An Fe–N–C catalyst with hierarchical hollow-interconnected porous architectures was fabricated via the self-assembly of TEMPO-modified carboxylated CNCs (T-CNCs) under the dual-confinement effect of wood framework and Fe³⁺ ions, followed by a calcination process, as illustrated in Fig. 1A. To enhance the coordination ability of CNCs with Fe³⁺ ions, a 2-step functionalization strategy was implemented to introduce –COOH groups onto the sulfuric acid-hydrolyzed CNCs (S-CNCs) [20]. This process involved hydrochloric acid-mediated surface activation at 80 °C, followed by TEMPO-mediated oxidation, as depicted in Fig. S1a. The x-ray diffraction (XRD) patterns of S-CNCs and T-CNCs (Fig. S1b and c) exhibit 3 peaks at 14.8°, 16.1°, and 22.8°, which correspond to the \left(1\overline{1}0\right), (110), and (020) crystallographic planes [21], respectively. Compared with that in S-CNCs, the reduced intensity of the diffraction peak at 16.1° in T-CNCs indicates that the crystalline structure of S-CNCs is perturbed by carboxylation treatment [20]. Fourier transform infrared spectroscopy (Fig. S1d) provides further evidence for the successful incorporation of –COOH. Relative to S-CNCs, T-CNCs show a diminished intensity of the –OH peak at 3,328 cm⁻¹, which arises from the replacement of partial –OH groups with –COOH groups. Concomitantly, a new peak emerges at 1,731 cm⁻¹, corresponding to the –C=O stretching vibration of –COOH, providing direct evidence for the oxidation of –OH groups. Notably, both T-CNCs and S-CNCs exhibit a peak at 812 cm⁻¹, corresponding to the C–O–S stretching vibration of –OSO₃H groups [22]. This result confirms that the –OSO₃H groups remain intact on the T-CNC surface after the 2-step functionalization, providing a S source for subsequent in situ S doping. Transmission electron microscopy (TEM) images (Fig. 1B and Fig. S2) show that T-CNCs and S-CNCs both maintain good dispersion in aqueous solution, presenting as rodlike nanostructures with uniform dimensions. The well-dispersed nature of CNCs is critical for their subsequent self-assembly, as it ensures sufficient interaction with surrounding coordination sites (e.g., Fe³⁺ ions and wood cell wall functional groups) to form a homogeneous and stable network structure. Subsequently, the wood framework impregnated with T-CNCs was transferred to an ethanol solution containing Fe³⁺ ions. As Fe³⁺ ions diffuse through the natural hierarchical channels of the wood, the dual-confinement-driven self-assembly of T-CNCs is initiated, and the wood framework provides spatial confinement to guide the macroscopic arrangement of T-CNCs, while Fe³⁺ ions act as coordination cross-linkers to mediate the microscopic assembly of T-CNCs via electrostatic interactions and coordinate bonds (e.g., C–O···Fe³⁺). To eliminate residual unanchored Fe³⁺ ions and avoid structural damage caused by ice crystal growth during freeze-drying, the ethanol within the wood framework was replaced with a tert-butanol aqueous solution [23], yielding the intermediate composite W-TCNC-Fe. The W-TCNC-Fe composite was first subjected to low-temperature carbonization at 300 °C. This step serves 2 key purposes, which are stabilizing the layered porous network formed by T-CNCs within the wood pores by promoting the partial carbonization of cellulose and the consolidation of hydrogen bonds between T-CNCs and wood cell walls, and facilitating the homogeneous mixing of the carbonaceous matrix with NH₄Cl (a nitrogen source) during subsequent cogrinding to lay the foundation for uniform N doping. Finally, high-temperature carbonization at 1,000 °C was performed to fabricate the Fe–N₅ catalyst. During this process, NH₄Cl decomposes to release NH₃ and HCl gases. NH₃ acts as a nitrogen source to dope N into the carbon network (forming pyridinic N, pyrrolic N, and graphitic N), while HCl etches the carbon frameworks to generate abundant micropores. These micropores function as anchoring sites for Fe atoms, effectively suppressing Fe agglomeration and enabling the generation of atomically dispersed Fe sites [24]. To decouple the contributions of each component (wood framework, CNCs, Fe³⁺ ions, and N source) to the hierarchical porous structure and the microenvironment of single-atom Fe, a series of control samples were prepared, which were N@TCW, Fe@TCW, Fe@CW, Fe@W, Fe@H80, Fe@H100, and Fe–N(100). Specifically, N@TCW was synthesized without adding Fe³⁺ ions (i.e., no Fe source). Fe@TCW was synthesized without adding NH₄Cl (i.e., no N source). Fe@CW was synthesized using S-CNCs instead of T-CNCs. Fe@W was synthesized by directly introducing Fe³⁺ ions and NH₄Cl into the wood framework (without CNCs). Fe@H80 and Fe@H100 were synthesized by replacing T-CNCs with CNC-H80 and CNC-H100, respectively. Fe–N(100) was synthesized using CNC-100 instead of T-CNCs.

