Nitrogen-doped tungsten carbide nanoarray as an efficient bifunctional electrocatalyst for water splitting in acid

Tungsten carbide is one of the most promising electrocatalysts for the hydrogen evolution reaction, although it exhibits sluggish kinetics due to a strong tungsten-hydrogen bond. In addition, tungsten carbide’s catalytic activity toward the oxygen evolution reaction has yet to be reported. Here, we introduce a superaerophobic nitrogen-doped tungsten carbide nanoarray electrode exhibiting high stability and activity toward hydrogen evolution reaction as well as driving oxygen evolution efficiently in acid. Nitrogen-doping and nanoarray structure accelerate hydrogen gas release from the electrode, realizing a current density of −200 mA cm−2 at the potential of −190 mV vs. reversible hydrogen electrode, which manifest one of the best non-noble metal catalysts for hydrogen evolution reaction. Under acidic conditions (0.5 M sulfuric acid), water splitting catalyzed by nitrogen-doped tungsten carbide nanoarray starts from about 1.4 V, and outperforms most other water splitting catalysts.

H ydrogen gas has long been regarded as a clean and sustainable energy carrier for replacing traditional fossil fuels. Cost-effective manufacture of hydrogen gas is one of the key points that remain to be addressed for a successful hydrogen economy [1][2][3] . Water electrolysis driven by electricity from sustainable energy sources (e.g., wind, or solar) is an environmentally friendly scheme to produce hydrogen gas with high purity [4][5][6][7][8][9][10][11] . Compared with alkaline electrolysis, hydrogen production in acids like in polymer electrolyte membrane electrolysis has the advantages of simplicity, high-current density, and high-pressure compatibility. However, high-performance catalysts for HER and OER in acid media are mostly limited to noble metals (e.g., platinum (Pt), iridium (Ir), and ruthenium (Ru)). The expensive cost and limited reserve of noble metals have restricted their wide application. Remarkable endeavors have been made to develop non-noble metal catalysts for HER in acid 12 , such as metal sulfides [13][14][15][16][17] , metal carbides 5,[18][19][20] , and metal phosphide [21][22][23] . Unfortunately, most non-noble metal catalysts for HER in acid are either of large onset overpotential [24][25][26][27] or lack of working stability especially at high current densities 11,[16][17][18][19] . As for OER in acid electrolyte, the effective catalysts are still limited to noble metal catalysts 28,29 , while only a few non-noble metal catalysts (such as Co 3 O 4 30 , MnO x 31 , and CoMnO x 32 ) have been reported, but with large onset potentials and poor stability.
Tungsten carbide (WC) is an earth-abundant and inexpensive catalyst, which is stable in acid and behaves like Pt in hydrogenolysis due to its electron configuration around the Fermi level 33,34 . These properties make WC a promising candidate for applications as a non-noble metal electrocatalyst for HER in acid 5,18,35,36 . Despite its popularity, our theoretical calculations show that adsorbed hydrogen atoms bind strongly to WC favoring H + reduction but hindering hydrogen desorption. It is therefore important to explore materials based on WC with better performance. Tungsten carbide nanorods have been synthesized without template 37,38 . In 2013, Hashimoto et al. synthesized a Fe-WCN electrocatalyst 39 . They suggested that the electron density of W atoms was abated by W-N bond and therefore improved the HER catalytic activity. Next, Chen et al. reported an electrocatalyst that combined tungsten carbide-nitride and graphene nanoplatelets, showing that synergistic effects between W 2 C and WN phases promote outstanding hydrogen evolution catalytic activity 40 . However, the effect was limited to only low current densities and the underlying mechanisms remained ambiguous. As a result, none of the materials displayed current densities as large as −200 mA cm −2 . Furthermore, besides the catalyst's intrinsic activity, gas bubbles' fast removal from the surface of the electrode is important for efficient gas evolution. Strong bubble adherence to the electrode surface interferes with charge and mass transfer between the catalyst and the electrolyte. Catalysts may also be peeled off upon departure of bubbles due to strong adhesion. Therefore, the activity of catalysts for gas evolution reactions can be significantly improved with superaerophobic electrodes 20,21 . Electrode design based on nanostructuring and atomic level modulation is thus a promising approach to accelerate the activity and stability for hydrogen evolution reaction.
