Metal-nitrogen-carbon catalysts towards acidic ORR in PEMFC: fundamentals, durability challenges, and improvement strategies

The mechanism of ORR

The ORR mechanism usually includes two pathways in acid media. A direct 4e^- pathway O₂ reduction to H₂O, and 2e^- + 2e^- pathway O₂ reduction to H₂O₂ which further produces H₂O^[29]. The selectivity of the 4e^- pathway is an essential parameter for evaluating the performance of catalysts. Currently, our understanding of the ORR mechanism is mainly based on theoretical calculations. The proposed mechanism of ORR can be mainly divided into the following parts:^[30-32]

O₂ dissociation:

OOH dissociation:

HOOH dissociation:

Indirect 4e^- pathway with the yield of H₂O₂:

Multiple ORR intermediates (e.g., O*, OH*, OOH*) involve in ORR, which bind to the surface of catalysts via the O atoms. The rate-limiting step of ORR is determined by the adsorption energy of the ORR intermediates. Too strong binding will hinder the desorption of ORR intermediates, while too weak binding hampers the reaction of the intermediates. Following the Sabatier principle, the catalyst can achieve excellent ORR performance only if the intermediates interact properly with the surface of the catalyst^[33-35]. Therefore, adjusting the absorption energy to the optimum value is the process of improving the catalytic activity. The rate-limiting steps and kinetic pathways of ORR remain unclear and are significantly affected by the catalytic site and environment. The pH environment can influence the structure of double layer and further affect the electron transfer process^[36]. Pt-based catalysts always exhibited a higher overpotential in alkaline media than in acid media [Figure 2A]. It can be inferred that the Pt surface catalyzes the reduction of 4e^- ORR by chemically adsorbing O₂ in acidic media via an internal spherical electron transfer (ISET) mechanism. However, ORR proceeds through both inner-sphere electron transfer (ISET) and outer-sphere electron transfer (OSET) mechanisms in alkaline media [Figure 2B]. Notably, PGM-free catalysts generally perform better in alkaline media than in acid media, even outperforming Pt/C catalysts sometimes^[37]. For PGM-free catalyst, the OSET mechanism is inhibited in alkaline media and the OH adsorption step is different with Pt/C. The OH desorption process of FeN₄ site involves the reduction of (H)O-Fe³⁺-N₄ to Fe²⁺-N₄, which facilitates the chemisorption of O₂ and the ISET mechanism. In contrast, the ISET mechanism requires more negative potential due to the lower Fe^2+/3+ redox potential in acid media. This above mechanism could be used to explain PGM-free catalysts perform better in alkaline media. PGM-free catalysts behave differently in different pH environments and still need further research. Although most PGM-free catalysts have better ORR performance in alkaline media, it is meaningful to develop PGM-free catalysts with high performance in acid media due to the higher maturity of PEMFC than alkaline anion exchange membrane fuel cells^[38].

Figure 2. (A) (a) Ring current measured during ORR, (b) Mass transport corrected Tafel plots; (c) ORR polarization curves in O₂ saturated electrolytes; 0.1 M NaOH, and 0.1 M HClO₄. (B) Schematic illustration of the double-layer structure during ORR in acidic (left) and alkaline (right) conditions. Insets (a) and (b) illustrate the ISET and OSET processes. These figures are quoted with permission from Ramaswamy et al.^[36].

Identification of catalytic active sites

Understanding the structure of active sites of M-N-C catalysts is a prerequisite for designing excellent catalysts. However, synthetic procedures of M-N-C catalyst are usually involved high-temperature pyrolysis of different precursors. The complex Fe-N bond formation process and uncontrolled changes in the carbon matrix during the pyrolysis make it hard to elucidate the formation of active sites. Initially, metal particles encapsulated in carbon shells were regarded as active sites due to the presence of metal nanoparticles observed in many M-N-C catalysts^[39]. With the development of advanced characterization techniques, the M-N_x structure is gradually recognized as the actual active site^[40]. Lefèvre et al. first detected different ions including FeN₄C_y⁺, FeN₃C_y⁺, FeN₂C_y⁺, and FeN₁C_y⁺ by Time-of-Flight Secondary Ion Mass Spectrometry^[41]. Furthermore, Chung et al. used high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) to determine the ratio of Fe and N atoms by the electron energy loss spectrum as 1:4^[40], which strongly supported the MN_x structure of active sites. Hitherto several types of active sites can be summarized from previous works, including MN₄, MN₂₊₂, MN₅, M₂N₅, M-N₄/M’N₄, MM’N₆, etc^[42]. Herein, we take Fe-N-C catalyst as an example to discuss the structure-stability relationship of the active site. Recently, Jaouen et al. suggested that Fe-N-C catalysts initially contained two different FeN_x sites of S1 (High-spin FeN₄C₁₂ moiety) and S2 (low- or intermediate-spin FeN₄C₁₀ moiety) by in situ ⁵⁷Fe Mössbauer spectroscopy [Figure 3A]^[43]. While S2 sites are stable during long-time Membrane electrode assemblies (MEAs) tests, S1 sites were unstable and easily degraded to ferric oxides. They proposed two possible mechanisms for the formation of ferric oxides. S1 sites of FeN_x can be directly oxidized by O₂ or triggered by protonation of nitrogen and carbon surface oxidation. Following further studies on the structural evolution of the active site, Xu et al. used in situ ⁵⁷Fe Mössbauer spectroscopy to reveal the relationship between the oxidized intermediates, the electric field in the electric double layer, and the Fe-N bond strength^[44]. They fitted the spectrum into four doublets (note as D1, D2, D3, and D4). D1 was commonly assigned to FeN₄C₈ (low-spin) or FeN₄C₁₂ (high-spin) moieties. D2 was most often assigned to FeN₄C₁₀ (low or intermediate-spin) moieties. D3 might be attributed to FeN₄ sites coordinating with two weak ligands. D4 might originate from the FeN₄ site that adsorbed oxygen species. As shown in Figure 3B, the demetallation rate of D1 and D2 sites became faster at a high potential which corresponded to the loss of active sites density in the chronoamperometric curve, and half-wave potential in the linear sweep voltammetry (LSV) curve was more significant in relative high potential. The faster loss of D1 than D2 strongly suggested the demetallation influence of the FeN₄ site. The increased content of D4 and the decreased content of D1 and D2 implied that the bare FeN₄ sites could spontaneously adsorb oxygen species to form a five or sixfold-coordinate active center [Figure 3C]. Theoretical calculations demonstrated that both oxidation intermediate adsorption and electric field might lead to longer and less robust Fe-N bonds in FeN₄ sites during ORR [Figure 3D]. On this basis, several new directions for improving catalyst durability were found: (1) increase the ratio and density of S2 (or D2) sites; (2) reduce the quantity of S1 or change S1 (or D1) into more stable active sites (S2); and (3) build more robust M-N bonds to reduce the effect of stretched M-N bonds.

Figure 3. (A) Coordination or structural changes of the S1 and S2 under in situ conditions. This figure is quoted with permission from Li et al.^[43]. (B) The ⁵⁷Fe Mössbauer spectra on Fe-N-C after degradation at different potentials for 150 h and relative contents of different types of FeN₄. (C) The in situ Mössbauer spectra of the ⁵⁷Fe-N-C and relative contents of different types of FeN₄. (Red: D1; Purple: D2; Blue: D3; Green: D4). (D) The relationship between structural evolution and binding energy or bond length of Fe-N illustrates the bond-breaking process. These figures are quoted with permission from Xu et al.^[44].

Mössbauer spectroscopy is of great importance for the structural analysis of complex Fe-N-C pyrolysis catalysts. However, the complex structure of M-N-C catalyst and limited time-space resolution characterization methods impeded the accurate identification of active moieties. The ⁵⁷Fe Mössbauer spectroscopy can identify the multiplex composition in Fe-N-C catalysts, but the FeN_x active sites in different oxidation and spin states still lack a theoretical basis to be distinguished^[45]. Detailed analysis of various doublets from different Fe structures could not be unambiguously assigned in the experiment due to no reference materials with a single kind of active site. The differentiation of active centers and the degradation of active sites under the ORR condition are still under debate and further insights are promising gained by in situ and operando Mössbauer spectroscopy.

