at low freqeuency. The equivalent circuit is shown inset in with the data and fits in Figure 1 below. R_Ω was fixed at the intercept of the arc with the real axis, and the parameters R⁻_ct and C⁻, the charge transfer resistance and capacitance of the anode, respectively, were fixed at the values of obtained for the response of one electrode in an unbiased symmetric cell measurement in H₂/H₂ at 250 °C (0.03 Ω and 0.002 F, respectively) and 75 °C dew point. In a biased symmetric cell measurement, the total cell impedance at a current density of 175 mA/cm² was identical to the unbiased impedance, indicating that negligible error is introduced by the subtraction of an unbiased anode response from the full cell impedance. The anode response is shown inset in the lower portion of Figure 1. The charge-transfer process on the cathode is modeled by the resistance R⁺_ct in parallel with a constant phase element Q⁺. A third circuit element was required to accurately reproduce the low-frequency behavior of the curves; here we used the generalized RC circuit characterized by the variables R⁺_g and C⁺_g. In the low-frequency limit, capacitors approach infinite real impedance and the total charge transfer impedance is the sum of the fitted resistances, less R_Ω.

Equivalent circuits are not unique descriptors of electrochemical processes, and there are many possible equivalent circuits that can describe the impedance response observed in SAFCs. For example, a finite-length Warburg diffusion element can be substituted for the RC circuit used with little effect on fit quality. For this reason, we have not assigned a particular process to the third circuit element in the absence of more detailed parametric studies.

We observe that high activity shown by the experimental cathode compositions emerges at a cell potential lower than 0.9 V, and at a relatively high current density. At an iR_Ω-corrected cell potential higher than 0.9 V, the Pt control electrode is at least as active as the experimental cathodes. This low-overpotential regime is commonly associated with oxygen activation in low/intermediate temperature fuel cells, and efficiency losses at greater overpotentials are usually attributed to mass-transport losses from reactant diffusion limitations in the porous electrode. Also notable is the depression of the PEIS arcs in Figure 1 below the real axis, a profile characteristic of distributed networks such as within three-dimensional porous electrodes.²³

Taken together, these observations prompted an investigation of the gross character of the experimental electrodes with scanning electron microscopy to observe cathode thicknesses and porosity. Representative micrographs of opposite ends of the compositional range are shown in Figure 3. Electrodes with a catalyst composition of Pd₁₁Pt₈₉ are virtually identical to those with a composition of Pd₈₄Pt₁₆, but the latter electrode is nearly 5 times more active at 0.8 V. We observed no structural difference in the MEA-level characteristics of the electrodes tested, and concluded that electrode structure was not a source of the observed activity enhancement.

Zoom In Zoom Out Reset image size

To investigate oxygen concentration effects on the experimental electrodes, the testing protocols summarized in Figure 1 and Table I were repeated on a Pd₈₄Pt₁₆ electrode and a Pt control with cathode feeds of 100% oxygen to a terminal voltage of 0 V. All other parameters were identical to previous tests (250 °C, gases hydrated to 75 °C dew point). Polarization curves are plotted in Tafel form (logarithmic current scale) in Figure 4. Immediately apparent is the lower open circuit voltage of the the Pd-rich cathode, an indicator of a mixed potential due to anodic metal oxidation processes on the electrode, and likely to be enhanced by carbon oxidation. The presence and role of carbon in the electrodes is elaborated upon in later sections. In the case of Pt cathodes, the gain in current density at a given overpotential η is a factor of approximately 2. This gain factor is also reflected in the doubling of the charge-transfer resistance of the Pt electrode after switching from pure oxygen to air. For electrodes with first-order oxygen reduction kinetics, a gain factor of 5 is expected, and lower values have been attributed to ionic transport limitations²⁴ in phosphoric acid and polymer electrolyte membrane fuel cells.²⁵ The existence of these limitations is not surprising for the case of SAFC electrodes, since CDP has a relatively low proton conductivity.²⁶ In comparison with Pt, Pd₈₄Pt₁₆ electrodes show no appreciable oxygen gain until approximately an iR-free voltage of 0.85 V. At a real cell voltage of 0.8 V, air operation results in an increase of charge transfer resistance of less than 20%. The lack of an appreciable oxygen gain in the Pd₈₄Pt₁₆ system is surprising; virtually no change in the rate of oxygen reduction results from a fivefold increase in oxygen concentration.

