Converting CO2 back to carbon (C)-based chemicals/fuels is a critical step en route to realizing green environment and energy sustainability1,2,3,4,5. CO2 can in principle be reduced to active forms of C-products, but the activated CO2 intermediates tend to undergo different reaction pathways, making it difficult to efficiently control the reaction selectivity6,7,8. Recently, the electrochemical CO2 reduction reaction (CO2RR) has been studied extensively due to its appealing potential to reduce CO2 with renewable electricity9,10. The center of these studies lies in the development of robust catalysts to lower the reduction overpotentials and to direct reaction pathways towards specific C-products. Among all possible products generated from the CO2RR, CO is an important precursor that is widely used for preparing other chemicals and fuels11,12,13. In the CO2RR process, CO is produced via two electron reduction of CO214,15, and this reaction can be catalyzed by many different catalysts, including not only noble metals (Au, Ag, and Pd)16, but also non-noble metals (Cu, CuIn alloys)17,18, and metal−N4 (M−N4) complex structures19,20,21, as summarized in Supplementary Table 1. The M−N4 structures, especially Ni−N- and Fe−N-based structures, have attracted particular interest as a class of catalysts for CO2RR to CO due to increased electron conductivity with decreased band gap of M−N4 structures and intensive bonding interactions between *COOH intermediates and metal centers, thus lowering the overall reaction barriers and promoting CO generation. However, the Ni−N-complex catalysis is limited by slow kinetics of CO2 + H+ + e → *COOH and weak *H binding for necessary hydrogenation steps, while the Fe−N-complex catalysis is hindered by the strong *CO and *H binding to the Fe−N sites, which prevent CO from being released as a gas form, or from being further activated for the following hydrogenation steps20,22,23,24. Therefore, it is important to stabilize the chemical environment of metal-N-catalysts and to optimize their catalysis towards hydrogenation or desorption of CO at low overpotentials for successful CO2RR.

Here, we report a strategy to improve Fe−N catalysis efficiency for CO2RR. The catalyst is based on FeN/Fe3N heterostructure nanoparticles (NPs) in which Fe is coordinated by 4 N’s and 2 N’s respectively, forming a unique Fe−N4/Fe−N2 interface on the NP surface. This interface offers the desired synergetic effect on CO2 adsorption on FeN, and electron transfer from Fe3N to FeN, enhancing the CO2 reduction to CO. In the CO2-saturated 0.5 M KHCO3 solution, the catalyst shows a high CO Faradaic efficiency (FE) of 98% at −0.4 V (vs. RHE) without noticeable efficiency change in the 100 h electrolysis test between −0.4 V and −0.9 V. The catalysis produces CO with a high single-pass conversion rate up to 56%, which is unprecedented. Our study presents an interface design strategy for improving CO2RR catalysis efficiency.


The FeN/Fe3N NPs were prepared via nitridation of an Fe salt that was pre-deposited on the commercial Ketjen (KJ) carbon support. In the process, Fe(NO3)3·9H2O urea and carbon support were mixed in water under ultrasonication, separated from the solution, washed with ethanol and deionized (DI) water, and dried under vacuum at 60 °C. The dried sample was then annealed under a flow of ammonia gas at different temperatures (300, 500, and 700 °C) to initiate iron nitride NP formation (see details in methods, and Supplementary Fig. 1). The NP crystal structure was analyzed by powder X-ray diffraction (XRD). Annealing at 300 °C led to the formation of FeN NPs with a cubic structure (space group F-43m, JCPDS No. 50 − 8017, a = b = c = 4.307 Å), annealing at 700 °C gave Fe3N NPs with a hexagonal structure (space group of P312, JCPDS No. 72 − 2125, a = b = 4.789 Å, c = 4.710 Å), while annealing at 500 °C yielded FeN/Fe3N NPs (Fig. 1a). The FeN/Fe3N, FeN, and Fe3N NPs are spherical and uniform, as shown in the scanning transmission electron microscopy (STEM) image (Fig. 1b and Supplementary Fig. 2). STEM mapping images display the similar Fe and N elements distribution across each FeN/Fe3N (Fig. 1c), FeN and Fe3N NP (Supplementary Fig. 2). The high-resolution TEM (HRTEM) image of the FeN/Fe3N NPs (Fig. 1d) shows clear FeN/Fe3N domains and interfaces with the lattice mismatch at 5.12% (detail in Methods). The detailed analysis of the FeN/Fe3N area by high resolution high-angle annular dark field aberration-corrected STEM (HAADF-STEM) reveals that the interface is created by the junction between cubic FeN and hexagonal Fe3N (Fig. 1e) with obvious atomic dislocation and two distinct diffraction spots at the junction (Supplementary Fig. 3). Away from the junction, the FeN (Supplementary Fig. 4) and Fe3N (Supplementary Fig. 5) regions are in single crystalline, showing no atomic dislocation. Additionally, the fast Fourier transform (FFT) images (Supplementary Fig. 6) of different regions in Fig.1e further confirm the FeN/Fe3N interface is formed with slightly distorted cubic and hexagonal structure. The N2 and CO2 adsorption ability of the Fe–N structures were measured by the Brunner–Emmet–Teller (BET) method under N2 and CO2 atmosphere (Supplementary Fig. 7). The FeN/Fe3N structure adsorbs more N2 (736.1 m2 g−1) than the FeN (314.5 m2 g−1) and Fe3N (586.0 m2 g−1) ones, while the FeN structure adsorbs more CO2 (25.8 cm2 g−1) than the FeN/Fe3N (24.7 cm2 g−1) and Fe3N (11.2 cm2 g−1) ones, suggesting that the FeN/Fe3N structure may be more porous, but it is the presence of the Fe–N4 coordination sites in the FeN structure that shows higher CO2 adsorption power.