In short, the design of M-N-C catalysts with a single type of active site will help to analyze the structure of FeN₄ sites in detail. Advanced in situ characterization techniques combined with theoretical calculations are very powerful in predicting active centers and explaining the experimental results.

Methods of evaluating ORR performance

Evaluating the ORR performance of catalysts is an essential part of designing catalysts. Only by acquiring the precise consequence can we well explain the experimental conclusions. However, accurate reflection of intrinsic ORR activity and durability remains a problem. A prevailing issue is that the catalysts exhibit excellent ORR performance in the rotating disk electrode test, but the corresponding MEA performance is still poor. The evaluation methods and test conditions can significantly influence the ORR performance. Here, we concisely discuss common methods for evaluating ORR performance from the ORR activity and ORR durability sections.

ORR activity

Highly active catalysts often suffer severe activity loss after durability testing^[46-48]. However, high-durability catalysts usually have poor activity issues. The high activity of the catalysts is the precondition for durability improvement. Achieving both high activity and durability in one catalyst is our ultimate goal for practical applications. Herein, we briefly introduce the different methods for evaluating the ORR activity of catalysts. Rotating disk electrode (RDE) is the most common technique to characterize the ORR performance of the catalysts^[49] and the half-wave potential is a key metric for evaluating the ORR activity of PGM-free catalysts^[50-54]. The experimental operating parameters would strikingly influence the consequence. Typically, M-N-C catalysts have a catalyst loading of 4 mg/cm^-2 and a thickness of 100 μm in MEAs. The catalyst loading of M-N-C catalysts ranges from 0.2-0.8 mg/cm² in RDE and the thickness (5-20 μm) is similar to the thickness of PGM cathodes in MEAs. The higher catalyst loading usually correlates with higher half-wave potential in RDE test^[53]. In contrast, the higher catalysts loading will induce more transport losses in MEA test. Thus, mass activity is also an important metric to evaluate catalytic performance^[55-57]. The catalysts loading can also influence the consequence of rotating ring-disk electrode studies. The H₂O₂ produced from the 2-electron pathway can be reduced to water internally rather than at the ring electrode if the electrodes are thick enough^[58]. Therefore, it will be more accurate to use mass activity and low loading to characterize the ORR activity and 4-electron selectivity. The electrolyte solution is also a significant parameter for ORR performance testing using RDE. The LSV curve of ORR usually measures with rotating disk electrodes in 0.1 M HClO₄ and 0.5 M H₂SO₄ solution. The performance of M-N-C catalysts often rivals or surpasses that of Pt-based catalysts in alkaline solutions. However, their performance remains insufficient in acid solutions, due to the self-adjustment of FeN₄ sites to *O-FeN₄C in alkaline media^[59]. Overall, the universal measurement of the RDE has been identified as simple to investigate the ORR performance of electrocatalysts.

MEAs testing and RDE testing are the most common methods to evaluate electrocatalysts activity. The MEA testing was more complicated than the RDE testing and represented practical fuel cell performances. During the test, RDE can provide a diffusion layer (5-50 µm) to simulate the fuel cell model. MEA is commonly composed of proton exchange membranes (PEM), two catalyst layers (CL), and two gas diffusion layers (GDL). Gas diffusion electrode (GDE) and catalyst coat membrane (CCM) are two typical MEA fabrication processes. Gas diffusion electrode is constructed by depositing catalyst ink on the microporous layer (MPL) of gas diffusion layers. The catalyst coat membrane is formed by depositing the catalyst ink on the membrane^[60]. MEA contains a much thicker catalyst layer (5-20 µm), which is hard for mass transportation. Both O₂ and H₂ should pass through the flow fields, gas diffusion layers, and catalyst layers, and finally reach the catalyst surface, which will reduce the concentration of soluble gas (O₂, H₂)^[61]. However, RDE has thinner catalyst layers (0.1-2 µm) and shorter diffusion lengths compared with MEA. Therefore, a prevalent problem with ORR catalysts was excellent activity in RDE but poor performance in the MEA testing. Translating catalyst activity and stability into MEA performance remains a challenge and requires further research^[52]. Several parameters were used in fuel cell testing to evaluate catalytic performance: (1) the current density at 0.90 V_iR in the H₂-O₂ fuel cell reflected the catalytic activity; (2) the current density at 0.80 V expressed the H₂-air fuel cell performance; and (3) high current-density performance in the H₂-air fuel cell in terms of a voltage at 0.8 A cm⁻².

Turnover frequency (TOF) and site density (SD) are also essential indexes to evaluate the intrinsic activity of M-N-C catalyst. There are two typical methods to measure SD, including the nitrite stripping technique and the CO chemisorption technique. Primbs et al. first developed the ex situ CO cryo-adsorption and in situ reductive nitrite stripping^[62]. A novel relationship was established between SD and structural information of Fe-N-C catalysts. The CO cryo-chemisorption measurement was introduced as follows: six different peak areas (△A) reflecting the amounts of CO absorption were respectively calculated from the six baseline-corrected integral pulse areas after pre-treatment, followed by six CO pluses injected into the sample loop. The calculation formula of △A was as follows:

A calibration constant c_f was used to convert peak areas to molar CO amounts:

The mass-based sites density with CO chemisorption could be calculated via Avogadro’s constant (N_A):

From the catalyst mass-based kinetic current, TOF could be further calculated.

Another method was in situ reductive nitrite stripping the calculation according to:

Q_strip was coulometric charge in units of coulomb associated with the NO stripping peak; n_strip was the number of electrons associated with the reduction of one nitrite ion; m_cat was the mass of the catalyst; SD_mass (NO₂) was mass-based Fe surface site density. In recent years, Primbs et al. reported a novel in situ method for SD quantification, which employed a cyanide anion probe^[62]. The mechanism was based on the relationship between cyanide-poisoned Fe³⁺N_x surface sites and decreased ORR activity. Irreversible poisoning of Fe³⁺N_x sites leads to the reduction in cyanide content, so the active sites could be quantified by ultraviolet−visible spectrophotometry. The linear correlation between cyanide concentration and absorbance supported SD and TOF calculation. It is worth noting that these different kinds of methods may not get the same SD or TOF due to the different adsorption abilities of FeN_x sites. Therefore, we recommend that the same method should be used to compare the TOF values of different catalysts^[63].

Methods of predicting ORR performance

Current approaches to designing M-N-C catalysts mainly rely on trial and error, mostly due to the unclear relationship between the structure and performance of ORR catalysts. A theoretical understanding of the ORR catalytic process is a major requirement for rational interpretation of experimental observations and also a prerequisite for rational improving the durability of M-N-C catalysts. Here, we briefly describe the current understanding of the relationship between catalyst theory and ORR performance.

ORR activity

The d-band theory was established to combine the surface adsorption strength with the d-state of transition metals. Since all transition metals have similar broad sp-band, the difference in absorption behavior of metal surface is almost determined by the interaction between the d-band states of metals and valance states of adsorbates^[84,85]. The d-band center (ε_d) was used to descript the average energy of the d-band and correlated with the adsorption energy and ORR activity. Cao's group presented a design principle to predict the performance of Single-atom catalysts^[86]. The free energy of each step is determined by the adsorption energy of the ORR intermediate. They found the linear relationship between electronegativity and the corresponding descriptor (φ) and adsorption energy of ΔG_OH* and ΔG_H*. Higher φ indicates more d-sate valence electrons in the single metal atom. The φ descriptor could be used to predict the adsorption energy of intermediates and catalytic performance.

