ELECTROCATALYSIS

Direct atomic-level insight into theactive sites of a high-performancePGM-free ORR catalystHoon T. Chung,1 David A. Cullen,2 Drew Higgins,1 Brian T. Sneed,3 Edward F. Holby,4

Karren L. More,3 Piotr Zelenay1*

Platinum group metal–free (PGM-free) metal-nitrogen-carbon catalysts have emerged as apromising alternative to their costly platinum (Pt)–based counterparts in polymerelectrolyte fuel cells (PEFCs) but still face some major challenges, including (i) theidentification of the most relevant catalytic site for the oxygen reduction reaction (ORR)and (ii) demonstration of competitive PEFC performance under automotive-applicationconditions in the hydrogen (H2)–air fuel cell. Herein, we demonstrate H2-air performancegains achieved with an iron-nitrogen-carbon catalyst synthesized with two nitrogenprecursors that developed hierarchical porosity. Current densities recorded in the kineticregion of cathode operation, at fuel cell voltages greater than ~0.75 V, were the same asthose obtained with a Pt cathode at a loading of 0.1 milligram of Pt per centimeter squared.The proposed catalytic active site, carbon-embedded nitrogen-coordinated iron (FeN4),was directly visualized with aberration-corrected scanning transmission electronmicroscopy, and the contributions of these active sites associated with specific lattice-level carbon structures were explored computationally.

The widespread integration of polymer elec-trolyte fuel cells (PEFCs) into vehicles willrequire substantial reductions in overallstack cost (1). The main stack cost contrib-utor (~46%) is the expensive Pt-based elec-

trocatalyst (2). The oxygen reduction reaction(ORR) at the cathode is inherently slower by sixorders of magnitude than the hydrogen oxida-tion reaction at the anode (3) and contributesmore to this cost. Thus, the implementation ofhigh-performance and durable platinum groupmetal–free (PGM-free) ORR catalysts could great-ly enable large-scale commercialization of fuelcell–powered vehicles.Toward this goal, heat-treated metal-nitrogen-

carbon (M-N-C) catalysts have been developedwhere themetal component is usually inexpen-sive and earth-abundant Fe or Co (4–6). A carbon-hosted FeN4 structure is the most commonlyproposed active site, arrived at on the basis ofboth theory and experimental data from x-rayabsorption, x-ray photoelectron spectroscopy,and Mössbauer spectroscopies (7–11), but thisactive site structure has not been directly ob-served and confirmed in the activated cata-lysts. Fuel cell performance has been improvedthrough better control of the pore structure ofPGM-free catalysts (12, 13) to improve masstransport properties, as well as accessibility ofthe active site(s). Serov et al. (13) incorporated

commercial silica powder as a template in thehigh-temperature synthesis of M-N-C catalysts,followed by removal of the porosity-inducingtemplate species with hydrofluoric acid. Othershave used metal-organic frameworks (MOFs) asprecursors for PGM-free catalysts, starting with aCo-imidazolate framework (14). Among severalMOF approaches, only Zn-MOF–derived PGM-free catalysts demonstrated high fuel cell per-formance (11, 12, 15), mainly attributed to thecatalyst porosity resulting from Zn evaporationduring the heat treatment (12).Despite improvements in fuel cell performance

with these approaches, mostM-N-C PEFC cathodestudies have been performed under H2-O2 con-ditions. The use of O2 likely masks the trueeffects of concentration polarization, commonlyobserved in H2-air cells (stacks) for automotiveapplications. Thus, demonstration of high PGM-free catalyst performance under practical H2-airconditions is required to validate such catalystsfor practical systems.Previously, our group has developed high-

activity Fe-N-C catalysts derived from polymersand simple organic compounds, such as polyani-line (PANI) (16–18) or cyanamide (CM) (19, 20).Here, PANI and CMprecursors were deliberatelycombined to synthesize a catalyst that exhibitsboth hierarchical pore structures and remark-able activity, reflected by the current densitiesrecorded in the kinetic region of the air cathodeoperation, at fuel cell voltages greater than~0.75 V.These current densities are the same as those ob-tainedwith aPt cathodeat a loading of 0.1mgPt cm

