Mitigation of PEM Fuel Cell Catalyst Degradation with Porous Carbon Supports

Maintaining high performance after extensive use remains a key challenge for low-Pt proton exchange membrane fuel cells for transportation applications. Strategically improving catalyst durability requires better understanding of the relationship between degradation mechanisms and catalyst structure. To investigate the effects of the carbon support morphology, we compare the electrochemical performance and durability of membrane electrode assemblies (MEAs) using Pt and PtCox catalysts with a range of porous, solid, and intermediate carbon supports (HSC, Vulcan, and acetylene black). We find that electrochemical surface area (ECSA) retention after a catalyst-targeted durability test tends to improve with increasing support porosity. Using electron microscopy, we investigate microstructural changes in the catalysts and reveal the underlying degradation mechanisms in MEA specimens. Pt migration to the membrane and catalyst coarsening, measured microscopically, together were quantitatively consistent with the ECSA loss, indicating that these were the only two significant degradation pathways. Changes in catalyst particle size, morphology, and PtCo core-shell structure indicate that Ostwald ripening is a significant coarsening mechanism for catalysts on all carbons, while particle coalescence is only significant on the more solid carbon supports. Porous carbon supports thus appear to protect against particle coalescence, providing an effective strategy for mitigating catalyst coarsening. © The Author(s) 2019. Published by ECS. This is an open access article distributed under the terms of the Creative Commons Attribution 4.0 License (CC BY, http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse of the work in any medium, provided the original work is properly cited. [DOI: 10.1149/2.0371904jes]

Hydrogen fuel cells are emerging as a useful technology for powering electric vehicles, although cost and durability remain key limitations to the widespread adoption of automotive fuel cells. While carbon-supported, platinum-based nanoparticles are the most promising demonstrated catalysts for automotive applications, which demand high power density, the high cost and limited availability of Pt necessitate development of low-Pt fuel cells (<10 g Pt per vehicle). 1,2 Such low platinum loadings, which leave little margin for failure, present new challenges for meeting the strict performance demands of automotive applications. 2,3 In addition to a high mass activity, catalysts must minimize reactant transport resistances, which limit high power performance and ultimately determine the overall system size and cost. The performance criteria for automotive fuel cells must be met throughout the intended life of the cell, including after cell materials experience extensive periods of corrosive conditions. Detailed understanding of degradation pathways in membrane electrode assemblies (MEAs) and careful design of catalyst materials is needed to maximize durability and minimize loss of performance.
Because oxygen reduction reaction (ORR) kinetics are the greatest source of voltage loss in well-optimized MEAs, improving catalyst ORR mass activities (MA) has been the primary focus of catalyst research as an effective strategy for decreasing platinum use. 1 The mass activity is the product of two more fundamental parameters: the electrochemically active surface area (ECSA) per platinum mass and the surface-area-normalized specific activity (SA). In addition to impacting MA, catalyst ECSA is an important parameter in its own right. In recent years, there has been growing awareness of the important role played by catalyst accessibility and reactant transport resistance in low-loaded fuel cells. In particular, researchers have identified oxygen transport resistance localized near the Pt surface to be a major limiting factor at high currents. The local oxygen transport resistance generally scales inversely with the Pt surface area and may be caused by poor dispersion of Pt particles, thin ionomer layers, ionomer interfaces, or constrictive pore structures. 2,4 Oxygen transport losses can be mitigated by designing accessible carbon support structures 5 and by ensuring that catalysts have high ECSA. 2 When considering catalyst durability, the oxygen transport problem highlights the importance of retaining high ECSA as well as mass activity to maintain adequate power performance over time. 3,6,7 The mechanisms of SA and ECSA degradation have been the subject of investigation and debate in the literature for many years. Many processes take place simultaneously, which we will summarize separately for SA and ECSA in the following paragraphs.
The specific activity is determined by the chemical properties of the catalytic surface, including the type and density of active sites, the strength of chemical bonding to reaction intermediates, and adsorbed species such as ionomer. Pure Pt catalysts typically experience an increase in specific activity over time, 8 due at least in part to increasing particle size, which increases the relative number of the most active surface sites. 9 This trend partly mitigates the loss in mass activity from ECSA losses, discussed below. Platinum alloy catalysts show more variable trends, either losing or gaining specific activity with electrochemical aging. 8,10,11 Loss of specific activity would be expected due to leaching of the secondary metal or increasing Pt shell thicknesses, which would decrease catalytic enhancements from either strain or electronic "ligand" effects.
For fuel cell electrodes, the ECSA is calculated from the electrochemically measured surface area (most commonly using either hydrogen underpotential deposition or carbon-monoxide electro-oxidation) divided by the initial platinum mass. Thus the measured ECSA can decrease from two broad categories of mechanisms: (1) Pt mass loss, where Pt becomes electrically disconnected and thereby electrochemically inactive, and (2) catalyst coarsening, where Pt nanoparticles grow in size, decreasing their specific surface area. For each of these there are multiple distinct mechanistic pathways.
Platinum mass loss primarily occurs as a consequence of Pt dissolution. Dissolved platinum may be chemically reduced by crossover hydrogen gas, depositing in either the membrane or the electrode ionomer. Many researchers have reported Pt reduction in the membrane forming a "Pt band". 8,[11][12][13][14][15][16][17] Significant Pt reduction in the electrode ionomer has been reported in fewer cases. 12 This may occur as the result of relatively severe hydrogen crossover and decreased hydrogen oxidation in the cathode due to passivation of the cathode Pt surface as Pt-O at some operating potentials. 8 Pt particles may also F199 lose electrical connectivity by detachment from the carbon support or by disruption of the percolating conductive carbon network in the electrode, especially in cases of severe carbon corrosion. 18,19 Two coarsening mechanisms have been described for fuel cell catalysts: Ostwald ripening and particle coalescence. Ostwald ripening is driven by the electrochemically-induced dissolution and redeposition of dissolved Pt on the surface of Pt particles. This process tends to increase particle sizes because smaller particles have lower stability and experience more rapid dissolution than larger, more stable particles. 20 Coalescence is a result of particle migration on the carbon support, leading to the collision and subsequent reshaping of particles.
