Nitrogen-doped mesoporous carbon of extraordinary capacitance for electrochemical energy storage

See allHide authors and affiliations

Science  18 Dec 2015:
Vol. 350, Issue 6267, pp. 1508-1513
DOI: 10.1126/science.aab3798

Store more energy with a touch of nitrogen

In contrast to batteries, capacitors typically can store less power, but they can capture and release that power much more quickly. Lin et al. fabricated a porous carbon material that was then doped with nitrogen. This raised the energy density of the carbon more than threefold—an increase that was retained in full capacitors, without losing their ability to deliver power quickly.

Science, this issue p. 1508


Carbon-based supercapacitors can provide high electrical power, but they do not have sufficient energy density to directly compete with batteries. We found that a nitrogen-doped ordered mesoporous few-layer carbon has a capacitance of 855 farads per gram in aqueous electrolytes and can be bipolarly charged or discharged at a fast, carbon-like speed. The improvement mostly stems from robust redox reactions at nitrogen-associated defects that transform inert graphene-like layered carbon into an electrochemically active substance without affecting its electric conductivity. These bipolar aqueous-electrolyte electrochemical cells offer power densities and lifetimes similar to those of carbon-based supercapacitors and can store a specific energy of 41 watt-hours per kilogram (19.5 watt-hours per liter).

Carbon supercapacitors have outstanding attributes of low weight, very fast charging/discharging kinetics, and bipolar operational flexibility. For carbon-based materials, only electrical double-layer capacitance (EDLC) is available; thus, surface area is the key concern. But even at a very large surface area (~2180 to 3100 m2 g−1), their specific capacitance is still relatively low (~250 F g−1), which has limited their appeal (14). Meanwhile, graphene has a theoretical EDLC of ~550 F g−1 (5, 6) because of its extraordinary conductivity and specific surface area (∼2630 m2 g−1). In practice, however, its capacitance has also been limited to ~300 F g−1, about the same as the best carbon-based EDLC (2, 57). Therefore, efforts have been made to enable redox reactions in ordered mesoporous carbon (OMC) (8, 9) and conducting polymers by N doping, which via proton incorporation can theoretically endow a capacitance of ∼2000 F g−1 to conducting polymer polyaniline (10). Nevertheless, such efforts have failed because conducting polymers are too unstable for practical electrochemical cells, while stable OMC is too resistive to deliver a high capacitance or power.

We demonstrate that N doping can turn inert graphene-like layered carbon into an electrochemically active substance. The preparation method is described in the supplementary materials, starting with a sacrificial mesoporous silica template, which contains self-assembled tubes (11) later covered by few-layer carbon. After etching away silica, a self-supported ordered mesoporous few-layer carbon (OMFLC) superstructure in Fig. 1A remained. Various N-doped OMFLC (OMFLC-N) having N incorporated at several OMFLC locations in Fig. 1B were also obtained, some further modified by a HNO3 oxidation treatment that partially converted N into N−O. This set of OMFLC-N samples (table S1) includes 8.2 atomic percent (at%) N before and after oxidation treatment (samples S1 and S2) and 11.9 at% N before and after treatment (samples S3 and S4). For comparison, an ordered mesoporous (amorphous) carbon and a commercial activated carbon (YP-50, Kuraray Chemical) were also studied. To demonstrate the relevance to practical applications, we further implemented the idea using a simplified, template-free, scalable method producing essentially the same N-doped mesoporous few-layer carbon materials with the same overall performance.

Fig. 1 Structure of N-doped ordered mesoporous few-layer carbon and related materials.

(A) Fabrication schematic of ordered mesoporous few-layer carbon (OMFLC). (B) Possible locations for N incorporation into a few-layer carbon network. (C and D) High-angle annular dark-field transmission electron microscopy (TEM) images of ordered mesoporous carbon (OMC) (C) and OMFLC (D); dark regions indicate connected pore channels. (E) High-resolution TEM image of OMFLC; nanoporous walls consist of few-layered carbon sheets. (F) Low-angle x-ray diffraction patterns of OMC, OMFLC, and OMFLC-N (S1), showing characteristic (100), (110), and (200) peaks of hexagonal packing. (G) Pore size distributions of OMC, OMFLC, and OMFLC-N (S1). (H) Wetting angles of 0.5 M H2SO4 droplet on OMFLC (85°) and OMFLC-N (S1) (21°) substrates.