Zoom In Zoom Out Reset image size

Reaction rates in electrochemical systems far from equilibrium obey what are known as Tafel kinetics. Both the Pt control and the Pd-rich experimental electrode show Tafel behavior, that is, a linear relationship between cell voltage and the logarithm of the current. From the perspective of classical electrochemistry,²⁷ the appearance of a Tafel slope implies the existence of a rate-limiting electron transfer step at an electrode. While this may be strictly true under ideal conditions, Gottesfeld¹² and Adler²⁸ have reviewed scenarios in which the Tafel slope may not be indicative of the intrinsic kinetics of the system. With these admonitions in mind, we can consider the obvious disparity in the Tafelian behavior of the electrodes and the unusual oxygen concentration dependence of Pd₈₄Pt₁₆ shown in Figure 4. Two well-defined linear regions are present in the oxygen-fed curve for Pd₈₄Pt₁₆. Fits to the empirical Tafel equation

with a and the Tafel slope, b, as adjustable parameters and the current density j as the independent variable returned a slope of 54 mV/decade in the low current region (approximately 2 mA cm ^{− 2} to 200 mA cm ^{− 2}, 0.93 V to 0.83 V), and 128 mV/decade at high currents (approximately 300 mA cm ^{− 2} to 950 mA cm ^{− 2}). The Pt electrode is linear through only one region, with a slope of 285 mV/decade from 100 mA cm ^{− 2} to 1000 mA cm ^{− 2}. At lower overpotentials, the Pt curve is smoothly varying and Tafel fits over a decade of current density were unsatisfactory. We stress that these values are "apparent" Tafel slopes rather than true kinetic parameters. In acidic media²⁹ and in the PEMFC system,³⁰ the intrinsic kinetics of the oxygen reduction reaction on Pt were shown to be modified by adsorptive coverage of the metal surface by chemisorbed oxide species, with a net effect of reducing the observed Tafel slope. Here, "oxide" is the common term for chemisorbed hydroxyls resulting from the water discharge reaction

and distinct from the bulk oxide. Surface coverage by water discharge products is potential-dependent, and upon stripping of these adsorbates at cathodic potentials a transition to the intrinsic Tafel slope is observed. The accepted intrinsic value of the Tafel slope for the ORR on an oxide-free Pt metal surface at 25° C is 118 mV/decade of current density.²⁹ In the case of first-order oxygen reduction kinetics and a temperature-independent transfer coefficient, we can scale this value to higher temperatures by simply multiplying by the ratio of the absolute temperatures of interest, for a apparent Tafel slope of 207 mV/decade at 250° C. The value of 285 mV/decade measured for Pt in an SAFC at 250° C is greater than this theoretical value, but in the absence of a proton transport correction, this represents fair agreement. The slopes derived for Pd₈₄Pt₁₆, however, are considerably lower than what might be expected from a Pt-like catalyst at 250° C.

Cyclic voltammetry (CV) was used to probe electrochemical oxide formation and reduction on the catalysts at 250 °C by purging the cathode feed with 75 sccm of ultrahigh-purity Ar (75 °C dew point) while maintaining the anode H₂ feed to create a pseudo-reference electrode. For each voltammogram the working electrode voltage was cycled three times at 50 mV s ^{− 1} from the open circuit potential (approximately 0.75 V) to a vertex potential of 1.2 V and then swept in back to 0.2 V. Voltammograms of Pt and Pd₈₄Pt₁₆ before and after a 12h test are shown in Figure 5. Before fuel cell tests, each cathode had received exposure to dry or humidified argon only. We note in passing the absence of hydrogen underpotential deposition (UPD) peaks in either voltammogram, a feature of SAFC electrodes that obviates the determination of electrochemical surface area via this method. The Pd-rich sample begins to form oxide at approximately 0.8 V during the anodic sweep, while the onset of oxidation in the Pt sample occurs closer to 0.9 V. Similarly, the maximum (absolute) current of the broad Pd₈₄Pt₁₆ oxide reduction peak appears at 0.74 V (and may be better described by two peaks), while the Pt reduction peak maximum is found at 0.85 V. The initial reduction peak charges were found to be 391 mC for the Pd₈₄Pt₁₆ sample and 220 mC for Pt.