Fig. 1: Fe-N NP characterization.
figure 1

a XRD patterns of FeN, FeN/Fe3N, and Fe3N NPs. b STEM image of FeN/Fe3N NPs. c Elemental mapping of Fe, N and C of a single FeN/Fe3N NP. d HRTEM image of FeN/Fe3N NPs showing the interface between FeN and Fe3N domains. e HRHAADF-STEM image of the FeN/Fe3N interface. f Schematic illustration of the FeN/Fe3N interface formed between FeN (100) and Fe3N (001). g 3D electron density difference distributions at the FeN/Fe3N interface. h The electrostatic potential of the FeN/Fe3N interface. i The electron localization function of the FeN/Fe3N interface. j 2D cross-section of electron density difference at the FeN/Fe3N interface. k The calculated work functions (Wf) of FeN and Fe3N.

We used the density functional theory (DFT) to further investigate the FeN/Fe3N structure (Fig. 1f–k). The (100) planes of FeN (pink) (interplanar distance 4.230 Å) and (001) planes of Fe3N (green) (interplanar distance 4.656 Å) match well to form the junction (Fig. 1f). We calculated electron density difference distributions25 between the FeN/Fe3N interface. As shown in Fig. 1g, the electron clouds around N and Fe in the FeN region (yellow) are much more enriched than that in the Fe3N region (blue), indicating the electron transfer from Fe3N to FeN. Due to this electron transfer, the FeN region becomes nucleophilic (Fig. 1h red and yellow region), while the Fe3N region is electrophilic (Fig. 1h blue region). The distribution of electrons across the interface structure was studied by 2D electron density distribution and electron location function, as shown in Fig. 1i, j. More electrons are localized over the FeN domain than over the Fe3N one. This is further supported by the calculated electrostatic potential26 (Fig. 1k) with the work function of FeN at 6.56 eV and Fe3N at 5.41 eV. Therefore, in the FeN/Fe3N interface structure, there exists a built-in electric field from Fe3N to FeN.

The Fe K-edge of the extended X-ray absorption fine structure spectroscopy (EXAFS) of the Fe3N/FeN NPs show the increased energy of the fingerprint peak from Fe3N (7096.8 eV) to FeN (7108.5 eV) (Fig. 2a), indicating the Fe valence state change from +1 in Fe3N to +3 in FeN27,28. There is also a fingerprint peak (7106.4 eV) that is located in the middle of the Fe3N and FeN, implying that the Fe valence in the FeN/Fe3N interface region is between +1 and +3. The pre-edge peaks at 7089.7 and 7088.2 eV are from Fe−N4 with tetrahedral coordination (FeTd, Insert in Fig. 2a right)29,30, not the octahedral coordination (FeOct, Insert in Fig. 2a left) in the Fe3N structure31. The Fourier transform k3-weighted EXAFS (FT-EXAFS) spectra in Fig. 2b (R space) show main peaks in the range from 1 to 3 Å. With more N binding to Fe, the Fe–Fe distance gets longer and longer (in the 2 − 3 Å range), but the Fe–N distance is shorter (in the 1–2 Å range), showing a double peak for the FeOct − N (in the Fe3N structure) and a broad peak for the FeTd − N (in the FeN structure). The main peak at 2.33 Å in the Fe3N spectrum is close to the Fe−Fe bond (2.23 Å) measured from the Fe foil. Fittings of the FeN (Supplementary Fig. 8 and Supplementary Table 2) and Fe3N (Supplementary Fig. 9 and Supplementary Table 3) spectra suggest that Fe has a coordination number (CN) of 3.92 and 1.40 respectively. The FeN/Fe3N structure shows more complex peaks in the R-space with Fe−N4 peak at 1.60 Å and Fe–N2 peak at 2.33 Å. The corresponding fitting gives the CN of 3.60 for the Fe–N4 and 1.80 for the Fe–N2 structure (Supplementary Fig. 10 and Supplementary Table 4). The Fe–N structures were further analyzed by wavelet transform of the EXAFS26. As shown in Fig. 2c, both FeN and FeN/Fe3N have the K values centered around 4.8 Å−1, while those values from Fe3N and FeN/Fe3N locate at 6.5 (FeN/Fe3N) and 7.5 Å−1 (FeN), closer to 8.2 Å−1 observed from the Fe foil. All these analyses confirm that the Fe3N structure has more metal character than the FeN structure.