Figure S3 shows the cross-sectional architecture of delignified balsa wood. Its inherent hierarchical honeycomb-like pores act as a 3D spatial template to create a confined microenvironment that guides the self-assembly of T-CNCs. To clarify the role of dual confinement from the wood framework and Fe³⁺ ions in structure formation, control experiments on T-CNCs’ self-assembly were conducted under different conditions. Without both the wood framework and Fe³⁺ ions, T-CNCs self-assemble into a dense layered structure with no observable pores (Fig. S4). This outcome stems from random stacking driven by unregulated intermolecular hydrogen bonding. With only the wood framework and no Fe³⁺ ions, T-CNCs form a loose fibrous network (Fig. S5) but lack Fe³⁺-mediated cross-linking, which leads to poor structural integrity and irregular pore distribution. With only Fe³⁺ ions and no wood framework, T-CNCs aggregate into a compact structure without large-scale pores (Fig. S6). This is because Fe³⁺ coordination alone causes excessive cross-linking without spatial guidance. These results confirm the indispensable role of the wood framework in directing T-CNCs’ self-assembly into a porous network. Specifically, under the dual-confinement effect, the wood framework provides spatial guidance and Fe³⁺ ions act as coordination cross-linkers, enabling T-CNCs to assemble into a uniformly distributed and ordered porous architecture (Fig. 1C). Abundant hydrogen bonds also form between the –OH groups on wood cell walls and T-CNCs’ functional groups including –OH, –COOH, and –OSO₃H. These bonds reinforce the structural stability of the T-CNC network and preserve its porous structure during subsequent high-temperature carbonization.

After carbonization at 1,000 °C, the Fe–N₅ catalyst retains the hierarchical 3D interconnected porous architecture (Fig. 1D), a key feature for ORR performance. In contrast, Fe@W, which is prepared by directly loading Fe³⁺ ions onto wood without T-CNCs, has no observable porous structure (Fig. S7). This confirms the necessity of T-CNCs for constructing the porous carbon matrix. Notably, Fe@TCW synthesized without the N precursor NH₄Cl still forms abundant pores (Fig. S8a). These pores are derived from the preformed porous architecture of W-TCNC-Fe, which is stabilized by the dual-confinement effect of the wood framework and Fe³⁺. This indicates that the porous structure is primarily determined by the dual-confinement effect of the wood framework and Fe³⁺ rather than N doping. In stark contrast, N@TCW prepared without Fe³⁺ ions shows no discernible porosity (Fig. S8b). Without Fe³⁺-mediated cross-linking, T-CNCs fail to form a stable porous network within the wood framework and collapse during carbonization. Fe@CW synthesized using S-CNCs instead of T-CNCs displays a stacked and collapsed porous network (Fig. S9). This phenomenon arises from the weaker coordination ability of –OSO₃H groups on S-CNCs toward Fe³⁺ ions compared to –COOH groups on T-CNCs. Inductively coupled plasma mass spectrometry (ICP-MS) results (Table S1) support this observation. The Fe content of Fe–N₅ is 0.61%, approximately twice that of Fe@CW, which is 0.33%. This confirms that –COOH groups enable stronger Fe³⁺ ion coordination and higher Fe loading. Furthermore, pore structure of catalysts is then revealed by N₂ adsorption–desorption isotherms and pore size distribution. As shown in Fig. 1E, all catalysts, including Fe–N₅, Fe@CW, and Fe@W, show sharp increases in the low-relative-pressure region, indicating the existence of microporous structures that favor the anchoring of Fe single atoms. Different from Fe@W, both Fe–N₅ and Fe@CW exhibit representative type IV isotherms with H₄ hysteresis loops in the high-relative-pressure range, demonstrating the simultaneous presence of mesoporous channels. The pore size distribution provided by the Brunauer–Emmett–Teller and mercury intrusion porosimetry methods (Fig. 1F and Fig. S10) further demonstrates that Fe–N₅ possesses a hierarchical pore structure with micropores (0.46 to 3.00 nm), mesopores (5.00 to 80.00 nm), and macropores (121.00 nm). Consequently, this distinctive Fe–N₅ catalyst featuring micropores and mesopores that densely penetrate the carbon matrix rich in macropores is constructed, which could not only maximize the exposure of actives sites but also ensure efficient mass transport and electron transfer at the 3D interfaces [24].

The XRD patterns of Fe–N₅, Fe@TCW, Fe@CW, and Fe@W exhibit 2 broad peaks at 25° and 44° (Fig. S11), corresponding to the (002) and (101) crystal planes of graphitized carbon, respectively. No metallic Fe or Fe compounds in Fe–N₅, Fe@CW, and Fe@W are observed. In contrast, Fe@TCW displays a distinct metallic Fe peak near 44°, demonstrating that the atomic-level Fe dispersion requires a coordination structure formed with N atoms that are derived from NH₄Cl. High-resolution TEM (Fig. 1H) further confirms no Fe nanoparticles in Fe–N₅. The lattice fringe is estimated to be 0.34 nm, which corresponds to the (002) planes of graphitic carbon, in accordance with XRD results. As shown in Fig. S12, no obvious iron nanoparticles are observed in Fe@CW and Fe@W.