−2.Atomic-level images of FeN4 active sites wereobtained with low-voltage (60 kV), aberration-corrected scanning transmission electron micros-copy (AC-STEM). Electron energy-loss spectroscopy

(EELS)with single-atom resolution confirmed suchFeN4 structures embedded within carbon basalplanes. However, considerably higher concentra-tions of dispersed Fe single atoms were observedalong the surfaces of graphitic domains or step-edges in multilayer graphene. On the basis of theprevalenceof such edge-hosted structures observedin the catalyst, these specific configurations wereexplored by using theoretical approaches, whichpointed to a relatively highORR activity formulti-layergraphene (basal-plane)edge-hostedFeN4 struc-tures. Notably, these structures can be operational inthe catalysis of a variety of other electrochemicalreactions, someofwhichhavealreadybeen reportedto be very promising (21).In the synthesis of the dual nitrogen–precursor

catalyst, aniline and CM were first dissolvedin 1.5 MHCl solution, followed by the additionof iron (III) chloride as the iron precursor andammonium persulfate as oxidant for the oxida-tive PANIpolymerization. The solutionwas stirredat room temperature for ~4 hours to allow fullpolymerization of aniline and then heat treated(22). After completion of the (CM+PANI)-Fe-Ccatalyst synthesis, a surface area of ~1500 m2 g−1

was achieved, as determined by the Brunauer-Emmett-Teller (BET) method. The volumes ofmesopores (pore diameter 2 to 50 nm) and mi-cropores (pore diameter <2 nm), as measuredwith nitrogen adsorption measurements, were~0.25 and ~0.61 cm3 g−1, respectively (Fig. 1A).In their extensive PGM-free catalyst rotatingdisk electrode (RDE) studies involving five dif-ferent laboratories and four different synthe-ses, Jaouen et al. (23) found that microporesurface area was the primary factor governingtheORRactivity of PGM-free catalysts. Themicro-pore surface area of the (CM+PANI)-Fe-C was~1600 m2 g−1, which would predict high ORRactivity on the basis of Jaouen et al.’s findings.PANI-Fe-C, i.e., the catalyst obtainedwithout usingCM as a pore former, wasmuchmore aggregated(Fig. 1B) than the combined (CM+PANI)-Fe-C cat-alyst (Fig. 1C), which contained fewer pores withdiameter >50 nm and a BET surface area of only~1000 m2 g−1. These morphological features depictwhatwe refer to as a “hierarchical pore structure.”We attribute the resulting hierarchical pore

structure of (CM+PANI)-Fe-C to the relatively lowCM-decomposition temperature compared to thatof PANI. The decomposition of CM apparentlyformed gases responsible for pore formationin the PANI-rich phase at higher heat-treatmenttemperatures. This hypothesis is supported bythe results of thermogravimetric analysis (TGA)in fig. S1, pointing to a drastic loss (~50%byweight)of the (CM+PANI)-Fe-C precursor between 100°and 300°C. The TGA of PANI-Fe-C precursors(fig. S1) showed a more gradual weight loss (nomore than 20%) up to 300°C.The vital role of macropores in providing

greater accessibility to the catalytically activesites and establishing a more open frameworkfor improving the ionomer distribution withincatalyst layers was revealed by high-angle annu-lar dark-field (HAADF)–STEM imaging of amicro-tomed cross section of a (CM+PANI)-Fe-C catalyst

RESEARCH

Chung et al., Science 357, 479–484 (2017) 4 August 2017 1 of 5

1Materials Physics and Applications Division, Los AlamosNational Laboratory, Los Alamos, NM 87545, USA. 2MaterialsScience and Technology Division, Oak Ridge NationalLaboratory, Oak Ridge, TN 37831, USA. 3Center forNanophase Materials Sciences, Oak Ridge NationalLaboratory, Oak Ridge, TN 37831, USA. 4Sigma Division, LosAlamos National Laboratory, Los Alamos, NM 87545, USA.*Corresponding author. Email: zelenay@lanl.gov

on August 23, 2020

http://science.sciencem

ag.org/D

ownloaded from

electrode (fig. S2, A and B). Figure S2B shows afluorine energy-dispersive x-ray (EDX) spectros-copy elemental map acquired from the same areashown in fig. S2A. Fluorine EDX mapping is anestablished method to investigate ionomer dis-tributions within a catalyst layer (24). Ionomerimpregnated the macroporous catalyst regions(“C” in fig. S2, A and B) but not the dense catalystregions (“D” in fig. S2, A and B) of the electrode.A stark difference in the ionomer distributionbetween porous and dense catalyst regions ishighlighted in fig. S2, C and D.Rotating ring-disk electrode (RRDE) testing in