The free energy diagram of ORR was widely used to predict the rate-determining step. For a classic example, the free energy of 4e^- ORR over Pt is shown in Figure 4A^[87]. At U = 1.23 V (equilibrium potential), the positive free energy change (ΔG) of the OH* to H₂O reaction step means that the surface of Pt would be covered by OH* and lose its catalytic activity. The highest potential for a facile reaction process was named thermodynamic limiting potential. The difference between the equilibrium potential and the limiting potential is called the theoretical overpotential (η), which is usually used to measure the activity of catalysts. The lower theoretical overpotential often corresponds to higher catalytic performance^[88]. Therefore, the adsorption energy regulation of ORR intermediates was a key to rationally designing highly active catalysts. Similarly, the volcano plot was universally used to illustrate the bonding energy between the active sites and intermediates, which could be split into two legs from the apex representing the strong and weak binding [Figure 4B]^[89]. The complexes on the ascending segment bind strongly with the *O intermediates and catalyze the reduction of 4e^- ORR, and the release of -OH is the key step determining the final activity. While the complexes on the descending segment (weakly bound side) usually catalyze the 2e^- ORR^[90]. Tuning the binding energy to reach the apex of the ORR volcano plot has proved to increase the activity of the M-N-C catalyst effectively.

Figure 4. (A) Free energy diagram of 4e^- ORR on Pt. This figure is quoted with permission from Kulkarni et al.^[87]. (B) M-O₂ adsorption energy. This figure is quoted with permission from Zagal et al.^[89].

ORR durability

M-N-C catalysts have revealed huge potential in replacing PGM catalysts for ORR with respect to the initial activity, but the Achilles’ heel remained in their poor stability under an acid medium. Therefore, the universal procedures to research catalyst durability are necessary for settling the key problem. Carbon corrosion, as well as the durability of catalysts, are related to the concentration of H₂O₂ and ROS. The rotating ring-disk electrode test was widely used to measure the yield of H₂O₂. However, the generation of ROS by the Fenton reaction between Fe³⁺ and H₂O₂ leads to severe degradation of catalyst performance and is also a major factor in the degradation of PEM. The Fenton reaction is shown as follows:

So, the content of ROS is a more significant parameter for evaluating the durability of catalysts because ROS can oxidize and destroy catalysts easily. Chen et al. used 7-hydroxyl coumarin as the probe to detect the content of OH* by analyzing the fluorescence intensity [Figure 5A]^[91]. Similarly, Xie et al. carried out the 6-carboxy fluorescein (6-CFL) dye as a fluorescent molecular and detected the content of ROS by comparing the fluorescence decay^[78]. Moreover, 5,5-dimethyl-1-pyrroline-N-oxide (DMPO) was used as a spin trap for OH* to prolong the lifetime of OH* so electron paramagnetic resonance (EPR) could detect the content of ROS. Our group proposed an in situ experiment to detect the content of ROS^[25]. The mechanism is that ABTS (2, 20-azinobis (3-ethylbenzthiazoline-6-sulfonate) could react with ROS and convert to ox-ABTS which induced the color change documented by UV-vis absorption spectra [Figure 5B]. The Cr-N-C catalyst shows a much smaller change than that of the Fe-N-C system, so the stability is higher. Theoretical calculations can also be used to predict catalyst durability. Xie et al. used theoretical calculations to predict the adsorption ability of different sites for ROS and H₂O₂ [Figure 5C]^[78]. The adsorption energies of ROS and H₂O₂ on Fe-N-C were higher than TaO₂-OH, indicating that the TaO₂-OH moieties trended to be attacked by ROS and thereby protecting Fe-N-C sites.

Figure 5. (A) Schematic illustration of the conversion of non-fluorescent coumarins to fluorescent 7-hydroxycoumarins by OH*. This figure is quoted with permission from Chen et al.^[91]. (B) The reaction between ROS and ABTS, and the color change of the solution after the Fenton reaction. This figure is quoted with permission from Luo et al.^[25]. (C) The adsorption energy of H₂O₂ and ROS on the surface of Fe-N-C and TaO₂-OH. This figure is quoted with permission from Xie et al.^[78].

Carbon corrosion

Carbon corrosion can be primarily divided into chemical corrosion and electrochemical corrosion. These two typical types of carbon corrosion generally involve oxidation and structural damage of the carbon matrix, which are the primary degradation mechanisms of M-N-C catalysts. In general, electrochemical corrosion is always ignored because it only happens at high potential (E > 0.9 V) and the sluggish kinetic also hinders the occurrence^[92]. However, in the fuel test, severe electrochemical corrosion will happen in the following situation. The absence of H₂ and the presence of O₂ in start-up and shut-down states result in a cathode transient potential of around 1.4 V^[93]. Fuel starvation can cause a high anode potential of up to 1.6 V. The addition of oxygen evolution reaction (OER) catalysts on the electrode surface may be a feasible way to reduce the electrochemical corrosion of carbon matrix^[94]. Severe carbon corrosion would lead to the collapse of the carbon matrix and falling/aggregation of active sites, which will enormously impact the activity and stability of the catalysts. Carbon corrosion could also drive the demetallation of MN₄ sites next to oxidized carbon atoms, increase the mass-transport resistance around the remaining active sites, reduce the active site density and electron conductivity, and reduce the electrode porosity^[28]. Therefore, carbon corrosion was regarded as the main cause of activity loss. As shown in Figure 6A, the attack of ROS eventually induced the degradation of the active sites and the formation of FeO_x particles. However, another kind of mild oxidation at the carbon surface reduced the TOF of FeC_xN_y moieties by increasing the binding energy with O₂^[95]. This kind of mild oxidation was restricted to the top surface without the formation of volatile CO or CO₂. In conclusion, the attack of ROS induces carbon oxidation decomposition active centers, and the leached iron ions, in turn, promote the formation of ROS from H₂O₂ through the Fenton reaction.

Figure 6. (A) Schematic illustration of the carbon corrosion mechanisms. This figure is quoted with permission from Wan et al.^[99]. (B) Schematic demonstration of three models of Fe-N-C catalysts composed of Fe particles, Fe particles with a carbon layer, and FeN_x sites. This figure is quoted with permission from Choi et al.^[97].(C) Comparison of the quantity of leached metal after 200 cycles AST (O₂ or Ar environment) in alkaline media. (D) The amount of dissolved Fe normalized to the applied electric charge. Relative losses of the power density (PD) at 0.65 V RHE (E) and the Fe content (F) over the Ar-AST-200, O₂-AST-200, O₂-AST-2000, or Ar-AST-2000 procedures. These figures are quoted with permission from Ku et al.^[98].

Demetallation

The demetallation of active sites is the direct cause of activity decay during ORR. Distinguishing M-N_x/C active sites in metal/metal compounds and identifying which kind of mental leaching will directly lead to the loss of activity is the prerequisite for elucidating the demetallation mechanism during long-term ORR. Most of the Fe leaching during the ORR process originates from inactive and acid-soluble iron species such as Fe nanoparticles/oxides/sulfides/carbides^[96].

Choi et al. applied a scanning flow cell connected to an inductively coupled plasma mass spectrometer (SFC/ICP-MS) to reveal the demetallation mechanism^[97]. They synthesized three kinds of Fe-N-C catalysts. The first Fe-N-C catalysts included a mass of both FeN_xC_y active moieties and Fe particles as a reference (denoted as FeNC-wet-1). The second Fe-N-C catalysts had low amounts of Fe particles and a mass of FeN_xC_y active moieties (denoted as FeNC-dry-1). The third Fe-N-C catalysts contained FeN_xC_y active moieties and undetectable Fe particles [Figure 6B]. These different Fe particle contents in Fe-N-C catalysts were then analyzed using EXAFS and Mössbauer spectroscopy. For FeNC-wet-1 catalysts, Fe leaching could be observed evidently in the initial 20 fast cyclic voltammograms (CVs) (ca. 705 μg_Feg_catal⁻¹). However, the Fe leaching amount of FeNC-dry-1and FeNC-dry-0.5 was less, ca. 87μg_Feg_catal⁻¹and no more than ca.10 μg_Feg_catal⁻¹, respectively. Although there was a huge discrepancy in the amount of Fe precipitated in the different catalysts. Surprisingly, they had similar tendencies in fuel cell tests in the first 50 h. They both exhibited high initial ORR activity and decreased slowly during the fuel cell testing. Therefore, the leaching of Fe from Fe particles could not lead to the loss of initial activity, but the leached iron ions would catalyze the Fenton reaction, which further promoted the formation of ROS and reduced the activity of the catalysts.

