Oxygen Reduction Reaction Causes Iron Leaching from Fe-N-C Electrocatalysts

Yu-Ping Ku Forschungszentrum Jülich Konrad Ehelebe Forschungszentrum Jülich Markus Bierling Forschungszentrum Jülich Florian Speck Forschungszentrum Jülich Dominik Seeberger Forschungszentrum Jülich Karl Mayrhofer Forschungszentrum Jülich https://orcid.org/0000-0002-4248-0431 Simon Thiele Helmholtz-Institute Erlangen-Nürnberg for Renewable Energy https://orcid.org/0000-0002-4248-2752 Serhiy Cherevko (  s.cherevko@fz-juelich.de ) Forschungszentrum Jülich https://orcid.org/0000-0002-7188-4857


1
On the way to a climate neutral hydrogen economy, 1 energy efficient, durable and 2 affordable fuel cells (FCs) are needed. While proton exchange membrane fuel cells (PEMFCs)  3 have been commercialized, further development of this technology is hindered by its 4 dependency on Pt as the main electrocatalyst material for both hydrogen oxidation reaction 5 (HOR) and oxygen reduction reaction (ORR). 2 Therefore, research on finding highly active and 6 stable platinum group metal (PGM)-free catalysts is booming as summarized in recent 7 reviews. [3][4][5] Among all identified PGM-free candidates, iron-nitrogen-doped carbon (Fe-N-C) is 8 the most promising to catalyse the sluggish ORR. 6,7 Recent studies have theoretically and 9 experimentally demonstrated that Fe-N-C catalysts are more active and stable in alkaline than 10 in acidic media. [8][9][10] This indicates the unique possibility to use Fe-N-C as a relatively efficient, 11 earth abundant and cheap cathode catalyst material in alternative anion exchange membrane 12 fuel cell (AEMFCs) technology. Due to the recent ground-breaking developments in membrane 13 science, AEMFCs demonstrate great potential for successful commercialization. [11][12][13][14] With 14 regards to durability and longevity, however, many open questions remain. In particular, 15 understanding of Fe x+ leaching and consequent implementation of mitigation strategies to its 16 suppression are considered main challenges. 6,8,15,16 17 So far, four key Fe-N-C degradation mechanisms have been suggested to occur in parallel 18 or series in acidic media, 7 providing clues also to studies in alkaline electrolytes: 19 (i) Carbon corrosion. Corrosion of the carbon support leads to Fe demetallation and a drop 20 in electron conductivity. 7,15 It has been shown in various studies that this mechanism 21 depends on temperature, 15 electrochemical potential, 15,17 and, as was demonstrated 22 recently, on the presence or absence of O2 in aqueous electrolytes. 18 23 (ii) Reactive oxygen species (ROS). Formed as intermediates or by-products of ORR, ROS 24 have shown to cause Fe-N-C catalyst degradation in acidic conditions. 10,18-20 ROS can 25 oxidize carbon support, resulting in a decrease of the ORR turnover frequency (TOF) of 1 the existing Fe active sites. The presence, amount and effect of ROS have been reported 2 to be dependent on the electrochemical potential, temperature, and ORR. 10,20 In an 3 alkaline environment the effect of ROS attack on catalyst degradation was shown to be 4 minor. 10,20 5 (iii) Agglomeration. Agglomeration of active single atom FeN4C12 sites to inactive ferric 6 oxides Fe2O3 was shown to be affected by temperature 21 or exposure to air. 22 7 (iv) Fe dissolution. Direct Fe ions leaching from active sites is influenced by electrolyte pH, 8 8 potential, 15 temperature, 15 chemical environment, 8 and water-flux across the active sites 9 within micropores. 17 10 While it is very likely that several degradation mechanisms are operative and the extent 11 of each mechanism is different depending on the FC operational conditions, a comprehensive 12 understanding of the stability of Fe-N-C catalysts is still lacking. For instance, the origin of the 13 dissolved Fe species from Fe-N-C catalysts was found to be different in the various Fe-N-C 14 stability studies. 15-17, 22 As an example, in scanning flow cell (SFC) experiments coupled to 15 inductively coupled plasma mass spectroscopy (ICP-MS), Fe dissolution in an Ar-purged acidic 16 electrolyte was assigned to leaching of inactive Fe sites. 