Here, we report a superaerophobic catalytic electrode for both HER and OER in acid based on N-doped WC nanoarray structures. The design strategy is multifold. N-doping modulates the surface energy level to optimize hydrogen binding and thus promote HER kinetics. The nanoarray structure not only exposes more active sites for electrochemical reaction but also facilitates gas release by offering a superaerophobic interface under water. Therefore, the interface has weak bubble adhesion, detaching small bubbles with little contact area that favors gas evolution at heterogeneous catalytic surfaces. As expected, the N-WC nanoarray electrode offers excellent HER activity and stability to −200 mA cm −2 , outperforming the vast majority of reported   Fig. 1). The partial density of states (DOS) on individual surface atoms of WC are displayed in Fig. 1d, as compared to those in N-WC, to analyze the N-doping effect on WC's electronic structure. We note that the density of 5d states of W atoms on N-WC (001) surfaces is downshift relative to that on WC (001) surfaces, suggesting a more negative d-band center of W atoms on N-WC (001). Indeed, the d-band center of W on WC (001) is calculated to be −1.12 eV, while that on N-WC (001) is calculated to be −1.50 eV, close to that on Pt (111) surfaces. The lower d-band center results in weaker H adsorption to the catalytic surface ( Supplementary Fig. 2) 43 . These results provide valuable insights on N-doping effect on the electronic structure, and imply more suitable H adsorption on the N-WC surface for promoting HER activity.
Synthesis of N-WC nanoarray electrode. The N-doped WC nanoarray electrodes were prepared via a sequential two-step approach (Fig. 2a). In the first step, a controlled hydrothermal reaction of tungstic acid was used to grow WO 3 nanoarray on carbon fiber paper (CFP). Subsequently, the WO 3 nanoarray was subjected to reduction and simultaneously carbonization and Ndoping at high temperature, using melamine as both C and N source. The chemical vapor deposition conditions were optimized to achieve high HER activity. The current densities for an applied potential of −200 mV vs. RHE were measured as indices. The optimal conditions that afford highest catalytic current density were found to be 850°C for 3 h ( Supplementary Fig. 3). The Ndoped WC powder without nanoarray structure (N-WC), WC nanoarray without N-doping (WC nanoarray), and WC powder (WC) were also prepared at the optimized temperature and time to serve as control samples (see details in "Methods" section). Figure 2b shows the scanning electron microscopy (SEM) of WO 3 grown on CFP, which exhibits a nanoarray structure composed of thin belts, with about 5 μm long,~400 nm wide, and 10 nm thick. The morphology of WO 3 was well maintained after being converted to N-WC nanoarray (Fig. 2c) except for becoming 2-3 nm thinner. The perfect maintenance of nanoarray structure could be attributed to the thin-belt morphology that most oxygen atoms were near the surface, so oxygen atoms could be easily taken out and substituted by carbon or nitrogen atoms without damage. Figure 3a displays the X-ray diffraction (XRD) data of CFP, WO 3 nanoarray, and N-WC nanoarray. In the XRD profile of CFP, the peaks at 26.4°and 42.5°are features of graphite (PDF#41-1487). The XRD pattern of WO 3 nanoarray matches well with hexagonal WO 3 (PDF#75-2187). After reduction and carbonization of WO 3 , hexagonal WC phase (PDF#51-0939) was successfully prepared as shown in the XRD pattern of N-WC nanoarray, in which the peak at 42.5°refers to extra tetragonal carbon (PDF#54-0501). The transmission electron microscopy (TEM) images of Fig. 3b show that each N-WC belt consists of parallel narrow belts arrayed side by side.  Supplementary Fig. 7 shows the absence of nitrogen. X-ray photoelectron spectroscopy (XPS) data was performed to get more insights on the chemical environment of N-WC nanoarray. From the XPS data in Fig. 3c, the N element can be identified and the content at the surface of the sample is about 8.0 at.%. Figure 3d displays N 1s high-resolution XPS profile. The fitting peaks at 399.7 and 401.3 eV can be assigned to features of N-C bonds, and the peak at 398.7 eV has been reported to represent N-W bonding 44 . The presence of the N-W bond is also supported by the W 4f XPS in Fig. 3e. The peak at higher binding energy of 41.3 eV results from the inevitable surface oxidation of N-WC nanoarray upon exposure to air. Binding energy at 35.5 and 38 eV are supposed to be W-N bonds 45 , and the peaks centered at 32.5 and 34.6 eV are attributed to be W-C bonds 18,46 .  Table 1 show the TEM-EDS result and the corresponding element distribution of N-WC nanoarray. As shown in elemental mapping, N is uniformly distributed along the N-WC nanobelts. The TEM-EDS spectrum of N-WC nanoarray is displayed in Supplementary Fig. 8. According to the TEM-EDS result, the N content is about 6.6 at. %, which is close to the XPS survey result, indicating that the N element is uniformly doped inside the sample.  