Riku Shibuya1, Takahiro Kondo2,3, and Junji Nakamura2,3
1University of Tsukuba, Graduate School of Pure and Applied Sciences, 1‐1‐1 Tennodai, Tsukuba, Ibaraki, 305‐8573, Japan
2University of Tsukuba, Faculty of Pure and Applied Sciences, 1‐1‐1 Tennodai, Tsukuba, Ibaraki, 305‐8573, Japan
3University of Tsukuba, Tsukuba Research Center for Energy Materials Science, 1‐1‐1 Tennodai, Tsukuba, Ibaraki, 305‐8573, Japan
The oxygen reduction reaction (ORR; O2 + 4H+ + 4e− → 2H2O) is a key reaction at the cathode of polymer electrolyte fuel cells, for which platinum (or Pt alloys such as PtCo) has been regarded as the best catalyst and has been widely used in commercial fuel cells [1, 2]. However, due to the limited resources and the high cost of platinum, developing highly active catalysts using low cost and abundant materials is now desired for the wide spread usage of fuel cell.
Nitrogen‐doped carbon materials are one of the best candidate catalysts as the Pt‐substitute material for the ORR [3–10], as these nonmetal catalysts exhibit high electrocatalytic activity for the ORR [11] and high durability under both basic and acidic conditions [12]. However, the catalytic activity is still lower than that of a state‐of‐the‐art Pt catalyst particularly in acidic media. Thus, to alternate Pt in commercial fuel cells, further developments are required.
To enhance the catalytic activity, the nitrogen‐doped carbon materials should be engineered to contain a high concentration of active sites without inactive components. The active sites of the nitrogen‐doped carbon materials have, however, been debated for a long time [13–22] until the appearance of our recent experimental works with model catalysts, in which the pyridinic nitrogen (nitrogen having two NC bonds) has been found to create the active sites for the ORR [23]. Here, the recent progress in the understanding of active sites is reviewed together with our works. The perspective and future directions for nitrogen‐doped carbon materials are then described.
Nitrogen‐doped carbon materials have been mostly prepared by heating mixtures of nitrogen‐containing molecules, carbon sources, and/or transition metal (Fe or Co) precursors at 800–1000 °C. Heating carbon samples in ammonia has also been often conducted as an additional nitrogen‐doping method, where the amount of oxygen‐containing functional groups in carbon materials is known to affect the amount of doped nitrogen in the process, because nitrogen is effectively doped by so‐called ammoxidation reaction [24]. The catalytic activity for the ORR significantly depends on the precursors, heat treatment temperature, carbon morphology, and synthesis conditions [3–10]. The difference in the catalytic activities can be ascribed to the differences in the electronic conductivity, proton conductivity, and oxygen diffusion efficiency. Concerning the catalytic activity, the role of metal species including in the catalysts is often pointed out and discussed. On the other hand, it has been shown that the nitrogen‐doped carbon catalysts show excellent catalytic activity in both acidic and basic media without inclusion of any metal species [3, 12]. Thus, the nitrogen species should at least create the active site of the ORR in carbon materials.
As schematically shown in Figure 8.1, there are various types of nitrogen species in nitrogen‐doped carbon materials. Among them, pyridinic nitrogen (pyridinic‐N, N bonded to two carbon atoms) and graphitic nitrogen (graphitic‐N, N bonded to three carbon atoms, also often called substituted N or quaternary‐N) are the major nitrogen species for nitrogen‐doped carbon materials especially after enough heat treatments [3–10]. The catalytic active sites are thus created by either pyridinic‐N and/or graphitic‐N, which has been a long debate [13–22]. The controversy can be ascribed to two reasons. One is the mixing of different types of nitrogen species in the carbon materials, which is inevitable in the doping processes, e.g. by annealing under NH3 atmosphere. The other lies in the inhomogeneities associated with the morphology and the graphitization level of the evaluated samples, which leads to inhomogeneous sizes of the π‐conjugated system. Indeed, the size of nitrogen‐doped graphene quantum dots has been reported to affect the ORR performance [25]. Thus, it is difficult to determine which type of nitrogen creates the active site for the ORR by comparing samples subjected to treatment or pyrolysis at different temperatures because the size of the π‐conjugated system is also dependent on the annealing temperature. To determine the active site conclusively, analysis with several well‐defined model catalysts are indispensable, where the model catalysts should have the same size of π‐conjugated system and are treated by the same pyrolysis temperature but possess different type of nitrogen species.