0.5 M H2SO4 revealed a half-wave potentialof 0.80 V versus the reversible hydrogen elec-trode (RHE), with a well-defined mass transport–limited current density (Fig. 2A). Ring-currentvalues from RRDE measurements verified thatthe H2O2 yield remained <2.5% at all electrodepotentials. This lowH2O2 yield corresponds to anaverage electron-transfer number perO2molecule(ne) of >3.95. The ORR activity and selectivity bymeans of RRDE rank among the highest for PGM-free catalysts reported (11, 13, 17, 25, 26). The cata-lyst performance loss after 30,000 RDE potentialcycles was ~5% (fig. S3).

The catalyst’s hierarchical pore structure plays acrucial role in exposing a large fraction of graphite(002) basal-plane edges of the carbon phasescomprising the catalyst to dioxygen. These surface-terminated basal planes are dominant in thecarbon structures and are most notably associatedwith small, randomly oriented graphitic domainsthat form thepredominant carbon phase, withFepresent at the exposed plane edges on the sur-faces. The high ORR activity measured throughRRDE testing cannot be directly correlated to highfuel cell performance because of the fundamentaldifferences in operating environment, temper-ature, and electrode structures between RDE andfuel cells. Thus, evaluation of PGM-free catalystsunder realistic fuel cell operation, i.e., H2-air con-ditions, is critical to assess performance from apractical standpoint, especially because the con-ventionally used H2-O2 test conditions do notcapture the important effects of mass transportin the cathode.Performance of (CM+PANI)-Fe-C cathode cata-

lyst layers (CCLs) in the membrane electrode as-sembly (MEA) was investigated under realisticconditions for the cathode operated on air (Fig.2B). The maximum power density reached with

an electrode containing 35 weight % (wt %) Nafionionomer was 0.39 W cm−2 at 1.0 bar partial pres-sure of H2 and air (the sum of the partial pres-sures of O2 and N2). Electrodes with increasedionomer content (50 and 60 wt % Nafion) werealso fabricated and exhibited enhanced currentdensities at fuel cell voltages >0.7 V resulting frommore effective filling and lining of macropores bythe ionomer that improved catalyst utilization(27). However, the performance improvement atlow current densities came at the expense ofhindered water removal, manifesting itself as aperformance decrease in themass transport region(below 0.6 V). The current density of ~75 mA cm−2

at a reference voltage of 0.8 V [~90 mA cm−2,iR corrected (i, current; R, resistance), fig. S4A]was 1.5 times the highest current density reportedto date, despite the use of much lower air pressurerelative to previous studies (11). We speculate thatthe large number of macropores formed by the in-clusion of CM during catalyst synthesis, coupledwith an improved ionomer dispersion that re-sulted from the presence of these macropores inthe CCL (fig. S2), contributed considerably to thissuperior H2-air performance.By comparison, also shown in Fig. 2B is the

performance of a 10 wt % Pt/C cathode (0.1 mgPtcm−2) prepared by a decal transfermethod. Underidentical test conditions, nearly the same currentdensities as those measured for (CM+PANI)-Fe-Ccatalyst were obtained in the kinetic region(>0.75V). Comparedwith a state-of-the-art 50wt%Pt/CMEAwitha cathodePt loadingof0.1mgPt cm

