Atomically dispersed Fe3+ sites catalyze efficient CO2 electroreduction to CO

See allHide authors and affiliations

Science  14 Jun 2019:
Vol. 364, Issue 6445, pp. 1091-1094
DOI: 10.1126/science.aaw7515

Three's a charm for iron and CO2

Large-scale electrochemical reduction of CO2 to CO could be a promising first step in sustainable conversion of the greenhouse gas to commodity chemicals. Currently, gold and silver are the most active catalysts for this process, whereas more abundant, less expensive metals tend to require impractically high potentials. Jun Gu et al. now report an iron catalyst with activity equaling or exceeding that of the precious metals. The key proved to be stabilization of the dispersed single iron ions in the +3 oxidation state.

Science, this issue p. 1091


Currently, the most active electrocatalysts for the conversion of CO2 to CO are gold-based nanomaterials, whereas non–precious metal catalysts have shown low to modest activity. Here, we report a catalyst of dispersed single-atom iron sites that produces CO at an overpotential as low as 80 millivolts. Partial current density reaches 94 milliamperes per square centimeter at an overpotential of 340 millivolts. Operando x-ray absorption spectroscopy revealed the active sites to be discrete Fe3+ ions, coordinated to pyrrolic nitrogen (N) atoms of the N-doped carbon support, that maintain their +3 oxidation state during electrocatalysis, probably through electronic coupling to the conductive carbon support. Electrochemical data suggest that the Fe3+ sites derive their superior activity from faster CO2 adsorption and weaker CO absorption than that of conventional Fe2+ sites.

Electrochemical reduction of carbon dioxide (CO2) is a promising approach to store intermittent renewable solar and wind energy in carbon-based fuels and chemicals, leading to reduced anthropogenic CO2 emission (1). To achieve high energy efficiency and scalability, the reaction must occur rapidly and selectively at low overpotentials. Numerous electrocatalysts have been developed for CO2 reduction (2), among which gold (Au) and, to a lesser degree, silver (Ag) are the most efficient at low overpotentials. For example, the Faradaic efficiency of carbon monoxide (CO) formation can exceed 90% with Au- (35) and Ag-based (6, 7) catalysts. On certain Au nanostructures, the partial current density of CO (denoted as jCO) reached 10 mA cm−2 at overpotentials even lower than 300 mV (4, 5). Catalysts composed solely of Earth-abundant elements typically have low selectivity for CO2 reduction (812). Recently, many single-atom catalysts (13, 14) have been developed, in which numerous catalytic metal sites separated from each other were chemically and electronically constrained on solid supports. These catalysts exhibit properties and activity distinct from both nanoparticles and molecular complexes of the same metal elements. Among them, iron (Fe) (15, 16), cobalt (Co), (17) and nickel (Ni) (1820) catalysts were reported to exhibit Faradaic efficiency of CO formation comparable with those of Au and Ag catalysts. However, with these non–precious metal catalysts, much larger overpotentials were required to obtain the same jCO. Here, we report a catalyst with dispersed single-atom Fe sites with ultrahigh activity for CO2 electroreduction to CO.