The ordered mesoporous nature of OMC (Fig. 1C) and OMFLC (Fig. 1D) was confirmed by electron microscopy. In both, OMFLC tubes appear as bright strips 4 to 6 nm wide. The tubes are porous, containing pores 1 to 2 nm in diameter (dark regions in strips), and are separated by aligned pore channels (dark regions between strips) of about the same size or diameter. High-resolution imaging of OMFLC’s tube walls further revealed graphene-like sheets with ≤5 layers (Fig. 1E). These relatively homogeneous and uniform mesoporous textures were largely preserved in OMFLC-N (fig. S1). The silica tubes in the template are known to form a two-dimensional (2D) hexagonal “crystal” with space group p6mm (11). The same superstructure was confirmed in OMC, OMFLC, and OMFLC-N by their diffraction patterns (Fig. 1F), which show decreasing peak intensities in the above order, indicating a progressive distortion of the superstructure.

Nitrogen adsorption-desorption suggests a bimodal pore size distribution (Fig. 1G) centered around 1.8 nm and 3.5 to 4.0 nm in all three mesoporous structures. They share similar N2 adsorption-desorption isotherms with a Langmuir hysteresis (fig. S2A) typical of well-defined mesopores. Among them, OMFLC-N (S1) has the largest surface area (1580 m2 g−1), the largest total pore volume (2.20 cm3 g−1), the smallest average pore width (2.25 nm), and the most prominent pores smaller than 2 nm (Fig. 1G). The characteristic Raman 2D band (fig. S2B) verified the formation of local graphene-like structure with ≤5 layers in OMFLC and OMFLC-N.

Local graphene-like structure formation and N doping profoundly altered other physical properties as well. Whereas OMC is clearly an insulator, OMFLC and OMFLC-N display much lower room-temperature resistance with much weaker temperature dependence (fig. S2C), indicating an improvement in the structural order of carbon (12). Meanwhile, whereas both OMC and OMFLC are hydrophobic, OMFLC-N is hydrophilic, wetting a 0.5 M H2SO4 droplet in Fig. 1H. This is consistent with the zeta potential: Nearly neutral OMFLC (zeta potential = −6 mV, almost the same as OMC’s −4 mV) becomes more nucleophilic OMFLC-N (−20 mV) as a result of lone-pair N 2pz electrons; these improved physical properties of OMFLC-N are generally conducive to the supercapacitive performance (see below).

Spectroscopy studies identified C-bonding and N-C locations in the carbon network. The presence of sp2 bonding expected for graphene and local graphene-like structure was evident from the high ratio of π* bonding to π* + σ* bonding (fig. S3A), giving 98% (±2%) sp2 bonding in OMFLC versus 86% (±2%) in OMFLC-N, with reference to graphite (100%). Evidence for N substitution in OMFLC-N was also detected (fig. S3B), and the N/O content and bonding of OMFLC-N quantified by x-ray photoelectron spectroscopy (XPS) (fig. S3, C to F, and table S1) provided the following picture: (i) Deconvoluted N 1s XPS contains three characteristic peaks at 398, 400, and 401 eV, corresponding to pyridinic (N-6), pyrrolic (N-5), and graphitic (N-Q) nitrogen, respectively, as shown in Fig. 1B (13, 14). (ii) As the N content increases from ~8.2 at% in sample S1 to 11.9 at% in S3, N substitution at “regular” graphitic C sites (N-Q) instead of defective sites (N-5 and N-6) becomes more abundant. (iii) Oxidative HNO3 treatment caused the least stable N-5 to substantially convert to N−O (N associated with an O, shown in fig. S3D at 403.2 eV) (15) without affecting the most stable N-Q, as suggested by the correlation of N-O percentage (%N-O) to the decrement of N-5 percentage (%N-5), denoted by Δ%N-5 in table S1. (iv) Non–N-Q fractions (i.e., %N-5 + %N-6 in table S1) decrease in the order of samples S3, S1, S2, and S4; their redox potentials also increase in the same order.