Zoom In Zoom Out Reset image size

After testing, the relative relationships of the peaks are preserved (reduction peak maxima at 0.77 V for Pd₈₄Pt₁₆ and 0.88 for Pt) but a decrease in peak charge is observed (37% loss for Pd₈₄Pt₁₆, 27% for Pt). The second, more cathodic Pd₈₄Pt₁₆ reduction peak is no longer present after operation. The observed decreases in peak charge, accompanied by similar decreases in double layer capacitance, are consequences of catalyst particle coarsening, an effect confirmed by XRD and SEM as described below. The cathodic offsets of Pd₈₄Pt₁₆ surface redox with respect to Pt are in accordance with those observed for oxygen reduction on Pd catalysts in acidic media, implying an absence of a heavily Pt-enriched surface phase.

Returning to the polarization curves shown in Figure 4 (and, by extension Figure 1) with context from the cyclic voltammograms in Figure 5, it is clear that the Pd-rich catalyst favors a coverage of oxo-type species through a potential range disparate from the Pt control catalyst. This coverage is consistent with a reduced Tafel slope, and can explain a low oxygen gain in the case of active sites that are nearly saturated with adsorbed oxide at mildly anodic potentials, even in air.

Under constant voltage operation at 0.6 V in H₂/air at 250 °C, the current density produced by Pd-rich cathodes rapidly declines. Subsequent to the 12 hour testing protocol outlined above, the Pd₈₄Pt₁₆ sample loses 50% of its maximum current density at 0.6 V. Electrodes with less than 60 at% Pd do not exhibit such rapid degradation, but rather have a stability similar to that of Pt. Current density at 0.6 V and area specific ohmic resistance (ASR) for selected samples during the first 12 hours of testing are plotted in Figure 6. The maximum current density at 0.6 V (j_max(0.6V)) and the current density at 0.6 V after 12 hours of testing (j_12h(0.6V)) are summarized for each sample in Figure I. Measured changes in the ASR of the Pd-rich samples are 10% or less (as shown in Figure 6) and are insufficient to explain the large change in current density at 0.6 V, implying the existence of another degradation channel.

Zoom In Zoom Out Reset image size

The initial enhancements in ORR activity, the loss of current density at 0.6 V, and the disappearance of the second oxide reduction peak for Pd₈₄Pt₁₆ (Figure 5) are correlated, prompting us to investigate structural, chemical, and morphological origins of these phenomena. Aside from SEM imaging, the analytical techniques used were applied to catalysts that had been isolated from their supporting CDP by dissolution of the electrolyte in water with the intent of increasing signal ratios.

FESEM micrographs of Pt, Pd₅₃Pt₄₇, and Pd₈₄Pt₁₆ catalysts on CDP supports before (pre-FC) and after (post-FC) fuel cell testing are shown in Figure 7. While pre-FC Pd₅₃Pt₄₇ appears very similar to Pt, pre-FC Pd₈₄Pt₁₆ displays a distinct morphology of large ( ∼ 100 nm diameter) spherical particles decorating the conformal nanoparticle film that coats the CDP support. The contrast of both the large particles and smaller particles appears enhanced relative to the rest of the image. These features may be enriched in Pd beyond the average bulk Pd content of Pd₈₄Pt₁₆; Pd has a higher secondary electron yield than Pt, and Pd-rich zones should appear brighter in an electron image. After testing, a coarsening and reorganization of the catalytic films is evident, consistent with an effect identified in Pt films in SAFCs previously.²⁰ The Pd₅₃Pt₄₇ electrode appears to have coarsened less than the Pt sample, but the Pd₈₄Pt₁₆ catalyst shows considerably more agglomeration and a greater density of discontinuities between particles than either of the other samples. Large agglomerates remain present in post-FC Pd₈₄Pt₁₆ and are approximately the same size as in the pre-FC sample. EDS spectra of washed post-FC catalysts indicated bulk Pt:Pd ratios within ± 2 at% of the pre-FC catalysts.