The Fe catalyst (Fe3+–N–C) was prepared through the pyrolysis of Fe-doped zinc (Zn) 2-methylimidazolate framework (ZIF-8) (21) under N2 at 900°C. The precursor adopts the same crystal structure as that of undoped ZIF-8 (fig. S2A), with a mole ratio of Fe:Zn of 4:96 (fig. S2E). Fe ions occupy Zn sites and are coordinated by four pyrrolic-type nitrogens (N), as revealed by the fitting of the Fe K-edge extended x-ray absorption fine structure (EXAFS) spectrum (fig. S2, G to I, and table S1). Fe3+–N–C is porous, with a Brunauer-Emmett-Teller surface area of 772 m2 g−1 (fig. S3A) and an electrochemical (double-layer) surface area of 554 m2 g−1 (fig. S3D). The porosity was confirmed by means of high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) (Fig. 1A). Inductively coupled plasma optical emission spectrometry (ICP-OES) analysis showed the weight fractions of Fe and Zn to be 2.8 and 3.4%, respectively, corresponding to a nearly equal mole ratio of Fe:Zn. A similar Fe:Zn mole ratio was found by means of x-ray photoelectron spectrometry (XPS) (fig. S3E) and energy dispersive x-ray spectroscopy (EDS) (fig. S4F) measurements. The majority of Zn ions in the ZIF-8 precursor were presumably reduced to Zn particles, which then evaporated during pyrolysis. The x-ray diffraction (XRD) pattern of Fe3+–N–C (fig. S3G) showed a broad feature at ~25° corresponding to the interlayer distance of the carbon matrix (with a d value of ~0.35 nm). No diffraction peaks of any crystalline species of Fe and Zn were observed. Likewise, no nanoparticles were found in the transmission electron microscopy (TEM) and high-resolution TEM (HRTEM) images; only curved fringes of the layered carbon matrix were observed (fig. S4, D and E). The EDS mappings (Fig. 1, B and C) revealed the homogeneous distributions of Fe and N in the carbon matrix. In the aberration-corrected HAADF-STEM image with atomic resolution (Fig. 1D), the bright spots with size of ~0.2 nm correspond to atomically dispersed Fe and Zn sites. The Fe 2p3/2 XPS spectrum (fig. S3F) and the Fe K-edge x-ray absorption near-edge structure (XANES) spectrum (Fig. 1F) showed binding and edge energies close to those of Fe2O3 and Fe3+-tetraphenylporphyrin-Cl (Fe3+TPPCl), indicating that the Fe ions in the as-synthesized Fe3+–N–C were in the +3 oxidation state. Thus, the Fe ions were oxidized from +2 to +3 during the pyrolysis, which is in agreement with previous reports of pyrolysis of Fe-containing organic precursors (15, 22). The oxidants might be of the same species as protons or residual oxygens that oxidized the carbon skeleton of the ZIF precursor. Fe K-edge EXAFS (Fig. 1H) supported the atomic dispersion of Fe sites in Fe3+–N–C. The fitting of the spectrum (table S1) indicated that the Fe center adopts a planar Fe–X4 (X = N or C) structure. The average coordination numbers of Fe–N and Fe–C paths were 3.4 and 0.5, respectively. No Fe–Fe bond was detected.

Fig. 1 Characterizations of Fe3+–N–C.

(A) HAADF-STEM image and the corresponding EDS mappings of (B) Fe and (C) N of the region enclosed by the red square. (D) Aberration-corrected HAADF-STEM image and (E) EDS spectrum of the red square region. (F) Fe K-edge XANES spectra of Fe3+–N–C (black), Fe2O3 (blue dashed), Fe3+TPPCl (green dashed), FeO (pink dashed), and Fe foil (orange dashed). (Inset) The enlargement of the main edges. (G) k-space and (H) R-space Fe K-edge EXAFS spectra. Shown are data (black) and fitting curves (red).

As shown by the linear sweep voltammetry (LSV) curve of Fe3+–N–C in CO2-saturated 0.5 M potassium bicarbonate (KHCO3) electrolyte (fig. S5), the onset potential was more positive than –0.20 V versus reversible hydrogen electrode (RHE). Compared with Fe3+–N–C, the current density of the Fe-free control sample, Zn–N–C (prepared by pyrolysis of undoped ZIF-8), was negligible, indicating that the electrocatalytic activity of Fe3+–N–C originates from Fe sites. We first tested the electrocatalytic activity in CO2 reduction using carbon paper electrodes in an H-cell (fig. S6A). CO and H2 were the only gas-phase products, and no solution-phase product was detected (fig. S6, B to D). CO was detected after electrolysis at –0.19 V versus RHE, equivalent to an overpotential of 80 mV (fig. S6, E and F). The Faradaic efficiency of CO was higher than 80% between –0.2 and –0.5 V versus RHE (Fig. 2A). The jCO reached 20 mA cm−2 at –0.47 V versus RHE (overpotential of 360 mV) (Fig. 2B). The rate of CO2 reduction might be limited by mass transport in an H-cell (23). Thus, we deposited Fe3+–N–C on a gas diffusion electrode (GDE) (24). The electrolysis was then conducted in a flow cell with N2-saturated 0.5 M KHCO3 as the catholyte, and CO2 gas was fed behind the GDE (fig. S7A). Ni–Fe layered double hydroxide (LDH) nanosheets (25) were used as the electrocatalyst for the anodic reaction (oxygen evolution). At –0.45 V versus RHE (overpotential of 340 mV), jCO reached 94 mA cm−2 (corresponding to 1.75 mmolCO hour−1 cm−2) (Fig. 2B), with Faradaic efficiency of CO on the cathode higher than 90% (Fig. 2A). Good reproducibility was observed in measurements of three independently prepared samples, giving a small standard deviation (Fig. 2, A and B).