To elucidate the mechanism of demetallation, Xie et al. applied DFT calculation to predict the demetallation process of MN₄ sites^[22]. The proposed metal leaching mechanism is that the metal center of MN₄ sites moves away from the state coordinating with N₄ to the carbon surface and combines with two N atoms, the other two N atoms bond with H in acid media. Then the Fe center adsorbed O₂ moves away from the plane and further leaches off the active sites.

Recently, Ku et al. proposed that different gas environments might induce different rates of demetallation of active sites in alkaline media^[98]. By comparing the content of iron leaching in the different environments (O₂, Ar), more iron leaching in the O₂ environment than in the Ar environment was observed after 200 cycles AST [Figure 6C]. The linear relationship between the leaching rate of Fe and the number of electric charges indicated the correlation of current density with Fe dissolution at relatively high current densities [Figure 6D]. On the contrary, no correlation was observed between Fe dissolution and the number of potential cycles. Furthermore, the relation between the Fe leaching rate and the power density change was further discussed. Around 3.4% loss of Fe content after O₂-AST-2000 induced the 34% loss of power density [Figure 6E and F]. This prominent power density in the O₂ environment loss could not be fully attributed to the demetallation of FeN₄ moieties, which might be related to the Fe leaching of the inactive site. An apparent decrease of the C-C signal in XPS C 1s proved that the carbon shells of Fe₃C were destroyed, exposing the Fe₃C moieties, which further induced the Fe leaching of Fe₃C and the decrease of TOF. Fe₃C might participate in the ORR process, as Jiang et al. reported^[39]. The Fe₃C nanoparticles might facilitate the adsorption of oxygen molecules. Thus, the coexistence of high content of FeN₄ and sufficient metallic iron nanoparticles was essential for the high ORR activity. Furthermore, the S-number (the reciprocal of the charge-normalized dissolution) was used to account for charge-normalized dissolution. During the first 1500 cycles AST, the S-number stabilizes around 10^-6 (around 10⁶ electrons transfer for a leached Fe atom). The S-number increased significantly after 1500 cycles, which was consistent with the rapid decay of the initial catalytic performance and subsequent stabilization. In previous reports, the demetallation of the catalysts was widely recognized as the primary factor contributing to catalyst instability issues. The demetallation of active sites will directly induce the degradation of catalytic performance. The demetallation of inactive Fe particles will also lead to the decrease of TOF and enhanced Fenton reaction. Therefore, improving the intrinsic demetallation resistance of active sites is an effective way to improve catalyst stability.

Nitrogen protonation

The nitrogen protonation process is the N atoms of MN₄ moieties combine with H to form N-H bonds, which is considered to be the rate-determining step in the degradation of the MN₄ moieties^[46]. After nitrogen protonation, the M-N bonds tend to break and the metal center will move away from the N₄ plane and leach out during ORR. The protonation process involves attaching one or two hydrogen atoms to the surface of active moiety. Nitrogen protonation affected the C-N bond stability of the pyridinic-N doped graphene (PNG) and oxidized PNG (PNG-O). After protonation, the Fe-N bond would considerably stretch^[100]. PNG and PNG-O exhibited different degrees of Fe-N bond elongation after nitrogen protonation, indicating that the adsorption of O₂ could promote the stretching of the Fe-N bond. Moreover, OH* could also promote the stretching of Fe-N bonds and demetallation of active sites; if two OH* bound to the metal site on the same side, the Fe centers would form the fragile *M(OH)₂ oxidation intermediates^[101]. The *M(OH)₂ intermediates were susceptible to successive N protonation that eventually caused irreversible damage to active sites [Figure 7]. The nitrogen protonation process further induced the demetallation and reduction of catalytic activity. Herranz et al. especially investigated N-protonation and anion-binding effects on Fe-N-C catalysts^[102]. They concluded with two hypotheses. The FeN₄ sites had high TOF when the active center was near the unprotonated N-groups and bound to the anions during ORR. Once the basic N-groups were protonated, the TOF of the FeN₄ site would be extremely decreased. The pace of the step is dependent on the speed of polymeric chains in reaching the N-groups. In this hypothesis, the protonation of pyridinic nitrogen might be the main reason for the activity decay of PGM-free catalysts. Another hypothesis was that the FeN₄ site kept high TOF when the near N-groups were protonated without anion bonds. However, the TOF of FeN₄ site will decrease when the anion bond with FeN₄ center. In this case, protons can rapidly access the basic N-groups, but the anions will not immediately follow with the protons. This hypothesis could be used to explain the catalytic performance of M-N-C catalysts, which exhibited high initial activity followed by a rapid decrease. In general, the nitrogen protonation process is the main reason for the decrease of TOF and the key step in the demetallation of the MN₄ sites.

Figure 7. Mechanism of the demetallation induced by N protonation. This figure is quoted with permission from Yang et al.^[101].

Water flooding

M-N-C catalysts with excellent performance often possess many micropores on the carbon basis in which most active sites are embedded. When the water is trapped by the micropores, the formation of three-phase interfaces is hindered and the diffusion of O₂ to the active sites is difficult. This undesired process is called water flooding.

Li et al. have shown that severe water flooding occurred in thick electrode layers, which might be the main cause of the initial activity loss in the MEAs^[103]. Due to the hydrophilicity of the porous carbon matrix and the chemisorption capacity of the catalyst, the micropores of the M-N-C catalyst tended to be blocked. The blocked micropores would further hinder access to oxygen and the loss of activity sites in the micropores^[104]. However, Choi et al. proposed the opposite view for microporous flooding^[105]. They compared the degree of micropore flooding with the observed activity loss in MEA tests. During the in situ stability test, this negligible submerged micropore was unlikely to be the main reason for performance degradation. Hence, the mechanism of micropores flooding is still unclear and needs further research.

Stable carbon matrix

Previous studies have demonstrated that highly graphitized catalysts could significantly improve the stability of catalysts^[106]. The highly graphitized carbon substrate can enhance the hydrophobicity of the catalyst, protect the catalyst from severe micropore flooding, improve the active site stability, and further improve the demetallation and corrosion resistance of the catalyst^[107]. Therefore, designing a more graphitized local structure of catalysts is a feasible way to improve catalyst stability. Higher pyrolysis temperature could induce higher a degree of carbon-based graphitization^[17]. Qu et al. designed Fe-N_x/C catalysts via a stepwise pyrolysis treatment which achieved a highly graphitized carbon and superior stability^[108]. Only 25 mV loss of half-wave potential after 30000 cycles ADT and high retention rate of activity after 100 h fuel cell test proved the role of high levels of graphitized carbon substrates in enhancing catalyst stability. Similarly, Yan et al. designed a nitrogen-doped bimetallic carbide-graphite catalyst achieving ultra-high stability in alkaline media^[77]. The catalyst (noted as N-Fe₂MoC-GC) was synthesized at high temperatures (1300-1400 °C), which achieved a high degree of graphitization of carbon matrix. After the durability test, no apparently increasing in particle size in N-Fe₂MoC-GC was observed. Therefore, high degree graphitization of carbon matrix could exceedingly improve the stability of catalysts. Recently, Li et al. synthesized ordered and interconnected macropores of carbon matrix [Figure 8A and Figure 8B] with FeCo alloy nanoparticles^[74]. The interconnected microporous and hierarchical porous carbon matrix provided channels for mass transfer and exceedingly facilitated the exposure of active sites. The relationship between Fe and Co was elucidated by DFT calculation, and the catalysts achieved a lower energy barrier for the adsorption of intermediates. The introduction of Fe into Co induced the d-band center of Co closer to the Fermi level, which accelerated the adsorption of intermediates. Furthermore, the high stability of the catalyst was mainly due to the strong interaction of FeCo alloy nanoparticles and porous carbon matrix, and only a small amount of metal leaching was measured after the stability test. The high durability performance and minimal metal leaching of the catalysts manifested that the activity sites were firmly anchored on the ordered porous carbon matrix.