The differences between pyridinic‐N and graphitic‐N in terms of its local electronic structure are first described, before showing the model catalyst study to determine the ORR active sites.
To study the local electronic structure of the nitrogen‐doped carbon materials, we have first prepared model systems of nitrogen‐doped graphite using a highly oriented pyrolytic graphite (HOPG) and examined it by X‐ray photoelectron spectroscopy (XPS) and scanning tunneling microscopy (STM)/scanning tunneling spectroscopy (STS) [26]. Figure 8.2 shows the XPS N 1s spectrum of nitrogen‐doped HOPG, which was prepared by N2+ bombardment with 200 eV at 300 K followed by annealing at 900 K under ultrahigh vacuum [26, 27]. There are four N 1s peak components at 398.5, 399.9, 401.1, and 403.2 eV, each of which has been assigned to pyridinic‐N, pyrrolic‐N (N part of a pentagon ring connected to two C and one H), graphitic‐N, and oxide‐N species, respectively [28–31]. At a low nitrogen concentration of 1.9 at.%, the dominant species is graphitic‐N (over 80%). With an increasing nitrogen concentration up to 7–9 at.%, the amount of pyridinic‐N increases and it reaches to be comparable with that of graphitic‐N (Figure 8.2).
An important point here is that there is a significant difference in the core level (N 1s) binding energy as much as 2.6 eV between pyridinic‐N and graphitic‐N. Higher and lower binding energies of graphitic‐N and pyridinic‐N are originated from positive and negative charges of nitrogen, respectively. For pyridinic‐N donating four electrons for the sp2 configuration including a lone pair, the higher electronegativity of nitrogen compared with that of carbon pulls π‐electrons from π‐conjugated system, resulting in the negative charge of nitrogen atom. On the other hand, graphitic‐N donating three electrons for the sp2 configuration, the remaining two electrons flow to some extent to the stable π‐conjugated system of graphite, resulting in the positive charge of graphitic‐N [26].
The local electronic structures of carbon surfaces near pyridinic‐N and graphitic‐N are studied by STM/STS measurements. As shown in Figure 8.3, STS spectrum at the carbon atoms around pyridinic‐N (which was identified by the comparison with the simulated STM image by density functional theory (DFT) calculations, Figure 8.3 [26]) shows localized states at the occupied region of −370 meV. Such localized states are known as a non‐bonding pz orbital of carbon formed as a result of the breaking of the aromaticity (π conjugated system) of carbon. For example, such states are known to appear at the zigzag edge of graphene or graphite with the propagation toward perpendicular direction with respect to the edge (the states are then called as edge states) [32]. The non‐bonding pz orbital also appears at the carbon atoms around the vacancy defects of graphene or graphite [33, 34]. In the same mechanism, the localized states thus appear at the carbon atoms around doped nitrogen. Here, due to the negative charge of pyridinic‐N, the surrounding carbon atoms would be charged positively as a result of the screening effect. The positive charge of carbon can explain a shift of the localized states of carbon from the Fermi level to the lower energy level. Consequently, the π electrons at this carbon are deficient and the carbon atom is positively charged but possesses lone pair at a relatively deeper energy level originated from non‐bonding pz orbital. Following this picture, we have proposed that the states (non‐bonding pz orbital) corresponding to the STS peak in the occupied region near the Fermi level, i.e. carbon atoms around the pyridinic‐N, may act as “Lewis base” [26]. In the case of graphitic‐N, localized states are observed at the unoccupied region of +500 meV as shown in Figure 8.4. Because graphitic‐N is positively charged, the surrounding carbon atoms should be charged negatively as a result of the screening effect. The negative charge of carbon can explain a shift of the localized states of carbon from the Fermi level to the upper energy level. Here, we have also proposed that the states with STS peak in the unoccupied region near the Fermi level, i.e. carbon atoms around the graphitic‐N, may act as “Lewis acid” [26].