Studies have disagreed on the attribution of observed coarsening between these two mechanisms. Many papers have provided evidence for Ostwald ripening as a coarsening mechanism due to the presence of significant amounts of dissolved Pt 12,20 and observations of the growth Pt shells on PtCo catalysts caused by Pt redeposition. 21,22 While some papers argue for Ostwald ripening as the only mechanism, others provide evidence that coalescence is also a significant mechanism, based on lognormal-like particle size distribution statistics with a long tail of larger particles that is not theoretically expected from Ostwald ripening, 19,23,24 or based on the observation of particles in aged PtCo catalysts with multiple PtCo cores contained in the same Pt shell, formed as a result of particle collisions. 22 There is now significant evidence that both mechanisms can occur, but less comprehensive understanding of the factors determining their relative significance.
Because coalescence involves the migration of particles over the carbon support surface, it is reasonable to expect that the morphology and surface properties of the carbon should impact the frequency of coalescence. An examination of the literature reveals that most studies reporting Ostwald ripening as the dominant or only mechanism used a porous, high surface area carbon (HSC) such as Ketjen black, 12,13,20,21,25,26 while most reporting the occurrence of coalescence as well used a solid carbon, such as Vulcan. 19,22,24,27,28 It is perhaps surprising that only recently has research been reported that attempts to systematically determine the impact of the carbon support on catalyst coarsening mechanisms. One recent study reported suppression of coalescence on porous carbon supports in a comparison of carbons in an aqueous environment using in situ small angle X-ray scattering (SAXS) measurements of particle size distributions. 28 Another examined catalysts extracted from MEA cathodes after stability testing using electron tomography and lattice imaging reported higher rates of coalescence on Vulcan and graphitized carbon based on observations of the particle morphology and comparisons of particle/crystallite size. 27 Given these indications that the carbon support morphology does impact the extent and mechanisms of catalyst degradation, selection and engineering of the carbon support is a promising pathway for improving catalyst durability. It is known that carbon support morphology also impacts the catalyst activity, with porous carbons providing higher activities by limiting detrimental ionomer adsorption on the Pt surface, [29][30][31][32] and reactant accessibility, by determining the pathways available for oxygen diffusion and proton conduction. 5,29,33 It is therefore important to understand the role of carbon support morphology on catalyst durability in realistic MEA environments under extended stability tests. Structurally sensitive but statistically robust methods are needed to investigate the changes taking place and the underlying degradation mechanisms in the catalyst and electrode.
In this paper we investigate the role of carbon support morphology in the durability of Pt and PtCo catalysts in fuel cell MEAs using electrochemical measurements and scanning transmission electron microscopy (STEM) analysis of post-mortem MEA specimens. Vulcan (Vu) and Ketjen Black (HSC) carbons were chosen for this study as prototypical solid and porous carbon supports, respectively. Vulcan carbon has a moderate surface area among carbon blacks and has been shown to support Pt almost exclusively on its exterior, while HSC has a higher surface area and a majority of supported Pt is embedded within its internal pores. An intermediate porosity Acetylene Black carbon (AB) was also included for the electrochemical experiments. Electrochemical measurements of the catalyst activity and ECSA were made before and after catalyst-targeting accelerated stability tests and showed a general trend of improved activity and durability with increasing support porosity.
Analysis of the MEAs with Vulcan and HSC catalysts by STEM imaging and elemental mapping was used to investigate structural changes and reveal the underlying degradation mechanisms. We used automated image processing to ensure that we obtained statistically robust and representative results from STEM imaging, including thousands of particles per sample for size distribution measurements. Our image analysis procedure was designed to more accurately reflect the surface area to volume ratios of irregular Pt particles in comparison to the common spherical particle approximation. This approach allows us to address traditional limitations to the statistical representativeness of (S)TEM microanalysis and make quantitative comparison to electrochemical measurements to infer and quantify degradation mechanisms. Together, these microscopic observations and electrochemical measurements provide a means to quantitatively explain the degradation pathways for different catalysts.

Methods
Materials and MEA assembly.-Fuel cell and electrochemical evaluations of each catalyst were performed in a membrane electrode assembly with the electrocatalysts of interest on the cathode. The active area of the fuel cell single cell was 50 cm 2 . The metal weight percent on carbon black supports of Pt in Pt and PtCo electrocatalysts was 20 and 30 respectively. The electrode Pt loadings were 0.1 and 0.025 mg Pt /cm 2 for the cathode and anode, respectively. Perfluorosulfonic acid (PFSA) Nafion D2020 was used in the electrode at an ionomer to carbon weight ratio of 0.8. 18 μm thick PFSA-based membranes were used to fabricate the MEAs. A 240 μm thick carbon paper with a 30 μm thick microporous layer (MPL) coated on top was used as the gas diffusion layer (GDL). The MEAs were fabricated using a catalyst-coated-membrane approach. The lamination procedure was discussed in detail elsewhere. 1 The same ionomer to carbon weight ratio was used for MEA assembly for all catalysts. While the BET surface area varies between the different carbon supports, the higher surface area of the porous carbon supports is largely due to their internal pores, which are not infiltrated by ionomer. 33 Consequently, the distribution of ionomer on the carbon support exterior is not expected to vary dramatically.