15,16 In contrast, for realistic cathodes 17 in PEMFC tests, the decrease of ORR activity was correlated to a loss of FeN4 active sites using 18 ex-situ Mössbauer spectroscopy. 17,22 This discrepancy illustrates that direct comparisons of the 19 results from thin catalyst layers in aqueous model systems (AMS), e.g. rotating disk electrode 20 (RDE) or SFC, and realistic cathodes in PEMFCs or AEMFCs are challenging, as highlighted 21 in our recent review. 23 Such differences may affect the intensity and the mechanisms of 22 degradation significantly. A major discrepancy between AMS and real devices is the direct 23 contact of the catalyst to gaseous O2, which cannot be easily realized in RDE or SFC. Moreover, 24 even in AMS, the influence of O2 on Fe-N-C stability is sporadically addressed as the stability 25 has traditionally been tested in de-aerated acidic environment. However, degradation of PGM-1 free materials can be significantly affected by the presence of O2. 18,24 Thus, using online ICP-2 MS, Speck et al. 24 identified different dissolution mechanisms for MnOx in Ar-and O2-purged 3 alkaline media. On the one hand, Mn transient-dissolution occurs in both Ar and O2 4 environment, which was attributed to the redox transitions of Mn. On the other hand, Mn total 5 dissolution is drastically increased in O2 environment. This was ascribed to the presence of 6 ROS, such as HO2generated during the ORR. A similar trend has been detected also for Fe-7 N-C catalysts degradation in acidic media. Combining RDE measurements with Raman 8 spectroscopy, Kumar et al. 18 showed that both deterioration of ORR activity and the extent of 9 carbon corrosion are significantly more severe in O2-compared to Ar-saturated electrolytes. 10 In order to bridge the gap between the fundamental and applied researches, gas diffusion 11 electrode (GDE) half-cell setups have emerged as a tool to combine the advantages of AMS 12 (three electrode setup, fast and comparable testing at standard operating conditions) and FCs 13 (realistic catalyst layers and current densities) systems. [25][26][27][28][29] In the current work, we use a GDE 14 half-cell setup coupled with online ICP-MS (GDE-ICP-MS) 30 to systematically investigate the 15 impact of O2 atmosphere and a solid electrolyte interface on Fe demetallation in realistic Fe-N-16 C alkaline catalyst layers (CLs) at current densities up to 125 mA·cm -2 . 17

18
In order to gain insights into the Fe-N-C demetallation in realistic alkaline CLs during 19 ORR, Fe dissolution in Ar-and O2-environments is compared during the application of similar 20 potential steps. Figure 1

10
In Figure 1A and 1B, the electrochemical protocol in Ar and O2 environment is 11 displayed. In the presence of O2, -125 mA·cm -2 can be reached due to ORR at around 0.6 VRHE. 12 In an Ar environment the current density is smaller than -1 mA·cm -2 at similar potentials. Under 13 these electrochemical conditions, the Fe dissolution profiles (see Figure 1C, left axis, solid 14 lines) show that significantly more Fe is dissolved in the presence of O2 than in the absence of 15 O2. Overall, the total amount of Fe dissolved (see Figure 1C, right axis, dotted lines) in O2 is 16 about one order of magnitude higher than that in Ar in the potential range between 0.8 -17 0.6 VRHE. Also, at 0.87 VRHE (the lowest applied ORR overpotential) the Fe dissolution is 18 already higher in O2 than in Ar. Hence, it is clear that the amount of Fe dissolution would be 1 hugely underestimated if the electrochemical testing is executed in an environment without O2 2 as often done in AMS. 8, 15 3 In addition to the impact of the presence of O2 or ORR, the following two characteristic 4 features of Fe-N-C demetallation can be observed for both Ar and O2 cases. Firstly, the Fe 5 dissolution starts right when the cathodic current or potential steps are applied. After moving 6 back to open circuit potential (OCP), also the dissolution drops immediately. In general, the 7 amount of dissolved Fe species rises with increasing the ORR overpotential or/and current 8 density. Additionally, the dissolution at the regularly repeated steps to similar potentials is 9 almost constant (see Figure S1C in the supporting information (SI)), indicating that the order 10 of the applied steps does not affect dissolution behaviour in the chosen protocol. Besides Fe 11 dissolution, also the activity measured at low overpotentials does not drop over the whole 12 protocol before the accelerated stress test (AST) (see Figure S1B). 