Fig. 2: Fe-N NP coordination and electron structure characterization.
figure 2

a The normalized Fe K-edge EXAFS spectra. Inset: the local FeN structure showing tetrahedral coordination (right) and Fe3N structure showing octahedral coordination (left). b Fe K-edge FT-EXAFS in R space and c wavelet transforms for the k3-weighted EXAFS signals for FeN, FeN/Fe3N, Fe3N, and Fe foil. d The N 1 s XPS spectra, e the normalized area of Fe–N bond, f the Fe 2p XPS spectra, and g the normalized area of Fe3+/Fe1+ for FeN, FeN/Fe3N, and Fe3N, respectively.

The Fe–N network structure was characterized by Raman spectroscopy (Supplementary Fig. 11). The same D and G peak position with slightly different ID/IG values (1.26 for FeN/Fe3N, 1.15 for FeN, and 1.21 for Fe3N) suggest that more defects exist in the FeN/Fe3N structure32, which supports what is concluded from the BET N2-adsorption analysis. X-ray photoelectron spectroscopy (XPS) shows that the NP sample contains C (C 1 s spectra at 284.6 eV, Supplementary Fig. 12) and three different N’s (N 1 s spectra, Fig. 2d)33. The FeN/Fe3N interface structure has the largest Fe–N bonding area (Fig. 2e). Two spin-orbit doublets of Fe 2p can be found in Fe 2p spectra (Fig. 2f). The Fe 2p3/2 peaks shift from 711.3 eV (for FeN) to 708.6 eV (for Fe3N) while the peak for the FeN/Fe3N interface structure locates at 710.3 eV, indicating that the Fe valence state in the FeN/Fe3N structure is between FeN and Fe3N and electrons are drawn from Fe3N to FeN. Comparing with the normalized Fe−N bonding area for FeN and Fe3N, we calculated that the FeN/Fe3N interface structure contained 66% FeN and 34% Fe3N (Fig. 2g). The Fe binding environment was finally characterized by Mössbauer spectroscopy34,35, as shown in Supplementary Fig. 13. The fitting matches well with the experimental spectra, indicating that the FeN, FeN/Fe3N, and Fe3N structures have the Fe−N4, Fe−N4/Fe−N2, and Fe−N2 coordination, respectively.