To determine the active site conclusively, we develop four types of model catalysts with well‐defined π‐conjugation using HOPG [23]: (i) pyridinic‐N dominated HOPG (pyri‐HOPG), (ii) graphitic‐N dominated HOPG (grap‐HOPG), (iii) edges patterned on the surface without N (edge‐HOPG), and (iv) clean‐HOPG. Of the four types of prepared HOPG model catalysts, preparation of the pyridinic‐N‐dominated HOPG model catalyst is the most challenging because pyridinic‐N atoms are preferentially located at the edges of graphite. We thus designed an edge‐patterned surface by Ar+ etching through a mask. Figure 8.5a–d shows surface morphological characterization of a typical edge‐patterned model catalyst. The ordered uniform rectangular groove structures were distributed over the surface in a wide range. The atomic force microscopy (AFM) image presented in Figure 8.5b shows dark regions corresponding to the grooves etched through the slits of the mask by ion beam and bright regions that correspond to non‐etched surfaces. The surface of the bright region is intact and is basically flat. The profile of the line in Figure 8.5b shows that the depth of the grooves is about 1200 ± 80 nm for this sample (Figure 8.5d), which could be varied from about 100 nm to more than 2 μm by manipulating the etching energy and duration.
Prepared model catalysts dominantly possess pyridinic‐N for pyri‐HOPG and graphitic‐N for grap‐HOPG as shown in the XPS spectra of Figure 8.6a. Figure 8.6b shows the corresponding ORR curves, where the pyri‐HOPG model catalyst shows high activity at high voltages, compared with the very low ORR activities of the N‐free model catalysts. The pyri‐HOPG sample with lower N concentration (N: 0.60 at.%) is much more active than the grap‐HOPG sample with higher N concentration (N: 0.73 at.%). As the pyri‐HOPG sample is nearly free of graphitic‐N, the ORR results indicate that it is pyridinic‐N rather than the graphitic‐N that reduces the ORR overpotential and creates the active site. The activity of the grap‐HOPG sample could also be ascribed to the presence of pyridinic‐N as a minor component (0.04 at.%) [23].
The dependence of the ORR activity of the catalyst on the concentration of pyridinic‐N was then investigated as shown in Figure 8.7a. The linear relationship was obtained, as shown in Figure 8.7b, between the current densities and concentration of pyridinic‐N at each investigated potential, independent of the preparation method (described in inset [23]), indicating that the ORR activity is determined solely by the pyridinic‐N concentration. This linear dependence also suggests that the inhomogeneities in the graphitization and size of the π‐conjugation system in the samples were overcome successfully in the HOPG model catalysts by applying the same annealing temperature of 973 K. As a result, the onset potential (potential versus RHE at current density of 1 μA) increases with increasing concentration of pyridinic‐N (Figure 8.7c).
We further investigated the intermediates of the ORR by ex situ post‐ORR XPS measurements of the HOPG model, which reflects the steady‐state surface of the N‐HOPG model catalyst under ORR and provides mechanistic information about the active sites. Figure 8.8a illustrates a significant change in the N 1s peak after the ORR half‐cell measurement in acidic condition, in which the concentration of pyridinic‐N (398.5 eV) decreased from 54% to 38%, and the concentration of the component corresponding to the 400.1 eV peak (either pyrrolic‐N or pyridonic‐N) increased from 11% to 29%, while the sum of the pyridinic‐N and pyrrolic/pyridonic‐N components remained largely constant (from 65% to 67%). On the contrary, an N‐HOPG sample after immersion in 0.1 M H2SO4 solution without the ORR scanning exhibited a negligible change in N 1s spectrum [23]. The difference in the composition of nitrogen species before and after the ORR suggests that the carbon atoms next to pyridinic‐N react with OH species with consequent transformation of the pyridinic‐N to pyridonic‐N, as shown in Figure 8.8b, suggesting that the active sites are the carbon atoms next to the pyridinic‐N rather than pyridinic‐N themselves.
Furthermore, the relationship between the basicity and the activity of the HOPG model catalysts was examined because our STM/STS results suggested that pyridinic‐N creates the Lewis base site as described in Section 8.3. Figure 8.8c shows the profiles for temperature‐programmed desorption (TPD) of CO2 from the HOPG model catalysts on which CO2 was adsorbed at room temperature in ultrahigh vacuum. It is found that acidic CO2 molecule is adsorbed only on the ORR‐active pyri‐HOPG catalyst, indicating that the Lewis basic site is created by pyridinic‐N on the HOPG surface. It is generally known that oxygen molecules can be adsorbed on Lewis base sites [35]. Because O2 adsorption is the initial step of the ORR, the Lewis base site created by pyridinic‐N is thus suggested to be the active site for ORR.