Electrochemical measurements.-Two accelerated stability tests (ASTs) recommended by the U.S. Department of Energy (DOE) 34 were used in this study: one accelerates the metal particle decay and one accelerates the catalyst support decay. The catalyst AST consists of 30,000 trapezoidal voltage cycles between 0.6 and 0.95 V. The dwell time at each voltage was 2.5 s and the ramp time was 0.5 s. Each cycle takes 6 s. The catalyst support AST consists of 5000 voltage cycles between 1 and 1.5V at a scan rate of 0.5 V/s. Both ASTs were performed at 80°C, 100% relative humidity, and ambient pressure. The fuel cell performance, electrochemically active surface area, and oxygen reduction reaction mass activity were measured at 0, 10,000 and 30,000 cycles during catalyst AST and at 0, 100, 500, and 5000 cycles during catalyst support AST. ORR activity is reported at 0.9 V RHE at 80°C, 100% relative humidity, and 1 bar of O 2 . 1 Pt ECSA was measured by CO stripping in an MEA. 35 X-ray scattering measurements.-The X-ray scattering data were collected on a combined Bonse-Hart (USAXS) and pinhole (SAXS/WAXS) instrument at beamline 9-ID-C at the Advanced Photon Source located at Argonne National Laboratory. Details regarding the optics and instrumentation have been previously reported. 36 The X-ray beam was monochromated via a pair of Si(220) crystals to an energy of 24 keV. The beam spot size for USAXS was 0.8 × 0.6 mm (horizontal x vertical) and 0.8 × 0.2 mm for SAXS/WAXS. The X-ray beam exposure times for each sample were 120 seconds for USAXS, 30 seconds for SAXS, and 30 seconds for WAXS. The samples were prepared by lifting off a section of the applied cathode catalyst layer from the membrane side with single-sided, transparent tape. Spectra collected on a blank piece of tape were utilized as a blank for data reduction. The data were corrected and reduced with the NIKA software package, 37 and analysis was conducted with the IRENA package. 38 Both packages were run on IGOR Pro 7.0 (Wavemetrics). Because only the USAXS/SAXS regions are appropriate for determining the particle size distributions of these types of nanoparticles at an X-ray energy of 24 keV, the WAXS data are not presented.
(S)TEM sample preparation.-MEA samples for (scanning) transmission electron microscopy ((S)TEM) were prepared by crosssectioning with an ultramicrotome. Strips cut from MEAs were embedded in EMbed 812 Resin (Electron Microscopy Sciences) and cured at 60°C overnight. Sections were cut using a Leica Ultracut UCT Ultramicrotome at thicknesses from 40 nm to 150 nm. Sections were collected on slotted Cu TEM grids and square-mesh Cu TEM grids with lacy carbon, which provide greater stability for high resolution imaging and EELS measurements. Samples were cleaned with oxygen-argon plasma prior to TEM measurements.

STEM EDXS measurements.-Energy-dispersive
X-ray spectroscopy (EDXS) quantification of Pt band losses was performed on a FEI Tecnai T12 Spirit S/TEM with a LaB 6 filament thermionic electron source operated at 120 kV in STEM mode. Spectroscopic maps were collected using an EDAX Genesis X-ray detector along with simultaneous bright-field and dark-field STEM images in a single scan per map covering the entire cathode and the region of the membrane including the Pt band. Measurements were made on 70 nm thick sections of EOL MEAs, and maps were collected from 3-5 different regions in each sample. Masks selecting the Pt band and cathode regions were drawn manually in ImageJ2 39 by inspection of the simultaneous STEM images along with higher resolution STEM images acquired prior to EDXS map acquisition. Pt that was included in the membrane by design was excluded from the Pt band masks. The relative Pt content of each region was determined from the summed spectra in the region by integrating the Pt L α1,2 peak after subtracting a constant fitted background value. Calculations were performed in Matlab.
Pt-Co composition, maps, and profiles were measured on a FEI Tecnai F20 S/TEM with a Schottky field emission gun operated at 200 kV in STEM mode using an Oxford X-Max 80 mm 2 X-ray detector and the INCA software. Measurements were made on ∼150 nm thick MEA cross-sections and catalyst powders supported on lacy carbon, which provides better signal averaging and sample stability. The average composition in MEA cathodes was measured by scanning over the entire cathode in at least 5 distinct regions per sample. Further details are provided in the supplemental information.
STEM imaging.-STEM images of catalyst morphology were acquired using a FEI Tecnai F20 S/TEM with a Schottky field emission gun operated at 200 kV. Images were acquired with a convergence semi-angle of ∼9.6 mrad and the image signal was collected on an annular detector at a 100 mm camera length to provide high angle annular dark field (HAADF) conditions. HAADF conditions provide approximately mass-thickness contrast which is sensitive to the atomic number Z with a scaling of approximately Z 1.7 , making Pt and PtCo particles easily distinguishable from the background of lower-contrast carbon, ionomer, and embedding medium.
Quantitative analysis of STEM images.-STEM images for quantitative analysis of Pt and PtCo particle sizes were acquired with fields of view ranging from 360 to 730 nm with a pixel size of approximately 3.5 Å. The images were prepared for segmentation of metal particles with a light Gaussian smoothing to reduce noise. The slowly-varying background from the carbon support and embedding medium was fit using a greyscale morphological open operation with a 30 nm radius and removed. Particles were identified with a threshold selected to be low enough so that the smallest particles are included. The particle segmentation is then refined using an active contour with the Chan- Vese method 40 for 300 iterations. Particles touching the edge of the image are not included for analysis. The segmented set of particles included some particles that are overlapping in projection. It is difficult for an automated routine to reliably distinguish single irregular particles from particles that are overlapping in projection. However, a reasonable approximation for the size of both, as well as single spherical particles, can be calculated using a local radius approximation rather than the more common spherical particle approximation. We implement this by using a watershed transform to determine local regions on each particle (or set of overlapping particles) and then assign each region a local radius equal to the maximum distance from the particle edge in that region. Each region is assigned a statistical weighting factor equal to the area of the region divided by the area of a circle of the same local radius. Size distributions are calculated using the local diameters and weighting factors for all particles. This procedure is described in further detail in the supplemental information and illustrated in Figure S3. Calculations were performed in Matlab using functions in the image processing toolbox.