13 To mimic fuel cell load cycles, we conducted AST between -0.05 and -125 mA·cm -2 in 14 O2 environment, which correspond to 0.88 ± 0.03 and 0.56 ± 0.04 VRHE, respectively. For 15 comparison, AST in Ar was conducted in a similar potential range between 1.0 and 0.573 VRHE. 16 Also using this protocol, the total amount of dissolved Fe species during O2-AST is almost one 17 order of magnitude higher than that during Ar-AST (see Figure 2A  Interestingly, despite the dramatically higher Fe dissolution during ORR compared to 5 the one in Ar environment, the charge-normalized dissolution rates during ASTs in O2 and Ar 6 are similar (see Figure 2B). This implies that Fe dissolution in Fe-N-C could be associated with 7 a charge-transfer-related process that commonly occurs in both Ar-AST and ORR. Detailed 8 evaluation of the degradation mechanism can be found in the discussion section. To further 9 validate this coherence, Fe dissolution in O2 for different holding periods at -125 mA cm -2 is 10 compared in Figure S2. The results confirm that Fe dissolution is fairly constant when 11 normalized to the applied charge. In contrast, when the same Fe dissolution data is normalized 12 to the number of applied cycles, no correlation can be seen. Therefore, it is unlikely that solely 13 the change of potential causes all the Fe dissolution. All these results imply that Fe dissolution 14 in alkaline media is directly proportional to the applied current in the applied potential range, 15 which will be discussed more in detail in the discussion section. 16 In oxygen evolution reaction (OER) research, the reciprocal of the charge-normalized 17 dissolution, so-called S-number, has been well established as a metric for catalyst stability. 33 18 Due to the correlation between applied charge and Fe dissolution, a similar concept can be 19 applied here. The S-number of the studied commercial Fe-N-C would be around 10 6 . This 20 means that in the studied potential region between 0.57 and 0.87 VRHE, a Fe site can averagely 1 undergo one million desired charge transfer reactions before it is dissolved. Accordingly, the 2 average probability of losing a Fe atom at each charge transfer event is approximately 10 -6 . 3 Further discussion on how applicable this value is for AEMFCs can be found in the section 2 4 in the SI. Although ORR is a complicated multi-step 4-electron-transfer reaction with several 5 intermediate states, 34,35 where the coordinated Fe cation could be variously unstable, the S-6 number could be a suitable quantification metric to study overall stability of Fe-N-C catalysts 7 in various conditions and to compare different materials. 8 To further approach more realistic AEMFC conditions, we additionally compare Fe 9 dissolution in GDEs without and with a thin AEM attached in an O2-saturated environment. For 10 Pt/C catalysts in acidic media, a significant stabilizing impact of Nafion membranes on the net 11 Pt dissolution was recently observed. 30 This effect was attributed to the impeded mass-transport 12 of dissolved catalyst species through the membrane. However, for our AEM system, we cannot 13 observe any significant impact of the AEM on the Fe dissolution (see Figure 3C and S3). The 14 diverging effects of the Nafion membranes and the AEM can be attributed to the following 15 differences between the two systems. Firstly, the thickness ratio of CL to the membrane in our 16 AEM system is two to three orders of magnitude higher than that in the work mentioned above 17 (see Table S1 and Figure S4). In our case, the doctor-blade coated AEM membrane is less than 18 1 μm thick, whereas in the literature membranes of 25 and 52 μm were used. 30   power density at 0.6 VRHE decreased by more than 10 %, although less than 1 % of Fe species 2 were dissolved. This indicates that during ORR, the detected Fe species were especially 3 dissolved from the Fe active sites, which is consistent with the works of Chenitz et al. 17 and Li 4 et al. 22 , using Mössbauer spectroscopy and PEMFCs running with H2/air or O2 at 80 ℃. This 5 comparison unveils the impact of O2 or ORR on the origins of dissolved Fe from Fe-N-C in 6 alkaline media, and helps explaining the existing discrepancy between observations from RDE 7 and FC setups. With the presented data, some derivations on the degradation mechanism of Fe-8 N-C catalysts in realistic alkaline catalyst layers are made in the further course. 9