To compare the HOPG model catalysts with powder catalysts, nitrogen‐doped graphene nanosheets (N‐GNSs) were prepared and their ORR activities were measured by the rotating disc method in 0.1 M H2SO4. Here, the N‐GNS catalysts were prepared by the reaction of GNS with NH3 at 973 K, which is the same temperature applied in the preparation of the HOPG model catalysts. Figure 8.9a shows the N 1s XPS profiles of the prepared N‐GNS powder catalysts. The powder catalysts have high percentages of pyridinic‐N, and the pyridinic‐N concentration increases from N‐GNS‐1 (0.7 at.%) to N‐GNS‐2 (1.9 at.%) to N‐GNS‐3 (6.3 at.%), whereas the graphitic‐N concentrations are as low as 0.4–0.8 at.%. Figure 8.9b shows the ORR performances of the N‐GNS powder catalysts, in which the currents are divided by the geometric electrode surface area (0.283 cm2), with a loading amount of 0.02 mg. The ORR activity increases with an increasing nitrogen concentration, e.g. the onset potential increases from 0.77 V for N‐GNS‐1 to 0.91 V for N‐GNS‐3. The ORR activity was further examined in terms of the current densities at different potentials (0.5, 0.6, and 0.7 V versus RHE). Linear relationships between the ORR activities at three different potentials and the concentration of pyridinic‐N were obtained (Figure 8.9c), consistent with the linear relationships for the HOPG model catalysts (Figure 8.7). It is thus confirmed that pyridinic‐N creates the active site for ORR in the N‐GNS powder catalysts.
We further compared the ORR‐specific activities of the N‐HOPG model catalysts and the N‐GNS powder catalysts. The current densities at 0.5 V for the N‐HOPG model catalysts are approximately 3 orders of magnitude lower than those for the N‐GNS powder catalysts, attributed to the difference in the absolute number of active sites per 1 cm2 of the geometric surface area of the electrodes. The N‐HOPG model catalyst is simply a plate with a very low graphite surface area of about 0.1 cm2, identical to the geometric electrode surface area. On the other hand, the BET (Brunauer–Emmett–Teller) surface areas of the N‐GNS catalysts on the electrode (0.283 cm2) are about 80 cm2. As the pyridinic‐N creates the active site for ORR, the specific activities per pyridinic‐N for the model and powder catalysts were calculated and compared by taking into account the BET surface area (400 m2 g−1) for N‐GNS [23]. The specific activities per pyridinic‐N are similar at about 0.1 e− s−1 pyri‐N−1 for both types of catalysts (0.07–0.14 e− s−1 pyri‐N−1). The agreement in the specific activity per pyridinic‐N manifests that, in general, the active sites of ORR for various kinds of nitrogen‐doped carbon materials are created by pyridinic‐N.
Finally, our proposed possible mechanism for the ORR on nitrogen‐doped carbon materials is shown in Figure 8.10. As the Lewis base site is created by pyridinic‐N, the oxygen molecule is first adsorbed at the carbon atom next to the pyridinic‐N followed by protonation of the adsorbed O2. Two pathways are then possible: one is the four‐electron mechanism taking place at a single site and the other is 2 + 2‐electron mechanism, which does not always take place at a single site. In the four‐electron mechanism, the other two protons attach to the two oxygen atoms, leading to breakage of the OOH bond and formation of OH species (d in Figure 8.10), as observed in post‐ORR XPS (Figure 8.8). The additional proton then reacts with the adsorbed OH to form H2O (e in Figure 8.10). In the 2 + 2‐electron pathway, H2O2 is formed by reaction of the adsorbed OOH species with another proton (f in Figure 8.10), followed by re‐adsorption of H2O2 and its reduction by two protons to generate H2O. The OH species detected in the post‐ORR XPS measurement may arise from the four‐electron mechanism, but it is also possible that the OH species next to the pyridinic‐N may arise from the reaction with H2O2 in the 2 + 2‐electron mechanism. In either pathway, the carbon atoms next to pyridinic‐N with Lewis basicity play an important role as the active sites at which oxygen molecules are adsorbed as the initial step of the ORR.