STEM EELS composition maps.-STEM electron energy loss spectroscopy (EELS) measurements were performed on an FEI Titan Themis S/TEM operated at 300kV, equipped with a Gatan GIF Quantum spectrometer in single-range EELS mode. A convergence semi-angle of 21.4 mrad and a beam current of around 250 pA were used to acquire spectroscopic images with a pixel size of approximately 1 Å. Composition maps were extracted by integrating the signal from the Co L 2,3 and Pt M 4,5 edges after background subtraction using exponential and linear combination of power laws background fits, respectively. After integration, EELS maps were drift-corrected using an affine correction determined by comparing the simultaneous ADF signal to the fast-scanned overview image. The affine correction was constrained to only shear and stretch components consistent with uniform drift during STEM imaging. Calculations were performed in Matlab using functions in the image processing toolbox.

Results and Discussion
Catalyst and catalyst support properties.-Three carbon blacks with varying internal porosity were chosen as catalyst supports for this study: Vulcan (Vu), a typical solid carbon, Ketjen Black (HSC), a typical porous carbon, and an acetylene black (AB) carbon with intermediate porosity. The BET surface area and the Raman G/D band ratio, indicating the degree of graphitization, are summarized in Table I. HSC and AB carbons both have high BET surface areas and low graphitization. AB has a slightly lower surface area and is slightly more graphitic than HSC. Vulcan has a significantly lower surface area and is somewhat more graphitic than HSC and AB.
MEAs made with Pt catalysts on each of these carbon supports, along with a broader set of carbon black supports with higher graphitization, were subjected to a support-targeting accelerated stability test intended to simulate unusual or unintentional operating conditions. The test consists of 5000 cycles between 1.0 V and 1.5 V to induce carbon oxidation, which may lead to electrode failure due to collapse of the electrode macropore structure. The results, reported in detail in the supplemental information ( Figure S1), confirmed previous findings of higher corrosion resistance for more graphitized carbons. [41][42][43][44][45] The less graphitic carbons also tend to have higher surface areas, which accelerate carbon corrosion. Differences in the surface functionalization may also impact carbon corrosion or the anchoring of Pt particles on the support. Unfortunately, even the compact carbons with the highest graphitization did not satisfy the DOE target of < 30 mV loss at 1.5 A/cm 2 after 5000 cycles. The relatively low carbon surface area and small number of surface functional group of graphitized carbons make preparing a supported catalyst with good particle size distribution very difficult. Furthermore, recent engineering effort in fuel cell system control has substantially mitigated the exposure of the electrode to high potential, making the tolerance of the cathode support to carbon oxidation less important. Consequently, going forward in this study we will focus on the three types of carbons listed in Table I to investigate the effect of carbon morphology on catalyst stability.
Pt and PtCo nanoparticles were synthesized on each carbon support, at 20wt% for Pt catalysts and 30wt% for PtCo catalysts, using wet impregnation synthesis methods. Table II summarizes measurements of the size and some electrochemical properties of these catalysts. Mean particle diameters were measured by SAXS and STEM (detailed STEM results are discussed below). The SAXS and STEM measurements of the mean particle diameters are generally similar in magnitude, with discrepancies of less than 15% for each sample. The bulk SAXS measurements provide validation for the STEM measurement methods, which provide more detailed results that are discussed below. The Pt catalysts were selected to have a similar particle size to control for the impact of particle size effects in their performance and durability. All three have mean particle diameters in the 2-3 nm range, as indicated by both SAXS and STEM, providing ECSAs of around 85-100 m 2 /g Pt . The PtCo catalysts have larger particle sizes because of the higher temperatures required for their synthesis. PtCo/HSC and PtCo/AB have similar mean particle diameters in the 4-5 nm range and ECSAs around 65-75 m 2 /g Pt . Synthesis of a similar quality alloy on Vulcan carbon resulted in a larger 5-6 nm mean particle size and a somewhat lower ECSA of 50 m 2 /g Pt . The initial Pt-Co composition, measured by ICP-AES, was Pt 0.77 Co 0.23 for PtCo/HSC and Pt 0.70 Co 0.30 for PtCo/Vu.
Porous carbon supports are distinct from solid supports because they have a significant fraction of the metal nanoparticles embedded in pores inside the carbon primary particles. The fraction of particles embedded in the carbon can be determined by electron tomography, although this is an arduous process and provides limited statistical sampling. Previous electron tomography experiments have shown that HSC typically supports a majority of Pt particles inside interior pores, while Vulcan supports almost all Pt particles on the carbon exterior. 33,46,47 A useful proxy to infer the fraction of particles on the carbon exterior is the utilization of Pt measured in a MEA with COstripping under dry conditions (10−20% RH). 33,48 Measurements of the dry-cell utilization are reported in Table II, with representative utilization curves presented in Figure S2. Both Pt/Vu and PtCo/Vu have very high dry-cell utilization (>95%), indicating that nearly all the metal particles are on the carbon exterior, as expected. Pt/HSC has a low dry-cell utilization of 35%, suggesting a majority of interior particles, while PtCo/HSC has a somewhat higher dry cell utilization of 55%, suggesting that around half of PtCo particles are on the carbon interior. Both Pt/AB and PtCo/AB have intermediate dry cell utiliza-tions around 49%, suggesting that about half of the metal particles are on the carbon exterior. The measurement of the dry cell utilization can be imprecise because the large-scale proton conductivity of the electrode may become disrupted before the carbon-interior pores are fully dried. This may cause the utilization to change rapidly in the 10-20% RH range. This was the case for the AB-supported catalysts, so their measurements should be considered very approximate. However, they showed significantly higher utilization under moderately dry conditions (20-40% RH) compared to the HSC-supported catalysts, suggesting that they support more particles on the carbon exterior.
