The Challenge of Achieving a High Density of Fe-Based Active Sites in a Highly Graphitic Carbon Matrix

As one of the most promising platinum group metal-free (PGM-free) catalysts for oxygen reduction reaction (ORR), Fe–N–C catalysts with a high density of FeNx moieties integrated in a highly graphitic carbon matrix with a proper porous structure have attracted extensive attention to combine the high activity, high stability and high accessibility of active sites. Herein, we investigated a ZnCl2/NaCl eutectic salts-assisted ionothermal carbonization method (ICM) to synthesize Fe–N–C catalysts with tailored porous structure, high specific surface area and a high degree of graphitization. However, it was found to be challenging to anchor a high density of FeNx sites onto highly graphitized carbon. Iron precursors with preexisting Fe–N coordination were required to form FeNx sites in the nitrogen-doped carbon with a high degree of graphitization, while individual Fe and N precursors led to a Fe–N–C catalyst with poor-ORR activity. This provides valuable insights into the synthesis-structure relationship. Moreover, the FeNx moieties were identified as the major active sites in acidic conditions, while both FeNx sites and Fe2O3 were found to be active in alkaline medium.


Introduction
The oxygen reduction reaction (ORR) is a cornerstone of various electrochemical energy storage and conversion devices [1][2][3], including metal-air batteries and fuel cells. Proton exchange membrane fuel cell (PEMFC) is a key technology to provide clean and sustainable energy conversion with potential applications in portable electronics, electric vehicles, and residential power generation. Recently, hydroxide exchange membrane fuel cells (HEMFCs) have attracted significant attention due to the tremendous progress in developing highly conductive and stable membranes and ionomers against hydroxide attack [4]. However, the significant amount of Pt needed at the cathode side to accelerate the sluggish ORR kinetics remains a long-standing impediment for the widespread commercialization of PEMFCs and HEMFCs [4,5]. Thus, extensive efforts have been devoted to reducing or eliminating the use of precious metals at the cathode side in the past decades [6][7][8][9][10]. Fe(Co)-N-C catalysts represent the most promising precious metal-free (PGM-free) catalysts, with their ORR activity approaching or

ORR Activity
The FeNC-ICM catalysts were prepared via pyrolysis at 900 • C in inert atmosphere of a catalyst precursor including a dual carbon and nitrogen source (adenine), an iron source and the ZnCl 2 /NaCl eutectic salt. Three iron precursors were studied, namely iron(II) phthalocyanine (FePc), 1,10-phenanthroline iron(II) perchlorate (FePhen 3 ) and iron(II) acetate (FeAc 2 ), keeping all other synthesis parameters unchanged. The catalyst precursors were prepared with a fixed mass ratio of adenine/salts of 1/10, while the mass of each Fe precursor was adjusted to reach 4 wt.% Fe relative to the sum of (nitrogen + carbon mass) in the catalyst precursors. The detailed synthesis protocol is given in Materials and Methods. When using FeAc 2 as Fe precursor, the amount of 1,10-phenanthroline necessary to reach a molar ratio Phen/Fe of 3 was also added, in order to ensure all three precursors possessed similar nitrogen contents. The catalysts obtained after pyrolysis and water-wash (to remove NaCl) are labelled as ICM-FePc, ICM-FePhen 3 and ICM-FeAc 2 , the second part of the label identifying the iron precursor used.
The ORR activity of FeNC-ICM catalysts was evaluated with a rotating disk electrode (RDE) setup in both acidic and alkaline media. First, the effect of the nature of the iron precursor in the ICM approach on the ORR activity in acidic electrolyte was studied ( Figure 1a). The ICM-FeAc 2 catalyst showed a low onset potential for ORR in 0.1 M HClO 4 and poorly-defined diffusion-limited current density. In contrast, the ICM-FePc and ICM-FePhen 3 catalysts exhibited a half-wave potential (E 1/2 ) of 0.75 and 0.77 V vs. a reversible hydrogen electrode (RHE), respectively. They also showed a well-defined redox peak at~0.73 V vs. RHE in cyclic voltammograms (CVs) recorded in N 2 -saturated acidic medium, while no redox peak was observed for ICM-FeAc 2 (Figure 1b). This redox peak has previously been attributed to the Fe 2+/3+ -N 4 redox transition in pyrolyzed Fe-N-C materials [17,30]. This suggests that, with the present synthesis method, the FeN x moieties could only form at high temperature when using Fe precursors with pre-existing Fe-N 4 or Fe-N 6 coordinations, while the FeAc 2 in the presence of Phen did probably not form the Fe(phen) 3 complex during the dry ball-milling before pyrolysis, and could apparently not form FeN x moieties during the pyrolysis either. This result is different from what has been observed for other synthesis approaches, where separate FeAc 2 and 1,10-phenanthroline precursors, mixed with ZIF-8 as a sacrificial carbon precursor, entirely converted into FeN x moieties in the ZIF-8 derived Fe-N-C [11][12][13][14]17]. Figure 1b further showed that the CV of ICM-FeAc 2 was tilted and strongly deviated from a square signal, typical for samples with low electronic conductivity. This suggests that the nature of the iron precursor not only affected the formation of FeN x moieties, but also changed the nature of the carbon matrix obtained during pyrolysis. As will be shown later in this study, the situation is however complex here, with the presence of three different Fe-based species in ICM-FePhen 3 , and its redox assignment is further discussed after advanced spectroscopic characterization.  The XRD and Raman spectra of Fe0.5NC (a ZIF-8 derived Fe-N-C pyrolyzed material [11]) are also presented as references for a material that comprises only FeNx moieties integrated in amorphous carbon.