These redox potentials in aqueous electrolytes were determined in three-electrode electrochemical cells in 0.5 M H2SO4 (pH 0) electrolyte using an Ag/AgCl reference electrode and a Pt counterelectrode. The working electrode was prepared by pressing together active-material powders (at a mass loading of 0.5 mg cm−2) and an inactive, highly compressible graphene foam (3D-graphene, specific capacitance = 30 F g−1) without any other additive. In cyclic voltammetry (CV) at 2 mV s−1 (Fig. 2A), cells with both OMC and OMFLC working electrodes have nearly rectangular CV curves representative of an ideal efficient EDLC. With OMFLC-N electrodes, the curves may be deconvoluted into (i) a nearly rectangular EDLC-like curve, albeit with a substantially higher charging/discharging current not seen with OMC and OMFLC, and (ii) a set of symmetric Faradaic charging/discharging peaks. In (ii), the charging peaks are located at ~0.25 V to ~0.5 V, increasing in the order of S3, S1, S2, and S4 (S4 data omitted in Fig. 2A but listed in table S1), which is exactly the same order that non–N-Q fractions decrease, thus strongly suggesting that the redox potential is related to N-5 and/or N-6. The above shape and symmetry features were maintained when the scan rate increased to 100 mV s−1, as shown for S1 in fig. S4A. This indicates that both EDLC-like and redox reactions have fast charging/discharging kinetics.

Fig. 2 Electrochemical evaluation.

(A) Cyclic voltammetry (CV test, at 2 mV s−1) from the first cycle for OMC, OMFLC, and OMFLC-N (S1 to S3) and for mixed OMFLC-N (SM). (B) Galvanostatic charge/discharge (CC test at 1.0 A g−1) from the first cycle for OMC, OMFLC, and OMFLC-N (S1 and SM). (C) Complex-plane plots of AC impedance. Inset shows phase angle versus frequency. (D) Capacity versus square root of half-cycle time. Solid symbols, CV test data from 2 to 500 mV s−1; open symbols, CC test data from 1 to 40 A g−1. Extrapolated intercept capacitance is rate-independent capacitance k1, the remainder diffusion-controlled capacitance. (E) Tafel plots of electrode potential against pH at steady-state current density of 10 μA cm−2. (F) Tafel plots of electrode potential against current i at pH 6.8 for OMFLC-N (S1 and SM). All potentials are relative to Ag/AgCl reference electrode; all electrolytes except (E) and (F) are 0.5 M H2SO4 aqueous solution. In (E) and (F), theoretical slope (–59.2 mV/decade) is shown as a straight line to suggest reasonable agreement with the data (see text).

To proceed further, we note that pseudocapacitive materials with a pronounced redox peak are usually inefficient electrodes in a symmetric electrochemical cell, which renders the effort of incorporating faradaic capacitance ineffective. This is because a symmetric cell is electrically equivalent to two serial capacitors, C1 and C2, so its total capacitance C1C2/(C1 + C2) is optimized when C1 = C2. This condition is usually impossible to satisfy at all potentials unless the CV curve is rectangular. We found that the following simple method can overcome this problem, however. By mixing three OMFLC-N powders at the ratio of S1:S2:S3 = 0.3:0.3:0.4 to form another OMFLC-N powder (SM), we obtain a new material that is capable of supporting multiple faradaic peaks. It exhibits a rectangular EDLC-like CV curve at a very large current (Fig. 2A)—a feature that should prove useful for constructing high-performance symmetric electrochemical cells. Because this is a qualitatively different CV curve from those of other OMFLC-N electrodes as well as OMC and OMFLC, we made further performance comparisons between SM, S1, OMC, OMFLC, and YP-50.

Consistent with the CV results, all galvanostatic charge/discharge tests (the CC test in Fig. 2B) show symmetric features with a fairly linear slope. A specific capacitance as high as 855 F g−1 at a current density of 1 A g−1 was obtained for SM, versus 715 F g−1 for S1 (fig. S4, C and D) and 175 F g−1 for YP-50 (fig. S5). Over a wide range of current densities, SM continued to provide a well-behaving CC curve and high capacitance (fig. S4, E and F), achieving 615 F g−1 at 40 A g−1, which is much higher than known EDLC and quite comparable to the capacitance of transition metal–oxide faradaic pseudocapacitors (1618).

Electrochemical impedance spectroscopy (Fig. 2C, enlarged in fig. S6A) found OMFLC-N (S1) to have the lowest equivalent series resistance of ~0.8 ohms, better than that of OMFLC and OMC. This may be attributed to better wetting on OMFLC-N, which lowers the interface resistance, because OMFLC has at least comparable, if not lower, resistivity than OMFLC-N (fig. S2C). The >45° (negative) phase angle of both OMFLC-N (S1 and SM; inset of Fig. 2C) at relatively high frequencies confirms their capacitive behavior at fast rates. Specifically, the frequency (of −45°) when the resistance and reactance have equal magnitudes is 0.48 Hz for OMFLC-N, giving a relaxation time (τ0 = 1/f0) of 2.1 s.