Table 2). Such a catalytic activity outperforms previously reported carbide-based electrodes (Supplementary Table 3). The performance of N-WC nanoarray is also outstanding among the best of non-precious metal-based HER electrocatalyst in acid (Supplementary Table 4). The electrochemical active surface areas (ECSA) data were derived from CV curves at non-Faraday area (see details in "Methods" section and Supplementary Fig. 9). Both the nanoarray samples show significant larger ECSA than the powder samples (Supplementary Fig. 10). After ECSA normalization, the superiority of N-WC samples is still remarkable, further demonstrating the improvement of intrinsic activity by N-doping ( Supplementary  Fig. 11). Also, the nanoarray samples show much higher activity than their powder counterparts, which implies that the nanoarray structure favors HER not only because of the enlarged surface area, but also due to the accelerated gas bubble departure. As displayed in Fig. 4b, the Tafel slope of N-WC nanoarray for HER is −75 mV dec −1 , which is the smallest among all the WCbased electrocatalysts, demonstrating a more favorable HER process of the N-WC nanoarray electrode. The fast HER reaction kinetics of N-WC is also reflected by a small charge transfer resistance of 0.5 ohms at the potential of −0.5 V vs. reversible hydrogen electrode (Fig. 4c).
The stability of N-WC nanoarray was tested at an overpotential of −0.144 V (~−20 mA cm −2 ) for the first 10 h, and then −0.156 V (~−50 mA cm −2 ) for another 10 h, as displayed in Fig. 4d. The current density increased by 2.9% at the first stage and then declined 1.9% at higher overpotential in the following 10 h of electrolysis. The excellent stability at −50 mA cm −2 has never been reported for tungsten carbide 18,35,36 . The increase of current density at −0.144 mV implies that the electrode is activated during HER first. Then, the decay of catalytic current appears in the next period. To verify this process, we further tested the stability of a fresh sample at about −60 mA cm −2 . As depicted in Supplementary Fig. 12, the current density increased in the first 6 h and then decreased a little to an equilibrium state that the current remained unchanged throughout the rest 12 h. The current density decreased about 0.5% after the total 22 h test at −60 mA cm −2 . To simulate the industrial production of H 2 at large current density with interruptions, we performed interrupted stability test at −100 mA cm −2 (Supplementary Fig. 13).
The current density decreased about 14% after 30 h electrolysis of water. The outstanding catalytic stability can be attributed to its high chemical stability and robustness of the nanoarray structure for bubble releasing. From the postmortem XPS analysis ( Supplementary Fig. 14), excellent stability of N-WC nanoarray is further confirmed by the negligible structural differences before and after long-term electrolysis of water.
Bubble behavior. The bubble releasing behaviors during the electrocatalytic reaction at −200 mV vs. RHE were recorded with a high-speed camera system. It is clearly seen that WC and WC nanoarray exhibit less and slower H 2 evolution than N-WC and  Fig. 5a). The observed significant difference shows that N-doping increases the intrinsic catalytic activity by modulating the electronic structure of WC as suggested by our computational predictions. Meanwhile, on the surfaces of WC and N-WC electrodes, we could see large areas of the light-reflecting gas films which result from H 2 gas strongly adhered on the electrode surface. Once the active sites are isolated from the aqueous electrolyte by adhered bubbles or gas films, the electrocatalytic reaction slows down. In contrast, there was no obvious gas film or large bubbles adhered on the surface of WC nanoarray or N-WC nanoarray, thus the catalytic HER continues without interruption. Furthermore, the bubbles on N-WC nanoarray are much smaller and leave much faster than bubbles on the other electrodes. The significant differences in the bubble behaviors suggest a weaker adhesion of bubbles at the nanoarray electrode than on the flat electrode, which is quantitatively described by the contact angle (CA) and adhesion force (AF). As shown in Fig. 5b and c, non-nanostructured electrodes show CA = 146°and AF = 19 μN for WC, while CA = 148°and AF = 12 μN for N-WC. In contrast, the two nanoarray electrodes show significant increases in CA (156°for WC nanoarray and 163°for N-WC nanoarray) and decreases in AF (5 μN for WC nanoarray, and almost undetectable for N-WC nanoarray). These data support the fact that surface roughening can improve the catalytic HER activity by weakening the interactions of the bubbles with the electrode surface, thus favoring rapid release of H 2 bubbles. Furthermore, the N-doped WC electrodes show a slightly larger CA and weaker AF than their non-N-doped WC counterparts of the same structure, showing further evidence that N-doping weakens the force between H 2 bubbles and the electrode surface.