The electrocatalytic performance of these Fe−N NP catalysts for CO2RR was evaluated by linear scan voltammetry (LSV, Supplementary Fig. 14) and cyclic voltammetry (CV, Supplementary Fig. 15) in an H-type cell filled with the Ar- and CO2-saturated 0.5 M KHCO3 electrolyte at room temperature36,37. The FeN/Fe3N structure is more responsive to the voltammetry tests, showing higher current density than either the FeN or Fe3N structure, suggesting that it initiates more CO2RR. Under a constant reduction potential, CO2RR products were characterized by gas chromatograph (GC) and nuclear magnetic resonance (NMR). In the current electrocatalytic reaction condition, the FeN/Fe3N structure is the most active catalyst for reducing CO2 to CO with the CO formation FE reaching 98% at −0.4 V and above 90% at −0.5 to −0.9 V (Fig. 3a). The H2 was another gas product (Supplementary Fig. 16) and no liquid products were detected (Supplementary Fig. 17). The FeN/Fe3N catalyst shows excellent CO selectivity compared with the recently reported catalysts (Supplementary Table 5). When the reduction potential more negative than −0.9 V, more HER was observed, which was further confirmed by DFT calculations that the H desorption energy was more than 0.9 eV on the FeN/Fe3N surface (Supplementary Fig. 18). This is also consistent with what was observed previously on *H binding energy increase at high potentials on Fe−N-based catalyst surfaces24,38. Based on the GC data from the closed reaction system, the single-pass CO2 conversion39 reached up to 56% at −0.8 V (Fig. 3b). The FeN/Fe3N catalyst is much more selective than N doped KJ carbon (CO FE is <20%), FeN (CO FE peaks at ~80% at −0.6 V) and Fe3N (CO FE peaks at ~70% at −0.6 V) (Supplementary Fig. 19), which indicates that the FeN/Fe3N interface structure is key to this CO selectivity enhancement. Its highest CO partial current density is 21 mA cmgeo−2 (Fig. 3c) and the corresponding mass activity is 400 mA mgFe−1 at −0.9 V, at which its turnover frequency (TOF, Supplementary Fig. 20) is 116 s−1 and turnover number (TON, Supplementary Fig. 20) is 99. The current density data were also normalized by geometric area of working electrode and CO2-BET area (Supplementary Fig. 21), all of which indicate the high catalytic performance of the FeN/Fe3N heterostructure. The FeN/Fe3N catalyst is superior to other representative Au, Ag, and M − N catalysts recently reported for catalyzing CO2RR to CO (Supplementary Table 5). The stability of the FeN/Fe3N catalyst was evaluated by time-dependent current change in 0.5 M KHCO3 at different potentials −0.4, −0.6, −0.8 and −0.9 V (Fig. 3e), which shows negligible decrease in both current density and CO FE in the 100 h testing period. It is worth to note that the CO FE stays above 90% during the tests, which is much better than the stability data obtained from other catalysts previously reported (Supplementary Table 5). We re-evaluated the FeN/Fe3N structure after the stability test. The XRD pattern (Supplementary Fig. 22a) still shows typical FeN and Fe3N peaks, the NP morphology is well-maintained in the TEM image (Supplementary Fig. 22b), and the interface structure is intact as evidenced by corresponding auto correlation and FFT images (Supplementary Fig. 22c). The XPS spectra of Fe 2p (Supplementary Fig. 23a) and N 1 s (Supplementary Fig. 23b), the EXAFS Fe K-edge spectra of the FeN/Fe3N (Supplementary Fig. 24a), the pre-edge peak in Fe K-edge (Supplementary Fig. 24a) and FT-EXAFS in R-space (Supplementary Fig. 24b) all indicate there is no obvious valence state change on Fe in the Fe−N structures, further confirming that the FeN/Fe3N heterostructure is well-preserved after the stability test.

Fig. 3: CO2RR performance of the FeN/Fe3N catalyst.
figure 3

a Reduction potential dependent FEs of CO and H2. b GC spectrum of the gas products upon a single pass CO2 conversion (Inset: reduction potential dependent CO2 conversion to CO). c CO partial current density and d corresponding mass activity. e CO2RR current density and CO FE changes at different reduction potentials over different electrolysis times. Error bars in ad represent the standard deviation of three independent measurements.

The evolution of the adsorbed CO2RR intermediates on Fe−N4, Fe−N2, and Fe−N4/Fe−N2 sites was studied by in situ electrochemical spectroscopies. The in situ electrochemical impedance spectroscopy (EIS)40,41 gives frequency-dependent changes of the Bode phase plots of the Fe−N-coordination sites, which is sensitive to the charge transfer and surface adsorption. The peaks of the FeN structure shift from middle frequency (100 − 102 Hz) to high-frequency (102 − 104 Hz) region (Fig. 4a) due to the low charge transport ability of the Fe−N4 coordination sites, while those of the Fe3N structure shift from high-frequency (102 − 104 Hz) to middle frequency region (100 − 102 Hz) (Fig. 4b) due to the poor surface adsorption of CO on the Fe–N2 coordination sites. In comparison, the FeN/Fe3N structure shows both FeN- and Fe3N-type Bode phase peaks, but these peaks shift to lower frequency (Fig. 4c) due to the optimized surface adsorption energy and enhanced charge transport ability from FeN3 to FeN.

Fig. 4: In situ spectroscopic studies of Fe-N catalysts and CO2RR.
figure 4

ac Bode phase plots of FeN a, Fe3N b, and FeN/Fe3N c at different reduction potentials. In situ DEMS plots showing the increased presence of CO with increased flowing CO2 gas d and CO signal under different potentials e of the FeN/Fe3N-catalyzed CO2RR. f The in situ SEIR spectra of the FeN/Fe3N for CO2RR.