−2

(28), the Pt/C MEA data shown in Fig. 2B dem-onstrated the same performance in the kineticregion, albeit with lower current densities inthe mass transport region, likely because of themuchgreater thickness of the 10wt%Pt/C cathode,effectively illustrating the importance of catalyst-layer thickness on mass transport properties.The lower performance of the (CM+PANI)-Fe-Ccatalyst in the mass transport region was dueto the thickness of the (CM+PANI)-Fe-C layer,which was approximately four times as thick asthe 10 wt % Pt/C layer (on the basis of a totalcarbon loading of ~4 mg cm−2 for PGM-freeversus 0.9 mg cm−2 for Pt/C). These results arecorroborated by an improvement in fuel cellperformance of the (CM+PANI)-Fe-C cathodewith 35 wt % Nafion when the air flow wasincreased from 200 to 760 standard cubic cen-timeters per minute (Fig. 2C). Higher currentdensities were observed across the entire rangeof investigated voltages, including an increase inmaximumpowerdensity from0.39 to 0.42Wcm−2.All these results further validate that (CM+PANI)-Fe-C–based cathodes suffer from mass transportlimitations, which must be addressed by furtherimprovements to the intrinsicORRactivity of PGM-free catalysts tomatch the performance of Pt-basedelectrodes.The performance of this (CM+PANI)-Fe-CMEA

can be compared favorably to that of other MEAsreported in peer-reviewed literature for PGM-freecatalysts (11, 29, 30) under realistic PEFC operat-ing conditions. Among reported results, the highestpower density of ~0.4 W cm−2 (11) was obtained

Chung et al., Science 357, 479–484 (2017) 4 August 2017 2 of 5

Fig. 1. Hierarchical pore structure. (A) Micropore and mesopore size distributions. The microporesize distribution had peaks associated with three pore widths: 0.8, 1.1, and 1.5 nm. dV/dlog (D) isthe differential pore volume distribution, where V is pore volume and D is pore diameter. (B and C) SEMimages of PANI-Fe-C and (CM+PANI)-Fe-C catalysts, respectively, demonstrating the effect of CM inmacropore formation.

RESEARCH | REPORTon A

ugust 23, 2020

http://science.sciencemag.org/

Dow

nloaded from

at a pair of 2.5 bar that wouldbe reduced by more than30% if measured at a pair of1.0 bar, the condition usedthroughout our testing, whichwas based on Serov et al.’spower density dependence onpressure measurements (29).Experiments under H2-O2

conditions were also per-formed to minimize masstransport effects and achievebetter insight into the trueactivity of the (CM+PANI)-Fe-C catalyst in the fuel cell(Fig. 2D, iR-corrected polar-ization plots shown in fig.S4B). Three different par-tial pressures of O2—0.3, 1.0,and 2.0 bar—were applied.At all of these pressures, theperformance losses causedby mass transport were bare-ly noticeable down to vol-tages as low as 0.2 V. Thisperformance reflected much-improvedmass transportwith-in the catalyst layer versusH2-air conditions. Themax-imum power density valuesof ~0.87 and 0.94 W cm−2

reached at pO2 of 1.0 and2.0 bar, respectively, were thehighest ever achieved withPGM-free ORR catalysts op-erating on oxygen (31).To better understand the

source of high ORR activity,the atomic-level structure andchemistry of the (CM+PANI)-Fe-C catalyst were studied indetail by AC-STEM and EELS. Figure 3A showsa bright-field (BF)–STEM image of the overallmorphology of the principal structures presentin the (CM+PANI)-Fe-C catalyst. The catalystconsisted of primary fibrous carbon particlesand secondary few-layer graphene sheets. Car-bon tubes can be produced from CM (20, 32) andgraphene-like structures produced from PANI(16, 17) in the presence of Fe, although no nano-tubes were observed in the (CM+PANI)-Fe-C.These two carbon phases each contained sim-ilar amounts of N [3.5 atomic % (at %)] and Fe(0.2 at %), as determined by EELS. The nitrogenand iron contents within the near-surface re-gions of the catalyst measured by x-ray photo-electron spectroscopy (XPS) were 5.0 and 0.3at %, respectively, which were slightly highervalues than those obtained by highly localizedSTEM-EELS measurements. Analysis of the XPSN1s peak indicated the presence of pyridinic–,pyrrolic–, and graphitic–nitrogen bonded spe-cies (fig. S5). The dense fibrous carbon particleswere composed of randomly oriented, inter-twined, turbostratic graphitic domains on theorder of a few nanometers in size, as shown inFig. 3B and fig. S6A.