The size distributions of bubbles detached from the electrodes are quantified statistically in Fig. 5d. The average bubble sizes on WC nanoarray and N-WC nanoarray are not distinctly different, and similar observation applies to WC and N-WC electrodes. However, when the nanoarray electrodes are compared to their non-structured counterparts, the advantages are clear-smaller bubble size along with larger CA and weaker AF. Accordingly, the WC nanoarray exhibits much better catalytic activity than WC, while N-WC nanoarray outperforms N-WC. Therefore, the accelerating effect of nanoarray structuring toward HER is further confirmed. RHE. Although the OER activity of N-WC nanoarray is impressive, the stability of N-WC nanoarray for OER in acid still needs to be improved, as shown in Fig. 6c. The potential increases from 1.35 to 1.54 V vs. RHE after 1 h of continuous water splitting at 10 mA cm −2 . To acquire the reason for the instability of N-WC nanoarray, we recorded the XRD patterns of N-WC nanoarray before and after OER test, showing that a small amount of tungsten oxides formed on N-WC nanoarray during OER test (Fig. 6d). Although N-WC nanoarray is not stable enough for practical application for now, the N-WC-based anode would represent an important research avenue for non-precious OER catalyst in acid.
Electrocatalytic water splitting. With N-WC nanoarray as the cathode, either N-WC nanoarray (electrolysis cell noted as W-W set) or commercial Ir/C (loading of 5 mg cm −2 ) electrode (electrolysis cell noted as W-IrC set) as the anode, the overall water splitting performance was measured. At room temperature (20°C ), the experiment was carried out in 0.5 M H 2 SO 4 , and the corresponding Volt-ampere curves are displayed in Fig. 7a. In terms of the W-IrC set, the overall water splitting reaction begins at about 1.5 V. We conducted a chronoamperometry test at 1.5 V without iR compensation and recorded the video of the electrodes (Supplementary Movie 2). From the snapshot displayed in Fig. 7b, bubbles can be seen clearly on both the anode and the cathode, consistent with water splitting starting at about 1.5 V. The N-WC nanoarray thus show promise for practical applications as a HER cathode. Surprisingly, the overall water splitting of the W-W set begins at about 1.4 V. Before 1.7 V vs. RHE, the current density reaches up to 30 mA cm −2 . These results are significant since such an excellent performance for catalytic water electrolysis in acid has never been reported with non-noble metal catalysts. The chronoamperometry test of W-W set at 1.4 V without iR compensation was conducted. A recorded video of the electrodes (Supplementary Movie 3) shows bubbling vigorously on the electrodes (the snapshot in Fig. 7c), confirming highefficient catalytic water splitting with N-WC nanoarrays as the bifunctional catalytic electrode for both the anode and the cathode. Furthermore, we tested water splitting performance of the W-IrC set and the W-W set powered by a commercial AA battery, as shown in Fig. 7d and corresponding videos (Supplementary Movies 4 and 5). Bubbles evolve from the electrodes of W-IrC set (Supplementary Movie 6, Fig. 7e), and more violently on the electrodes of W-W set (Supplementary Movie 7, Fig. 7f).

Discussion
In summary, we have designed, constructed, and demonstrated a superaerophobic N-WC nanoarray catalyst for water splitting into H 2 and O 2 via controlled synthesis of WO 3 nanoarray, followed by simultaneous reduction, carbonization, and N-doping with melamine at high temperature. To make a clear cut, Ndoping plays the most important role in elevating the HER activity of N-WC nanoarray by improving the intrinsic activity at atomic level. At the same time, nanostructuring also boots the  WO 3 powder without nanoarray structure was also synthesized by the same method without CFP.