Then the ORR activity of ICM-FePhen3 was evaluated in 0.1 M HClO4 and 0.1 M KOH, as well as after acid wash (denoted as ICM-FePhen3-AW) (Figure 1c). The acid wash did not affect the ORR activity of ICM-FePhen3 in acidic medium (same on-set potential and half-wave potential as ICM-FePhen3), but resulted in a higher diffusion-limited current density, possibly due to the removal of poorly-active Fe species favoring the 2-electron reduction of ORR to H2O2. Moreover, ICM-FePhen3-AW exhibited a redox (attributed to Fe 2+/3+ N4) at similar potentials (but much lower intensity) than The most active ORR-catalyst in acid medium, ICM-FePhen 3 , was further characterized with X-ray diffraction (XRD) and Raman spectroscopy in order to identify the structure of the carbon matrix. Compared to the broad peaks (24 • and 43 • ) associated with amorphous carbon in the XRD patterns of a ZIF-8 derived Fe-N-C catalyst (labelled Fe 0.5 NC, used here as reference Fe-N-C catalyst from our laboratory that was demonstrated to comprise only FeN x moieties [3]), ICM-FePhen 3 exhibited narrower graphite-related peaks in XRD (peaks at 26 • , 43 • and 45 • , Figure 2a). The first-order Raman spectrum of ICM-FePhen 3 showed two peaks: the G band at~1590 cm −1 and the disorder-induced D band at~1360 cm −1 (Figure 2b). In the second-order Raman spectrum, three less intense peaks were observed, namely the G' (2D) band at~2710 cm −1 attributed to the D-band overtone, the peak at 2918 cm −1 due to a combination of D and G band, and the peak at~3240 cm −1 ascribed to the G-band overtone ( Figure 2b) [31]. ICM-FePhen 3 showed a lower I D /I G ratio (0.67 vs. 0.84) and a much more intense peak at~2710 cm −1 (widely observed in graphite and graphene) [32] compared to Fe 0.5 NC (Figure 2b), further revealing its higher degree of graphitization. Graphitic carbon is believed to have improved corrosion resistance relative to amorphous carbon, which could result in improved stability of Fe-N-C catalysts in PEMFCs or HEMFCs [33]. However, a synthetic contradiction between the desire for a high density of FeN x sites and a high degree of graphitization of carbon might occur, as discussed later.
polarization curves of ICM-FePhen3 catalyst before and after an acid wash (AW), measured in O2saturated 0.1 M HClO4 or 0.1 M KOH with a scan rate of 20 mV/s at 1600 rpm. (d) CVs measured in N2-saturated 0.1 M HClO4 or 0.1 M KOH with a scan rate of 20 mV/s. Results obtained with a commercial Pt/C catalyst (Tanaka Kikinzoku Kogyo (TKK, Japan), 46.7 wt.% Pt, anodic scan) are presented as a reference. The catalyst loading for Fe-N-C is 800 μg/cm 2 . The Pt loading is 10 μgPt/cm 2 . The XRD and Raman spectra of Fe0.5NC (a ZIF-8 derived Fe-N-C pyrolyzed material [11]) are also presented as references for a material that comprises only FeNx moieties integrated in amorphous carbon.
Then the ORR activity of ICM-FePhen3 was evaluated in 0.1 M HClO4 and 0.1 M KOH, as well as after acid wash (denoted as ICM-FePhen3-AW) ( Figure 1c). The acid wash did not affect the ORR activity of ICM-FePhen3 in acidic medium (same on-set potential and half-wave potential as ICM-FePhen3), but resulted in a higher diffusion-limited current density, possibly due to the removal of poorly-active Fe species favoring the 2-electron reduction of ORR to H2O2. Moreover, ICM-FePhen3-AW exhibited a redox (attributed to Fe 2+/3+ N4) at similar potentials (but much lower intensity) than  [11]) are also presented as references for a material that comprises only FeN x moieties integrated in amorphous carbon.