Fig. 2 CO2 electroreduction performance.

(A) Faradaic efficiency of CO (solid lines) and H2 (dashed lines) production and (B) jCO of Fe3+–N–C in an H-cell (red) and on a GDE (blue), and of Fe2+–N–C in an H-cell (black). Data from the H-cell were obtained by means of chronoamperometry, whereas data from the GDE were obtained by means of chronopotentiometry. Each error bar was the standard deviation determined based on tests of three individual electrodes. Loading was 0.6 mg cm−2 for Fe3+–N–C and Fe2+–N–C; 2.5 mg cm−2 for Fe3+–N–C/DGE. (C and D) Comparison of (C) jCO and (D) apparent TOFs of CO production of Fe3+–N–C in an H-cell (red circles) and on a GDE (red stars) and of Fe2+–N–C in an H-cell (red squares), to that of other reported catalysts: other Fe-N-C catalysts [Fe-0.5d (15) and Fe–N–C (16)], a Co–N–C catalyst with two coordinating nitrogen atoms (Co–N2) (17), atomically dispersed Ni on nitrogen-sulfur codoped graphene (A–Ni–NSG) (19), oxide-derived Au electrode (OD-Au) (3), carbon black supported Au nanowires with a length of 500 nm (C–Au-500) (4), needle-shape Au nanostructures (Au needles) (5), nanoporous Ag electrode (np-Ag) (7), and Au-polymer-multiwall carbon nanotubes composite loaded on GDE (Au/GDE) (34) in bicarbonate electrolytes. (E) Chronoamperometry curve and Faradaic efficiency of CO production (dots) by Fe3+–N–C in H-cell at –0.37 V versus RHE. The electrolytes were prepared from potassium carbonate (K2CO3) (99.999%) and deionized water (18.2 megohms cm) (black), KHCO3 (99.5%) and deionized water (red), and KHCO3 (99.5%) and tap water (blue), respectively.

We compare the jCO of Fe3+–N–C to other state-of-the-art catalysts in Fig. 2C and fig. S8. The jCO of Fe3+–N–C measured in an H-cell between –0.2 and –0.5 V versus RHE is considerably higher than that attained by other non–precious metal catalysts and even Ag catalysts (7), reaching comparable levels with those of oxide-derived Au catalysts (3). The mole-normalized current of Fe3+–N–C is significantly higher than that of Au catalysts in this potential range (fig. S8B). Assuming all Fe atoms to be catalytically active, the apparent turnover frequencies (TOFs) of Fe3+–N–C (Fig. 2D) are comparable with those of Au catalysts (3, 4, 26) and greatly exceed those of other non–precious metal catalysts (15, 16, 19). Unlike for catalysts based on copper (Cu), Ag, and Au (27), ultrapure electrolyte solutions were not necessary for Fe3+–N–C. When KHCO3 with a purity of 99.5% was used to prepare electrolyte, or even tap water was used in place of deionized water (18.2 megohm cm), the Faradaic efficiency and jCO show no obvious change, and the performance was stable for at least 12 hours (Fig. 2E). After 12 hours of electrolysis, the weight fraction of Fe in Fe3+–N–C (measured with ICP-OES) was 2.6%, indicating no appreciable leaching of Fe ions from the catalyst. No aggregation of Fe or Zn species was detected in TEM images, and a high density of discrete Fe and Zn atoms was still observed (fig. S9).