Reduction, carbonization, and N-doping. WO 3 nanoarrays on a combustion boat that containing melamine was heated at various temperatures (T) and time (t) periods, with Ar gas flowing in a tube furnace. To optimize the catalytic performance, we tried T = 650, 750, 850, and 950°C, and t = 0.5, 1, 3, and 6 h. The current densities (without iR correction) at overpotential of −200 mV were measured as indices. The optimal condition to afford highest catalytic current density was found to be 850°C for 3 h (Supplementary Fig. 3). The N-WC powder without nanoarray structure was using WO 3 powder as the precursor, carbonized and Ndoped at the same condition with the best performed N-WC nanoarrays. WC nanoarrays and WC was made at the same temperature and time with N-WC nanoarrays separately, except using Camphora as the carbon source.  Fig. 7 Overall water splitting. a The Volt-ampere curves of overall water splitting with N-WC nanoarray as the cathode, while using Ir/C (noted as W-IrC set) or N-WC nanoarray (noted as W-W set) as the anode. b The video snapshot of water electrolysis by W-IrC set with voltage at 1.5 V. c The video snapshot of water electrolysis by W-W set with voltage at 1.4 V. d Setup for overall water splitting powered by a commercial AA battery, and the corresponding snapshots of gas bubbling with water electrolysis of W-IrC set (e) or W-W set (f) of the WC-based catalysts are all of 10 mg cm −2 . Pt/C electrode was made by the same procedure except that 2 μl Nafion dispersion and 0.4 mg 20 wt.% Pt/C were mixed with 0.5 ml isopropanol to form catalyst ink. The loading of Pt/C electrode was 0.4 mg cm −2 . Following the same procedure, Ir/C and IrO 2 electrodes were made. 50 mg 20 wt.% Ir/C and 100 μl Nafion dispersion were mixed with 5 ml isopropanol to form Ir/C catalyst ink. IrO 2 catalyst ink was made by 11.7 mg IrO 2 and 20 μl Nafion dispersion mixed with 5 ml isopropanol. The metal loadings of Ir/ C and IrO 2 electrode were 10 mg cm −2 . LSV for HER was performed at 1 mV s −1 . LSV for OER was performed at 5 mV s −1 . ECSA data were derived from CV curves of WC-based catalysts at non-Faraday area with different scan rates. At −0.5 V vs. RHE, EIS was measured from 100 kHz to 0.005 Hz with amplitude of 10 mV. Loadings of all catalysts are listed in Supplementary Table 5.
Bubble behavior study. At room temperature, CA of hydrogen bubbles on different electrodes in 0.5 M H 2 SO 4 were measured in ambient air by the captivebubble method 47 and the bubble size was controlled at 2 μl. AF between hydrogen bubbles and electrodes in 0.5 M H 2 SO 4 was tested by Dataphysics DCAT21, Germany 20. The video for hydrogen evolution process was recorded by Microscope (SZ-CTC, OLYMPUS) with X-Motion, AOS Technologies. Statistics. In terms of Fig. 5d, 50 bubbles detaching from the surface of each electrode were randomly recorded and used directly in the box plots.
DFT calculations. All periodic boundary calculations were performed by Vienna ab initio simulation package [48][49][50][51] . To describe the electron-ion interactions, projector augmented plane wave method 52,53 and the PBE exchange-correlation functional 54 were used. For the plane wave basis set in all calculations, a cutoff of 450 eV was chosen. For the optimization of bulk Pt and WC, a 7 × 7 × 7 Monckhorst-Pack type k-point grid was chosen. For the calculation of Pt (111) and WC (001) slabs, a 5 × 5 × 1 Monckhorst-Pack type k-point grid was chosen. The second-order Methfessel-Paxton was used for Brillouin-zone integration 29 and the σ-value was chosen to be 0.1 eV. For structure optimization, the force convergence criterion of 0.01 eV Å −1 was used. The energy convergence criterion was set to be 10 −4 eV per unit cell. To model the Pt (111) surface, five layers of Pt atoms and four Pt atom per layers were used. The bottom three layers were fixed during optimization. Four layers of W and four layers of C atoms and four atoms per layer were used to model the WC (001) surface. To model the N-WC (001) surface, we replaced half of the subsurface C atoms with N atoms, which corresponding to an atomic percentage of N atoms of 6.25%, close to the experimental atomic percentage of 6.61%. The bottom two layers of W and C atoms were fixed during geometry optimization for WC (001) and N-WC (001) surfaces. A vacuum slab of about 15 Å was inserted between the surface slab. A supercell of 15 Å × 15 Å × 15 Å and a 1 × 1 × 1 Monkhorst-Pack k-point mesh was employed for the calculation of H 2 . Only one hydrogen atom was included in the calculation of the hydrogen absorbed surfaces, corresponding to a surface coverage of ¼. The HBE was calculated as HBE = E total -0.5E H2 -E surf . E total is the absorbed system's total energy, E surf is the optimized bare surface's energy, and E ads is the energy of the adsorbate (H 2 ) in vacuum.
Data availability. Upon reasonable request, more data are available from the corresponding author.