Then the ORR activity of ICM-FePhen 3 was evaluated in 0.1 M HClO 4 and 0.1 M KOH, as well as after acid wash (denoted as ICM-FePhen 3 -AW) ( Figure 1c). The acid wash did not affect the ORR activity of ICM-FePhen 3 in acidic medium (same on-set potential and half-wave potential as ICM-FePhen 3 ), but resulted in a higher diffusion-limited current density, possibly due to the removal of poorly-active Fe species favoring the 2-electron reduction of ORR to H 2 O 2 . Moreover, ICM-FePhen 3 -AW exhibited a redox (attributed to Fe 2+/3+ N 4 ) at similar potentials (but much lower intensity) than observed for ICM-FePhen 3 , suggesting that FeN x moieties survived the acid wash, and probably worked as the major active sites for ORR in acid (Figure 1d). In alkaline medium, ICM-FePhen 3 and ICM-FePhen 3 -AW showed however different ORR activities, contrasting with the case in acidic medium. ICM-FePhen 3 -AW had an E 1/2 value negatively shifted by 60 mV vs. ICM-FePhen 3 . These two observations strongly suggest that the acid-wash removed some Fe species that are ORR-inactive in acidic medium but ORR-active in alkaline medium. The iron speciation is reported and discussed further below. For the most active sample, ICM-FePhen 3 , its E 1/2 value at a catalyst loading of 800 µg FeNC /cm 2 in alkali was 0.86 V vs. RHE, surpassing that of the commercial Pt/C (TKK, 10 µg Pt /cm 2 ). This is among the best ORR activities reported for Fe-N-C catalysts in alkali [34][35][36][37].

Iron Speciation
In order to identify the Fe species present in ICM-FePhen 3 and ICM-FePhen 3 -AW, XRD, 57 Fe Mössbauer spectroscopy, and X-ray absorption spectroscopy (XAS) measurements were performed. The presence of Fe 2 O 3 was detected by XRD in ICM-FePhen 3 (main peak at ca 36 • and secondary peaks at ca 30 • , JCPDS #39-1346), but not detected in ICM-FePhen 3 -AW ( Figure 2a). With room-temperature 57 Fe Mössbauer spectroscopy, four major spectral components were detected in ICM-FePhen 3 ( Figure 3a and Table 1): a sextet assigned to iron carbide (Fe 3 C, 30%), two sextets assigned to iron oxide (ε-Fe 2 O 3 , 12%) [38], a broad singlet (12%) and a doublet (46%). The assignment of the singlet will be discussed later. The Mössbauer parameters of that doublet (Table 1) allow the following straightforward but ambiguous assignment: i) either the D1 component typical for low-spin ferrous FeN 4 moiety in pyrolyzed Fe-N-C catalysts [11,17,39] or ii) a high-spin Fe 3+ species such as amorphous or β-Fe 2 O 3 with a small particle size [40,41]. It is known that the Mössbauer spectrum of nanometric ferric oxides changes from a doublet to a sextet spectral component below a certain critical temperature [42,43], while that of D1 (FeN x moiety) is independent of temperature, remaining a doublet even at 5 K [44]. In order to distinguish between the two possible assignments, the 57 Fe Mössbauer spectrum of ICM-FePhen 3 was measured at 5 K ( Figure 3b). In that spectrum, the relative area (RA) of the two sextets assigned to Fe 2 O 3 (both ε + β) increased to 62%, while the doublet disappeared. This indicates that the well-defined narrow doublet observed at room temperature was arising from nanometric β-Fe 2 O 3 . This is also independently confirmed later with high-resolution transmission electron microscopy (HR-TEM) images.  (1) Unresolved from Mössbauer spectroscopy alone (see Table 2 for the assignment of that broad singlet).; (2) Those RA % are only indicative, due to the low signal-to-noise ratio of that spectrum.  (1) Unresolved from Mössbauer spectroscopy alone (see Table 2 for the assignment of that broad singlet); (2) Those RA % are only indicative, due to the low signal-to-noise ratio of that spectrum. NA Fe 3 C 32 1 (1) Both the sextet and doublet components assigned to Fe 2 O 3 particles of different size and/or polymorphs; (2) Those RA % are only indicative, and poorly reliable due to the low signal-to-noise ratio of that spectrum.
For ICM-FePhen 3 -AW, the analysis of its room-temperature Mössbauer spectrum (Figure 3c) showed first that the magnitude of the γ-rays absorption was much decreased (synonym with much lower total Fe content in the sample) and second, that both the sextets of ε-Fe 2 O 3 and the intense doublet of β-Fe 2 O 3 observed in ICM-FePhen 3 were no longer visible. Instead, the spectrum clearly showed only the sextet typical for Fe 3 C and the broad singlet component already observed in ICM-FePhen 3 (Figure 3c). It however cannot be excluded that a small fraction of the narrow doublet assigned to nanosized Fe 2 O 3 is still present in ICM-FePhen 3 -AW, embedded within the broad singlet. These results can be interpreted as the complete or nearly-complete removal of the Fe 2 O 3 (nano)particles during the acid wash, while a significant fraction of Fe 3 C was not dissolved. Last, the singlet with unusually large line width (in green color in Figure 3) was present in all three Mössbauer spectra and is a spectral component previously identified in complex Fe-N-C materials comprising numerous Fe species [45]. It represents only 12% of the Fe signal in ICM-FePhen 3 but 55% in ICM-FePhen 3 -AW (Table 1), suggesting it corresponds to a species that is stable in acidic medium. As will be seen later, its assignment can only be made after discussing the XAS results.