10
The degradation of Fe-N-C proceeds through diverse mechanisms in different 11 electrochemical potential ranges or pH environments. 8,15,20 It has to be noted that the current 12 work only allows derivations for the potential range between 1.0 and approximately 0.6 VRHE 13 in alkaline media. This, however, corresponds well to a usual potential range of AEMFCs under 14 load. 36 Under the investigated conditions, Fe dissolution increases with decreasing potential. In 15 contrast, carbon corrosion is expected to increase with potential. 15,37 Therefore, carbon 16 corrosion is not considered as the dominating degradation mechanism under the studied 17 conditions. On the other hand, the main outcome of this work emphasizes that the Fe dissolution 18 is directly proportional to the applied charge. In other words, Fe dissolution can be largely 19 correlated to charge transfer events, which are part of the ORR catalytic cycle. This implies that 20 the observed Fe dissolution may be attributed to the destabilization of Fe active sites at one or 21 more step(s) during the ORR catalytic cycle, namely the adsorption of O2 on Fe, the reduction 22 of reaction intermediates and the Fe 3+ /Fe 2+ redox transition. 20,34,38-40 Those three steps and their 23 potential impacts on Fe dissolution will be discussed below. 24 Firstly, according to density functional theory (DFT) calculations conducted by Aoyama 1 et al. 38 , the adsorption of O2 molecules on Fe active sites can shift the Fe position from in-plane 2 to out-of-plane. This leads to destabilization of the coordinated Fe cation and therefore 3 pronounced dissolution. Although it is a study for an acidic condition, we could not rule out the 4 possibility that a similar destabilizing effect due to this step is also occurring in alkaline media, 5 as O2 adsorption is a part of the ORR catalytic cycle in both environments. performed H2O2-treatment on Fe-N-C in acidic or alkaline media, and found notable declines 10 of the ORR activity after H2O2-treatments in acidic yet a negligible decrease after a H2O2-11 treatment in alkaline. This suggests that compared to H2O2 in acidic, HO2is less harmful to Fe-12 N-C in alkaline environment, and thus its impact on the degradation process is potentially less 13 pronounced. Indeed, a great discrepancy between the potential dependencies of the charge-14 normalized Fe dissolution detected in this work and the HO2faraday efficiency reported for a 15 similar catalyst in a previous work, 32 can be observed (see Figure 4). Yet, it has to be considered 16 that the faradaic efficiency was determined in a rotating ring-disk electrode (RRDE) setup, 17 where ORR is severely limited by the restricted mass transport of dissolved O2 when the 18 potential is below 0.8 VRHE. This could lead to deviations when the RRDE result is compared 19 to the dissolution data from this work, gathered from GDE half-cell measurements, where those 20 mass transport limitations do not play a significant role. 25-28 Hence, the results point against a 21 dominant impact of HO2or ROS on the stability of Fe-N-C in alkaline media and O2 22 environment; however, with currently available data it cannot be conclusively excluded. From the Fe dissolution data during ORR, any destabilizing step during the ORR 17 catalytic cycle could be suspected to be responsible for the Fe demetallation in Fe-N-C. 18 However, more insights may be gained from the surprisingly similar values of the charge-19 normalized Fe dissolution during O2-AST and Ar-AST (see Figure 2B). This implies that the 20 key destabilizing factor to the Fe dissolution in Fe-N-C could be one that is related to charge 21 transfer events and also occurs in both O2-AST and