High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) and corresponding energy-dispersive x-ray spectroscopy (EDX) elemental mapping (Fig. 1G and Figs. S13 and S14) also indicate that Fe atoms in Fe–N₅, Fe@CW, and Fe@W are uniformly dispersed within the N, S-codoped carbon matrix without agglomeration. The aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (AC HAADF-STEM) images (Fig. 1I) provide direct visualization of atomically dispersed Fe as bright spots. The local magnified view (Fig. 1J) shows that Fe atoms are localized at the edges of micropores within the carbon matrix, which might modulate the charge density distribution and electronic structures of active centers, thus optimizing intermediates’ adsorption configuration [25]. The disordered structure is observed by Raman spectroscopy (Fig. S15). The D band at 1,350 cm⁻¹ is attributed to sp³-hybridized defective carbon, and the G band at 1,580 cm⁻¹ is attributed to sp²-hybridized graphitic carbon. Fe–N₅ exhibits an I_D/I_G ratio of 1.04, which is similar to that of Fe@TCW (1.05) but significantly higher than that of N@TCW (1.01). This higher ratio indicates a greater density of carbon defects in the T-CNC-Fe³⁺ coordination network. Electron paramagnetic resonance spectra (Fig. 1K) show a characteristic signal at g = 2.005 for Fe–N₅, Fe@CW, and Fe@W, indicating structural defects in all catalysts. Notably, Fe–N₅ exhibits a much higher signal intensity than Fe@CW and Fe@W, confirming more defect sites [26].

The chemical composition and oxidation states of Fe–N₅ and related samples were investigated by high-resolution x-ray photoelectron spectroscopy (XPS). The high-resolution N 1s spectra (Fig. S16a) of Fe–N₅, Fe@CW, and Fe@W reveal 5 peaks centered at 398.28, 399.26, 400.50, 401.56, and 403.40 eV, which are attributed to Fe–N, pyridinic N, pyrrolic N, graphitic N, and oxidized N, respectively. As shown in Fig. S16b, oxidized N remains relatively constant across the 3 samples. Compared with Fe@CW and Fe@W, Fe–N₅ exhibits increased graphitic N, of which graphitic N facilitates electron transfer within the carbon framework. Compared to Fe@W, both Fe–N₅ and Fe@CW exhibit increased Fe–N content and reduced proportions of pyridinic N and pyrrolic N. This arises from the increased amount of Fe due to the coordination of functional groups on CNCs with Fe³⁺ ions, which requires more pyridinic N and pyrrolic N to generate isolated Fe–N_x sites. Moreover, the carboxylation of S-CNCs further facilitates the involvement of both pyridinic N and pyrrolic N in the coordination with Fe atoms. The S 2p spectrum (Fig. S17) of Fe–N₅ exhibits 2 distinct peaks at 164.29 and 165.54 eV, assigned to thiophene-like C–S–C species (2p_3/2 and 2p_1/2), along with weak peaks at 168.53 and 169.59 eV, corresponding to oxidized –SO_x⁻ species. Compared with Fe@CW, no signal of Fe–S bonds is detected in Fe–N₅, confirming the absence of Fe–S coordination. Furthermore, both the high-resolution N 1s and S 2p XPS spectra of the Fe–N₅ moiety exhibit additional characteristic peaks, which are assigned to N–S bonding species. This spectral feature provides direct evidence that the S atom is spatially adjacent to the N atom in the carbon matrix, an arrangement that originates from the in situ doping of S (derived from the residual –OSO₃H groups of T-CNCs) and its subsequent interaction with N-coordinated Fe sites during high-temperature carbonization. The characteristic peaks in these spectra correspond to S species bonded to carbon, further validating the in situ sulfur doping process derived from residual –OSO₃H groups on T-CNCs. Meanwhile, O 1s XPS further confirms the absence of Fe–O bonding in Fe–N₅ (Fig. S18). Fe 2p XPS spectra reveal that the 2p_3/2 peaks of Fe–N₅, Fe@CW, and Fe@W are located at 709.29 eV (Fe²⁺) and 712.10 eV (Fe³⁺), 710.36 eV (Fe²⁺) and 713.40 eV (Fe³⁺), and 709.68 eV (Fe²⁺) and 713.09 eV (Fe³⁺), respectively, accompanied by satellite peaks—these features indicate the partially oxidized state of Fe (Fig. S19). Compared with Fe@CW, no characteristic peak corresponding to Fe–S species is detected in Fe–N₅, demonstrating that TEMPO-mediated carboxylation effectively suppresses the formation of Fe–S coordination structures. Notably, the Fe 2p_3/2 peak of Fe–N₅ exhibits the lowest binding energy compared to those of Fe@CW and Fe@W. This observation suggests that the coordination structure of Fe–N₅ facilitates greater electron accumulation around the Fe centers, which favors oxygen intermediate uptake and release throughout the ORR process, thus optimizing catalytic kinetics. Furthermore, peaks indicative of metallic Fe-based species are absent in Fe–N₅, evidencing the successful formation of atomically dispersed Fe sites within the carbon matrix.