In order to further study the coordination environments of iron, XAS measurements at the Fe K-edge were performed on ICM-FePhen 3 and ICM-FePhen 3 -AW, as well as on selected reference compounds. The similar X-ray absorption near edge structure (XANES) spectra (Figure 4a In order to go beyond the limitations of Mössbauer spectroscopy (unresolved broad singlet component), we then performed linear combination fitting (LCF) of the XANES spectra of ICM-FePhen3 ( Figure 4c) and ICM-FePhen3-AW (Figure 4d) with three reference spectra, namely Fe2O3, Fe3C and Fe0.5NC (the latter comprising exclusively Fe as atomically dispersed FeNx moieties [3]). The XANES spectrum of ICM-FePhen3 was well fitted with a combination involving the above three components. Their relative amounts are shown in Table 2, and agree well with the RA % of Fe2O3, Fe3C and of the broad singlet quantified from the 5K Mössbauer-spectra fittings. Thus, for ICM-FePhen3, the broad singlet identified in the Mössbauer spectrum can be entirely assigned to FeNx moieties.
For ICM-FePhen3-AW, the three components of Fe2O3, Fe3C and Fe0.5NC were also needed to get the best LCF fit of its XANES spectrum (Figure 4d), including 30% of the reference Fe2O3 spectrum and 38% of the reference Fe0.5NC spectrum. LCF with only the Fe3C and Fe0.5NC references resulted in a clearly poorer fit quality (especially at the white line intensity region), confirming the presence of Fe2O3 in ICM-FePhen3-AW (Supporting Figure S1). This is in contrast with the Mössbauer analysis, which could not identify the presence of Fe2O3 or nano Fe2O3, probably due to the low signal-to-noise ratio. The presence of Fe2O3 in ICM-FePhen3-AW is also supported by the peak at ca 2.5 Å in the FT-EXAFS spectrum (Figure 3b) and by elemental mapping results (shown later).
The LCF XANES analysis and FT-EXAFS spectra thus indicate that the ICM-FePhen3-AW sample contains three Fe species: Fe3C, FeNx moieties and nanosized Fe2O3. Since only two spectral components have been used for the Mössbauer fitting of that sample, the comparison of RA% and XANES LCF % is less straightforward. The broad singlet used for the Mössbauer fitting of that catalyst probably includes both the signal from FeNx moieties and from nano-Fe2O3, and this indicates that the RA% for the singlet and for Fe3C indicated in Table 1 for ICM-FePhen3-AW is not reliable (as initially expected also from the high noise level of that spectrum). In conclusion, the combined Mössbauer, XANES and EXAFS analysis indicates that both ICM-FePhen3 and ICM-FePhen3-AW contain FeNx moieties. These sites were present already in ICM-FePhen3 and survived the acid wash, In order to go beyond the limitations of Mössbauer spectroscopy (unresolved broad singlet component), we then performed linear combination fitting (LCF) of the XANES spectra of ICM-FePhen 3 ( Figure 4c) and ICM-FePhen 3 -AW (Figure 4d) with three reference spectra, namely Fe 2 O 3 , Fe 3 C and Fe 0.5 NC (the latter comprising exclusively Fe as atomically dispersed FeN x moieties [3]). The XANES spectrum of ICM-FePhen 3 was well fitted with a combination involving the above three components. Their relative amounts are shown in Table 2, and agree well with the RA % of Fe 2 O 3 , Fe 3 C and of the broad singlet quantified from the 5K Mössbauer-spectra fittings. Thus, for ICM-FePhen 3 , the broad singlet identified in the Mössbauer spectrum can be entirely assigned to FeN x moieties.
For ICM-FePhen 3 -AW, the three components of Fe 2 O 3 , Fe 3 C and Fe 0.5 NC were also needed to get the best LCF fit of its XANES spectrum (Figure 4d), including 30% of the reference Fe 2 O 3 spectrum and 38% of the reference Fe 0.5 NC spectrum. LCF with only the Fe 3 C and Fe 0.5 NC references resulted in a clearly poorer fit quality (especially at the white line intensity region), confirming the presence of Fe 2 O 3 in ICM-FePhen 3 -AW (Supporting Figure S1). This is in contrast with the Mössbauer analysis, which could not identify the presence of Fe 2 O 3 or nano Fe 2 O 3 , probably due to the low signal-to-noise ratio. The presence of Fe 2 O 3 in ICM-FePhen 3 -AW is also supported by the peak at ca 2.5 Å in the FT-EXAFS spectrum (Figure 3b) and by elemental mapping results (shown later).
