Catalytic activity of graphene-covered non-noble metals governed by proton penetration in electrochemical hydrogen evolution reaction

Graphene-covering is a promising approach for achieving an acid-stable, non-noble-metal-catalysed hydrogen evolution reaction (HER). Optimization of the number of graphene-covering layers and the density of defects generated by chemical doping is crucial for achieving a balance between corrosion resistance and catalytic activity. Here, we investigate the influence of charge transfer and proton penetration through the graphene layers on the HER mechanisms of the non-noble metals Ni and Cu in an acidic electrolyte. We find that increasing the number of graphene-covering layers significantly alters the HER performances of Ni and Cu. The proton penetration explored through electrochemical experiments and simulations reveals that the HER activity of the graphene-covered catalysts is governed by the degree of proton penetration, as determined by the number of graphene-covering layers.


Supplementary methods
Fabrication of graphene-covered Cu and Ni sheets. After the monolayer graphene growth on Cu foils by a standard chemical vapor deposition (CVD) process, O2 plasma treatment was operated to remove the graphene on one side of Cu foil, and then Nafion sheet was coated on the other side through a spin coater. The Cu foils were dissolved by 0.25 M Fe(NO3)3 solution at 24 °C for 12 h, and then the ultrapure water (resistivity: 18.2 MΩ) was used to replace the Fe(NO3)3 solution to remove ions several times completely.
After transferring a graphene sheet onto a Nafion-attached Cu or Ni sheet, the Nafion/monolayer graphene/Nafion on the Cu or Ni sheet was achieved. Finally, the asprepared sheet was heated at 130 °C for 2 h and coated by an insulator liquid gasket (ThreeBond 1211) to prevent the contact with acidic electrolyte (Supplementary Fig. 1).
Typical size of exposed reaction area was 2.0 × 2.0 mm. For preparing the Cu or Ni sheet with multilayer graphene covering, the process was as same as the steps shown in Supplementary Fig. 3.

Fabrication of a Si3N4 chip supported Nafion/graphene/Nafion membrane.
After the monolayer graphene growth on Cu foils, O2 plasma treatment was operated to remove the graphene on one side of Cu foils, and then Nafion sheet was coated on the other side with graphene through a spin coater. The Cu foils were dissolved in 0.25 M Fe(NO3)3 solution at 24 °C for 12 h, and then the ultrapure water (resistivity: 18.2 MΩ) was used to replace the Fe(NO3)3 solution to wash ions several times. After transferring a graphene sheet to a Si3N4 chip with an attached Nafion sheet on window area, the Nafion/monolayer graphene/Nafion membrane on a Si3N4 chip was achieved. Finally, the chip was heated at 130 °C for 2 h before using.
For the bilayer graphene membrane, one additional step was needed: transfer a Nafion protective monolayer graphene onto a Cu foil with monolayer graphene on one side (step 6 in Supplementary Fig. 3). The chip was heated at 130 °C for 2 h after graphene transfer every time. For preparing the membrane with graphene more than bilayer, the process from step 4 to 7 was simply repeated.
Synthesis of graphene-covered Ni nanoparticles. The 3D porous graphene substrate was achieved by etching the NiMo substrate of graphene-covered NiMo alloy using 2.0 M HNO3 solution mixed with isopropanol (volume ratio of HNO3 to IPA was 4:1) at 80 °C.
The deposition of NiO nanoparticles on 3D porous graphene was followed a reported literature with slight modifications 1 . 20 mL ultrapure water (Millipore, resistivity: 18.2 MΩ) contained 3D porous graphene, 20 mL ethanol, 4.5 g Ni(NO3)2•6H2O (FUJIFILM Wako Pure Chemical Corp., 98%), and 0.7 g urea (FUJIFILM Wako Pure Chemical Corp., 99%) were added into a 50 mL Teflon-lined stainless steel autoclave, and then treated at 120 °C for 24 h in a muffle furnace. After cooling down to 24 °C, the as-prepared products were washed by ultrapure water and ethanol several times, and further dried at 120 °C for 12 h. 10.0 mg of 3D porous graphene supported NiO nanoparticles was loaded on a corundum boat and inserted into the center of a quartz tube (φ30 × φ27 × 1000 mm) in a furnace. Through annealing at 300 °C for 20 minutes under an atmosphere of H2 (99.9999%, 100 sccm) and Ar (99.999%, 200 sccm), the deposited NiO nanoparticles were reduced to Ni nanoparticles. After the reduction, the furnace temperature increased to 700 °C for the N-doped graphene growth on Ni nanoparticles by a CVD method under a mixed atmosphere of H2 (100 sccm)/Ar (200 sccm)/pyridine (1.0 mbar, Ardrich, 99.8%, anhydrous). The layer number of encapsulating graphene was adjusted through controlling the CVD time. For example, deposition times were 1.0, 4.0, and 10.0 s for 1−2, 3, and 6−7 graphene layers, respectively. The average layer numbers were 1.9, 3.2, and 6.1 for the above three deposition times. After the CVD process, the furnace was cooled to 24 °C by using a fan and the resulting samples were stored for characterizations and measurements. anhydrous), respectively. The number of graphene covering layers was adjusted through controlling the CVD time. For example, deposition times were 1.0, 4.0, and 10.0 s for 1−2, 3, and 6−7 N-doped graphene layers, respectively. The average layer numbers were 1.7, 2.9, and 6.4 for the above three deposition times. After the CVD process, the furnace was cooled to 24 °C by using a fan and the resulting samples were stored for characterizations and measurements.