Figure 8. (A, B) Field-emission scanning electron microscopy (FESEM) image of 3D ordered porous ZIF-67 taken from different directions. These figures are quoted with permission from Li et al.^[74]. Thin (C) and thick (D) carbon shell models. These figures are quoted with permission from Yoo et al.^[109]. (E) Schematic demonstration of Fe-N-C catalyst with NH₄Cl treatment and deposit of a carbon layer on the Fe-AC catalyst. (F, G) ORR polarization curves after different cycles AST text in O₂-saturated 0.5 H₂SO₄. These figures are quoted with permission from Liu et al.^[46] (H) H₂-O₂ PEMFCs polarization and power density curves. Cathode, 3.5 mg_cat cm^-2 for Fe-N-C; anode, 0.4 mg_Pt cm^-2. This figure is quoted with permission from Yin et al.^[81].

The carbon layer could effectively improve the durability of catalysts by preventing the catalysts from being attacked by ROS. Yoo et al. summarized the strategy of designing carbon shells on catalyst surfaces and discussed the relationship between the structure of carbon shells and the durability of catalysts^[109]. On the one hand, the carbon shell provided a physical parclose that could protect the catalysts from ROS and H₂O₂ attack. On the other hand, the carbon shell maintained active sites and prevented metal sites from agglomeration during the ORR, thus keeping high accessible surface area. The thickness of the carbon shell was a significant influencing parameter for catalytic performance, and a thin carbon shell (< 1 nm) allowed the permeation of reactants during the ORR [Figure 8C]. However, a thick carbon shell (> 3 nm) had redundant coverage of active sites, which would hinder mass transport [Figure 8D]. Liu et al. developed a novel strategy to synthesize a more stable Fe-N-C catalyst via chemical vapor deposition (CVD), depositing an N-doped carbon layer on the Fe−N−C catalyst, resulting in excellent stability^[73]. Gaseous deposition of 2-methylimidazole on Fe-doped ZnO substrates followed by thermal activation at 1000 °C to from the thin carbon layer (< 1 nm). Recently, Liu et al. reported a method for chemical vapor deposition of ZIF-8 onto Fe-AC^[46]. As shown in Figure 8E, NH₄Cl treatment could create amounts of defects in the carbon matrix. Further, the CVD method was used to improve the durability of the Fe-AC catalyst. On the one hand, the deposited C-N source on the catalyst could form a thin carbon layer that physically protects the active site from ROS attack. On the other hand, the CVD process could promote the transfer of S1 sites to S2 sites. As shown in Figure 8F, the Fe-AC exhibited the best ORR activity in RDE test with a half-wave potential of 0.915 V. Unfortunately, such high ORR activity decreased strikingly (E_1/2: 0.915 V-0.791 V) after 30000 cycles AST test. Therefore, the CVD technique was used to solve the unstable issue. With a decrease in ORR activity, the stability of Fe-AD-CVD was increased considerably [Figure 8G]. Achieving both high activity and high durability in one catalyst remains difficult, and it is important to unravel the degradation mechanisms. The CVD technique can be used not only to deposit C-N sources to improve catalyst stability but also to deposit metal sources. Sun's group report a CVD method that deposits FeCl₂ on a surface-rich pyridinic-N carbon matrix, achieving a high density and accessibility of FeN₄ sites^[81]. The ORR activity of Fe_g-NC/Phen in PEMFC even reached the DOE (2025) target with a current density of 46 mA cm^-2@0.9 V and a P_max of 1.53 Wcm^-2(H₂-O₂), which is almost the best reported performance of M-N-C catalysts in MEA [Figure 8H]. The performance of MEAs is not only related to the ORR capability of the catalyst but also related to the MEA preparation process, catalyst ink, and test conditions. Seizing gaseous Fe²⁺ on the carbon matrix increased the density of surface FeN₄ sites with high utilization. The introduction of defects in the carbon matrix can also improve the accessibility of the active substance. Kim et al. embedded FeCo alloy nanoparticles into a defect-rich N-doped carbon matrix^[110]. The carbon matrix was oxidized by the appropriate amount of H₂O₂. A great number of defect sites and a high surface area carbon matrix in the catalyst improved the accessibility of active sites. The encapsulated alloy nanoparticles were protected by the carbon layer that prevented corrosion and aggregation, achieving high stability during the ORR. Therefore, regulating the structure of carbon support is essential to improve catalytic performance. The carbon matrix could not only anchor the active site on the carbon matrix but also change the electronic structure of the metal centers, which was related to the intrinsic activity of the catalysts. Therefore, designing the carbon matrix is a feasible method to improve the performance of catalysts.

Regulation of the M-N_x bonds

The length of the Fe-N bond is an essential parameter affecting the stability of catalysts. Building robust Fe-N bonds can reduce the prolongation caused by absorption of oxygenated intermediates, further increasing the resistance of demetallation. In a recent study by Ohyama et al., in situ XAS was used to analyze the durability of the 14-mcembered hexaaza macrocyclic Fe complex (Fe-14MR) [Figure 9A] and Fe phthalocyanine (FePc) [Figure 9B]^[111]. As shown in Figure 9C and Figure 9D, the CV peak of Fe-14MR/C was weakened to a lower degree than FePc, indicating that Fe-14MR/C was more resistant to demetallation than FePc during ORR. Furthermore, the Fe concentrations of catalysts were monitored by Fe-derived X-ray fluorescence. The Fe leaching rate of 14-membered ring ligands was much lower than FePc in acid, which also indicated the higher durability of the catalysts [Figure 9E]. Therefore, regulating the length of Fe-N bonds was key to the stability of the Fe-14MR during the ORR. It was the shorter Fe-N bond and more compact ligands that led to the high stability of Fe-14MR. Changing the binding constant and chelation of the metal center was a feasible way to regulate Fe-N bond length and further reduced the demetallation of FeN₄ active sites. Increased temperature also resulted in shorter Fe-N bonds^[112]. Miao et al. successfully synthesized P(AA-MA) Fe-N catalysts and regulated the lengths of Fe-N bonds by manipulating the rate of acrylic acid and maleic acid during the copolymerizing process^[67]. PAA-Fe-N catalyst was used as a reference and was synthesized from homopolymerization of acrylic acid and other precursors with shorter Fe-N bonds. Fe-N bonds of different lengths for the P(AA-MA)Fe-N and PAA Fe-N catalysts were verified by XAS [Figure 9F] and Mössbauer spectroscopy. Moreover, isothermal titration calorimetry (ITC) was applied to measure the binding constant stoichiometry of ligands and metal ions. The binding constant of Fe³⁺ to P(AA-MA) was higher than Fe³⁺ with PAA or MA. As expected, the P(AA-MA) Fe-N catalysts had higher stability and activity than PAA Fe-N catalysts during the durability test. A similar consequence was achieved by Chen et al.^[68]. Hierarchically porous N-doped catalysts were synthesized by pyrolysis followed by HF etching to remove Zn and SiO₂ from the substrate (note as surf-FeN₄-HPC). The catalysts achieved high accessibility of active sites by anchoring abundant FeN₄ sites at the carbon edges. As shown in EXAFS spectra [Figure 9G], surf-FeN₄-HPC had a longer length of Fe-N bonds than FeN₄-HPC (anchoring on basal-plane) and achieved high durability, with only 20 mV loss of half-wave potential after 20000 cyclic voltammetry scans. These two above consequences seemed to conflict with other studies. Li et al. synthesized Fe-pyridinic N-C catalysts with a higher ratio of pyridinic N than Fe-N-C catalysts^[69]. Fe-pyridinic N-C catalysts achieved both better activity and stability than Fe-N-C catalysts. DFT calculation showed that pyridine-type FeN₄ had a higher Gibbs free energy during the Fe leaching process. The bond lengths of Fe-N to pyridine nitrogen and pyrrolic nitrogen were 1.88 and 2.08 Å, respectively, which also supported that the shorter Fe-N bond was stable. A similar consequence could also observe in CoN₄ sites, and the Co-N-C catalysts exhibited outstanding resistance to metal leaching, which might be related to the shorter Co-N bond length^[22]. The Co-N bond of Co-N-C (1.94 Å) was shorter and more stable than porphyrin-like CoN₄C₁₂ (2.01 Å)^[72].