The mechanism of the ORR at the molecular level has been studied using well‐defined pyridinic‐N containing conjugated molecules. Li et al. have used graphene quantum dots with two pyridinic‐N atoms (N‐doped graphene quantum dot (N‐GQD) 1 in Figure 8.11) [36]. They constructed a so‐called Pourbaix diagram shown in Figure 8.12 based on the measurements of the reduction potentials at 25 °C and various pH values. The reduction of N‐GQD 1 is clearly pH dependent (Figure 8.12a), indicating the participation of protons in the reduction process. The Pourbaix diagram (Figure 8.12b) delineates the potential and pH ranges in which either N‐GQD 1 or its reduction products are thermodynamically stable. In the range between pH 4 and 13, straight line with a slope of −27 mV per pH unit was consistent with a concerted one proton, two‐electron process, where the following reaction takes place.
They thought that species 4 is responsible for O2 adsorption site. Because of delocalization of the negative charge over the conjugated system, 4 possesses the characters of both an amide anion and carbanion (a part of resonance structure as described in Figure 8.12c). Here, the carbanion next to protonated pyridinic‐N is the optimized oxygen adsorption site according to the DFT calculation (Figure 8.13). Their DFT calculation suggests that adsorbed oxygen molecule subsequently reacts with H atom bound to pyridinic‐N atom followed by further reduction to form water as a final product. They described that this carbanion mechanism for oxygen activation bears remarkable resemblance to the oxygen activation mechanism of flavin. In the case of flavin, the fully reduced flavin (Figure 8.14a) also possesses carbanion next to protonated pyridinic nitrogen (IV in Figure 8.14a), which shows the resonance structure (Figure 8.14b) same as 4. The molecular oxygen attacks this carbanion to form hydroperoxyflavin (Figure 8.14c) [37].
Li et al. have reported that the conjugation size in these carbon‐based materials (N‐GQD) was an important parameter determining the ORR activity [25, 38]. As shown in Figure 8.15, N‐GQD 5 having lowest N:C ratio shows the highest ORR onset potential among all N‐GQDs (N‐GQD 5, 6, and 7). Namely, larger conjugation size of N‐GQDs shows higher ORR activity, indicating the importance of size of π system around pyridinic‐N.
Importance of local information near the pyridinic nitrogen in ORR at molecular level has been recently clarified by Surendranath and coworkers [39]. Specifically, they showed distinct difference in the catalytic activities depending on the local structure around pyridinic‐N; three different types of graphite‐conjugated pyrazine (GCP) moieties having different functional groups (Figure 8.16) show different ORR activities.
As shown in Figure 8.16, three samples (GCPs 1, 2, and 3) were prepared by condensation reaction, where each three ortho‐phenylenediamine derivatives with different functional group graphitized onto edges at pre‐oxidized glassy carbon electrode. GCPs 1 and 2 show XPS peak at 398.9 eV in N 1s region, indicating the presence of pyridinic‐N. In the case of GCP 3, additional peak at 401.7 eV, attributed to pyridinium species, was observed together with a peak at 398.8 eV (pyridinic‐N). ORR activity measured in 0.1 M KOH exhibits different catalytic activity with an order of GCP 3 > GCP 1 > GCP 2. Here, GCP 1 shows onset potential of 0.75 V versus RHE, GCP 2 (with electron‐donating methyl group) shows higher over potential of +18 mV shift at 1 mA cm−2 compared with that of GCP 1, and GCP 3 (with electron‐withdrawing pyridinium moiety) shows lower overpotential of −24 mV shift at 1 mA cm−2 compared with that of GCP 1.
It is of interest that ORR activity is not only determined solely by the presence of pyridinic‐N but also influenced by local structure with electrophilicity and nucleophilicity. The former may improve and the latter may suppress the ORR activity, based on the results by Surendranath and coworkers [39].