Because of the extremely high density of gra-phitic domains that terminate on the surfaces ofthe fibrous carbon particles, the surfaces of theseparticles were dominated by exposed edges andsteps of graphite (002) basal planes. We considerthe exposure of such basal-plane edges in the(CM+PANI)-Fe-C catalyst to be the most impor-tant factor contributing to its high ORR activity.Raman spectra for (CM+PANI)-Fe-C (fig. S7)confirmed an abundance of the (002) basal-planeedges as verified by a high D band–to–G bandintensity ratio (ID/IG) (1.07). This value was usedas a metric to evaluate structural disorder withinthe catalyst. A broad (002) peak in the x-raydiffraction pattern (fig. S8) also indicated a verysmall graphitic domain size coupled with turbo-stratic alignment of the basal planes within thedomains.Atomic-resolution STEM images of the fibrous

phase (Fig. 3B and fig. S6B) showed single atomsdispersed across the carbon surface (dots exhibit-ing bright contrast), which were confirmed to beprimarily Fe by EELS (fig. S6C). On the basis ofthe weak signal acquired in EEL spectra (due todata acquisition from a very limited area of ~1 to2 Å2 combinedwith the instability of the individual

Fe under the electron beam), it was not possibleto determine the valence state of the individualFe from EEL fine-structure analysis. However,according to Li et al.’s work (11), it is likely thatatomic Fe is present both in Fe2+ and Fe3+ states.Besides being present as highly dispersed singleatoms, excess Fe also existed in larger, isolatedparticulate form as either iron or iron sulfide (fig.S9), distributed randomly within the catalyst.These particulates were typically embedded orencased within few-layer graphene or graphiticshells. These structural datawere consistent withresults observed previously for heat-treatedM-N-Ccatalysts (16, 20, 26) and are not expected tocontribute to the catalytic activity of the catalyst.Individual Fe atoms were also observed to beembedded in the few-layer graphene sheetphase (Fig. 3C and fig. S10). The Moiré patternoriginating from overlapping and rotated layersof the graphene honeycomb lattice is visible inthese images.The stacked, few-layers of graphene provided a

stabilizing effect for these Fe atoms and allowedfor more detailed spectroscopic analysis. A sim-ilar graphene-stabilizing effect was previouslyobserved for MoS2 (33). EEL spectra obtained

Chung et al., Science 357, 479–484 (2017) 4 August 2017 3 of 5

Fig. 2. Electrochemical and fuel cell performances. (A) ORR performance of (CM+PANI)-Fe-C catalyst. Steady-state RDEpolarization plots were obtained by using a 20-mV potential step and 25-s potential hold time at every step. Electrolyte 0.5 MH2SO4, temperature 25 ± 1°C, rotation rate 900 rpm. (B and C) H2-air fuel cell polarization plots. Cathode: ~4.0 mg cm−2 of(CM+PANI)-Fe-C; air 200 ml min−1 (2.5 stoichiometry at 1.0 A cm−2) and 760 ml min−1 (9.5 stoichiometry at 1.0 A cm−2); 100%relative humidity (RH); and 1.0 bar partial pressure. Anode: 2.0mgPt cm

−2 Pt/C; H2 200mlmin−1; 100% RH; and 1.0 bar partialpressure. Membrane Nafion 211, cell 80°C, electrode area 5 cm2. (D) H2-O2 fuel cell polarization plots. Cathode: ~4.0 mg cm−2

of (CM+PANI)-Fe-C; O2 200mlmin−1 (40mlmin–1 cm–2); 100%RH; 0.3, 1.0, and 2.0 bar partial pressures. Anode: 2.0mgPtcm−2 Pt/C; H2 200 ml min−1; 100% RH; 1.0 bar partial pressure. Membrane Nafion 211, cell 80°C; 5 cm2 electrode area.

RESEARCH | REPORTon A

ugust 23, 2020

http://science.sciencemag.org/

Dow

nloaded from

directly around the Fe (Fig. 3D) showed that Nwas associated with the Fe, which was other-wise absent in the surrounding graphene-onlyregions (Fig. 3D). High-resolution EEL spectrumimaging (fig. S11) confirmed this N-Fe associa-

tion. Quantification of the Fe-to-N ratio from theEELS data acquired for several of these sitesyielded an average composition of FeN4, con-sistent with previously proposed active sites forthis type of PGM-free catalyst (7, 9–11). However,

previous active-site determinations were basedon bulk material analyses with Mössbauer, XPS,and x-ray absorption spectroscopy, which aver-age the obtained signal, versus direct microscopicevidence for the formation of individual FeN4