Synthesis of non-doped graphene (GL)-covered and N-doped graphene (NGL)-
covered Ni sheets for generated H2 bubbles observation. The Ni sheets were loaded on a corundum plate and inserted into the center of a quartz tube (φ30 × φ27 × 1000 mm) in a furnace. Through pre-annealing at 1000 °C for 30 minutes under an atmosphere of H2 (99.9999%, 100 sccm) and Ar (99.999%, 200 sccm), the covering GL on Ni sheet was synthesized through introducing CH4 flow (20 sccm, 99.995%) for 30 minutes at 1000 °C, while NGL was synthesized with an additional flow of pyridine (0.5 m bar) at 800 °C.
Finally, the as-prepared Ni sheets were fully coated by an insulator liquid gasket (ThreeBond 1211) to prevent the contact with acidic electrolyte. The size of exposed reaction area was 1.0 × 1.0 cm.  DFT calculations. We also performed DFT calculations to estimate the energy barrier for hopping of proton from graphene to the NiMo surface by using the VASP 2 code. We used the projected augmented wave (PAW) method 3 and the Perdew-Burke-Ernzerhof (PBE) functional 4 . The plane wave energy cutoff was set to 400 Ry. The dispersion correction was included using the Grimme's D3 (BJ) method 5, 6 .
The NiMo surface covered by a graphene with SV-3N defect was prepared based on our previous paper 7 . Since the structure of the 1:1 NiMo could not be well characterized in the experiment, we assumed that the NiMo system forms the δ-phase NiMo: Ni24(Ni4Mo16)Mo12 8,9 , and its (100) face forms the surface of the NiMo system. The We also calculated the energy barriers for proton penetration in the presence of aqueous environment using the CP2K program. The simulation cell lengths in the x-, y-, and z-directions were 25.56, 24.595, and 70 Å, which means that the z-length was increased by 20 Å to include water molecules. We added 300 water molecules on both sides of a defect-free graphene and a graphene with an SV-3N defect. After the 1 ns of equilibration using classical force fields 10 , we further performed 5 ps of equilibration using DFT molecular dynamics simulation. The temperature was kept to be 320 K using the canonical sampling velocity rescaling method. DZVP/TZV2P basis was used for graphene and water, respectively. The auxiliary plane-wave cutoff was set to 320 Ry.
Subsequently, we added H + to a water molecule near the interface and optimized the whole structure. Based on the optimized structures, the nudged elastic band method was applied to calculate the penetration barrier. The proton penetration barrier for the pristine graphene does not affected by the explicit water (from 3.16 to 2.97 eV) and the barrier height is consistent with a similar report 11 . On the other hand, the proton penetration barrier for the graphene with an SV-3N defect decreased from 3.30 to 1.93 eV, because the H3O + carries the proton close to the defect. Those snapshots are shown in Supplementary Fig. 38. We note that the proton penetration barriers are still higher than estimated values from experiments, which would be attributed to the atomic defects or bias potential 11 .
We further calculated the proton penetration barrier from H3O + in water to bilayer graphene with SV-3N defects for step 1−3, because the similar reduction of barrier was expected. We added 400 water molecules to one side of bilayer graphene with SV-3N defects ( Supplementary Fig. 40) and followed the same procedure described above. The energy barrier reduced from 4.63 to 2.00 eV. Finally, the activation energy of the interlayer proton transfer was calculated using the same system. The calculated value was 1.56 eV for water/graphene → graphene and 1.61 eV for graphene → water/graphene.
Since the activation barrier without water was 1.53 eV ( Supplementary Fig. 40), the effect of water molecules on the interlayer proton transfer is limited. Note, however, that the calculated proton penetration barrier in the presence of water molecules is a sample from various configurations of explicit water molecules.
Finally, we employed the above water/graphene/water structure and after equilibration and calculated the electrostatic potential under the electric field using the SIESTA code 12 . From the difference in the electrostatic potentials with and without electric field, the x-y averaged (z-direction is perpendicular to the graphene layer) electric field. Supplementary Fig. 23 shows that the voltage drop is dominated at the interface between water layer and graphene surface.