The ORR activities of all samples were evaluated in an O₂-saturated 0.1 M KOH electrolyte. First, the cyclic voltammetry (CV) curves of the catalysts (Fig. S20) exhibit distinct oxygen reduction peaks. Among these, Fe–N₅ shows the most positive cathodic peak potential at 0.938 V, a key indicator of its superior ORR performance relative to those of the other catalysts. The linear sweep voltammetry curves (Fig. 2A and Fig. S21a) further confirm the outstanding ORR activity of Fe–N₅, displaying a half-wave potential (E_1/2) of 0.964 V and an onset potential (E_onset) of 1.110 V, which surpass those of commercial Pt/C (E_1/2 of 0.884 V and E_onset of 1.040 V), Fe@CW (0.925 and 1.080 V), Fe@W (0.898 and 1.080 V), and Fe–N(100) (0.940 and 1.080 V). Notably, the superior performance of Fe–N₅ over Fe–N(100) validates the positive contribution of in situ S doping toward ORR activity. Fe–N(100) is derived from the confined self-assembly of CNC-100 without –SO₃H moieties (Fig. S21b). Although CNC-100 contains abundant –COOH groups and achieves a higher Fe loading of 0.85% (Table S1), the lack of S doping renders its catalytic activity inferior to that of Fe–N₅. In addition, Fe@H80 exhibits a lower Fe loading of 0.45% than that of Fe@H100 (0.51%) (Table S1), while both catalysts deliver a comparable E_1/2 of 0.95 V (Fig. S21d). This phenomenon mainly originates from the much higher S content in Fe@H80 (0.85%) than that in Fe@H100 (0.02%), further indicating that S doping effectively enhances ORR activity by modulating the electronic structure of active sites. Furthermore, Fe–N(100) exhibits a markedly higher activity than Fe@W, highlighting the indispensable role of the hierarchical porous structure (Fig. S21c) fabricated via the dual-confinement strategy in mass transport and exposure of active sites. Notably, a comprehensive comparison with previously reported M–N–C catalysts (Fig. 2B and Table S9) reveals that the ORR performance of Fe–N₅ stands among the top tier of non-precious-metal catalysts. The kinetic current density (J_k) (Fig. 2C) of Fe–N₅ at 0.95 V reaches 6.90 mA cm⁻², which is 7.6 times that of Pt/C (0.91 mA cm⁻²) and significantly exceeds those of Fe@CW (1.39 mA cm⁻²) and Fe@W (1.27 mA cm⁻²). This demonstrates its outstanding catalytic performance and the superiority of the dual-confinement effect for establishing Fe–N–C active sites. The fast ORR kinetics of Fe–N₅ is further evidenced by its Tafel slope of 61.9 mV dec⁻¹ as shown in Fig. 2D, which is lower than those of Fe@CW (77.3 mV dec⁻¹), Fe@W (93.1 mV dec⁻¹), and Pt/C (76.5 mV dec⁻¹). Moreover, Fig. S22 illustrates that Fe–N₅ possesses the smallest electrochemical double-layer capacitance (C_dl) of 17.91 mF cm⁻², while Fe@CW shows the highest C_dl of 23.93 mF cm⁻², followed by Fe@W (19.32 mF cm⁻²). This result confirms that the electrochemical active surface area (ECSA) plays a nonprimary role in driving the improved ORR performance of Fe–N₅, pointing instead to the superior intrinsic activity of its Fe–N active sites.

Rotating ring-disk electrode (RRDE) measurements (Fig. 2E) demonstrate that the electron transfer number (n) of Fe–N₅ is 3.99 to 4.00, accompanied by an exceptionally low H₂O₂ yield of less than 0.5%, unambiguously demonstrating the high selectivity for the 4e⁻ ORR process. This is further confirmed by the Koutecký-Levich (K-L) plots in Fig. S23, of which n is calculated to be 4.02 over the potential range of 0.3 to 0.7 V. To evaluate the intrinsic activity of the catalysts, the mass activity (MA) and turnover frequency (TOF) were calculated. The MA values (Fig. 2F and Table S2) of Fe–N₅ at 0.85 and 0.90 V are 2.40 × 10⁴ and 1.40 × 10⁴ A g_Fe⁻¹, respectively, which are 184 and 259 times as much as those of Pt/C (130.00 and 54.00 A g_Pt⁻¹) and exceed those of Fe@CW (2.17 × 10⁴ and 7.50 × 10³ A g_Fe⁻¹) and Fe@W (1.89 × 10⁴ and 6.70 × 10³ A g_Fe⁻¹). The TOF values (Fig. 2G and Table S3) of Fe–N₅ at 0.85 and 0.90 V are 3.680 and 2.040 s⁻¹, respectively, corroborating the higher activity than those of Pt/C (0.053 and 0.022 s⁻¹), Fe@CW (3.150 and 1.090 s⁻¹), and Fe@W (2.750 and 0.980 s⁻¹). Electrochemical nitrite stripping (Fig. S24 and Table S4) is employed to determine the active site density and more precise TOF values [27]. The site density values for Fe–N₅, Fe@CW, and Fe@W are determined to be 1.42 × 10¹⁹, 7.84 × 10¹⁸, and 1.12 × 10¹⁹ sites g⁻¹, respectively. The TOF value of Fe–N₅ is estimated to be 3.04 e⁻ site⁻¹ s⁻¹, higher than those of Fe@CW (1.11 e⁻ site⁻¹ s⁻¹) and Fe@W (0.71 e⁻ site⁻¹ s⁻¹). These results collectively confirm that the unique Fe–N moiety, constructed via the dual-confinement effect of T-CNCs, is responsible for the enhanced ORR efficiency.