Chung et al., Science 357, 479–484 (2017) 4 August 2017 4 of 5

Fig. 3. STEM images and EEL spectra of the(CM+PANI)-Fe-C catalyst. (A) BF-STEM image of atypical (CM+PANI)-Fe-C catalyst showing primaryfibrous carbons and secondary graphene sheets.(B) Atomic-resolution HAADF-STEM image of Featoms distributed across the surface of fibrouscarbon phase showing randomly oriented, intertwinedgraphitic domains. (C) HAADF-STEM image of individualFe atoms (labeled 1, 2, and 3) in a few-layer graphenesheet. (D) EEL spectra of the N k-edge (Nk) and Fe L-edge(FeL) acquired from single atoms (1 and 2) and few-layergraphene (3), demonstrating the presence of N aroundthe Fe atoms.

Fig. 4. Model structures used in theoreticalstudies with spontaneously formed OHligand. Views from above (A and B), from side(C and D), and from tilted perspective (E andF). (A), (C), and (E) show bulk-hosted FeN4

and (B), (D), and (F) show zigzag edge-hostedFeN4 structures with OH ligands. C, gray; Fe,bronze; H, white; N, blue; O, red.

RESEARCH | REPORTon A

ugust 23, 2020

http://science.sciencemag.org/

Dow

nloaded from

complexes embedded in few-layer graphene,away from any exposed edges. Recently Fei et al.(34) observed atomic cobalt on nitrogen-dopedgraphene; however, the Co-N bond proposedwasbased on results from an indirect bulk technique,i.e., wavelet transform of extended x-ray absorp-tion fine structure, as opposed to our direct obser-vations of single complexes with STEM-EELSreported here.It must be emphasized that the FeN4 com-

plexes observed away from exposed basal-planeedges (bulk hosted) represent a relatively smallfraction of the total Fe and N observed by STEMimaging of the catalyst; most of the highly dis-persedFe atomswere predominantly positioned atexposed basal-plane edges and steps (edge hosted)in both of the carbon phases in (CM+PANI)-Fe-Ccatalyst, fibrous carbon particles (primary phase),and few-layer graphene sheets (secondary phase).The tendency of Fe to occupy edge and step siteswas further demonstrated for the (CM+PANI)-Fe-Ccatalyst in small regions of single-layer grapheneidentified within the mostly few-layer graphenesheets (fig. S12). However, the specific N coordi-nation of these edge-positioned Fe atoms couldnot be easily verified with STEM-EELS becauseof the instability of these atoms under even thelow-energy electron beamused (60 keV). Althoughmany candidate active sites could account forthe observed Fe distribution, edge-hosted FeN4

structures would simultaneously satisfy the in-direct evidence of FeN4 structures identified byMössbauer, XPS, and x-ray absorption spectros-copies, as well as the predominance of Fe observedat the basal-plane edges for both the fibrousand graphene-sheet phases present.Previous theoretical studies (8, 35) suggest

that edge-hosted FexNy structures had lowerformation energies—i.e., an increased relativestability—than bulk-hosted structures. Thus, theequilibrium concentration of the edge-hostedstructures is expected to be substantially higherthan bulk-hosted structures. Because most of theFe observed by AC-STEM imaging is present atthe basal-plane edges and steps, and theoreticalstudies suggest higher concentration of edge-hosted FeNx structures, we used quantum chem-ical modeling to study the relative ORR pathwaysof edge-hosted FeN4 sites versus bulk-hostedFeN4 sites to gain further insight into the possiblepredominant active sites in the (CM+PANI)-Fe-Ccatalyst.In addition to relative equilibrium concentra-

tions, previous quantum chemistry studies ofactive site structures of M-N-C catalysts suggestthat edge-adjacent Fe2N5 configurations have high