The CV and CC tests are in broad agreement with each other when compared at the same half-cycle time T as seen in in Fig. 2D, which also provides insight into the charging/discharging kinetics. (In the CV test, T is the time to sweep over the voltage window. In the CC test, it is the time to discharge.) In general, the capacitance C may contain a rate-independent component k1 (classically attributed to EDLC) and a diffusion-limited component controlled by the scanning rate, ν = T−1, taking the form (19, 20)Embedded Image (1)In Fig. 2D, the k2v−1/2 term represents the long-T data, which extrapolate to k1 at the T1/2 = v−1/2 = 0 intercept. (In the CV test, Fig. 2D reduces to the standard C-v−1/2 plot in fig. S6B, from which one can also obtain k1.) Apparently, k1 dominates in OMFLC-N, exceeding 700 F g−1 in SM and 545 F g−1 in S1. Dominance of rate-independent capacitances is common for EDLC, but it nevertheless holds here in redox reactions of the above materials because (i) OMFLC is a low-dimensional, fast-conducting, high-surface-area, few-layered material, and (ii) OMFLC-N is mesoporous (fig. S1) and hydrophilic (Fig. 1H). Therefore, they allow facile reactions both outside and inside the few-layer carbon tubes, as well as across the tube thickness.

The data from the slowest, near-equilibrium tests allowed us to construct the Tafel plots in Fig. 2, E and F, to compare the energetics of faradaic reactions and reveal a fundamental difference between OMFLC-N and OMC or OMFLC. For both S1 and SM, the potential required to sustain a constant current density of 10 μA cm−2 from pH 4.0 to 7.0 (Fig. 2E) lies close to the theoretical Tafel line with a slope of 2.3 × RT/F (−59.2 mV/pH) (21). Likewise, measuring the potential required for different current densities (Fig. 2F) at pH 6.8 gives again a slope in close coincidence with 2.3 × RT/F. Because both sets of Tafel lines imply a one-electron reaction, the pH dependence must arise from the concurrent incorporation of one proton and one electron. In contrast, for OMC and OMFLC, the slope is very flat, suggesting little redox activity. Recalling that the redox potential decreases with increasing concentration of nongraphitic N (Fig. 2A and table S1), we believe the redox reaction previously proposed for pseudocapacitance in N-containing polypyrrole (22) and carbon nanotubes (8, 9)—that each pyrrolic (N-5) and pyridine (N-6) nitrogen can incorporate an electron and a proton—is also operational here: It fits all of the above descriptions. According to N 1s XPS that can “see” through tubes less than 2 nm thick, there is 8.2 at% N in OMFLC-N (S1), which may store an additional faradaic charge of 660 F g−1. This is more than enough to account for the storage difference (390 F g−1) between OMFLC-N (S1) and OMFLC.

The proposed N-H mechanism dictates that an acidic condition is more favorable for redox reactions. This was verified for S1 in the three-electrode CC test at 1 A g−1: It has a larger capacitance in 0.5 M H2SO4 (715 F g−1, Fig. 2B) than in 1 M KOH (pH 14) electrolyte (405 F g−1; fig. S7, A to D, also confirmed by the CV tests). In contrast, OMFLC, which solely relies on EDLC, has very similar capacitances in the two electrolytes (fig. S7E). These results lend further support to the proposed N-H redox mechanism that makes OMFLC-N a superior supercapacitor.