The LCF XANES analysis and FT-EXAFS spectra thus indicate that the ICM-FePhen 3 -AW sample contains three Fe species: Fe 3 C, FeN x moieties and nanosized Fe 2 O 3 . Since only two spectral components have been used for the Mössbauer fitting of that sample, the comparison of RA% and XANES LCF % is less straightforward. The broad singlet used for the Mössbauer fitting of that catalyst probably includes both the signal from FeN x moieties and from nano-Fe 2 O 3 , and this indicates that the RA% for the singlet and for Fe 3 C indicated in Table 1 for ICM-FePhen 3 -AW is not reliable (as initially expected also from the high noise level of that spectrum). In conclusion, the combined Mössbauer, XANES and EXAFS analysis indicates that both ICM-FePhen 3 and ICM-FePhen 3 -AW contain FeN x moieties. These sites were present already in ICM-FePhen 3 and survived the acid wash, and probably Catalysts 2019, 9, 144 9 of 16 worked as the major active sites for ORR in acidic medium. This hypothesis would explain the similar ORR activity of ICM-FePhen 3 and ICM-FePhen 3 -AW in acidic medium.

Porous Structure and Morphology
Following the carbon bulk structure and Fe speciation characterization, the porous structure of ICM-FePhen 3 was investigated with N 2 adsorption/desorption isotherms. The hysteresis loop (assigned to small mesopores) and the slope of the N 2 adsorption/desorption isotherm of ICM-FePhen 3 (assigned to large mesopores) (Figure 5a), as well as the corresponding pore size distribution (calculated from density functional theory (DFT) method, Figure 5b) indicated the mesoporous structure of the carbon matrix, with a volume-average pore size around 3.4 nm ( Table 3) [46].
Catalysts 2019, 9 FOR PEER REVIEW 9 and probably worked as the major active sites for ORR in acidic medium. This hypothesis would explain the similar ORR activity of ICM-FePhen3 and ICM-FePhen3-AW in acidic medium.

Porous Structure and Morphology
Following the carbon bulk structure and Fe speciation characterization, the porous structure of ICM-FePhen3 was investigated with N2 adsorption/desorption isotherms. The hysteresis loop (assigned to small mesopores) and the slope of the N2 adsorption/desorption isotherm of ICM-FePhen3 (assigned to large mesopores) (Figure 5a), as well as the corresponding pore size distribution (calculated from density functional theory (DFT) method, Figure 5b) indicated the mesoporous structure of the carbon matrix, with a volume-average pore size around 3.4 nm ( Table 3) [46]. Last, the morphology of ICM-FePhen3-AW was further studied with scanning electron microscopy (SEM) and high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM). Tubular-shaped, hollow carbon matrixes with diameters around 60 nm were observed from the SEM image (Figure 6a). The HAADF-STEM images (Figure 6c-e) identified a highly porous structure of the catalyst, agreeing well with the high double layer capacitance from CVs (Figure 1b,d) and the high BET surface areas (~1500 m 2 /g, Table 3). The obtained hierarchically porous structure was arisen from the ZnCl2/NaCl eutectic salts introduced in the catalyst precursor [22]. The elemental compositions were investigated by energy dispersive X-ray spectroscopy (EDX). Carbon, oxygen and iron were present in the sample (Figure 6f). It should be noted that nitrogen was not analyzed due to the close energy between C and N, and the stronger signal of C. Moreover, trace amount of S (introduced from acid wash with H2SO4), Si (from the quartz tube and quartz boat used during pyrolysis) and Ni (from the TEM grid) were also detected by EDX. Aggregated Fe particles with different sizes were also observed in HAADF-STEM images for ICM-FePhen3-AW (Figure 6be). The EDX elemental mapping results suggested that these particles are iron oxides, or oxidized Fe3C (Figure 6g-j), and they survived the acid wash probably due to protection of the carbon shell. This is in line with the peak at ca 2.5 Å assigned to Fe-Fe backscattering from Fe oxides in the FT-EXAFS spectrum of ICM-FePhen3-AW, as well as the LCF of its XANES spectrum.   Last, the morphology of ICM-FePhen 3 -AW was further studied with scanning electron microscopy (SEM) and high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM). Tubular-shaped, hollow carbon matrixes with diameters around 60 nm were observed from the SEM image ( Figure 6a). The HAADF-STEM images (Figure 6c-e) identified a highly porous structure of the catalyst, agreeing well with the high double layer capacitance from CVs (Figure 1b,d) and the high BET surface areas (~1500 m 2 /g, Table 3). The obtained hierarchically porous structure was arisen from the ZnCl 2 /NaCl eutectic salts introduced in the catalyst precursor [22]. The elemental compositions were investigated by energy dispersive X-ray spectroscopy (EDX). Carbon, oxygen and iron were present in the sample (Figure 6f). It should be noted that nitrogen was not analyzed due to the close energy between C and N, and the stronger signal of C. Moreover, trace amount of S (introduced from acid wash with H 2 SO 4 ), Si (from the quartz tube and quartz boat used during pyrolysis) and Ni (from the TEM grid) were also detected by EDX. Aggregated Fe particles with different sizes were also observed in HAADF-STEM images for ICM-FePhen 3 -AW (Figure 6b-e). The EDX elemental mapping results suggested that these particles are iron oxides, or oxidized Fe 3 C (Figure 6g-j), and they survived the acid wash probably due to protection of the carbon shell. This is in line with the peak at ca 2.5 Å assigned to Fe-Fe backscattering from Fe oxides in the FT-EXAFS spectrum of ICM-FePhen 3 -AW, as well as the LCF of its XANES spectrum.