The performance of Fe3+–N–C was stable between –0.2 and –0.5 V versus RHE, although at potentials more negative than –0.5 V versus RHE, the activity became unstable (fig. S7B). As shown in fig. S7C, the current density at –0.41 V was stable during a 28-hour chronoamperometry test. Whereas at –0.51 V versus RHE, the initial jCO was much higher, it decreased rapidly to a value similar to that obtained at –0.41 V versus RHE. This result indicates some changes of Fe3+–N–C around –0.5 V versus RHE. To explore the nature of this change, we conducted operando XAS measurements in the CO2-saturated 0.5 M KHCO3 catholyte. Fe K-edge spectra were obtained on dry samples and on samples that were loaded on glassy carbon electrodes and immersed in the electrolyte at open circuit potential (OCP) as well as at –0.1 to –0.6 V versus RHE (Fig. 3A). For Fe3+–N–C, the Fe K-edge showed no obvious shift between the dry powder and the in situ sample at –0.4 V versus RHE. The edge energy was close to that of Fe3+TPPCl, indicating that the Fe ions in Fe3+–N–C remained in the +3 oxidation state during CO2 electroreduction at potentials as negative as –0.4 V versus RHE. When the applied potential was shifted further negative, to –0.5 V versus RHE and beyond, the Fe K-edge shifted to lower energies, which were comparable with that of FeO, suggesting the reduction of Fe3+ to Fe2+. This reduction process occurred at the same potential as the above-mentioned deactivation of Fe3+–N–C, implying that Fe3+ sites are more active for generating CO. Moreover, the fitting of EXAFS spectra (fig. S10C) indicates that the reduction of Fe3+ sites is accompanied by a change of local structure around the Fe ions. Before the reduction of Fe3+ sites, the first shell coordination number of Fe (Fe–N and Fe–C) was about 4, whereas as the Fe3+ sites were reduced to Fe2+ sites, the first shell coordination number of Fe decreased to about 3.

Fig. 3 Operando XAS characterization.

(A and B) Fe K-edge XANES spectra (left) and the first derivative of the spectra (right) of (A) Fe3+–N–C and (B) Fe2+–N–C as dry powder (black) and loaded on glassy carbon electrodes at open circuit potential (OCP) (blue), –0.1 V (light blue), –0.2 V (green), –0.3 V (dark green), –0.4 V (dark blue), –0.5 V (red), and –0.6 V (pink) versus RHE, with the spectra of Fe2O3 (blue dashed), Fe3+TPPCl (green dashed), FeO (pink dashed), and Fe foil (orange dashed) as references.

To investigate the origins of the improved activity of Fe3+–N–C as compared with previously reported single-atom Fe catalysts, we measured in situ Fe K-edge XANES spectra of Fe0.5d (fig. S11, E and F) (15). Under potentials between –0.2 and –0.5 V versus RHE, the energy of the Fe K-edge was close to that of FeO, indicating a +2 rather than +3 oxidation state for the Fe sites during CO2 reduction. This difference in oxidation state might be due to different ligand environments, particularly with regard to the N atoms. For Fe0.5d (15) and Fe–N–C (16), Fe ions coordinated with four pyridinic N were proposed as the active sites. For Fe3+–N–C, XANES and XPS spectra of N indicate that the Fe ions were coordinated to pyrrolic N. In the N K-edge XANES spectrum (fig. S12A), π* and σ* features of Fe3+–N–C were similar to those of a metal-porphyrin derivative (28). In the N 1s XPS spectrum (fig. S12B), the major peak at 398.6 eV was attributed to pyrrolic N coordinated to Fe, which is in agreement with the spectrum of Fe3+TPPCl. This assignment is consistent with the atomic fractions of N and metals and the coordination number of metal-N (table S4).