Zoom In Zoom Out Reset image size

Cu Kα X-ray diffraction patterns are shown in Figure 8 for pre-FC and post-FC catalysts absent of CDP. All samples show a crystalline fcc structure with no evidence of bulk oxides, with the exception of the post-FC Pd sample, which shows a weak PdO component. For strong (111) peaks, the Cu Kβ peak is present at approximately 36° in 2θ. The peaks of the as-sythesized particles are significantly broader than those extracted from tested cathodes. The coherence length of the respective particles was computed with the Scherrer equation applied to the fcc (111) peak. Post-FC samples displayed a domain size increase of 3 × to 4 × . The Scherrer width of the Pt control increased from 3.7 nm to 12.4 nm, while the Pd₈₄Pt₁₆ width increased from 4.3 nm to 17.2 nm. Complete results are collected in Table II.

Zoom In Zoom Out Reset image size

Lattice parameters were also calculated with Bragg's law using the fcc (111) peak. Palladium has a slightly lower lattice parameter than platinum (Pd: 0.3892 nm; Pt: 0.3923 nm), and Pt-Pd solid solutions exhibit reasonable agreement with Vegard's Law. However, as shown in Figure 8, significant departures from this relationship are observed in the as-synthesized catalysts. Rather than contracting with Pd addition, the lattices of the as-synthesized materials are dilated, and become increasingly so with increased Pd content. The lattice parameters of post-FC catalysts, on the other hand, display very good agreement with Vegard's law. However, in the Pt-Pd system, the Pd lattice expansion due to Pt addition is less than Vegard's law predicts,³¹ suggesting that a slight lattice dilatation in excess of that expected from alloying may still be present.

The lattice expansion deduced from the pre-FC patterns is a hallmark of interstitial impurity defects, and common to Pd^32–35 and Pd alloys^36,37 exposed to hydrogen or hydrocarbon environments. Both carbon and hydrogen are likely to occupy interstitial sites in Pd, and have been shown to competitively insert based on environmental conditions.^38–41 Carbon is a more likely impurity species in this case. Carbon impurities are common in thin films deposited from metallorganic precursors at low temperatures and in the absence of a strong reducing agent, and indeed, we have previously detected surface carbon species on Pt SAFC catalysts via XPS.²⁰ Carbon and hydrogen content of the samples were determined by CHN analysis (Galbraith Laboratories) with a detection limit of 0.5 wt%. No hydrogen was detected, but carbon was found in all as-synthesized samples, in amounts increasing with Pd content. Post-FC samples still contained carbon, but in reduced amounts. The Pd₈₄Pt₁₆ sample contained 2.85 wt% carbon pre-FC, and 0.92 wt% post-FC. In comparison, both pre-FC and post-FC platinum catalysts had nearly identical carbon content, 1.14 wt% and 1.16 wt%, respectively.

A linear relation empirically describes the lattice parameter expansion induced by interstitial carbon:³⁸

Here, a₀ is the carbon-free lattice parameter (in this case, calculated for each composition using Vegard's law) and x is the mole fraction of dissolved carbon. Returning to the case of Pd₈₄Pt₁₆, equation 3 yields an interstitial carbon fraction of 0.10 prior to fuel cell tests. Conversion of the CHN data from weight to a molar basis results in a mole fraction of 0.22, indicating that approximately half of the carbon present in the sample does not occupy lattice sites. Above 53at% Pd, interstitial carbon accounts for approximately 50% of the total carbon in each sample. Below this composition, dissolved carbon fractions are not constant, but rather track the Pd fraction. Table III summarizes carbon content derived by various methods.