Discussion
Based on the results and analyses, the ICM-FePhen3 catalyst comprises mainly Fe2O3 (60-68%) together with FeNx moieties (20-28%) and Fe3C (10-18%). After acid wash, the ICM-FePhen3-AW sample still comprises Fe in these three different environments, but now in comparable relative contents (26-34% Fe2O3, 35-41% FeNx moieties, 31-33% Fe3C). The majority of Fe2O3 particles in ICM-FePhen3 were thus removed by acid wash. The Fe2O3 particles still observed after acid wash may have been protected from the acid leaching by a surrounding carbon layer, and/or may have formed from the partial oxidation of Fe3C particles during the acid wash. The almost-superimposed ORR activity in acidic medium of ICM-FePhen3 and ICM-FePhen3-AW suggests that all or most of the FeNx moieties present in ICM-FePhen3 survived the acid wash, and are at the origin of the major part of the ORR activity of both samples, in line with the current consensus in the field that FeNx sites are the most active Fe-sites in acidic medium [11,14,17,47]. The slightly lower ORR activity of ICM-FePhen3-AW vs. ICM-FePhen3 may be assigned to a small contribution of Fe3C particles to the ORR activity of ICM-FePhen3, or to the removal by acid wash of a few FeNx moieties with low stability. For the former hypothesis, the acid wash is expected to remove Fe3C particles that are either unprotected by a carbon layer, or protected by a thin or porous N-doped carbon layer, while Fe3C particles with a thick graphite shell protection are expectedly stable to an acid wash. Since Fe3C particles surrounded by a thin N-doped carbon layer have been shown to possess some ORR activity in acidic medium, but also to be poorly stable due to defects in their carbon shell, it is possible that the small decrease in ORR activity of ICM-FePhen3-AW vs. ICM-FePhen3 comes from the removal of a fraction of the Fe3C particles. FeNx moieties were identified in both the ICM-FePhen3 and ICM-FePhen3-AW catalysts, with the former showing a clear redox peak in N2-saturated CV, while the latter exhibiting a redox peak at similar position with much lower intensity (Figure 1d). The redox signal of FeNx moieties in pyrolyzed Fe-N-C catalysts is often weak and broad, necessitating to resort to square-wave voltammetry to identify it. Thus, while the position of this redox peak fits well with

Discussion
Based on the results and analyses, the ICM-FePhen 3 catalyst comprises mainly Fe 2 O 3 (60-68%) together with FeN x moieties (20-28%) and Fe 3 C (10-18%). After acid wash, the ICM-FePhen 3 -AW sample still comprises Fe in these three different environments, but now in comparable relative contents (26-34% Fe 2 O 3 , 35-41% FeN x moieties, 31-33% Fe 3 C). The majority of Fe 2 O 3 particles in ICM-FePhen 3 were thus removed by acid wash. The Fe 2 O 3 particles still observed after acid wash may have been protected from the acid leaching by a surrounding carbon layer, and/or may have formed from the partial oxidation of Fe 3 C particles during the acid wash. The almost-superimposed ORR activity in acidic medium of ICM-FePhen 3 and ICM-FePhen 3 -AW suggests that all or most of the FeN x moieties present in ICM-FePhen 3 survived the acid wash, and are at the origin of the major part of the ORR activity of both samples, in line with the current consensus in the field that FeN x sites are the most active Fe-sites in acidic medium [11,14,17,47]. The slightly lower ORR activity of ICM-FePhen 3 -AW vs. ICM-FePhen 3 may be assigned to a small contribution of Fe 3 C particles to the ORR activity of ICM-FePhen 3 , or to the removal by acid wash of a few FeN x moieties with low stability. For the former hypothesis, the acid wash is expected to remove Fe 3 C particles that are either unprotected by a carbon layer, or protected by a thin or porous N-doped carbon layer, while Fe 3 C particles with a thick graphite shell protection are expectedly stable to an acid wash. Since Fe 3 C particles surrounded by a thin N-doped carbon layer have been shown to possess some ORR activity in acidic medium, but also to be poorly stable due to defects in their carbon shell, it is possible that the small decrease in ORR activity of ICM-FePhen 3 -AW vs. ICM-FePhen 3 comes from the removal of a fraction of the Fe 3 C particles. FeN x moieties were identified in both the ICM-FePhen 3 and ICM-FePhen 3 -AW catalysts, with the former showing a clear redox peak in N 2 -saturated CV, while the latter exhibiting a redox peak at similar position with much lower intensity (Figure 1d). The redox signal of FeN x moieties in pyrolyzed Fe-N-C catalysts is often weak and broad, necessitating to resort to square-wave voltammetry to identify it. Thus, while the position of this redox peak fits well with that assigned to FeN x moieties, the overall redox signal of ICM-FePhen 3 may originate from both FeN x moieties and, mostly, from Fe 2 O 3 particles. This would also explain the reduced peak intensity of ICM-FePhen 3 -AW, as the majority of Fe 2 O 3 was removed via acid wash. The residual redox signal after acid wash possibly originates from FeN x moieties and/or the residual Fe 2 O 3 .