Figure 9. Schematic demonstration of the structure of Fe14MR (A) and FePc (B). CV curves for Fe-14MR/C (C) and FePc(D). (E) Relative concentration changing of Fe in Fe-14MR/C and FePc during CV cycling under N₂ and O₂. These figures are quoted with permission from Ohyama et al.^[111]. (F) k³ -weighted FT-EXAFS spectra of P(AA-MA)(5-1)-Fe-N, PAA-Fe-N, and Fe foil samples. This figure is quoted with permission from Miao et al.^[67]. (G) k³ -weighted FT-EXAFS spectra of sur-FeN₄-HPC, FeN₄-HPC, FePc, and Fe foil. This figure is quoted with permission from Chen et al.^[68].

In a recent study by Wan et al., the introduction of Fe clusters near FeN₄ sites with two-dimensional porous carbon (noted as Fe_SA/Fe_AC-2DNPC) achieved excellent catalytic performance during the ORR^[72]. The strong interaction of Fe clusters and FeN₄ sites could change the electronic structure of active sites and lead to a much more suitable absorption strength of ORR intermediates [Figure 10A]. The schematic of ORR is shown in Figure 10B. Fe_SA/Fe_AC-2DNPC lost only 18 mV of half-wave potential after 10,000 potential cycles in an O₂-purged acidic medium. Then DFT calculation was used to elucidate the possible mechanism; the interaction between the Fe cluster and FeN₄ promoted the axial adsorption of OH at the Fe center while the other side of FeN₄ was used for ORR. Fe(OH)N₄ achieved a lower limiting energy barrier and enhanced ORR activity. Several models were built to elucidate the catalytic performance of FeN₄ sites near different Fe clusters (Fe₄-N₆, Fe₁₃-N₆, Fe₄-C₆). The different numbers of Fe atoms in the cluster have no obvious effect on the activity enhancement; on the contrary, the effect of N-coordinated Fe clusters on FeN₄ centers is greater than that of C-coordinated Fe clusters. Furthermore, molecular dynamic simulation was carried out to obtain the Fe-N radial distribution function, which could be used to illustrate the Fe-N bond length [Figure 10C]. The average Fe-N bond lengths in FeN₄ were 1.82-2.38 Å, which were slightly longer than that of Fe-N₄/Fe₄-N₆ (1.82-2.22 Å), indicating that FeN₄ sites near Fe₄-N₆ clusters are more stable than bare FeN₄ sites. The Fe-N bond lengths of bare FeN₄ sites were considerably stretched from 1.82-2.38 Å to 1.71-2.50 Å; on the contrary, Fe-N₄/Fe₄-N₆ still kept compact distribution, and short Fe-N bond lengths of 1.86-2.06 Å, which correlated with the high stability of Fe_SA/Fe_AC-2DNPC, even at the elevated temperature. The compact distribution of Fe-N bonds might be attributed to the incoherent vibrations of Fe clusters and single Fe atoms. The Fe₄-N₆ clusters increased the stability of the active center by building robust Fe-N bonds. However, the Fe³⁺ from the Fe clusters seemed to contradict the mitigates of the Fenton reactions to minimize ROS generation. Notably, the Fe_SA/Fe_AC-2DNPC demonstrated higher stability relative to references, which included the catalysts containing only bare FeN₄ sites (Fe_SA-2DNPC) and the catalysts containing Fe nanoparticles and FeN₄ sites (Fe_SA/Fe_NP-2DNPC). The atomic Fe clusters were chemically bound to the carbon matrix and kept stable in acid media^[113]. Some researchers reported that the metal cluster or nanoparticle in Fe-N-C catalysts improved the ORR activity in both RDE and MEA testing^[114,115]. However, the unstable Fe cluster would be leaching and transferred ions, which promoted the Fenton reaction and reduced the conductivity of Nafion under the test condition^[116]. Thus, the regulation M-N_x bonds by introducing metal clusters and nanoparticles should be cautious due to the degradation mechanism in PEMFCs.

Figure 10. (A) Free energy diagrams at 1.23 V for ORR over three types of active sites of Fe-N₄, Fe-N₄/Fe₄-N_6, and Fe-N₄-OH/Fe₄-N₆. (B) Schematic ORR process on the Fe-N₄ site of Fe-N₄/Fe₄-N₆. (C) Fe-N radical distribution function of the Fe-N₄ site in the models of Fe-N₄ near Fe₄N₆ and bare Fe-N₄ site at 25 °C and 80 °C. These figures are quoted with permission from Wan et al.^[72].

Introduction of the fifth ligand

The demetallation of active centers may involve the adsorption of ORR intermediates in the axial direction of metal center. The introduction of the fifth ligand can reduce the elongation of M-N bond. Recent studies have shown that Fe centers could spontaneously adsorb ORR intermediates to adjust the electronic structure of the FeN₄ active sites and catalyze ORR with O₂ at another side^[117]. Their calculation found that the Fe centers of Fe-N-C catalysts tend to be covered with OH* in wide potential rage (0.28-1.00 V) and the adsorption energy optimized by the OH* intermediates so that these sites achieved enhanced ORR performance [Figure 11A]. The theoretical result indicated that Fe(OH)N₄ structure exhibited higher half-wave potential (0.88 V) and was closer to the actual result than bare FeN₄ in catalyst [Figure 11B]. Our group concluded that RuN₄OH structure was the actual active site in the Ru-N-C catalysts^[118]. As shown in Figure 11C, RuN₄OH structure achieved downhill in free energy for each step at U=0 V due to the presence of the -OH, which demonstrated facile reaction progress for ORR. The introduction of -OH downshifted the d-band center of Ru and reduced the adsorption energy for *OH intermediates. As shown in Figure 11D and Figure 11E, the linear relationship between the adsorption energy, overpotential, and Bader charge of Metals proved that the -OH ligand changed the electronic structure of the metal center, so the Ru-N-C catalysts move close to the volcano apex and achieve the high ORR performance.

Figure 11. (A) Coverages of the ORR intermediates on the Fe center as a function of U. (B) Simulated ORR polarization curves of Fe-N-C catalysts based on FeN₄ and Fe(OH)N₄ centers. These figures are quoted with permission from Wang et al.^[117]. (C) Gibbs free energy diagram for ORR. (D) Linear relation between Bader charge of the mental and binding energy of ORR intermediates. (E) Linear relation between Bader charge of mental and overpotential. These figures are quoted with permission from Xiao et al.^[118].

The MN₄ site tends to be covered by molecules in the axial direction, forming a 5-coordinated or 6-coordinated structure, but these weak ligands may not keep stable in working conditions. In situ Mössbauer spectroscopy and XAS can be used to detect adsorbed species on FeN₄ surfaces. Unfortunately, the time resolution is not enough to monitor the fast changes of adsorbed species on the surface of the active sites, which hinders the cognition of the relationship with axial coordination and performance. Therefore, it is necessary to develop in-situ and operando characterization techniques to resolve the dynamic mechanism and design effective electrocatalysts.

To elucidate the influence of ORR intermediates on the durability of FeN₄ centers, Wei et al. used in situ spectroscopy to illustrate the interaction between FeN₄ active sites and ORR intermediates^[119]. During the ORR process, Fe centers would move away or toward the plane when the FeN₄ site adsorbs the ORR intermediates. As shown in Figure 12A, the 593 cm⁻¹ Raman band was attributed to the vibration of in-plane Fe-N₄^[120]. Notably, in acid media, the Raman band at 593 cm⁻¹ attenuated considerably at a lower potential compared with the spectroscopy at 1.0 V, implying that the Fe centers tend to move away from the plane at a relatively lower potential. This change was reversible when the potential was kept low for a short time and quickly switched back to high potential. Conversely, this change could also become irreversible when the catalyst is kept at a low potential for a long time (＞ 10 min) in acid media. The discrepancy in alkaline media [Figure 12B] was also observed; the enhanced intensity at 593 cm⁻¹ with a wide potential range matched with the higher stability of FePc/C catalysts in alkaline media than in acid media. The possible mechanism was that the axial of the FeN₄ site would spontaneously adsorb a hydroxyl to form a more catalytical and stable site. The axial OH* of the Fe(OH)N₄ site would desorb at low potential (< 0.5 V) in acid media, while Fe(OH)N₄ remained stable over a wide range of potentials in alkaline media^[40].