Electrochemical performance and durability.-The differences in MEA performance and durability between Pt and PtCo catalysts on solid Vulcan supports and porous HSC supports are clearly evident in fuel-cell polarization curves, shown in Figure 1. In the beginning of life (BOL) MEAs, the higher mass activity of PtCo catalysts provides a 10-15 mV improvement relative to Pt catalysts at low current densities. However, PtCo catalysts also suffer greater mass transport losses than Pt catalysts due to their lower ECSA, which approximately negates the benefit from mass activity at ∼2 A/cm 2 current densities. The higher mass activity of HSC-supported catalysts provides ∼30 mV better performance at low currents relative to Vulcan-supported catalysts. Vulcan can provide reduced mass transport losses (especially evident for Pt/V) due to the increased accessibility of particles on the carbon exterior. However, reduced mass transport losses only outweigh the activity gains provided by HSC at very high current densities around 2 A/cm 2 for Pt catalysts. Figure 1 also shows performance losses after 10K and 30K cycles of a catalyst-targeted accelerated stress test. After 30K cycles, MEAs are also denoted "end of life" (EOL). All catalysts show a decrease in performance at low current densities due to decreasing mass activity, but at high currents the additional mass transport losses are especially severe. Both Vulcan-supported catalysts suffer much more dramatic losses than the corresponding HSC-supported catalysts. Figure 2 shows a more detailed comparison of the electrochemical properties of the catalysts at beginning of life and after the catalyst AST. Figures 2a, 2b show the total ECSA of the cathode catalysts and the retained fraction of the BOL ECSA. As noted previously, all three Pt catalysts were selected to have similar ECSAs at BOL, and the PtCo alloy catalysts have somewhat lower ECSAs (especially PtCo/Vu) due to the higher temperatures required for alloy synthesis. The PtCo catalysts all retain a higher fraction of their initial ECSA after durability testing compared to the Pt catalysts, likely due to the stability provided by their larger particle sizes. It is notable that all the PtCo catalysts maintain total ECSAs equal to or higher than their Pt counterparts after either 10k or 30k cycles, so the higher ECSAs provided by Pt catalysts at BOL are not maintained throughout a practical fuel cell lifetime. For either Pt or PtCo catalysts, there is a general trend of higher ECSA retention for more porous supports. HSC-supported catalysts retain a significantly higher fraction of their ECSA than Vulcansupported catalysts (37 ± 2% for Pt/HSC vs. 24 ± 2% for Pt/Vu, 68 ± 3% for PtCo/HSC vs. 57 ± 2% for PtCo/Vu after 30K cycles). The intermediate AB support behaves more similar to HSC for the Pt catalysts, and more similar to Vulcan for the PtCo catalysts.  Figure 2c shows the trends in catalyst mass activity and the activity durability. The PtCo catalysts all display higher mass activities than Pt catalysts on the same supports both at BOL and after durability testing. There is also a strong trend of higher mass activity with increasing support porosity, before and after cycling, despite generally similar ECSA and particle sizes. This is consistent with prior results, and reflects the detrimental effect of adsorbed ionomer, which can be limited on porous supports with Pt surfaces embedded in pores. [29][30][31][32] The specific activity of the catalysts, shown in Figure 2d, reflects the same trends with support porosity and metal composition. The Pt catalysts all display a trend of increasing SA after durability testing, as previous researchers have observed, 8 which is at least partly the result of increasing particle size. 9 This counteracts some of the loss in mass activity caused by the decrease in ECSA. The PtCo catalysts either maintain a stable SA after stability testing, in the cases of PtCo/HSC and PtCo/AB, or lose SA, in the case of PtCo/Vu. The PtCo catalysts experience a smaller increase in particle sizes relative to the Pt catalysts, as will be shown below, which would provide a smaller increase to their SA that is likely counteracted by the effect of Co losses. After the full 30k cycle durability test, the Pt and PtCo catalysts tend to converge to similar SA values on the same support, likely because of the more rapidly increasing particle size for Pt catalysts and Co loss from PtCo catalysts.
Because ECSA losses after durability testing are responsible for most of the mass activity degradation, as well as loss in high power performance, it is important to understand the mechanisms and develop strategies for mitigating ECSA loss. The following sections will focus on understanding mechanisms through microscopic structural characterization and explaining the improved ECSA retention of the catalysts with more porous supports.

EDXS quantification of Pt mass in Pt band.
-Accounting for the observed ECSA loss in terms of the catalyst and electrode microstructure requires quantification of both Pt mass losses and catalyst coarsening, as discussed in the introduction. The fraction of Pt mass lost by migrating into the membrane and redepositing in the Pt band after 30K cycles was measured using STEM EDXS mapping in cross-sectional MEA samples, as shown in Figure 3.
Because the catalyst-targeted AST was conducted with hydrogen in the anode and nitrogen in the cathode, the Pt band forms in the membrane immediately beyond the cathode-membrane interface. 15,17 The cathode-membrane interface was observed to have micron-scale roughness, and so spectrum imaging with sufficient spatial resolution to identify the interface and separate Pt in the membrane from Pt remaining in the cathode was necessary. Fortunately, the Pt particles that make up the Pt band are very large (tens of nanometers) and often cubic, 12,49 and are thus easily distinguished from catalyst particles in the cathode based on their morphology.
To measure the relative amounts of Pt in the Pt band and remaining in the cathode, the Pt signal was integrated in each region of the EDXS maps, selected as shown in for a representative example in Figure 3a. Figure 3b shows the vertically integrated profile of Pt in the map (black) along with the profile of Pt in the Pt band (red) and in the cathode (blue). A gradient in the Pt concentration is visible in the cathode region, with less Pt remaining in the cathode in regions closer to the membrane. This general pattern in Pt distribution has been reported in previous studies 7,12,13 and was observed in all EOL MEA samples analyzed. Such a distribution is expected due to diffusion of dissolved Pt in the electrode where the membrane acts as a "sink" where dissolved Pt becomes permanently deposited after chemical reduction by crossed-over hydrogen. For EOL PtCo samples, the cobalt distribution within the cathode follows the same profile as the Pt distribution, resulting in a Pt-Co composition that is roughly constant across the electrode ( Figure S5).