The stability of the electrocatalysts is assessed by continuous 10,000 CV cycles across a potential window of 0.165 to 1.165 V. After 10,000 CV cycles, Fe–N₅ exhibits negligible decay in both E_1/2 and limiting current density (Fig. 2H). In contrast, commercial Pt/C displays a 24-mV decay in E_1/2 after only 5,000 CV cycles (Fig. S25). Chronoamperometric (i–t) tests (400 rpm, O₂-saturated 0.1 M KOH) (Fig. S26) further validate the excellent durability of Fe–N₅, which retains 92.3% of its initial current after 10 h, superior to that of Pt/C (78.7%). The outstanding catalytic activity and stability of Fe–N₅ suggest that both pyridinic N, which contributes to superior stability, and pyrrolic N, which is responsible for enhanced catalytic activity, participate in the coordination of the central Fe atom [28]. Additionally, methanol tolerance tests (Fig. 2I) demonstrate the remarkable methanol tolerance of Fe–N₅, demonstrating the potential application for direct methanol fuel cells.

To decode the chemical state and local coordination environment of the central Fe site at the atomic scale, synchrotron-based XAS measurements were conducted. Figure 3A presents the Fe K-edge x-ray absorption near-edge structure (XANES) spectrum of Fe–N₅ alongside those of reference samples including Fe foil, FeS, FePc, FeO, and Fe₂O₃. FePc is a model compound with square-planar Fe–N₄ coordination. A characteristic pre-edge peak is observed at 7,114 eV for FePc, and this peak corresponds to the 1s → 4pz electronic transition. It serves as a spectral fingerprint sensitive to the local symmetry of the central Fe atom [29], providing a benchmark for comparing Fe coordination environments. For Fe–N₅, the pre-edge peak (7,116 eV) exhibits a slight rightward shift and a reduced intensity, indicating the presence of a distorted Fe–N configuration [30]. This distortion may originate from adjustments to the Fe coordination number, heteroatom doping such as S doping, or defects in the carbon matrix [31]. The absorption edge energy of Fe–N₅ is intermediate between those of FeO (Fe²⁺) and Fe₂O₃ (Fe³⁺). It is also closer to that of FePc, suggesting an intermediate Fe oxidation state between +2 and +3. Compared with that in FePc, the Fe species in the Fe–N₅ catalyst exhibits a lower oxidation state. This reduction in Fe oxidation state results from the enhanced electron-donating effect of the N, S-codoped carbon matrix, where the introduction of in situ S doping modulates the electron delocalization within the carbon framework. Specifically, S atoms, with a lower electronegativity than N, can alter the charge distribution at Fe–N₅ active sites, promoting electron transfer away from the carbon matrix toward the central Fe atoms and thus lowering the Fe oxidation state [32]. This electron-enriched Fe environment is critical for optimizing the adsorption strength of oxygen-containing intermediates during the ORR, ultimately boosting catalytic kinetics. Quantitative fitting of Fe K-edge XANES by comparing with standard samples (FeO and Fe₂O₃) reveals an average oxidation state of approximately +2.6 for Fe–N₅ (inset of Fig. 3A), which coincides with XPS results (Fig. S19). The k²-weighted Fourier transform spectra from extended x-ray absorption fine structure (EXAFS) spectroscopy (Fig. 3B) further clarify a dominant peak of Fe–N₅ at 1.53 Å, slightly higher than that of FePc, which originates from the first-shell Fe–N scattering path. This observation suggests the possible presence of axial N coordination in Fe–N₅, which elongates the Fe–N bond length within the Fe–N₅ plane. The absence of Fe–S (1.87 Å) and Fe–Fe (2.20 Å) peaks can be observed, confirming the atomic dispersion of Fe, consistent with XRD (Fig. S11a) and HAADF-STEM (Fig. 1I) results. Quantitative fitting of the EXAFS spectra in R and k spaces gives an R-factor of only 0.010 with excellent fitting quality and indicates a first-shell Fe–N coordination number of 5.30 ± 0.2 with an average Fe–N distance of 2.03 Å, suggesting Fe–N₅ coordination (Fig. S27 and Table S5), which is distinctly different from the 4-coordinate Fe–N bond length (1.94 Å) in FePc. Combined with the pre-edge XANES features, the axial pyrrolic N coordinated to 4 pyridinic N in FeN₅ moieties is inferred, which coincides with the well-fitting results of K-edge EXAFS with the proposed model in both R and k spaces (Fig. 3C and D). The wavelet transform contour maps of Fe–N₅ reveal a peak at 3.60 Å⁻¹ (Fig. 3H), assigned to the Fe–N scattering path. While Fe foil exhibits a maximum intensity at 7.60 Å⁻¹ (Fig. 3E), aligning with the Fe–Fe bond, Fe₂O₃, FePc, and FeS show maxima at 3.30 Å⁻¹ (Fe–O) (Fig. S28), 3.40 Å⁻¹ (Fe–N) (Fig. 3F), and 4.75 Å⁻¹ (Fe–S) (Fig. 3G), respectively. These results conclusively corroborate that the atomically dispersed iron sites reside in a unique Fe–N₅ coordination configuration.