Ar-AST. Among the three above mentioned 22 AST as can be seen from the cyclic voltammograms in Ar environment (see Figure S5). Indeed, 2 the correlation between Fe 3+ /Fe 2+ redox transition and the stability of Fe active sites in the 3 absence of O2 could be supported by the findings revealed by Li et al. 22 They reported that the 4 Fe active sites that undergo reversible Fe 3+ /Fe 2+ redox transitions due to potential switches 5 between 0.2 and 0.8 VRHE (FeN4C12) are less stable than those where the charge of Fe ions (2+) 6 is constant and independent of the potential switches (FeN4C10). This supports the idea that the 7 coordinated Fe could be less stable during its redox transitions even in the absence of O2. 8 Based on this correlation between Fe 3+ /Fe 2+ redox transition and Fe dissolution, we 9 propose that Fe demetallation of the investigated Fe-N-C catalyst in alkaline media can be 10 attributed to the instability of the coordinated Fe during the redox transition. This hypothesis 11 can also explain the different amounts and varied origins of the dissolved Fe species in the 12 absence and presence of ORR: On the one hand, the Fe redox transition in a de-aerated 13 electrolyte would only be triggered by the change of electrochemical potential, occurring only 14 twice in each AST cycle. Also, the transitioned Fe species do not necessarily have high activity. 15 On the other hand, in the presence of O2, the active Fe sites continuously undergo redox 16 transitions when catalysing ORR in the first half of each AST cycle. Moreover, the more active 17 the Fe site is, the more frequent it undergoes the redox transition. Therefore, the Fe dissolution 18 from the active sites during O2-AST is more significant than that during Ar-AST. 19

20
In the present work, Fe demetallation in alkaline Fe-N-C CLs is studied at realistic 21 conditions, such as O2 environment, elevated current densities, and a thin AEM on the CL. We 22 show that Fe dissolution is significantly enhanced in O2 environment compared to Ar. 23 Additionally, Fe dissolution is shown to be directly proportional to the applied charge. By 24 comparing activity data with Fe dissolution, we can derive that Fe dissolution in Ar mainly 1 happens at inactive sites, whereas during ORR significantly more Fe active sites are dissolved. 2 Moreover, we discovered strong correlations between the Fe dissolution and the Fe 3+ /Fe 2+ redox 3 transition in the presence and absence of ORR in the studied potential range, 1.0 -0.57 VRHE. 4 This leads to our hypothesis that the instability of the coordinated Fe during the redox 5 transitions could be highly responsible for the Fe demetallation in Fe-N-C catalysts in alkaline 6 media. This hypothesis could help rationalizing the different scales and origins of the Fe 7 dissolution in the presence and absence of ORR. For future work, differently synthesized Fe-8 N-C catalysts in different potential regions should be evaluated. The Fe dissolution data should 9 be accompanied by ex-situ spectroscopic techniques, such as Mössbauer and Raman 10 spectroscopies, 17,18,22 to rationalize the structure-stability relationship of Fe-N-C catalysts. 11 Based on that, special strategies for improving the stability of Fe-N-C catalysts for AEMFCs 12 need to be developed, since the dominant degradation mechanism of Fe-N-C catalysts might be 13 different in alkaline compared to acidic environment. 14 can be found in section 3.1 of the SI. The loading of the CL is determined to be 22

Experimental
1.35 ± 0.05 mg·cm -2 . The thicknesses of the CL and AEM are 47.5 ± 5.2 μm and 0.5 ± 0.2 μm, 23 respectively, measured from images (see Figure S4 and Table S1) taken via cross-sectional 24 imaging with a focused ion beam scanning electron microscopy (Crossbeam 540 FIB-SEM,  1 Zeiss). More detailed information on the procedure can be found in section 3.2 of the SI. 2

GDE-ICP-MS measurements 3
The GDE-ICP-MS setup and methodology have been introduced in our previous work. 30  (O2), the data was 100 % post iR-corrected. The Ru was measured via EIS at each current step, 10 as proposed previously. 27,28 In all experiments, the electrolyte was always purged with 11 50 ml·min -1 Ar. 12