To quantify the fraction of Pt in the Pt band, we integrate all of the Pt signal in the Pt band region and divide by the total Pt signal in both the cathode and the Pt band. A comparison of the fraction of Pt in the Pt band for different catalysts is shown in Figure 3c. Very little variation in mass loss was observed between the samples, with all showing approximately 15% of Pt in the Pt band. PtCo/HSC may show slightly less Pt mass loss at ∼12%, although this is a small difference and not significant at the 2σ level. Separate measurements by electron probe micro-analysis and ICP support this finding with ∼14% of Pt mass loss to the membrane regardless of catalyst and support types.
Considering that the catalysts in this study were selected to control for the known effects of particle size on the Pt dissolution rate, this finding suggests that the amount of Pt that migrates into the membrane is generally insensitive to the morphology of the carbon support. This may imply that the rate of Pt dissolution is very similar between the catalysts as well. Although it may be expected that the PtCo catalysts would experience less Pt dissolution due to their larger particle size, the presence of Co in the alloy may also facilitate the dissolution of Pt atoms. This finding suggests that other macroscopic factors such as electrode thickness, Pt 2+ diffusion, and redeposition rates dominate. Catalyst particle morphology and particle size distribution.-ADF STEM imaging of cathode catalyst particles in MEA crosssections was performed to investigate the extent and mechanisms of catalyst coarsening. Figure 4 shows representative STEM images comparing the catalyst morphology for the different catalysts before and after voltage cycling, all with the same field of view.
At beginning of life, the Pt catalysts ( Figure 4, left half) both generally show a small 2-3nm particle size and an overall good dispersion of particles on the carbon support, although Pt/Vu appears to have slightly more irregular particle shapes, while the particles in Pt/HSC are generally roughly spherical. After voltage cycling, both catalysts experience a clear growth in particle sizes, although the morphologies of the Pt particles become quite distinct. In Pt/Vu, some particles retain a relatively small size around 5 nm and a spherical morphology, but a minority of very large, irregular particles also forms. By contrast, in Pt/HSC essentially all particles retain a generally spherical morphology, good dispersion, and relatively small size around 5 nm.
A similar trend is evident for the PtCo catalysts ( Figure 4, right half). Both PtCo/Vu and PtCo/HSC start with roughly spherical particles that are somewhat larger than those in their pure Pt counterparts, around 4-5 nm. Relative to their larger initial sizes, the coarsening of the PtCo catalysts appears somewhat subtler. However, PtCo/Vu shows the same formation of large, irregular particles and growth of the smaller, spherical particles observed in Pt/Vu. PtCo/HSC shows a similar uniform growth of spherical particles to Pt/HSC.
The difference in morphology between the large, irregular particles observed on Vulcan-supported catalysts and the smaller, spherical particles observed on all catalysts at end of life suggests a different primary coarsening mechanism for the two populations. The irregular shapes of the larger particles intuitively suggest formation by coalescence, while the uniform growth of particles maintaining a spherical geometry suggests coarsening by Ostwald ripening. A difference in the prevalence of these mechanisms on the different carbon supports may be expected based on the difference in primary porosity between the carbons.
Further insight into the possible coarsening mechanisms and impacts for the catalyst performance can be gained by examining and comparing statistical size distributions for the different catalysts before and after cycling. Figure 5 shows violin plots of the particle size distributions for each sample measured using automated image analysis of STEM images including thousands of nanoparticles per sample from a variety of electrode locations. (Tabulated values describing the size distributions are available in the supplemental information in Table S2 and Table S3.) A local diameter approximation was used instead of the more common spherical approximation to allow irregular shaped particles to be included in the analysis. While some other studies have deliberately excluded irregular particles formed on the carbon support, 12 this omission precludes study of the mechanisms forming these particles and their consequences. Figure 5a shows conventional size distributions for each sample, which are equivalent to histograms showing the number of particles of a given size. The Pt/Vu and Pt/HSC catalysts both start with a relatively narrow size distribution, with most particles in the size range of 2-3 nm. After cycling, both distributions see an increase in the mean size and a broadening of the distribution, but the distribution shapes also change dramatically. The size distribution for Pt/Vu develops a long tail of large particles above 10 nm, while the distribution for Pt/HSC remains roughly symmetric, with almost no particles above 10 nm. Similar trends hold for the PtCo catalysts, although the PtCo/Vu distribution begins with a significant tail of larger particles. The distribution tail of larger particles in PtCo/Vu grows after electrochemical cycling. The PtCo/HSC distribution has a more uniform upward shift after cycling, remaining symmetric, as the Pt/HSC distribution does, although with a smaller overall shift relative to its larger starting size.
Granqvist and Buhrman noted that coalescence and Ostwald ripening are theoretically expected to produce different shapes in the particle size distribution. 23 Coalescence is expected to produce a lognormallike size distribution with a heavy tail of larger particles. Ostwald ripening, by contrast, does not produce a tail of larger particles. Granqvist and Buhrman predicted that distributions formed by Ostwald ripening would have a heavy tail of smaller particles, although the relatively low stability of smaller Pt particles 20 under typical fuel cell conditions may remove this feature, which has not been observed for fuel cell catalysts, resulting in a roughly symmetric distribution.
Interpreting our observed size distributions with this framework suggests that the HSC-supported catalysts do not experience significant coalescence, but instead coarsen primarily by Ostwald ripening. The growth of heavy tails of large particles in the Vulcan-supported catalysts suggests that they do experience significant coarsening by coalescence. The shift in the distribution peaks to larger sizes and the loss of very small particles suggests that the Vulcan-supported catalysts coarsen by Ostwald ripening as well.