We further conducted in situ differential electrochemical mass spectroscopy (DEMS) test to characterize products separated by the FeN/Fe3N-catalyzed CO2RR, from which the concentration of CO increases with the flowing CO2 gas (Fig. 4d), suggesting the detected CO is due to CO2RR. The CO signal in DEMS tests is enhanced from −0.2 to −0.8 V vs. RHE (Fig. 4e), indicating that the CO is the main product for CO2RR (Supplementary Fig. 25). We performed in situ spectra in the CO2-flowed 0.5 M KHCO3 electrolyte at various reduction potentials to characterize CO2 reduction intermediates. The *CO and *H competition was evidenced from in situ surface enhancement infrared spectroscopy (SEIR)42,43. According to the electrochemical Stark effect44,45, the vibrational frequencies of the intermediates adsorbed on an electrode usually shift negatively. As shown in Fig. 4f, there is no peak observed at ~1900 cm−1 from open circuit voltage (OCV) to −0.2 V. However, at −0.4 V, a small peak at ~1920 cm−1 starts to appear, and shows a negative shift with 49.66 cm−1 per voltage from −0.4 to −1.2 V, which attributes to the adsorbed CO* peak46,47,48,49. At more negative potentials, the intensity of the CO peak increases initially, but starts to drop after −1.2 V due to the strong competition of H* generated from the enhanced HER50. From the Raman spectra (Supplementary Fig. 26a), the electrolyte, CO2, CO, and H2O peaks can be found. The two peaks for HCO3 at ~1010 cm−1 and CO32− at ~1060 cm−1 are found from the 0.5 M KHCO3 electrolyte51, while the ~1920 cm−1 peak has a negative shift from −0.4 to −0.8 V, which corresponds to the stretching vibration mode of the adsorbed CO45 (Supplementary Fig. 26b). In situ Raman spectra of CO2 adsorption peaks at ~2330 cm−1 for FeN (Supplementary Fig. 26c), Fe3N (Supplementary Fig. 26d), and FeN/Fe3N (Supplementary Fig. 26e) also intensify with enhanced voltage from OCV to −0.8 V vs. RHE (Supplementary Fig. 26f), demonstrating that the FeN/Fe3N heterostructure has an optimized interfacial adsorption ability.