It is widely known that carbon catalysts show high electrocatalytic activities for 4e pathway of ORR in alkaline media, whereas the activity is not so high in acidic media. Li and coworkers addressed the difference in the selectivity of ORR between acidic and basic conditions from a view point of solvation effect based on DFT calculation [40]. They have reported that local hydrophobic/hydrophilic properties near the active site of nitrogen‐containing graphene quantum dots significantly influence the selectivity, that is, the 2e or the 4e pathway. The hydrophobic/hydrophilic property was evaluated by a dielectric constant. As shown in Figure 8.17, they chose graphene quantum dots with a phenazine structure (1 for short) as a model structure. In the mechanism of Figure 8.18, the phenazine catalyst molecule 1 is first reduced with two electrons and one proton to form carbanion in the resonance structure at a certain electrochemical potential. Then, an oxygen molecule adsorbs on the carbanion to form peroxygraphene of 3. If the peroxygraphene is decomposed to O2H−, the 2e pathway will proceed. On the other hand, if the peroxy group reacts with the internal carbon atom (like 4 and 5 as shown in Figure 8.18), 4e pathway will proceed. That is, the branching point for 2e and 4e pathway can be attributed to the cleavage of CO bond and OO bond, respectively, as shown in Figure 8.18. The effect of the hydro property on the branching point was thus examined by DFT with varying dielectric constant.
They first simulated how easily ORR takes place in alkaline media, depending on solvation with varying the dielectric constant from ε = 1 (gas phase) to 78 (bulk of water) as extreme cases. At ε = 1 or a hydrophobic condition, the CO bond cleavage is thermodynamically unfavorable. That is, the 2e pathway does not proceed. As for the 4e pathway at ε = 1, the activation barrier is relatively high but the free energy change is enough downhill (−1.48 eV). It is thus regarded that the 4e pathway proceeds in a hydrophobic condition, which agrees with experimental observations. On the other hand, for a dielectric constant of 78 or a hydrophilic condition, the reaction barrier for the release of HO2− is very small (below 0.1 eV) and the free energy change indicates a thermodynamically favorable reaction. That is, the 2e pathway is possible rather than the 4e pathway if it is hydrophilic. This is not the case in experimental results.
As described above, the solvation condition influences the activation barrier and the free energy change in ORR. They have thus considered that the local solvation effect near the active sites should be of importance because N‐doped graphene quantum dot of 1 has three long alkyl chains, which generally offer the limits of water access. That is, nitrogen‐doped graphene quantum dots of 1 is hydrophobic with a lower dielectric constant compared with that for bulk of water. It is generally accepted that active sites in enzyme are buried in local hydrophobic interior where local dielectric constant lowers below 10 [41]. They thus calculated the free energy diagram at low dielectric constants from 2 to 8 as shown in Figure 8.19a. It was revealed that the 2e pathway or CO bond cleavage becomes energetically unfavorable above ε = 2.6, whereas the 4e pathway or the OO bond cleavage remains thermodynamically favorable. They thus concluded that the active site of quantum dots of 1 is highly hydrophobic with a lower dielectric constant below ε = 2.6.
They also discuss the ORR selectivity of their catalyst 1 in acidic media. Free energy diagram of branching point for 2e and 4e selectivity of ORR at pH = 1 is presented in Figure 8.19b, setting a dielectric constant of ε = 4. It is found that the barrier for the 2e pathway is lower than that for the 4e pathway, indicating that the 2e pathway is preferential.
As described in Section 8.4, active sites of nitrogen‐doped carbon materials are created by pyridinic‐N. However, the exact active site(s) created by pyridinic‐N has not been clarified at the atomic scale. Thus, the role of pyridinic‐N in the ORR mechanism and the effect of the local structure around pyridinic‐N on the ORR activity are open questions, although some indications have been reported recently as described in Sections 8.5–8.7. Many basic problems concerning nitrogen‐doped carbon catalysts still remain in order to design active non‐Pt fuel cell catalysts. To fully understand the mechanism of ORR at the atomic scale, model catalyst studies using pyridinic‐N‐containing molecules with well‐defined local structures are required. The molecules are, for example, phenanthroline‐type, pyrazine‐type, and acridine‐type pyridinic‐N, and so on.
Once the ideal local structure of pyridinic‐N is clarified, highly active carbon catalysts are expected by arranging high‐density active sites using bottom‐up methods. In addition to design the active catalysts, one should design ideal triple‐phase boundaries around the active sites by controlling hydrophobic/hydrophilic properties as well as the local electronic conductivity in the real fuel cell system.
With these scientific and technological efforts, a state‐of‐the‐art nitrogen‐doped carbon catalyst will be eventually realized as Pt substitute catalysts for the ORR in fuel cells.