To further test the above hypothesis, we directly compared the Fe3+–N–C catalyst with an analogous Fe–N–C catalyst in which the Fe ions were coordinated by pyridinic N atoms. Considering that pyrolysis of Fe precursors containing either pyridinic ligands or pyrrolic ligands seemed to conserve the pyridinic or pyrrolic nature of the N atoms, we prepared the reference sample (Fe2+–N–C) by pyrolysis of a composite containing a Fe-phenanthroline complex at 700°C (22). Aberration-corrected HAADF-STEM (fig. S13D) and Fe K-edge EXAFS (fig. S14E and table S1) confirmed the single-atom nature of the Fe sites in Fe2+–N–C. In the N 1s XPS spectrum (fig. S12C), the major feature at 399.7 eV was assigned to pyridinic N coordinated to Fe, which is in agreement with the assignments of the spectra of Fe-phenanthroline complexes and previously reported metal-N-C catalysts with pyridinic N ligands (16). This assignment is also consistent with the percentage of coordinated N measured with other methods (table S4). Thus, Fe2+–N–C contained Fe ions coordinated by pyridinic N atoms. A XANES spectrum (fig. S14D) showed that initially, the energy of Fe K-edge of Fe2+–N–C was considerably higher than that of FeO. The Fe 2p XPS (fig. S14C) spectrum showed that the binding energy of Fe ion was similar to that of Fe2O3. These data suggested an important number of Fe3+ sites in the as-prepared sample of Fe2+–N–C. The in situ XANES (Fig. 3B) showed that Fe3+ in the as-prepared Fe2+–N–C started to be reduced to Fe2+ at –0.1 to –0.2 V versus RHE. During CO2 electroreduction (at potentials more negative than –0.2 V versus RHE), the energy of the Fe K-edge was slightly lower than that of FeO. Thus, the Fe sites in Fe2+–N–C under reaction conditions had an oxidation state of +2 or lower. The TOF of CO production of Fe2+–N–C is more than an order of magnitude lower than that of Fe3+–N–C under the same potential (Fig. 2D). The current density of Fe2+–N–C decreased markedly during 2-hour choronoamperometry tests (fig. S6G), indicating its lower stability as compared with the Fe3+–N–C catalyst. These data suggest that pyrrolic type ligands are important to keep Fe sites in the +3 oxidation state during CO2 electroreduction and consequently maintain the high activity and stability of Fe3+ sites. This hypothesis is further supported by the different reactivity of Fe3+–N–C and Fe2+–N–C toward NaBH4. The Fe3+ ions coordinated by pyridinic N ligands in Fe2+–N–C could be reduced by NaBH4, whereas those coordinated by pyrrolic N ligands in Fe3+–N–C could not (fig. S15).

For Fe2+–N–C, the jCO at a fixed potential versus the standard hydrogen electrode (SHE) is largely independent of the concentration of the proton donor (fig. S16, A and B), indicating that the 1-electron reduction (adsorption) of CO2 is decoupled from a proton transfer (29, 30). At modest overpotentials, the jCO of Fe2+–N–C has a Tafel slope of 117 mV/decade (fig. S16E), suggesting that CO2 adsorption is slow and rate-limiting (supplementary materials, kinetic and mechanistic analysis). On the other hand, the jCO of Fe3+–N–C is approximately first-order in the concentration of HCO3 (fig. S16D) and has a Tafel slope of 64-71 mV/decade at low overpotentials (fig. S16E). These kinetic data suggest that for Fe3+–N–C, the 1 electron reduction of CO2 is also decoupled from a proton transfer. Moreover, CO2 adsorption is fast, and the rate-limiting step is the protonation of the adsorbed CO2 to form an adsorbed COOH intermediate (supplementary materials, kinetic and mechanistic analysis). These results indicate CO2 adsorption as a descriptor of catalytic activity at low to modest overpotentials. They also reveal a faster CO2 adsorption in Fe3+–N–C than in Fe2+–N–C, which explains why Fe3+–N–C has a lower onset overpotential. The CO2 electroreduction was conducted in the presence of CO (0.2 atm) (fig. S17). External CO did not influence the activity of Fe3+–N–C but largely decreased the activity of Fe2+–N–C. This result suggests that at high overpotentials, CO desorption becomes rate limiting for Fe2+–N–C. Because CO desorption is a non-Faradaic step, once it becomes rate limiting, the rate will hardly increase with increasing overpotentials. At higher overpotentials, the Tafel slope of Fe2+–N–C becomes enormous (546 mV/decade) (fig. S16E), and jCO cannot exceed 2 mA cm−2. On the other hand, the reaction at Fe3+ sites was not limited by CO desorption and could reach a very high current density. Thus, the higher activity of Fe3+–N–C compared with Fe2+–N–C at high overpotentials can be rationalized by a weaker CO binding at an Fe3+ center than at an Fe2+ center.