The formation of CO from CO2RR is generally believed to follow two proton-coupled electron-transfer steps to form COOH*, and then CO* that is desorbed from the catalyst surface52,53,54. In this process, the formation of COOH* usually is a potential-limiting step due to the high activation energy of CO2 molecule. To understand the Fe−N catalysis better, we used DFT to calculate the related activation energy on the catalyst surfaces. We compared the electronic density of states (DOS) and d-band center of Fe3N, FeN, and FeN/Fe3N structures55,56,57. As shown in Fig. 5a, the d-band center of the FeN/Fe3N heterostructure is at −0.506 eV, while that of FeN is at −0.523 eV and Fe3N at −0.500 eV, suggesting Fe3N is beneficial to improve the overall d-band center shift to the Fermi level (EF). Additionally, the d-band center of pure FeN (100), pure Fe3N (001), FeN cell, and Fe3N cell is −0.591, −0.503, −0.647, and −0.502 eV (Supplementary Fig. 27), further demonstrating Fe3N with rich electron is key for built-in electric field from Fe3N to FeN. As a result, the intensity of DOS (Supplementary Fig. 28), and the integral DOS (IDOS) (Supplementary Fig. 29) of the FeN/Fe3N structure is higher than that of the FeN or Fe3N structure, suggesting that electrons are more populated at EF level on the FeN/Fe3N surface for effective binding of reaction intermediates. The FeN/Fe3N surface potential change upon binding to COOH and CO was calculated by electrostatic potentials and shown as colored contours in Fig. 5b (for *COOH) and Fig. 5c (for *CO), which shows that electron transfer causes a decrease in the potential of CO and COOH (red means lower electrostatic potential and blue means higher electrostatic potential). The corresponding 3D electron density distributions on the FeN/Fe3N (Fig. 5d, e), FeN (Supplementary Fig. 30c, d), and Fe3N (Supplementary Fig. 31c, d) surfaces upon their binding to COOH and CO suggest that electrons are transferred from catalyst to COOH/CO. The electron localization of COOH and CO on the FeN, Fe3N, and FeN/Fe3N surface calculated by electron localization function is about 0.8, 0.7, 0.8 (Supplementary Fig. 30e, Supplementary Fig. 31e and Fig. 5f) for COOH and 0.7, 0.7, 0.7 (Supplementary Fig. 30f, Supplementary Fig. 31f and Fig. 5g) for CO, indicating that COOH and CO interact covalently with the catalyst surface58,59,60. Following the common CO2RR reaction pathway specifically on the FeN/Fe3N surface (Fig. 5h), we calculated the Gibbs free energy (ΔG) for the potential-limiting step (CO2ad to *COOH) on the FeN/Fe3N structure to be 0.317 eV (Fig. 5i), much lower than that on the FeN (0.540 eV) and Fe3N (1.10 eV) in the heterostructure. We also calculated the CO2RR performance on the other side of FeN/Fe3N heterostructure on the interface Fe3N (001) (Supplementary Fig. 32a) and other side of Fe3N (001) (Supplementary Fig. 32b). The ΔG for CO2 to COOH is 0.985 eV on the interface Fe3N (001), ΔG for *CO on other side of Fe3N (001) is −2.566 eV, which is too stable for the CO to dissociate from the plane. ΔG for CO2 to COOH is about 0.360 and 1.08 eV on the pure FeN (100) (Supplementary Fig. 32c) and pure Fe3N (001) (Supplementary Fig. 32d), indicating that CO2 adsorption and activation on the FeN (100) is more favorable for CO2RR. We further constructed the FeN (110)/Fe3N (001) heterostructure to reveal the role of the FeN (110) plays on catalyzing CO2RR. The electron cloud around N and Fe in the Fe3N region (yellow) is much more enriched than that in the FeN region (blue) (Supplementary Fig. 33a), while is also supported by the 2D electron localization function (Supplementary Fig. 33b). These indicate that the FeN (110)/Fe3N (001) structure has no electron transport across interface and is not active for CO2RR. Then, the ΔG for CO2ad to *COOH on the FeN (110)/Fe3N (001) structure is 1.189 eV (Supplementary Fig. 33c) much larger than that of FeN (100)/Fe3N (001) heterostructure, further demonstrating that FeN (100)/Fe3N (001) heterostructure is superior for CO2RR. Our calculations support what we concluded from both structure and catalysis analysis of the Fe−N catalysts that the FeN (100)/Fe3N (001) interface with the built-in electric field has abundant electrons closer to EF for CO2 adsorption and activation, showing much enhanced catalysis for CO2RR to CO.

Fig. 5: DFT calculations on the Fe−N catalyzed CO2RR.
figure 5

a The electronic density of states and d-band centers of Fe in the FeN, Fe3N, and FeN/Fe3N structures. The electrostatic potential for adsorption of COOH b and CO c on the FeN/Fe3N interface. 3D electron density difference distributions for adsorption of COOH d and CO e on the FeN/Fe3N interface. The electron localization for adsorption of COOH f and CO g on the FeN/Fe3N interface. h Schematic reaction CO2RR to CO pathway projected on the FeN/Fe3N surface. i The calculated Gibbs free energy diagrams for CO2RR on the FeN, Fe3N, and FeN/Fe3N surfaces.


In summary, we have reported a process for synthesizing heterostructured FeN/Fe3N NPs with the NP surface populated with Fe–N4/Fe–N2 interfaces. Electron polarization from Fe3N (Fe−N2 site) to FeN (Fe−N4 site) builds the desired electric field across the FeN/Fe3N interface, enhancing its power to adsorb CO2. In the electrochemical reduction condition, this FeN/Fe3N structure has an excellent mass and electron transfer ability in catalyzing CO2RR to CO with high FE (98% at −0.4 V) and much improved stability in the 100 h electrolysis period (FE > 90% in the broad −0.4 to −0.9 V potential range). The catalyst surface adsorption of CO2 and its activation intermediates is further characterized by various in situ spectroscopy techniques (EIS, DEMS, IR and Raman) from which the adsorbed CO* is identified. DFT calculations suggest that the FeN/Fe3N structure has a d-band center that is closer to its EF level where there are abundant electrons for binding to reaction intermediates (CO2ad, COOH*, CO*). This strong interaction reduces the ΔG of the CO2ad to COOH* conversion and provides a necessary force to lower the reduction overpotential and improve the catalytic efficiency. Our work demonstrates that creating a polarized field on the NP surface is a promising approach to the design of highly active catalysts for electrochemical CO2RR.