ORR activity (36). Using the same methodology(22), we calculated the reaction pathways (fig. S13)and thermodynamic limiting potential, Ul, thatserves as descriptor for relative ORR activity forthe graphene bulk-hosted and nanoribbon zigzagedge-hosted structures. Similar to the previouslyreported edge-hosted Fe2N5 structure (36), bothbulk-hosted (Fig. 4, A, C, and E) and zigzag edge-hosted (Fig. 4, B, D, and F) FeN4 structuresspontaneously evolved an OH ligand at relevantpotentials. With the OH ligand attached, thebulk-hosted FeN4 does not exothermically bindO2, and a bound OOH spontaneously dissociatesto a bound O and free OH. These results suggestthat such an OH-modified site may not act as asingle reaction site on anassociativeORRpathway.However, the zigzag edge-hosted FeN4 with OHliganddoesbindO2 and, onanassociativepathway,has a limiting potential of 0.80 V. ThisUl is equalto the highest reported value (35, 36), in whichthe initial chemical adsorption of O2 was con-sidered. Thus, quantum chemistry calculationsshow that the FeN4 structures follow differentORR reaction pathways, depending on whetherthey are hosted in the bulk or at the edge of thegraphene, and that the edge-hosted sites, whenspontaneously ligated by OH in the fuel cell en-vironment, lead to highly ORR-active structures.Though FeN4 sites at the exposed basal-plane

edges could not be directly discerned with AC-STEM, the considerably higher concentration ofFe associated with edges and steps observed byAC-STEM imaging combined with the quantumchemical calculations supported higher popula-tions as well as higher ORR activities with theedge-hosted FeN4 compared with the bulk-hostedFeN4. Thus, we can plausibly suggest that edge-hosted FeN4 sites are likely themajor contributorsto the overall high activity observed in both theRDE and MEA testing of the (CM+PANI)-Fe-Ccatalyst.

REFERENCES AND NOTES

1. T. Yoshida, K. Kojima, Electrochem. Soc. Interface 24, 45–49(2015).

2. B. D. James, J. M. Moton, W. G. Colella, “Fuel CellTransportation Cost Analysis,” U.S. Department of Energy’s2014 Annual Merit Review and Peer Evaluation Meeting,Washington, DC, 19 June 2014.

3. M. K. Debe, Nature 486, 43–51 (2012).4. J. Masa, W. Xia, M. Muhler, W. Schuhmann, Angew. Chem. Int.

Ed. 54, 10102–10120 (2015).5. K. Mamtani, U. S. Ozkan, Catal. Lett. 145, 436–450 (2015).6. G. Wu, P. Zelenay, Acc. Chem. Res. 46, 1878–1889 (2013).7. U. I. Kramm, M. Lefèvre, N. Larouche, D. Schmeisser,

J.-P. Dodelet, J. Am. Chem. Soc. 136, 978–985 (2014).8. E. F. Holby, G. Wu, P. Zelenay, C. D. Taylor, J. Phys. Chem. C

118, 14388–14393 (2014).9. A. Zitolo et al., Nat. Mater. 14, 937–942 (2015).

10. N. R. Sahraie et al., Nat. Commun. 6, 8618 (2015).11. J. K. Li et al., Energy Environ. Sci. 9, 2418–2432 (2016).12. E. Proietti et al., Nat. Commun. 2, 416 (2011).13. A. Serov, K. Artyushkova, P. Atanassov, Adv. Energy Mater. 4,

1301735 (2014).14. S. Ma, G. A. Goenaga, A. V. Call, D.-J. Liu, Chemistry 17,

2063–2067 (2011).15. J. Shui, C. Chen, L. Grabstanowicz, D. Zhao, D. J. Liu, Proc.

Natl. Acad. Sci. U.S.A. 112, 10629–10634 (2015).16. G. Wu et al., J. Mater. Chem. 21, 11392 (2011).17. G. Wu, K. L. More, C. M. Johnston, P. Zelenay, Science 332,

443–447 (2011).18. M. Ferrandon et al., J. Phys. Chem. C 116, 16001–16013

(2012).19. H. T. Chung et al., Electrochem. Commun. 12, 1792–1795

(2010).20. H. T. Chung, J. H. Won, P. Zelenay, Nat. Commun. 4, 1922

(2013).21. A. S. Varela et al., Angew. Chem. Int. Ed. 54, 10758–10762

(2015).22. See supplementary materials.23. F. Jaouen et al., ACS Appl. Mater. Interfaces 1, 1623–1639

(2009).24. D. A. Cullen et al., J. Electrochem. Soc. 161, F1111–F1117

(2014).25. D. Zhao et al., Adv. Mater. 26, 1093–1097 (2014).26. K. Strickland et al., Nat. Commun. 6, 7343 (2015).27. S. Komini Babu, H. T. Chung, P. Zelenay, S. Litster, ACS Appl.