To elucidate the atomic-level mechanism underlying the superior activity of the Fe–N₅ catalyst, density functional theory (DFT) calculations were conducted. Multiscale characterizations including scanning electron microscopy, HAADF imaging, Raman spectroscopy, XPS, and XAS reveal that Fe–N₅ possesses a hierarchically porous structure, with Fe atomic centers situated in a 5-coordination environment at the micropore edges, along with the presence of S doping. To systematically investigate the effect of structural geometry on the electronic properties and catalytic behavior of Fe active sites, 4 distinct configurations of Fe-coordinated moieties were constructed on the first carbon layer (denoted as layers 1a to 1d), with the classic 4-fold pyridinic N-coordinated Fe moiety (Fe–N₄) embedded in a pristine carbon network (layer 1a) serving as the reference system (Fig. S29). Layer 1b features 4 pyridinic N atoms coordinated to a central Fe atom, with 1 S atom incorporated into the same 5-membered ring as the Fe center (Fig. S29). For layer 1c, the S atom is positioned within the 6-membered ring that contains the central Fe atom (Fig. S29), differing from layer 1b in the cyclic environment of the doped S. Layer 1d is derived from layer 1b by introducing an adjacent micropore into the first carbon network, with the pore localized near the S atom (hereafter referred to as the S1 site) to mimic the hierarchical porous feature of the catalyst (Fig. S29). Correspondingly, 3 configurations are designed for the second carbon layer (light gray, Fig. S29) to explore the influence of axial coordination on Fe sites. Layer 2e shows a pristine carbon network without additional heteroatoms, used to replicate the nonaxial coordination environment (Fig. S29). Layer 2f exhibits a carbon network with an axial pyridinic N atom that coordinates to the central Fe, enabling the formation of an Fe–N₅ configuration (Fig. S29). Meanwhile, layer 2g is similar to layer 2f, but with an axial pyrrolic N atom instead of pyridinic N, to compare the effect of axial N species on Fe coordination (Fig. S29). These first and second carbon layer configurations are further combined to construct 5 representative models for subsequent analysis. For model A, it represents the traditional Fe–N₄ coordination structure, which is composed of layer 1a and layer 2e (Fig. 4A). Model B combines layer 1b and layer 2f, as presented in Fig. 4B. For model C, it is formed by layer 1b and layer 2g (Fig. 4C). Model D consists of layer 1c and layer 2g, as shown in Fig. 4D. Model E is composed of layer 1a and layer 2g, as shown in Fig. S30a. For the target configuration, that is, the Fe–N₅ model, it is constructed by layer 1d and layer 2g (Fig. 4E).

Figure 4F demonstrates a strong correlation between Fe site geometry and catalytic behavior. The 4e⁻ ORR process primarily involves the adsorption of 3 key intermediates: *OOH, *O, and *OH. At U = 1.23 V (Fig. 4F, Fig. S30b, and Table S6), the last step of ORR (*OH desorption) is the most endothermic on all catalyst models and thus is the reaction-determining step (RDS). The Fe–N₅ model shows the lowest energy barrier of 0.31 eV among the catalysts, that is, model A (0.96 eV), model B (0.76 eV), model C (0.42 eV), model D (0.73 eV), and model E (0.89 eV). In contrast to model C, Fe–N₅ shows a reduced energy barrier, confirming that micropores optimize intermediate adsorption via electronic structure regulation. Compared with model A, model E possesses a lower energy barrier, indicating that axial pyrrolic N coordination regulates the electronic structure of Fe sites and weakens oxygen-containing intermediate adsorption. Model C delivers a far lower barrier than model E, revealing strong electronic synergy of S doping with the FeN₅ framework to tune the active site electronic structure and boost intrinsic catalytic activity. As indicated in Fig. 4G, at U = 0 V, all ORR steps from O₂ to OH⁻ are spontaneous exothermic on Fe–N₅. The significant difference between the free energy for *OOH → *O (1.9 eV) and the H₂O₂ formation pathway (0.61 eV) further confirms the preferential selection of the 4e⁻ pathway on Fe–N₅ (Fig. 4J) and effective suppression of by-product generation (Table S7). Bond length analysis (Figs. S31, S32, and S34 and Table S8) provides insight into the stability of Fe–N coordination. The Fe–N₅ bond length in model B with axial pyridinic N coordination is measured as 2.40 Å, significantly longer than that in pyrrolic N-coordinated Fe (model C, 2.16 Å, and Fe–N₅, 2.12 Å), indicating the superior stability of the Fe–N₅ bond in the catalysts with axial pyrrolic N coordination.