In alkaline medium however, the ORR activity of ICM-FePhen 3 -AW is significantly lower than that of ICM-FePhen 3 (Figure 1c), and this is paralleled by the removal of a large fraction of Fe 2 O 3 during the acid wash. This suggests that one or several polymorphs of Fe 2 O 3 contributed largely to the overall ORR activity of ICM-FePhen 3 in alkaline. This is in line with the previously reported ORR activity of Fe oxide in alkaline medium [48]. Other inorganic iron species (FeSn 0.5 [49,50], FeS@Fe 3 C [51] and Fe 2 P@Fe 4 P [52]) have been reported to be ORR active as well, and they generally exhibit much superior ORR activity in alkali compared to that in acid, agreeing well with our results. Fe 3 C is also present in ICM-FePhen 3 and may have contributed to the overall ORR activity in alkaline medium. While in acid medium, the ORR activity of ICM-FePhen 3 is mostly assigned to FeN x moieties, the contribution to the overall ORR activity of Fe 3 C particles encapsulated in nitrogen-doped graphene may be higher in alkaline medium due to a stronger enhancement of the ORR turnover frequency with pH changing from 1 to 13 for N-doped carbon (with or without subsurface Fe 3 C) than for FeN x moieties. Both experimental results [37,53] and theoretical calculations [54] have proved the ORR activity of Fe 3 C particles encapsulated in nitrogen-doped graphene.
It should be noted that the major iron species in the present ICM-derived Fe-N-C catalysts is Fe oxides, and the site density of FeN x moieties is quite low. This explains the relatively low ORR activity in acidic medium of the present ICM-derived Fe-N-C catalysts compared to state-of-the-art Fe-N-C catalysts [11,14,17,39,47]. The low site density of FeN x moieties seems correlated to the high degree of graphitization of the carbon matrix derived from the ICM approach, as compared to the amorphous carbons usually obtained from the pyrolysis of ZIF-8. The present approach even requires the pre-existing Fe-N 4 (FePc) or Fe-N 6 (FePhen 3 ) coordination as iron precursors to obtain reasonable ORR activity after pyrolysis. This suggests that there is a synthetic contradiction between the desires for a high density of FeN x sites (necessary to reach a high ORR activity) and a high degree of graphitization of carbon (possibly leading to improved stability). Highly graphitized carbons with few defects typically result in low nitrogen content, in turn limiting the maximum number of FeN x moieties that may be hosted in the carbon matrix.

Materials and Methods
Sample preparation. All catalyst precursors were prepared via the dry ball-milling of Zn(II) chloride, Na(I) chloride, adenine and an iron precursor. A mass of 100 mg of adenine, 825 mg of ZnCl 2 , 175 mg of NaCl and certain amount of iron precursor (iron(II) phthalocyanine (FePc), 1,10-phenanthroline iron (II) perchlorate (FePhen 3 ), or iron (II) acetate (FeAc 2 ) corresponding to 4 wt.% Fe in the total precursor (the inorganic ZnCl 2 /NaCl salts not being counted) were weighed and poured into a ZrO 2 crucible with hundred zirconium-oxide balls of 5 mm diameter. When using FeAc 2 as Fe precursor, 1,10-phenanthroline was added to reach the stoichiometry Fe:Phen of 1:3, to ensure all catalyst precursors possessed similar nitrogen contents before pyrolysis. The ZrO 2 crucible was then sealed under air and placed in a planetary ball-miller (Fritsch Pulverisette 7 Premium, Fritsch, Idar-Oberstein, Germany). The powders were ball-milled for 30 min at 200 rpm milling speed, then pyrolyzed in flash mode at 900 • C in flowing Ar for 1 h. Then they were washed in deionized water for 48 hours at room temperature to remove NaCl and finally dried. The catalysts are labelled "ICM-Fe-precursor". In some cases, the washing after pyrolysis was performed in 0.5 M H 2 SO 4 rather than in water, in otherwise the same conditions. The catalysts are labelled in this case ICM-Fe-precursor-AW.