Mater. Interfaces 8, 32764–32777 (2016).28. A. Kongkanand et al., ACS Catal. 6, 1578–1583 (2016).29. A. Serov et al., J. Power Sources 327, 557–564 (2016).30. J. Tian et al., Angew. Chem. Int. Ed. 52, 6867–6870 (2013).31. H2-O2 fuel cell performance in this work was measured at a

relatively low O2 flow rate of 40 ml min–1 cm–2. The maximumpower density reported to date is ~0.9 W cm–2. It was obtainedat pO2

of 1.0 bar and a high O2 flow rate of 260 ml min–1 cm–2

(12) and, independently, at a high pO2of 2.0 bar and high O2

flow rate of 80 ml min–1 cm–2 (15).32. H. T. Chung, P. Zelenay, Chem. Commun. (Camb.) 51,

13546–13549 (2015).33. R. Zan et al., ACS Nano 7, 10167–10174 (2013).34. H. Fei et al., Nat. Commun. 6, 8668 (2015).35. E. F. Z. Holby, P. Zelenay, Nano Energy 29, 54–64 (2016).36. E. F. Holby, C. D. Taylor, Sci. Rep. 5, 9286 (2015).

ACKNOWLEDGMENTS

This work was supported by the Office of Energy Efficiency andRenewable Energy of the U.S. Department of Energy (DOE) through theFuel Cell Technologies Office. Microscopy was performed as part of auser project supported by Oak Ridge National Laboratory’s Center forNanophase Materials Sciences, which is a DOE Office of Science UserFacility. Computational resources were provided by the InstitutionalComputing program of Los Alamos National Laboratory. We thankA. Dattelbaum and J. Spendelow (Los Alamos National Laboratory),R. Adzic (Brookhaven National Laboratory), G. Wu (University atBuffalo), and P. Atanassov (University of New Mexico) for worthwhilediscussions. All results are presented in the main paper andsupplementary materials. The binary nitrogen precursors PGM-freecatalyst synthesis method has been patented by H.T.C. and P.Z. asU.S. patent number US20160351915 A9, Non-precious metal catalysts.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/357/6350/479/suppl/DC1Materials and MethodsFigs. S1 to S13References (37–47)DFT Data File S1

13 March 2017; accepted 6 July 201710.1126/science.aan2255

Chung et al., Science 357, 479–484 (2017) 4 August 2017 5 of 5

RESEARCH | REPORTon A

ugust 23, 2020

http://science.sciencemag.org/

Dow

nloaded from

Direct atomic-level insight into the active sites of a high-performance PGM-free ORR catalystHoon T. Chung, David A. Cullen, Drew Higgins, Brian T. Sneed, Edward F. Holby, Karren L. More and Piotr Zelenay

DOI: 10.1126/science.aan2255 (6350), 479-484.357Science 

, this issue p. 479Science.4structure and exhibits high ORR performance when running on air. The proposed catalytically active site is FeN

developed an iron-nitrogen-carbon catalyst from two nitrogen precursors that forms a high-porosityet al.oxygen. Chung Most of the candidate replacement catalysts that have shown high performance do so only when running on purereduction reaction (ORR) with ones based on non-noble metals would speed up the adoption of hydrogen fuel vehicles.

Replacing expensive and scarce platinum catalysts in polymer electrolyte membrane fuel cells for the oxygenReplacing platinum in air-fed fuel cells

ARTICLE TOOLS http://science.sciencemag.org/content/357/6350/479

MATERIALSSUPPLEMENTARY http://science.sciencemag.org/content/suppl/2017/08/03/357.6350.479.DC1

REFERENCES

http://science.sciencemag.org/content/357/6350/479#BIBLThis article cites 44 articles, 4 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 1200 New York Avenue NW, Washington, DC 20005. The title (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science. No claim to original U.S. Government WorksCopyright © 2017 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of

on August 23, 2020

http://science.sciencem

ag.org/D

ownloaded from