As revealed by the differential charge density plot (Fig. 4H), it shows large nonuniform charge distribution at the Fe sites on Fe–N₅. Compared to model A, S doping in Fe–N₅ is beneficial for adsorbing electrons [31] and donating electrons to Fe, resulting in a lower oxidation state of Fe, which is consistent with the XAS and XPS results. Meanwhile, compared to model C, the existence of a micropore near the S₁ site slightly decreases the charge density at the Fe sites, which may be due to the accelerated shutting of electrons by micropores [24]. The d-band center theory is employed to analyze the binding affinity of Fe sites for oxygen intermediates as shown in Fig. 4I. Compared to model A (−0.89 eV), the d-band centers of Fe–N₅, model B, and model C shift to −1.23, −0.92, and −1.35 eV, respectively. A downshift in the d-band center weakens the binding energy between the Fe site and *OH, thereby promoting *OH desorption. Notably, the introduction of a micropore near the S₁ site when comparing Fe–N₅ with model C induces a slight rightward shift of the d-band center. This shift balances the adsorption/desorption strength of oxygen intermediates at the Fe site. It avoids overly strong *OH adsorption, which would hinder desorption or overly weak adsorption, which would impair intermediate activation. Collectively, these DFT results demonstrate that the unique configuration of Fe–N₅ contributes to its excellent ORR performance. This configuration consists of an Fe–N₅ moiety with axial pyrrolic N coordinated to 4 equatorial pyridinic N, a S atom in the adjacent 5-membered ring, and a nearby micropore. This structure optimizes the energy band structure of Fe centers and accelerates the desorption of the RDS *OH intermediate. This synergy of structural features ultimately boosts ORR electrocatalytic kinetics.

Leveraging the excellent ORR properties of Fe–N₅, a proof-of-concept ZAB experiment was conducted to demonstrate its feasibility as an air cathode (Fig. 5A and Fig. S35), of which a zinc foil was used as the anode and 6.0 M KOH + 0.2 M Zn(Ac)₂ was applied as the electrolyte. The Fe–N₅-based ZAB exhibits a high OCV of 1.51 V (Fig. 5B), a peak power density of 150.0 mW cm⁻² at 228.0 mA cm⁻² (Fig. 5C), and a specific capacity of 830.0 mAh g⁻¹ at 5 mA cm⁻² (Fig. 5D), which are higher than those of commercial Pt/C (1.44 V; 128.0 mW cm⁻² at 212.0 mA cm⁻²; 794.5 mAh g⁻¹). Additionally, the Fe–N₅-based ZAB shows high stability with a minimal voltage fluctuation of 10 mV even when it fluctuates from 5 to 100 mA cm⁻² (Fig. 5E), highlighting its superior rate capability. The Fe–N₅-based ZAB exhibits only a 3.8% voltage loss after 300 h of continuous discharge at 5 mA cm⁻² (Fig. 5F), exceeding the performance of the state-of-the-art ZAB. Of greater significance, the ZAB fabricated using Fe–N₅ + RuO₂ achieves ultrastable cycling (>250 h), demonstrating the viability of Fe–N₅ for practical device implementation (Fig. 5G), which can be attributed to the strong FeN₅ coordination configuration within Fe–N₅ constructed by the dual-confinement effect of T-CNCs, suppressing Fe dissolution. Moreover, the Fe–N₅-based ZAB retains round-trip efficiencies of 59.11% and 55.53% after 50 and 250 h, respectively, with an initial round-trip efficiency of 61.42%. This performance surpasses that of the ZAB fabricated with commercial Pt/C, which exhibits a more pronounced efficiency decay during operation, decreasing from an initial value of 65.28% to 58.76% at 50 h and further plummeting to 55.55% after only 100 h of cycling (Fig. S36), unequivocally verifying the exceptional long-term durability of the Fe–N₅ catalyst and its substantial advantage over commercial Pt/C for practical energy storage applications.

Degreasing cotton was purchased from Alibaba Taobao (Hangzhou, China). Basswood woodblocks were purchased from Alibaba Taobao (Hangzhou, China). Sodium chlorite (NaClO₂), acetic acid (CH₃COOH), ethanol, sulfuric acid (H₂SO₄, 98%), iron(III) nitrate nonahydrate (Fe(NO₃)₃·9H₂O), ammonium chloride (NH₄Cl), tertiary butanol, sodium hydroxide (NaOH), sodium chloride (NaCl), concentrated hydrochloric acid (HCl, 36% to 38%), TEMPO, sodium hypochlorite (NaClO), and sodium bromide (NaBr) were purchased from Rhawn. All reagents were of analytical grade and used without further purification. A Nafion 117 membrane was purchased from Dupont.