Discovering significant carbon content in all vapor-grown catalysts and the presence of an interstitial carbide phase in Pd-rich samples motivated a deeper inquiry into the surface chemistry of selected samples using X-ray photoelectron spectroscopy. Due to the possibility of surface carbides that can shift C1s binding energy,^36,42 the Pt4f₇/2 peak was used as a reference for calibration of binding energy. Pre-FC and post-FC samples of Pt, Pd₈₄Pt₁₆, and Pd₅₃Pt₄₇ were selected for analysis.

XPS C1s peaks are shown in Figure 9. Pre-FC samples clearly show enhanced intensity near 288 eV characteristic of the C = O bond, a likely remnant of the diketonate precursor ligand and in agreement with spectra we have reported previously.²⁰ Pre-FC catalysts also show a systematic spectral shift toward lower binding energies, possible evidence of surface carbides. Nonlinear least-squares fitting was used determine the locations and intensities of the spectral components of the C1s peaks. Before testing, all samples show spectral shifts toward lower binding energies. The largest component of the Pd₈₄Pt₁₆ spectrum is found at 284.4 eV before fuel cell tests, and shifts to 284.7 eV after testing. Shifts to higher binding energies more often associated with graphitic carbon are also seen for Pt and Pd₅₃Pt₄₇ after testing.

Zoom In Zoom Out Reset image size

The possibility of surface carbides is supported by slight shifts in the Pd3d_5/2 peaks to a binding energy higher than is usually reported for Pd metal. In Figure 9(b), the measured Pd3d spectra exhibit shifts of +0.4 to +0.5 eV with respect to the accepted Pd⁰ binding energy at 335 eV. This is consistent with values reported for carbide formation during alkyne hydrogenation⁴¹ and CO exposure.³⁵ After fuel cell operation, we observed shifts to slightly higher binding energy (approximately +0.1 eV), possibly indicative of carbon migration from the lattice to surface sites.

The ratio of carbon to Pt and Pd was estimated from the XPS data using the equation

where x_C is the mole fraction of carbon, I_j is the integrated intensity of the strong photoelectron peak for element j of the set C, Pt, and Pd (for the 1s, 4f, and 3d peaks, respectively) and s_j are the associated atomic sensitivity factors. XPS offers a third perspective on carbon in this system, complementing the total carbon content obtained from CHN analyis and the lattice carbon from XRD with surface-sensitive compositional data.

All samples were significantly enriched in surface carbon in comparison with the values obtained via other methods. After fuel cell operation, the Pt-Pd alloys showed decreases in surface carbon of nearly 30%. Pt, on the other hand, showed no decrease in carbon content post-FC, in agreement with CHN analysis. This is good evidence that carbon is not electrochemically active in the Pt catalysts and is not oxidized in excursions to higher potentials. By the same token, the decrease in carbon content in Pd-containing catalysts, from both surface and lattice sites is likely to be a result of an anodic process.

The decarburization of Pd-rich samples during fuel cell operation, and the concomitant decrease in activity, suggests that carbon may play a role in oxygen reduction. Subsurface palladium carbides have been shown to play an essential role in hydrogenation processses,^40–43 and have a strong influence on the activity and selectivity of Pd catalysts for alkene formation. It is possible that carbon may have a similar promotional effect on the ORR in SAFCs.

Equation 4 was also applied to the Pt 4f and Pd 3d intensities and scattering factors to estimate the state of the Pd₈₄Pt₁₆, and Pd₅₃Pt₄₇ surfaces before and after fuel cell tests. The surface composition of the Pd₈₄Pt₁₆ sample from this method was Pd₇₂Pt₂₈ prior to testing, indicative of surface enrichment in Pt relative to the bulk material. Virtually the same value is found post-FC (Pd₇₃Pt₂₇). For Pd₅₃Pt₄₇ an XPS surface composition of Pd₄₁Pt₅₉ is found, again indicating relative surface enrichment of Pt. However, after testing, a surface composition of Pd₅₆Pt₄₄ results, very close to the bulk value from EDS.