Figure 12. In situ Raman spectroscopy of FePc/C during ORR in (A) acid solution and (B) alkaline solution. These figures are quoted with permission from Wei et al.^[119]. (C) ORR mechanism at fourfold coordination FeN₄C₈ and fivefold coordination FeN₄C₈. This figure is quoted with permission from Nematollahi et al.^[121].

Recently, Nematollahi et al. reported the effect of the fifth ligand including -NH₂, -OH, and -SO₄ on the stability of FeN₄ site^[121]. In the presence of the fifth ligand, the degree of Fe-N bond elongation during ORR was significantly reduced. The fifth ligand of the Fe center might improve the resistance of demetallation. The possible mechanism was that the fifth axial ligand at the FeN₄ site could pull the Fe center back into the plane rather than being pulled out of the plane by ORR intermediates [Figure 12C].

Introducing radical scavengers

The addition of radical scavengers to catalysts is an active defensive strategy to reduce the content of H₂O₂ and ROS. The free radicals from the Fenton reaction render catalysts and PEM less durable^[122]. The chemical degradations of PEMs are mainly caused by free radical attacks. Therefore, reducing the yield of free radicals was the direct way to improve the durability of PEMFCs. Radical scavengers have been used to improve the durability of PEMs. However, few attempts have been made to introduce radical scavengers into catalysts. Cerium oxide nanoparticles have the excellent scavenging capability of radicals and are widely used radical scavengers^[123,124]. Wei et al. introduced CeO₂ into the Fe-N-C catalyst, which achieved reduced H₂O₂ production^[125]. The CeO₂ could catalyze the reduction of H₂O₂.

The addition of CeO₂ significantly reduced the production of hydrogen peroxide and ROS, which further improved the stability of the catalyst.

Other metal oxides could also be used as radical scavengers. For example, Xie et al. reported a novel method to address the issues of carbon corrosion and demetallation, which are caused by ROS and H₂O₂, by introducing Ta-TiO_x to Fe-N-C catalysts [Figure 13A]^[78], where the Ta-TiO_x was used as a radical scavenger. To determine the radical scavenging ability of Ta-TiO_x, 6-carboxy fluorescein (6CFL) was used as a dye to measure the content of free radicals [Figure 13B]. Fenton reagent/Ta-TiO_x scavengers exhibited the lowest fluorescence decay, demonstrating the powerful scavenging ability of Ta-TiO_x nanoparticles. Moreover, 5,5-dimethyl-1-pyrroline-N-oxide (DMPO) was used as a spin trap for OH* to prolong the lifetime of free radicals, so electron paramagnetic resonance (EPR) could be used to further evaluate the scavenging ability. As shown in Figure 13C, the variation in EPR peak intensity of OH-DMPO with or without Ta-TiO_x/KB proved that Ta-TiO_x nanoparticles significantly decreased the content of free radicals. No apparent change of initial activity after the introduction of scavengers showed that the scavengers well matched up with the Fe-N-C catalysts [Figure 13D]. Ta-TiO_x was introduced into Fe-N-C catalysts as a radical scavenger to catalyze the oxidation from H₂O₂ to O₂ and H₂O and scavenge radicals.

Figure 13. (A) Schematic comparing the PGM-free catalyst in the ORR process without and with the radical scavengers. (B) Stern-Volmer plot obtained using 6CFL dye in the radical solution with Fenton’s reagent, Fenton’s reagent with Fe-N-C, and Fenton’s reagent with Ta-TiO_x/KB. (C) Electron paramagnetic resonance spectra of the OH-DMPO. (D) ORR performance of Fe-N-C catalysts with or without the TaTiOx-/KB radical scavengers. These figures are quoted with permission from Xie et al.^[78].

Compared with the introduction of radical scavengers to catalysts, previous strategies to improve durability were the passive way to protect active sites from free radical attacks, such as increasing the degree of graphitization of carbon matrix and etc. However, the addition of radical scavenger additives was a proactive approach to eliminating H₂O₂ and free radicals, which further improved the stability of catalysts.

Designing a diatomic catalyst

To address the instability caused by the Fenton reaction, numerous strategies have been proposed over the past few decades, such as changing the Fe metal center to another transition metal and designing diatomic catalysts. Mn-N-C^[26], Co-N-C^[22,70], Cu-N-C^[24,82], and Sn-N-C^[83] catalysts have presented higher stability due to these metal centers with weak Fenton reactivity. Thus, designing diatomic catalysts is an effective way to improve stability and activity of catalysts. The interaction between metal and metal always involves d-orbital hybridization, electronic structure change, and charge density change^[126]. In particular, the transition metals have highly variable d-orbitals and spatial geometries; thus, the introduction of another transition metal into Fe-N-C catalysts was an efficient way to regulate the electronic structure of active centers. Recently, our group reported a FeCoN₅/C catalyst with Fe-Co dual-atom center sites to decrease the Fenton reaction and boost ORR activity by adjusting the electronic structure of the mental site^[80]. The formation energy of FeCoN₅ (-11.801 eV) was calculated, which was much lower than FeN₄ (-8.48 eV) and CoN₄ (-7.7 eV), implying the spontaneous formation and higher stability of diatomic centers. FeCoN₅/C achieved high catalytic activity with a half-wave potential of 0.86 V and only 13 mV loss after 5000 potential cycles. Homonuclear diatomic catalysts were also a part of diatomic catalysts, our group designed Co₂N₅ diatomic activity sites, which exhibited higher ORR activity than CoN₄ sites^[71]. Regulating the neighbor atoms of Fe was an effective way to regulate the spin state of the FeN₄ site. For example, He et al., by neighboring Cu atoms with Fe atoms, successfully regulated the spin state of Fe, which achieved a low H₂O₂ yield (1.2%) and enhanced ORR durability^[79]. The interaction of different centers can be regulated via changing bond polarity, atomic distance, bonding coordination, etc. The correlated metal atoms can not only directly bind with the metal center in the first coordination shell, but also bond with the mental center via bridging atoms (O, N, S, etc.) in the second coordination shell. Indirect coordination exhibits complicated mutual effects and effective electronic structure adjustment. For example, Liu et al. built the bridge bond of Fe-O-Ti^[127], and the introduction of Ti into the second coordination shell of the Fe center changed the spin state of the Fe center, resulting in enhanced ORR activity. The introduction of bridge bonds of Fe-O-Fe into catalysts also achieved enhanced catalytic activity^[128]. Recently, Peng et al. reported a novel catalyst with axial coordination with N rather than another metal atom via the bridge of O atoms (noted as FeN₄-O-NCR)^[129]. By theoretical calculation, FeN₄-O-NCR sites were thermodynamically more stable than bare FeN₄ sites. From the PDOS, after the FeN₄ site was axial bridging with the O atom, the 3d-band center of the Fe atom shifted to a more negative energy direction compared to bare FeN₄ sites [Figure 14A]. Therefore, the appropriate absorption energy of ORR intermediates at active sites and the higher thermodynamically stability led to the enhanced ORR performance of the catalysts. CoN₄ sites exhibited higher resistance to the Fenton reaction and higher stability than FeN₄ sites^[22]. Intergrading CoN₄ and FeN₄ into one catalyst may achieve improved stability while maintaining high catalytic activity. Yang et al. integrated two single-metal sites, FeNC (shell) and CoNC (core), in the catalysts (noted as p-FeNC@CoNC)^[75]. From the result of Raman spectroscopy, p-FeNC@CoNC (0.93) had a medium ratio of I_D/I_G, implying a higher degree of graphitization carbon matrix than p-Fe-N-C (0.84). Therefore, the introduction of the CoN₄ metal center could induce the formation of graphitic carbon matrix. Evidently, the p-FeNC@CoNC catalysts achieved 0.98 V of onset potential and 0.87 V of half-wave potential, only 16mV loss of half-wave potential after 30000 cycles AST in acid media [Figure 14B].