Hoping to reduce these new mechanisms into practice, we investigated whether the three-electrode performance of OMFLC-N can be translated to electrochemical cells. Carbon-based electrodes are special in that they can be used as both cathodes and anodes in symmetric electrochemical cells, with a per-electrode specific capacitance nearly the same as that measured in the three-electrode test. This was confirmed for YP-50, OMC, and OMFLC (table S2). Here, we multiply the nominal specific capacitance of a symmetric EC by 4 to obtain the per-electrode specific capacitance (23). In contrast, as mentioned before, pseudocapacitive materials with a pronounced redox peak are usually inefficient electrodes in symmetric electrochemical cells (23, 24) because their two differential capacitances at the two electrodes, C1 and C2, are different. (Under the normal circumstance when one electrode has a suitable potential for the major redox peak and thus a larger differential capacitance, the other electrode is at a potential away from the major redox peak, hence having a smaller differential capacitance.) So their total capacitance C1C2/(C1 + C2) is lower than the maximum, which is ½C1 = ½C2 when C1 = C2. In contrast, despite predominant contributions of redox reactions, our SM electrode maintains a nearly rectangular CV curve (Fig. 2A)—that is, a constant differential capacitance—in the three-electrode test. So we expect its symmetric electrochemical cell to satisfy C1 = C2, thus to provide a per-electrode specific capacitance identical to that measured in the three-electrode test. Indeed, its symmetric-cell CV curve (Fig. 3A) in 0.5 M H2SO4 electrolyte is rectangular and rather symmetric, and its symmetric-cell CC test (Fig. 3B) gives a per-electrode capacitance of 840 F g−1 at 1 A g−1—within 2% of the three-electrode capacitance of 855 F g−1 (see table S2). In comparison, other OMFLC-N electrodes (S1 to S3) each having a distinct redox peak in the CV curve all suffered from capacitance losses of 10 to 15% when used in a symmetric electrochemical cell (table S2). All the symmetric-cell electrochemical measurements were conducted in 0.5 M H2SO4 electrolyte using an operating voltage of 1.2 V, which did not cause any detectable H2 or O2 evolution (fig. S8A).

Fig. 3 Electrochemical performance of symmetric cells.

OMFLC-N SM cathode and anode were used in two aqueous electrolytes, 0.5 M H2SO4 (pH 0) and 2 M Li2SO4 (pH 1.8). (A) Cyclic voltammetry from the first cycle at 2 mV s−1 scan rate. (B) Galvanostatic charge/discharge curves from the first cycle at 1 A g−1. (C) Symmetric electrochemical cell devices retain >92% after 100 hours of sustained loading (blue symbols, upper scale) at 1.2 V (in 0.5 M H2SO4) and 1.6 V (in 2 M Li2SO4), and retain >80% of their initial response after 50,000 cycles (black symbols, lower scale) from 0 to same peak voltages in two electrolytes. (D) Gravimetric (left) and volumetric (right) capacitance (at 1 A g−1) of symmetric electrochemical cell device (counting electrode weight and volume only) versus areal mass loading of OMFLC-N SM in two aqueous electrolytes. (E) Ragone plot of specific energy versus specific power for OMFLC-N SM symmetric devices (counting all-device weight) using 0.5 M H2SO4 (solid squares) and 2 M Li2SO4 (solid circles) electrolytes, as well as several standard devices: electrochemical capacitors (EC) (2, 28), lead-acid batteries (1, 26), nickel metal-hydride batteries (27), and lithium-ion batteries (28). Data counting electrode mass only are shown as open symbols. (F) Ragone plots of energy density versus power density for OMFLC-N SM packaged symmetric devices (counting all-device volume) using 0.5 M H2SO4 (solid squares) and 2 M Li2SO4 (solid circles) electrolytes, as well as several standard devices as in (E). Data counting electrode volume only are shown as open symbols. Dotted lines in (E) and (F) are current drain time, calculated by dividing specific energy by specific power.

The performance of OMFLC-N SM electrodes in symmetric aqueous electrochemical cells was further confirmed using another electrolyte, Li2SO4, which helps prevent carbon-electrode corroding and allows a higher operating voltage up to 1.9 V (25). Indeed, in 2 M Li2SO4 electrolyte at pH 1.8, a symmetric electrochemical cell with SM electrodes had a threshold water-splitting voltage of 1.8 V; at 1.6 V there was no detectable H2 or O2 evolution after 24 hours (fig. S8B). In acidic (pH 1.8) but not basic (pH 9.2) Li2SO4, pronounced redox was confirmed by the CV test (fig. S9). With this electrolyte, symmetric electrochemical cells obtained a specific capacitance of 740 F g−1 at 1 A g−1 from the CV and CC tests (Fig. 3, A and B), just 5% below the three-electrode capacitance of 780 F g−1 (table S2).

Having established the robust redox bipolar activities of OMFLC-N SM as both cathode and anode, we further evaluated its suitability for practical applications, starting with their stability in sustained and cyclic loading (Fig. 3C). After 100 hours of sustained loading, the capacitance retention was 93% at 1.2 V in 0.5 M H2SO4 electrolyte and 92% at 1.6 V in 2 M Li2SO4 electrolyte. The symmetric electrochemical cell withstood 50,000 cycles between 0 and 1.2 V in 0.5 M H2SO4 electrolyte with 82% of the capacitance remaining; a similarly cycled device between 0 and 1.6 V in 2 M Li2SO4 (pH 1.8) electrolyte retained 80%.