The spectroscopic data indicate that Fe3+–N–C comprises pyrrolic N ligands, whereas Fe2+–N–C comprises pyridinic N ligands. The pyrrolic N ligands may stabilize Fe3+ relative to Fe2+, whereas the pyridinic N ligands have the opposite effect. Thermodynamically, the respective reduction potentials support this hypothesis: The standard reduction potential of [Fe(phen)3]3+/2+ is 1.06 V versus SHE, whereas that of Fe3+/2+ couple in Fe-porphyrin complexes can reach as low as –0.4 V versus SHE (31). Preservation of the +3 oxidation state during CO2 electroreduction is counterintuitive because the formation of highly reduced, low-valent Fe species is necessary for molecular Fe catalysts (32). Once conjugated to a conductive carbon matrix, however, the Fe3+ site is electronically coupled to the conductive support so that the Fe3+/2+ reduction potential moves to the same degree as the Fermi level of the carbon support when applying an external bias (fig. S18, A and B). The Fe3+/2+ reduction potential remains more negative than the Fermi level of the carbon support, leading to the stabilization of the Fe3+ ions. An analogous “strong-coupling” effect was formulated to explain the lack of redox chemistry on conjugated molecular sites during potential cycling (33). The reduction of Fe3+ sites in Fe3+–N–C at –0.5 V versus RHE is probably enabled by a change of their coordination environment. In situ EXAFS (fig. S10C) revealed that the Fe ion lost one pyrrolic N ligand at this potential, possibly because of protonation or hydrogenation of the ligand driven by the electric field. The new coordination environment increases the Fe3+/2+ reduction potential to be more positive than the Fermi level of the carbon support, resulting in conjugated Fe2+ ions (fig. S18C).

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S18

Tables S1 to S4

References (3643)

References and Notes

Acknowledgments: We thank C. Corminboeuf and K.-h. Lin (EPFL) for discussion of theoretical analysis. We thank J. Luterbacher and F. Héroguel (EPFL) for their help in physical adsorption experiments. Funding: This work was supported by the GAZNAT SA and the European Research Council (grant 681292). We also acknowledge support from the Ministry of Science and Technology, Taiwan (contract MOST 107-2628-M-002-015-RSP). Author contributions: J.G. performed the majority of the synthesis, characterization, and electrochemical tests. L.B. contributed to the initial synthesis of catalysts and TEM measurement. C.-S.H. performed the in situ x-ray absorption experiments. J.G., C.-S.H., H.M.C., and X.H. analyzed the data. J.G. and X.H. wrote the paper, with input from all other co-authors. H.M.C. and X.H. directed the research. Competing interests: European priority patent applications (nos. 18156529.2 and 18193304.5) titled “Fe-N-C Catalyst, method of preparation and uses thereof” were filed by GAZNAT SA with J.G. and X.H. as inventors. Data and materials availability: All results are reported in the main text and supplementary materials. EXAFS, XANES, and microscopy data files are deposited in Zenodo (35). Unless restricted by patent, the materials are available for the purpose of reproducing or extending the analysis.

Stay Connected to Science

Navigate This Article