Electrochemical characterization. The catalyst powders were deposited on glassy carbon electrodes as working electrode. Catalyst ink was prepared by dispersing 10 mg of the catalyst powder in a mixture of Millipore water (36.5 µL, 18.2 MΩ cm) and ethanol (300 µL, Sigma-Aldrich, 99.8%, Lyon, France), into which 5 wt% Nafion solution (108.5 µL, Sigma-Aldrich, Lyon, France) was added as a binder phase. The resulting mixture was sonicated for 60 min, and then an aliquot of 7 µL was drop-cast onto the glassy carbon electrodes (0.196 cm 2 , Pine instrument, Grove City, PA, USA), resulting in a loading of 800 µg/cm 2 . The working electrode with the deposited catalyst layer was used in a three-electrode cell set-up connected to a bipotentiostat (Biologic SP 300, Bio-Logic Science Instruments, Isère, France) and rotator (Pine Instruments, Grove City, PA, USA). The graphite rod and reversible hydrogen electrode (RHE) were used as counter and reference electrodes, respectively. The XRD. X-ray diffraction patterns were conducted using a PANanalytical X'Pert Pro powder X-ray diffractometer (Malvern Panalytical Ltd, Malvern, UK) with Cu K α radiation.
Raman. Raman spectra were collected using a LabRAM ARAMIS Raman microscope (HORIBA, Kyoto, Japan) with a 473 nm laser.
BET. N 2 sorption analysis was performed at liquid nitrogen temperature (77 K) with a Micromeritics ASAP 2020 instrument (Micromeritics, Verneuil-en-Halatte, France). Prior to the measurements, FeNC-ICM catalysts were degassed at 200 • C for 5 h in flowing nitrogen to remove guest molecules or moisture. The pore size distributions were calculated by fitting the full isotherm with the quench solid density functional theory model with slit pore geometry from NovaWin (Quantachrome Instruments).
Mössbauer spectroscopy. 57 Fe Mössbauer spectroscopy was used to obtain information on iron speciation. 120 to 300 mg samples were mounted in a 2-cm 2 holder. Mössbauer spectra were measured at variable temperatures between 5 and 300 K in a helium flow cryostat (SHI-850 Series from Janis, Woburn, MA, USA). The Mössbauer spectrometer (Wissel, Germany) was operated in the transmission mode with a 57 Co: Rh source at room temperature. The velocity driver was operated in the constant acceleration mode with a triangular velocity waveform. The velocity scale was calibrated with the magnetically split sextet of a high-purity α-Fe foil at room temperature. The spectra were fitted to appropriate combinations of Lorentzian profiles representing quadrupole doublets, sextets by least-squares methods. Isomer shifts are given relative to α-Fe at room temperature.
XAS. Fe K-edge X-ray absorption spectra were collected at room temperature in transmission mode at SAMBA beamline (Synchrotron SOLEIL, Saint-Aubin, France). The beamline is equipped with a sagittally focusing Si 220 monochromator and two Pd-coated mirrors that were used to remove X-rays harmonics. The catalysts were pelletized as disks of 10-mm diameter with 1-mm thickness using Teflon powder (1-µm particle size) as a binder. Data treatment and linear combination fitting (LCF) were performed with the Athena software [55].
SEM. Scanning electron microscopy micrographs were obtained with a Hitachi S-4800 apparatus (Hitachi, Tokyo, Japan).
TEM. Transmission electron microscopy micrographs were obtained with a JEOL ARM-200 CF probe Cs-corrected scanning transmission electron microscope (JEOL, Tokyo, Japan), operated at 80 kV and equipped with Centurio 100 mm 2 SDD EDXS system and Gatan GIF Quantum ER Dual EELS system.

Conclusions
ZnCl 2 /NaCl eutectic salts assisted carbonization was demonstrated to be an effective approach to synthesize Fe-N-C catalysts when using iron precursors with pre-existing Fe-N coordination (FePc and FePhen 3 ). The obtained catalysts exhibited a highly porous structure and a high degree of graphitization. However, it is a challenge to form a high density of FeN x sites in graphitized carbon, probably due to the low nitrogen content in the graphitic matrix. Iron precursors with pre-existing Fe-N x coordination was required to form a small content of FeN x moieties at high temperature, while separate Fe and N precursors lead to an inactive Fe-N-C material. Even for the most active samples, a variety of iron species was detected in the FeNC-ICM catalysts, including FeN x moieties but also Fe carbides and, mostly, Fe oxides (different sizes and polymorphs). From a comparison of the ORR activity in acid and alkaline conditions for two catalysts (a fresh and an acid-washed version of the same starting material), the FeN x moieties were identified as the most active sites in acid medium, while both FeN x moieties and Fe 2 O 3 oxides are proposed to promote ORR in alkaline electrolyte. The ORR activity of Fe 3 C particles encapsulated in nitrogen-doped graphene cannot be excluded in either acid or alkali, with different contributions to the overall activity.