Synthesis of Fe−N based precursor

Firstly, the commercial Ketjen (KJ) carbon (0.1 g) was suspended in water (H2O, 50 mL) by ultrasound. Then, a mixture of iron (III) nitrate hexahydrate (Fe(NO3)3•9H2O, 0.01 mol, 4.04 g) and urea (CH4N2O, 0.1 mol, 6.00 g) was added into the carbon suspension and continued ultrasound for 8 h. Finally, the precipitate was separated by centrifuge (6000 rpm (4830 × g), 5 min) and washed with water and ethanol for three times and dried under vacuum at 60 °C for 6 h to give the Fe−N based precursor.

Synthesis of FeN, FeN/Fe3N, and Fe3N nanosparticles

100 mg Fe−N precursor was placed in crucible and put in the tube furnace (Supplementary Fig. 1). Then the tube was heated to 300, 500 or 700 °C with a rate of 5 °C min−1 for 2 h under the flowing NH3 atmosphere. Finally, the system was cooled to room temperature under a flowing NH3 gas. In this process, FeN, FeN/Fe3N and Fe3N were obtained at 300 °C, 500 °C, and 700 °C, respectively33.

Catalyst characterization

X-ray diffraction (XRD) experiments were conducted from 20° to 90° on an X’Pert ProX-ray diffractometer with Cu Ka radiation (λ = 0.1542 nm) under a voltage of 40 kV and a current of 40 mA. Sample compositions were determined by ICP-AES (HITACHI P-4010, Japan). Product morphology was imaged by field-emission scanning electron microscopy (FESEM, Zeiss) at an acceleration voltage of 5 kV. All samples were coated with a thin layer of gold prior to FESEM observations. More detailed product morphology and structure were analyzed by transmission electron microscopy (TEM) and high-resolution transmission electron microscopy (HRTEM) under an acceleration voltage of 200 kV with a JEOL JEM 2100 TEM. Atomic-scale STEM images were recorded on a probe aberration-corrected STEM (Cubed Titan G2 60-300, FEI, USA) operated at 300 kV. Product surface composition was analyzed by X-ray photoelectron spectroscopy (XPS) with a VG ESCALAB 220I-XL device. All XPS spectra were corrected using C1s line at 284.6 eV. The absorption spectra of Fe K-edge were collected in transmission mode using a Si (111) double-crystal monochromator at the BLW141 station of the Beijing Synchrotron Radiation Facility (BSRF). Transmission 57Fe Mössbauer spectra were recorded by using a conventional spectrometer working in constant acceleration mode. A 50 mCi of 57Co embedded in a Rh matrix moving at room temperature was used as the γ-ray source. The absorber was prepared with a surface density of ~8 mg cm−2 natural iron. The drive velocity was calibrated with sodium nitroprusside at room temperature and all the isomer shifts quoted in this work are relative to that of the α-Fe.

Lattice mismatch calculation

The lattice mismatch for the heterostructure can be calculated by lattice of different domains, as shown below:


where δ is lattice mismatch, a1 and a2 are the lattice parameters of different domains. We define that two lattices are well-matched when δ is <5%, semi-matched if δ is between 5% and 25%, and totally mismatched if δ is >25%. The δ value for the FeN/Fe3N heterostructure is 5.12% (δFeN/Fe3N = (2 × |aFeN – aFe3N | )/(aFeN + aFe3N) = (2 × |0.19 – 0.20 | )/(0.19 + 0.20) = 5.12%), therefore, the FeN and Fe3N lattices are well-matched, which is favorable for electron transfer at the interface.

Electrochemical test for CO2RR

Electrochemical measurements were carried out at room temperature using the three-electrode system directly connected to a CHI 760 E Electrochemical Workstation (CHI Instruments, Shanghai Chenhua Instrument Corp., China). The CHI workstation was used to conduct CO2 reduction experiments in aqueous 0.5 M KHCO3 (pH = 6.8 when saturated with CO2, pH = 8.3 when saturated with Ar). A platinum wire was used as counter electrode. All potentials were measured against an Ag/AgCl reference electrode (4.0 M KCl, Pine instrument) and were calibrated against a reversible hydrogen electrode (RHE). The experiments were performed in a gas-tight cell with two-compartments separated by an anion exchange membrane (Nafion® 212). Each compartment contained 10 mL electrolyte with approximately 10 mL headspace6,37.