A plausible structural explanation of this outcome is that solid carbon supports like Vulcan permit the particles on their exterior surfaces to migrate and collide, while porous supports like HSC constrain the particles embedded in their interior pores, limiting their migration and coalescence. The pores in HSC have previously been shown to have a constrictive, bottleneck geometry. 33 Furthermore, the porous structure of HSC supports a better dispersion of particles both inside and outside the carbon support, resulting in larger interparticle distances compared to Vulcan. 27 This is expected to lower the probability that particle migration leads to particle collision and coalescence. Ostwald F205 ripening, however, requires only the presence of dissolved Pt ions to be electrochemically redeposited. The consistency of the Pt band formation across catalysts shown in the previous section suggests that the different carbon supports result in similar rates of Pt dissolution, and likely similar degrees of Ostwald ripening. The greater interparticle spacing on HSC may also slow Ostwald ripening by lowering the rate of Pt redeposition on neighboring particles.
The heavy tail in the BOL PtCo/Vu size distribution is likely formed by particle coalescence at the higher temperatures required for alloy synthesis. It is intuitively reasonable to expect that the carbon morphology should impact thermal coalescence and electrochemical coalescence in a similar manner, with solid Vulcan permitting particle migration and coalescence and porous HSC constraining it. It is notable that the larger particles in PtCo/Vu at beginning of life do not have irregular shapes like those formed during electrochemical cycling, possibly because the high temperature in synthesis provides a stronger driving force for particle reshaping.
It is notable also that there is significant convergence after cycling in the particle sizes between the Pt catalysts and the PtCo catalysts. The Pt catalysts start with much smaller typical particle sizes than the PtCo catalysts but end with only slightly smaller particle sizes. This is attributable in part to the much higher stability of particles around 5 nm relative to particles around 2 nm, which would result in much faster coarsening by Ostwald ripening for the smaller Pt particles. 20,26 In addition to the conventional size distributions shown in Figure 5a, it can be useful to examine volume-weighted size distributions, which show the distribution of Pt mass between particles of different sizes, shown in Figure 5b. This clarifies the consequences of the formation of large particle tails in the distributions for Vulcansupported size distributions. Although the larger particles that form on these catalysts are relatively few in number, the cubic scaling of volume with particle diameter means that these particles can account for a large fraction of the Pt mass in the catalyst. For instance, for EOL Pt/Vu and EOL PtCo/Vu, particles larger than 15 nm in diameter are so few in number that they are difficult to present in conventional number-weighted size distributions such as in Figure 5a. However, these large particles account for roughly half of the Pt mass, as visible in the volume-weighted size distributions in Figure 5b, and are therefore far from negligible.
Identifying particle coalescence with EELS composition mapping.-Direct evidence of the coarsening mechanism for PtCo catalysts can be obtained by using composition mapping by electron energy loss spectroscopy to reveal the core-shell structure of particles formed after voltage cycling. 22 After dealloying or exposure to an acidic environment, PtCo particles form a shell of Pt surrounding a PtCo core. Spongey or hollow structures may also form in large or Co-rich particles, especially if the particle diameter is around 10 nm or greater. 50 Such structures were observed only rarely for the catalysts in this study, possibly because of the relatively narrow BOS size distributions, and were not investigated in depth. Particle size growth by Ostwald ripening (or additional Co leaching) causes the thickness of the Pt shell to grow. PtCo particles that collide and coalesce retain their PtCo cores, separated by a Pt layer. Particles with multiple PtCo cores contained within the same Pt envelope can thus be identified as having undergone coalescence. 22 Figure 6 shows EELS composition mapping of PtCo/Vu and PtCo/HSC catalysts before and after voltage cycling. At beginning of life, both catalysts show very similar structure, with uniform, thin (∼0.5 nm) Pt shells -formed during dealloying -surrounding generally uniform PtCo cores. After 30K voltage cycles, thicker Pt shells, due to either Ostwald ripening or Co leaching, have formed on particles in both catalysts, but distinct differences are apparent in their core shell structure. In EOL PtCo/Vu many particles contain multiple PtCo cores, again indicating that coalescence is common on the Vulcan carbon support. The large, irregular particles in EOL PtCo/Vu are especially likely to contain multiple cores providing evidence that the large irregular particles described in previous sections are formed by particle coalescence. Among the particles mapped in EOL PtCo/Vu, ∼35% of the particles larger than the median diameter of ∼7 nm had more than one core, while none of the particles smaller than 7 nm did. Many of the larger particles in EOL PtCo/Vu that contain only one core are likely formed by thermal coalescence during the catalyst synthesis and make up the heavy tail of larger particles already present at BOL, as discussed in the previous section.
By contrast, multi-core particles are much less common in PtCo/HSC, indicating that coalescence is rare on the porous HSC support. Only two of the 30 EOL PtCo/HSC particles shown in Figure 6 have more than one core, and these particles are also not unusually large. With so few observed occurrences it is not possible to precisely quantify the frequency of coalescence, but it appears that coalescence is not a significant coarsening mechanism on the HSC support.
One relevant question is whether dissolution and redeposition occurs differently for PtCo particles within the carbon pores from the ones on the carbon exterior in PtCo/HSC. Without ionomer contacting particles on the carbon interior, factors impacting degradation such as proton activity and chemical/electrochemical potentials are not well understood. To investigate this, we combined EELS mapping with 3D tomographic imaging for EOL PtCo/HSC ( Figure S6). This experiment revealed no significant difference between the core-shell structure of PtCo particles on the interior and exterior of the HSC support. Thus any differences that there may be in their degradation were not significant enough to be observed in this study.