Supplementary discussions (1) Electrochemical impedance spectroscopy
We used electrochemical impedance spectroscopy to understand the resistance of proton penetration in details. Nyquist plots of the graphene membranes exhibited a semicircular arc at the frequency range of 10 4 Hz < ω <10 6 Hz. The signal in this region is attributed to proton penetration through the graphene layers ( Supplementary Fig. 16b) Table 4). The R1 of the Nafion membrane is 20 times less than that of the graphene membranes, which confirms that the resistance to proton penetration originates from the graphene layers.

(2) Intercalation of the proton into graphene interlayers
The intercalation of proton into graphene interlayers can further reduce the energy barrier of proton penetration through multilayer graphene. The non-linear (i.e. logarithmic) relationship between the average proton current and the layer number of graphene ( Fig. 3f) demonstrates that the interaction between adjacent graphene layers reduces the energy barrier of penetration. In contrast, an alternating Nafion/graphene laminated structure provided a linear relationship between proton current and graphene layer number, due to individual adsorption and desorption steps of proton penetration through each isolated graphene layer ( Supplementary Fig. 18b). The differences in configuration provided a much lower energy barrier for the stacked graphene membrane than for the alternating laminated membrane, which is good agreement with DFT calculations ( Supplementary Fig. 40b). Thus, protons adsorbed at the defect sites on the outermost graphene layer can reach the NiMo surface by the intercalation into the multilayer graphene.  Table 8), which means that graphene covering reduces catalytic activity. However, after 1000 cyclic voltammetry (CV) cycles, the η10 required for the bare NiMo sample was 800% higher than the initial η10 value at the 1st CV cycle, while the required η10 values for the NiMoNP samples with 1-2NGL, 3NGL, and 6-7NGL were 70.3, 13.7, and 0.9% higher than the initial ones at the 1st CV cycle ( Supplementary Fig. 28c), respectively, which demonstrates that graphene covering dramatically suppresses the degradation of NiMoNP's catalytic activity.
The chemical corrosion of each sample after cycling was examined by inductively  Fig. 28e). The NiMoNP/1−2NGL sample retained only 51.1% of its initial current density, whereas the current of the NiMoNP/6−7NGL sample maintained >99% of its initial value.
The NiMoNP/3NGL sample displayed a current density of 44 mA cm -2 (92.5% of its initial value), which continued for 25 h with a low rate of leaching of catalysts.

(5) Mo2C formation on the graphene-covered NiMoNP
Mo2C is known to be as a predominant component at a high carbonization temperature under a carbon gas atmosphere. The contribution of the HER-active Mo2C species should influence the proton penetration and the HER activity on the graphenecovered NiMoNP samples. Therefore, the annealing temperature dependence of Mo2C formation during the CVD process were investigated. In the XRD patterns ( Supplementary Fig. 33 Figs. 27c and 27d). Thus, the metallic state of NiMo alloy is predominant for at the carbonization temperature of 700 °C, which was employed in the main text for NiMo/NGL samples.

(6) Influences of defects on graphene lattice toward the HER activities of the graphene-covered NiMoNP
The NiMoNP/6-7NGL sample with a higher defect density of graphene was synthesized at a lower carbonization temperature of 500 °C, which was used to investigate the relationship between the defect density and the HER activity. The higher defect density of graphene (ID/IG = 1.44) in the 500 °C-prepared NiMoNP sample was confirmed by Raman spectra as compared to the counterpart of the 700 °C-prepared NiMoNP sample (ID/IG = 0.87) (Supplementary Fig. 34a). The HER activity of the 500 °C-prepared NiMoNP sample showed an improved initial HER activity ( Supplementary Fig. 34b) in Ar-saturated 0.5 M H2SO4 in a three-electrode system, owing to the high density of structural defects. However, the poor stability of the 500 °C-prepared NiMoNP sample was also observed. Furthermore, a HER-activity comparison between NiMoNP/1-2NGL and NiMoNP/1-2GL (i.e., covered by 1-2 layers non-doped graphene) was also carried out in Ar-saturated 0.5 M H2SO4 in a three-electrode system. Compared with the NiMoNP/1-2NGL sample (ID/IG = 0.78), the NiMoNP/1-2GL sample showed a very small ID/IG value of 0.09 ( Supplementary Fig. 34c). After 1000 CV cycles testing, the NiMoNP/1-2GL sample exhibited a better stability than the NiMoNP/1-2NGL sample, owing to the lower number of structural defects on graphene covering layers ( Supplementary Fig. 34d). Taken together, these results mean that the higher defect density of graphene enhances both proton penetration and dissolution of NiMo through defect-rich regions on graphene, reducing the catalyst lifetime. Thus, the balance of defect density plays an important role in HER performances (activity and catalyst lifetime).
Complete encapsulation on the metal catalyst has been examined by following a similar acid-treatment method in reference 13 . We measured and compared the HER  Moreover, we evaluated the energy barrier and the order from high energy barrier to low energy barrier of hydrogen molecule penetration as follow.
Therefore, it is a dominant pathway that hydrogen molecules penetrate without the split and recombination.