Figure 14. (A) Scaling relationships between the η_ORR and d-band center of Fe-3d orbits for the FeN₄-based structures (pd, pr, and gr refer to pyridinic N, pyrrolic N, and graphitic N). This figure is quoted with permission from Peng et al.^[129]. (B) LSV after different cycles Potential cycling during low-potential range (0.6-1.0 V vs. RHE) p-FeNC@CoNC at 25 °C in O₂-saturated 0.5 M H₂SO₄ solution. This figure is quoted with permission from Yang et al.^[75]. (C) ΔG_1*H and (D) ΔG_2*H for Fe/M-N-C (M = Mn, Fe, Co, Ni, Cu, and Zn). (E) Protonation diagram of Fe/Zn-N-C. These figures are quoted with permission from Li et al.^[76].

The protonation of active sites is the main reason for the decrease in activity. The active sites trend to combine with H⁺, leading to the decrease of TOF in acid media. Recently, Li et al. designed a durable half-metallic diatomic catalyst to alleviate the influence of protonation^[76]. Six transition metals (M = Mn, Zn, Ni, Cu, Co, Fe) with more than half-filled 3d orbit were selected to separately form bimetal catalyst models with Fe. The free energies (G) of protonation (both one and two hydrogen pathways) were calculated in six bimetal models [Figure 14C and D]. Fe/Zn-N-C catalysts had the highest G of protonation, indicating that the coordination of Fe and Zn atoms achieved excellent resistance to protonation. The Fe/Zn-N-C catalysts showed remarkable durability, exhibiting a half-wave potential of 0.906 V and only 12 mV loss after 5000 potential cycles, which was lower than Fe-N-C (28 mV). Both adsorptions of O₂ and H⁺ led to the elongation of M-N bonds, with only a slight change of Fe-N bonds compared with Zn-N bonds. However, most Zn-N bonds were broken after protonation. The FeN₄ active centers were protected by the sacrifice of Zn-N bonds [Figure 14E]. Therefore, designing diatomic catalysts was the feasible way to solve the unstable issue of catalysts, including insufficient intrinsic stability, protonation of N-groups, excessive H₂O₂ yield, etc.

Water management

Large amounts of active sites are embedded in the micropores; thus, blocking the micropores induce high mass transfer resistance due to the prevention of O₂ and H⁺ approach to active sites. Micropore flooding always leads to activity loss of catalysts during the MEA test. Therefore, efficient water management is significant for improving the durability of PGM-free cathode in MEA. Regulating the structure of carbon matrix is a feasible method to alleviate water flooding. The hierarchical porous structure can maximize the effect of catalysts by facilitating mass transfer and alleviating micropore flooding. The introduction of macropores and microporous can offer sufficient water-draining channels for reactants transferring and alleviating micropore flooding. To elucidate the influence microporous layer in water management, Tabe et al. developed a freezing method to monitor water distribution on the catalyst layers and further illuminate the influence of microporous layer on the performance of fuel cells and the water content of gas diffusion layers^[130]. The water content was further measured by comparing the presence [Figure 15A] and absence of microporous layer [Figure 15B], where the presence of water in the catalyst layers was detected as ice. The interface of catalyst layers and gas diffusion layers provided a space for the accumulation of water. In contrast, the participation of microporous layer could boost water gasification and restrain water accumulation. The enhanced MEA performance also proved the effect of micropores [Figure 15C], which offered a buffer layer for water accumulation.

Figure 15. Schematic water distribution on a cathode with (A) or without(B) microporous layer. (C) Polarization curves for the MEAs at 35 °C operations (100% and 76% RH). These figures are quoted with permission from Tabe et al.^[130]. (D) The hydrophobic group (Ar-CF₃) was modified in Fe-N-C catalysts to boost the resistance to carbon corrosion and water flooding. This figure is quoted with permission from Wang et al.^[133]. (E) H₂-O₂ PEMFCs polarization and power density curves of the cells with Pt/C@Nafion and Pt/C@SDT-Nafion measured at 80 °C and 100% RH under 150 kPa. This figure is quoted with permission from Zhang et al.^[134]. (F, G) Schematic representation of ionomer distribution and gas transport at high and low RH in standard and nanofiber electrode layers. These figures are quoted with permission from Kabir et al.^[135]. (H) Schematic of the improvement of ORR activity caused by the enriched protons. This figure is quoted with permission from Liu et al.^[136].

The electrocatalytic reaction mainly occurs in the oxygen-water-catalyst three-phase microenvironment. The prevailing strategies for water management include increasing the density of exposed active sites and optimizing the three-phase environment by tailoring the catalyst and Nafion. Modification on carbon matrix surfaces such as -OH, -COOH (hydrophilic functional groups), or heteroatom could change the electronic structure and achieve high intrinsic activity^[131]. However, the introduction of hydrophilic function groups can prolong the water retention time in micropores or even block micropores, consequently resulting in high mass transfer resistance and carbon corrosion^[132]. If the carbon surface is modified with hydrophobic functional groups, hydrophobic channels can be provided to improve the durability of the catalyst. For example, Wang et al. modified the Fe-N-C catalysts with a hydrophobic group (Ar-CF₃) [Figure 15D]^[133]. Fe-N-C-F catalysts showed a high angle of 165° and kept not-wetting in the contact angles test, while for the pristine Fe-N-C catalysts, the droplet was absorbed too quickly to measure the contact angle. The stability in H₂-O₂ PEMFC was assessed by potentiostatic tests, which showed that the decay of Fe-N-C-F was less than that of pristine Fe-N-C, with only a 15% decay in initial performance after 100 h at 0.6 V. After 40 h of operation at 0.5 V, the content change of Fe was measured by UV-Vis spectrophotometry. The pristine Fe-N-C changed greatly (from 1.8% to 0.38%), while only a slight change of Fe content for Fe-N-C-F (from 1.2% to 1.1%) was observed. The higher hydrophobicity of carbon matrix led to less demetallation of Fe and enhanced stability of catalysts. After introducing the hydrophobic group (Ar-CF₃), the catalyst exhibited high hydrophobicity and enhanced carbon corrosion resistance. The tailoring of the catalyst and Nafion by incorporating hydrophobic functional groups or pore-forming additives can improve gas transport capability. Optimizing the three-phase environment by tailoring the Nafion was also an important strategy for water management. Zhang et al. incorporated a sulfonic acid functionalized covalent organic framework (COF) nanosheets with different pore sizes into Nafion, which facilitated the formation of the three-phase interface and improved the mass transfer of O₂^[134]. The ultra-high peak power density after water management (2.21 Wcm^-2) is 1.4 times higher compared to the performance of the MEA without COF [Figure 15E]. The durability of the fuel cell was also improved significantly, with a low rate of reduction (38%) in mass activity after 30,000 cycles AST, even meeting the DOE’s 2025 target (40%). Electrospinning techniques can also be used to promote the mass transport of O₂ and protons. Kabir et al. prepared the nanofiber electrode by electrospinning the mixture of Fe-N-C catalyst, Nafion, and Poly Acrylic acid^[135]. As shown in Figures 15F and 15G, nanofiber electrodes provided more mass transfer channels than standard electrodes, especially in test conditions with high relative humidity. Our group designed Czif-8-Fe-xPANI catalysts modified with polyaniline, which created a protophilic environment on the catalyst surface^[136]. Polyaniline can provide enough amine groups to facilitate the release of protons from the water cluster [Figure 15H] and improve the ORR performance, with the half-wave potential of the catalyst increased from 0.815 V to 0.835 V.

In conclusion, water management is an important method of catalytic performance improvement, especially in MEAs.