The increased thickness of Pt shells implies some degree of Co loss. EDXS measurements made over the entire cathode (see the supplemental information for additional details) indicated that the relative Co content of the PtCo/HSC catalyst decreased by about 46%, while the relative Co content of the PtCo/Vu catalyst decreased by about 31%. The smaller Co loss for PtCo/Vu can be attributed to its larger average particle size, which results in more stable surfaces and a lower fraction of the volume near the particle surface where Co may dissolve more easily. It is notable that PtCo/HSC shows better specific activity retention than PtCo/V (Figure 2d) given its more severe cobalt loss. Earlier studies have shown that enhanced specific activity can be maintained if Pt shell thickness growth is very small. 10,51 The thicker Pt layers formed in PtCo/V at the concave junctions between coalesced particles, as reported by Xin, Mundy, et al., 22 may result in more severe loss of surface strain. The relationship between core-shell structure and surface strain is the subject of ongoing investigation.
Quantitative account of ECSA degradation.-The quantitative STEM measurements of Pt mass loss to the Pt band and coarsening of the particle size distributions, presented in previous sections, may be considered together to attribute the observed ECSA loss between the different degradation mechanisms. Figure 7 shows the retained fraction of Pt mass in the electrode (blue triangles) and the retained specific surface area (SSA) of the catalyst in the electrode (green squares) measured with STEM techniques. Because the active catalyst surface area is the product of the SSA and the electrically connected Pt mass, the expected surface area retention after Pt mass loss and SSA loss is the product of the Pt mass retention and the catalyst SSA retention. The retained surface area calculated from STEM measurements (red circles), is quantitatively consistent with the ECSA retention observed electrochemically (black circles). This implies that Pt migration to the membrane and catalyst coarsening together account for all of the ECSA loss within the margin of error, and other mechanisms, such as mass loss in the cathode by particle detachment or chemical redeposition, are negligible.
The more severe coarsening and SSA loss of Vulcan-supported catalyst can be explained mechanistically by the results of the previous sections. Vulcan-supported catalysts suffer significant coarsening through both coalescence and Ostwald ripening mechanisms, while the porous morphology of HSC protects against coalescence, leaving only Ostwald ripening as a significant coarsening mechanism for HSCsupported catalysts. This results in a significantly improved ECSA retention for HSC-supported catalysts, leading to superior retention of mass activity and high power performance (Figure 1, Figure 2). Figure 6. EELS composition maps of PtCo catalysts before (BOL, top) and after (EOL, bottom) 30K cycle catalyst stability test. The Pt M 4,5 signal is shown in red and the Co L 2,3 signal is shown in green, combining to yellow. In PtCo/Vu (left) multi-core, coalesced particles are common, especially forming the largest particles, while in PtCo/HSC multi-core particles are rare. Figure 7. Comparison of observed ECSA retention (black circles) with quantitative STEM measurements of cathode Pt mass retention (blue triangles) and specific surface area retention (green squares) after 30k voltage cycles. The losses from Pt mass loss to the Pt band and SSA loss due to coarsening may be combined by taking the product of the mass retention and SSA retention, which is quantitatively consistent with the ECSA retention. Error bars represent one standard error of the mean.
The porous morphology of HSC that protects against particle coalescence also appears to result in superior activity by limiting ionomer adsorption, suggesting that improving HSC supports to limit their deficiencies may be a promising direction for future research. Success has already been reported in designing more accessible HSC supports that reduce the transport resistance associated with transport inside support micropores. 5 Efforts directed toward developing more graphitic, carbon-corrosion-resistant supports with a similar porous morphology may also be valuable. Mitigation of coarsening due to Ostwald ripening will likely require different approaches. The similarity of Pt band formation across the catalysts studied here suggests that the rate of Pt dissolution, and likely also Ostwald ripening, is similar regardless of support morphology. Efforts to improve the chemical stability of alloy catalysts, such as chemical ordering, 52 may prove effective at mitigating Ostwald ripening as well.

Conclusions
This investigation explored the role of the carbon support morphology in determining the degradation mechanisms occurring for Pt and PtCo fuel cell cathode catalysts. Electrochemical comparison of MEAs using a porous carbon support (HSC), a solid carbon support (Vulcan),

F207
and an intermediate carbon support (acetylene black) showed that the ECSA retention after a catalyst-targeting voltage cycling test tends to improve with increasing support porosity.
The mechanisms of ECSA loss underlying this trend were investigated using STEM, EDXS, and EELS measurements in post-mortem cross-section samples of the MEAs with HSC and Vulcan-supported Pt and PtCo catalysts. For all of these catalysts, EDXS quantification found similar Pt mass migrated from the electrode to the Pt band in the membrane: around 15%. The degree of catalyst coarsening, however, varied significantly between the different catalysts. Vulcansupported catalysts showed more severe SSA loss through coarsening, especially through the formation of large, irregularly shaped particles. Three lines of evidence indicate that these larger particles are formed primarily by coalescence: their irregular morphology, the lognormallike size distributions, and the frequent presence of multiple PtCo cores in these particles. By contrast, the HSC-supported catalysts contained few large, irregular particles, maintained more symmetric size distributions, and rarely had multi-core PtCo particles. Catalysts on both types of carbon supports also showed growth in the size of smaller, more spherical particles. These smaller particles showed an increase in Pt shell thickness in EELS measurements of PtCo samples, indicating that they grew through Ostwald ripening. The electrochemically observed ECSA loss was quantitatively consistent with the loss that would be expected based on STEM measurements of Pt mass in the Pt band and catalyst coarsening, indicating that these are the only significant degradation mechanisms under these test conditions.
Together, these lines of evidence demonstrate that Ostwald ripening takes place as a significant coarsening mechanism for all catalysts, regardless of the support morphology, while particle coalescence only contributes significantly for catalysts with solid carbon supports. Porous carbon supports thus suppress coalescence, likely by restricting the migration of particles embedded in carbon pores and providing greater interparticle spacing. This is an encouraging result, as it demonstrates that design of the catalyst support morphology can provide an effective strategy for mitigating catalyst degradation through coalescence.