(8) Ejection of generated molecular hydrogen through graphene
The ejection of H2 generated on the Ni surface through graphene layers was investigated by measuring CV with different scan speeds. The NGL-/GL-covered Ni sheets ( Supplementary Figs. 44a and 44b), a graphitic rod, and 0.5M H2SO4 solution were used as the working electrode, counter electrode, and electrolyte, respectively. Firstly, CV was performed for the NGL-covered Ni sheet under two scan speeds (slow one: 0.01 mV s −1 and fast one: 1.0 mV s −1 ). Through manual operation of suction by a pipette, we confirmed that the generated H2 bubbles (Supplementary Fig. 44c) were attached on the NGL under both scan speeds of 1.0 and 0.01 mV s −1 , and we could not find the bubbles in-between NGL and Ni surface after tossing all removable bubbles ( Supplementary Fig.   45a). We also observed that the Raman mapping of G band intensity and ID/IG intensity ratio before and after reaction did not show any significant change of the NGL ( Supplementary Fig. 46a). However, it is in stark contrast with the GL-covered case, where one can clearly see the bubble encapsulation was observed at the low potential range from 0.0 V to −0.2 V (vs. RHE) with a slow scan speed (0.01 mV/s) ( Supplementary   Fig. 45b). The slow CV scan speed (0.01 mV s -1 ) resulted in a slow H2 generation speed, thus the H2 bubbles encapsulated by defect-less GL could be observed. In another case, the fast CV scan speed (1.0 mV s -1 ) results in a fast H2 generation speed, which was much faster than the H2 ejection speed through GL. Finally, the GL that encapsulates H2 bubble burst, which confirmed by Raman mapping (Supplementary Fig. 46b). These differences between defect-rich NGL-and defect-less GL-covered cases can be attributed to the balance between the ejection of molecular hydrogen through graphene layers dominated by defects/nanopores and the generation speed of molecular hydrogen dominated by experimental parameters (i.e., scan speed and potential range). Indeed, the defect-rich regions in NGL contribute to the efficient ejection of molecular hydrogen due to the smaller energy barrier (Supplementary Table 11).
Recently, the H2 permeation through a graphene layer can be explained by the ripples and defects inducing a local curvature as catalytically active sites 14 . Our calculation data suggest that the defects/nanopores induced by N dopants reduced the activation energy barrier of H2 penetration, achieving a lower value than that for GL (Supplementary Table 11 and Supplementary Fig. 43). Thus, one can conclude that H2 prefer to be ejected through the nanopores/defects in NGL and then the gathered H2 molecules form a big bubble.

Supplementary Figures
Supplementary Figure 1

. Preparation of the transferred CVD graphene-covered Cu
and Ni sheets. For preparing the Cu or Ni sheet with multilayer graphene covering, the process was as same as the steps shown in Supplementary Fig. 3. sample with NiMo substrates. The 6−7NGL covered sample was a good target to verify the proton penetration effect on HER activity, because of two major reasons. One was the charge transfer effect from the underlying metal substrates to graphene covering layers was limited to three layers. 13 The 6−7NGL can totally exclude the charge transfer effect.

Supplementary
The other reason was that the phenomenon of proton penetration through graphene layers was not observed beyond 10NGL (Fig. 3e). Thus, we concluded that the 6−7NGL was the best sample to investigate proton penetration effects on HER in absence of the charge transfer effect. (2) the desorption of the proton from the graphene lattice. The necessary energy needed to be overcome for proton penetration is called "energy barrier". The negative energy value for "SV-3N" lattice means that the proton energetically prefers to adsorb on the lattice rather than penetration, therefore an energy of 3.30 eV is needed for desorption. As shown in the energy diagram, the energy barriers for defect-free lattice and SV-3N lattice were 3.16 eV and 3.30 eV, respectively. In addition, the positive or negative value means that the shape of the potential energy surface is peak-shape or trough-shape.