CNCs obtained via sulfuric acid hydrolysis were used as the pristine material and denoted as S-CNCs. TEMPO-mediated oxidation was employed to synthesize carboxylated CNCs (T-CNCs), following a 2-step functionalization protocol. In the first step, surface activation of S-CNCs was performed. An aqueous suspension of S-CNCs (0.5 wt%) was mixed with hydrochloric acid (HCl, 50.0 mM) and magnetically stirred at 80 °C for 24 h. After cooling to room temperature, the suspension was partially neutralized to a pH of approximately 4.0 using sodium hydroxide (NaOH) solution. The activated S-CNCs were then purified by dialysis against deionized water to remove residual electrolytes, followed by concentration via ultrafiltration to adjust the solid content for subsequent reactions. The second step involved TEMPO-mediated oxidation of the activated S-CNCs. To the concentrated S-CNC suspension (1.0 wt% in deionized water), TEMPO (0.1 g per gram of dry S-CNCs), sodium hypochlorite (NaClO, 1.9 g per gram of dry S-CNCs), and sodium bromide (NaBr, 1.0 g per gram of dry S-CNCs) were sequentially added under stirring until complete dissolution. The reaction mixture was stirred at room temperature for 4 h, during which the pH was maintained between 10.0 and 11.0 by dropwise addition of 1.0 M NaOH. This pH range was sustained until no further pH drift was observed, indicating the completion of oxidation. Subsequently, NaCl was added to flocculate the T-CNCs, and the product was washed by centrifugation and decanting (3 times with 0.5 M NaCl, 2 times with 0.1 M HCl, and once with deionized water) before final purification by dialysis.

CNCs-H80: S-CNCs were activated with 50 mM hydrochloric acid at 80 °C for 24 h, without TEMPO oxidation; all other preparation procedures are identical to those of T-CNCs. CNCs-H100: S-CNCs were activated with 250 mM hydrochloric acid at 100 °C for 48 h, without TEMPO oxidation; all other preparation procedures are identical to those of T-CNCs. CNC-100: S-CNCs were activated with 250 mM hydrochloric acid at 100 °C for 48 h, followed by TEMPO oxidation. All other preparation procedures were consistent with those of T-CNCs.

Balsa wood with dimensions of 20 mm × 20 mm × 3 mm (tangential × radial × axial) was treated with 2.0 wt% NaClO₂ at 70 °C, maintaining a pH of 4.6 for 18 h, followed by washing with deionized water to neutrality. It was then freeze-dried to obtain delignified balsa wood.

Two pieces of the delignified balsa wood were impregnated in a 0.9 wt% T-CNC aqueous solution for 30 min followed by ultrasonication for 100 min. The T-CNC-impregnated wood and residual solution were collectively stored in a refrigeration chamber at 8 °C overnight. Subsequently, the T-CNC-loaded scaffolds were transferred into 25.0 ml of Fe(NO₃)₃ (0.1 M) ethanolic solution and maintained at 8 °C for 3.5 h. The treated samples were then immersed in a 20.0 wt% tert-butanol aqueous solution at 8 °C for 6 h. Finally, freeze-drying yielded porous architectures within the wood frameworks. The obtained product was designated as W-TCNC-Fe.

The Fe–N₅ precursor (W-TCNC-Fe) was pre-treated under 300 °C for 2 h in a N₂ atmosphere. The resulting material was then thoroughly ground with NH₄Cl at a 20-fold mass ratio to form homogeneous powder, followed by immersion in 20.0 ml of deionized water and magnetically stirred at 60 °C for 24 h until complete dryness. Subsequently, the dried powder was directly heated to the target temperature (1,000 °C) at a heating rate of 5 °C min⁻¹ and kept for 4 h under a flow of N₂ in a tube furnace. At last, the Fe–N₅ catalyst was successfully synthesized, featuring hollow-interconnected hierarchical porous architectures anchored on wood cell walls.

The synthesis procedures described above are as follows:

Fe@CW was prepared by immersing 2 delignified balsa wood pieces into 0.9 wt% S-CNC solution, followed by identical experimental steps as those for Fe–N₅. N@TCW was synthesized by immersing the T-CNC-impregnated delignified balsa wood in ethanol (without the addition of Fe(NO₃)₃), with the remaining steps identical to those of Fe–N₅. Fe@TCW was obtained by carbonizing the Fe–N₅ precursor, followed by direct grinding without mixing with NH₄Cl, while maintaining the same synthesis parameters as those for Fe–N₅. Fe@W was synthesized by directly impregnating Fe(NO₃)₃ into the wood framework (without the T-CNC impregnation step in the delignified balsa wood), with all other procedures identical to those of Fe–N₅. Fe@H80 was prepared by immersing 2 delignified balsa wood pieces into 0.9 wt% CNC-H80 solution, followed by identical experimental steps as those for Fe–N₅. Fe@H100 was prepared by immersing 2 delignified balsa wood pieces into 0.9 wt% CNC-H100 solution, followed by identical experimental steps as those for Fe–N₅. Fe–N(100) was prepared by immersing 2 delignified balsa wood pieces into 0.9 wt% CNC-100 solution, followed by identical experimental steps as those for Fe–N₅. Moreover, the entire experimental process eliminated the tedious acid washing steps typically required in the preparation of single-atom Fe–N–C catalysts.

All calculations were performed using DFT as implemented in the Vienna Ab initio Simulation Package with the Perdew–Burke–Ernzerhof exchange-correlation functional [2]. The electron–ion interaction was addressed using the projector-augmented-wave method and a plane wave with a cutoff energy of 450 eV. Structural models were optimized until Hellmann–Feynman forces converged below 0.02 eV/Å and energy changes were less than 10⁻⁵ eV. Dispersion interactions between atoms in adsorption models were accounted for using Grimme’s DFT-D3 method. A gamma-centered k-point mesh of 2 × 2 × 1 was adopted. A vacuum space of 12 Å was imposed along the z-direction.