Product analysis

CO2 gas was delivered at an average rate of 30 mL min−1 (at room temperature and ambient pressure) and routed directly into the gas sampling loop of a gas chromatograph (Agilent 7890 A). The gas phase composition was analyzed by GC every 30 min. The GC analysis was set up to split the gas sample into two aliquots whereof one aliquot was routed through a packed MoleSieve 5 A column and a packed HP-PLOT Q column before passing a thermal conductivity detector (TCD) for H2 quantification. Argon (Corp Brother, 99.9999%) and Helium (Corp Brother, 99.9999%) were employed as carrier or make-up gases respectively. The second aliquot was routed through a packed HP-PLOT Q + PT column equipped with a flame ionization detector (FID) for analyzing CO and C1 to C3 hydrocarbons. The GC was calibrated using commercially available calibration standards from JJS Technical Services. 1H-NMR spectroscopy was employed to characterize and quantify CO2 reduction products at the end of the experiments. 1H-NMR spectra were recorded on Bruker DRX 400/600 Avance. A total of 0.5 ml electrolyte and 0.1 ml D2O (content 0.05 μL dimethyl sulfoxide (DMSO), as inner standard) were mixed together to get the NMR signal.

Theoretical calculations

The first-principles density functional theory (DFT) calculations were performed within the generalized gradient approximation (GGA) based on via Vienna abinitio simulation package (VASP)61,62,63. We constructed FeN/Fe3N interface structure, FeN and Fe3N surface model with periodicity in the x and y directions and the depth of the vacuum layer is greater than 20 Å in order to prevent self-interactions. In addition, the bottom two stoichiometric layers were need to be fixed while the top layer and adsorbates (CO*, COOH*) were allowed to relax in the optimization of the FeN/Fe3N and Fe3N, for FeN, only one stoichiometric layer needs to be fixed. The VASP was employed to perform all DFT calculations within the functional as proposed by Perdew-Burke-Ernzerh (PBE)61,64 of interactions is represented using the projector augmented wave (PAW)61,65,66 potential. The Kohn-Sham one-electron valence states were expanded on the basis of plane waves with a cutoff energy of 400 eV in the in the process of structure optimization. A geometry optimization was considered convergent when the electronic energy and Hellmann-Feynman forces convergence criterion was smaller than 10−5 eV and 0.03 eV Å−1, respectively. The K-point of 3 × 1 × 1 was used for the optimization of the FeN/Fe3N, FeN and Fe3N surface, and the corresponding adsorption structure (CO*, COOH*). Subsequently, we calculated the charge density, electron localization function (ELF), work function and electrostatic potential in the self-consistent process. The electronic energy was considered self-consistent while the energy variation was smaller than 10−5 eV. The value of K-point, cutoff energy is the same as the process of structure optimization. After the self-consistent calculation converges, the charge density distribution, electron localization, electrostatic potential and work function can be obtained, the self-consistent structure is used as the input file, and the wave function and charge density file are read to obtain the density of states of the heterojunction, FeN and Fe3N.

The d-band center was used to understand the adsorption capacity of the different catalyst (FeN/Fe3N, FeN and Fe3N) surface for CO2. The calculation formula of d-band center:

$${\varepsilon }_{d}=\frac{{\int }_{-\infty }^{\infty }{n}_{d}(\varepsilon )\varepsilon d\varepsilon }{{\int }_{-\infty }^{\infty }{n}_{d}(\varepsilon )d\varepsilon }$$

where nd(ɛ) represents the density of states and ɛ represents the energy of state.

The Gibbs free energy also was calculated in this study. The zero-point energy (ZPE) correction was performed referring to the approaches previously reported. In the DFT process, we calculated the Gibbs free energy according to the equations as follow:61


Where G0 is the Gibbs free energy, EDFT is total free energy, ZPE is the vibration energy; TS0 is the entropy change (T = 298.15 K).

TOF and TON calculation for FeN/Fe3N catalyst

The turnover frequency (TOF) was calculated by the following formula:67

$${{{{{\rm{TOF}}}}}}=\frac{j\times {{{{{\rm{S}}}}}}\times {{{{{\rm{FE}}}}}}}{2\times {{{{{\rm{n}}}}}}\times {{{{{\rm{F}}}}}}}$$

where j (mA cm−2) is the current density at −0.4 V, S is the surface area of the working electrode, FE is the Faraday efficiency of CO, the number 2 means 2 electrons mol–1 of CO, F is Faraday’s constant (96485.3 C mol–1), and n is the moles of coated Fe atoms on the electrode. The Fe content of catalysts was measured by the ICP-AES measurement.

Similarly, the turnover number (TON) of the catalyst was calculated based on:

$${{{{{\rm{TON}}}}}}={{{{{\rm{TOF}}}}}}\times {{{{{\rm{FE}}}}}}\times 0.01$$