To pack more energy and power into the device, we increased the mass loading to the limit of not sacrificing full electrochemical efficiency. (To aid electrode formation at >2.0 mg cm−2 OMFLC-N SM loading, we added 5 wt% PVDF to the OMFLC-N powders.) Up to 6.0 mg cm−2 (∼0.69 g cm−3), the gravimetric specific capacitance of the symmetric electrochemical cell changed minimally (Fig. 3D), indicating that OMFLC-N powders had full access to the electrolyte without geometric or electric hindrance or diffusion limitation. Such increased loading benefits the volumetric capacitance, which is important for practical applications. Peaking at 6.0 mg cm−2, the volumetric capacitance increases by more than a factor of 8, so that OMFLC-N SM can reach 560 F cm−3 and 810 F g−1 in 0.5 M H2SO4, and 490 F cm−3 and 710 F g−1 in 2 M Li2SO4 (pH 1.8).

The merit of our material relative to existing battery and supercapacitor materials was evaluated using Ragone plots (specific power versus specific energy) for symmetric electrochemical cells on both the device gravimetric basis (Fig. 3E) and the device volumetric basis (Fig. 3F). In 0.5 M H2SO4 electrolyte, our device has a specific energy E of 24.5 Wh kg−1 based on the device weight (corresponding to 39.5 Wh kgOMFLC-N−1 based on the active-material weight) or 12.0 Wh liter−1 based on the device volume (or 27.0 Wh literOMFLC-N−1 based on the electrode volume). The specific power P is 26.5 kW kg−1 (42.5 kW kgOMFLC-N−1) or 13.0 kW liter−1 (29.0 kW literOMFLC-N−1), with a current-drain time (E/P) of 3.4 s. In 2 M Li2SO4 electrolyte, E increases to 41.0 Wh kg−1 (63.0 Wh kgOMFLC-N−1) and 19.5 Wh liter−1 (43.5 Wh literOMFLC-N−1) and P stays at 26.0 kW kg−1 (44.0 kW kgOMFLC-N−1) and 12.5 kW liter−1 (30.0 kW literOMFLC-N−1), with a drain time of 5.7 s. For supercapacitor applications, these properties are notable in that high specific power can be simultaneously achieved along with high specific energy, thus making carbon-based supercapacitors potentially competitive against batteries, such as lead-acid batteries (3, 26), nickel metahydride batteries (27), and perhaps even lithium batteries (28) on a gravimetric basis.

Simplified fabrication of N-doped mesoporous few-layer carbon (omitting the sacrificial silica template and post–carbon deposition etching as described in supplementary materials) was finally implemented by combining chemical vapor deposition with a sol-gel process of inexpensive, environmentally friendly, Si-free precursors/catalysts. The material obtained is made of highly conductive (σ = 360 S/cm) mesoscopically ordered few-layer carbon with a large surface area (1900 m2 g−1). In 0.5 M H2SO4 electrolyte, its electrode has a specific capacitance of 790 F g−1 at 1 A g−1, and its packaged device has a specific energy of 23.0 Wh kg−1 and a specific power of 18.5 kW kg−1 based on the device weight; in 2 M Li2SO4 (pH 1.8) electrolyte, the corresponding values are 720 F g−1, 38.5 Wh kg−1, and 22.5 kW kg−1. Indeed, in all important respects (figs. S10 to S13), this material behaves within ~10% of the best OMFLC-N SM described above, thus providing an outstanding low-cost carbon-based material for electrochemical cells for electric power applications.

Supplementary Materials

Materials and Methods

Figs. S1 to S14

Tables S1 and S2

References (2932)

References and Notes

  1. Acknowledgments: Supported by National Natural Science Foundation of China grants 51125006, 91122034, 61376056, and 51402336 and Science and Technology Commission of Shanghai grant 14YF1406500. I.W.C. was supported by U.S. Department of Energy BES grant DE-FG02-11ER46814 and used the facilities (Laboratory for Research on the Structure of Matter) supported by NSF grant DMR-11-20901.
View Abstract

Stay Connected to Science

Navigate This Article