Introduction

Carbon materials have been extensively explored for various advanced electrochemical technologies such as ion batteries1,2, supercapacitors3, hybrid capacitors4, and capacitive deionization (CDI)5,6 owing to abundance, electrochemistry stabilization, good electrical conductivity, adjustable platforms, and environmental benignity. However, the commercial carbon electrodes (e.g., graphite, activated carbon) deliver extremely limited ion-capture properties (<300 mAh g−1 or <20 mgNaCl g−1). The working mechanism is established on reversible electrosorption of charged species at the carbon-electrolyte surface and interface based on electrical double-layer (EDL) process7. Essentially, the low EDL capacitance (EDLC) of carbon materials is mainly attributed to the attenuation of electrostatic interaction along with the thickness of EDLs increasing, leading to the limited hosting sites and/or insertion space. Tunings of pore distribution, functional groups, and surface defects are kinetically favorable to enhance the electrochemical performance due to the “surface-capacitive-dominated” charge-storage mechanism8. However, the outcomes of pore modulation are limited by inaccessibility and poor compatibilities between porous network and electrolyte ions. In the case of oxygen functionalities, the surface Faradaic behavior is commonly documented as the reversible reaction of hydroquinone/quinone groups. It is very interesting to note that Kyotani et al. revealed a positive correlation between carbon corrosion and the number of edge sites terminated by H and oxygen-functional groups9, and demonstrated the seamless, edge-free carbon structure being highly demanded for high EDLC and excellent stability10.

Nitrogen doping is an efficient way to increase active sites by utilizing fast bonding–debonding processes benefitting from its high electroactivity, plentifulness, and easy implantation into carbon frameworks11,12,13. Nitrogen-functionalized porous carbon, however, exhibits a so-called pseudocapacitance that involves electrochemically active nitrogen groups as well as an EDLC. Hulicova-Jurcakova et al. fabricated nonporous nitrogen-enriched carbon which showed much higher capacitance than that of porous carbon, directly demonstrating how nitrogen doping is crucial for high-performance carbon materials14. Intriguingly, they also observed that pyridinic and/or pyrrole-like (e.g., imids, lactams) nitrogen groups are the origin of pseudocapacitance for nitrogen-doped carbon. Importantly, the analysis of corresponding surface area based on the pore distribution above 10 Å was first developed to differentiate the pseudocapacitive contribution to the overall capacitance, which provides a unique insight to understand the role of doped nitrogen in nitrogen-functionalized porous carbon15. Additionally, it has been confirmed that nitrogen doping favors improving the adsorption capacity and cycling stability of carbon electrodes, which was due to positive effect of the enhanced surface negative charge on the CDI performance16. It is generally considered that nitrogen doping can not only promote the pseudocapacitive adsorption, but also enlarge the interlayer spacing, thus improving the electrochemical ion-hosting capacity11,17. As well established, the total N content is deemed to have a pivotal impact on enhanced amplitude of ion-capture performance18. However, the total N concentration is usually low as the nitrogen species are thermodynamically unstable and tend to be volatile at high temperature during pyrolysis process. In general, when pyrolyzed temperature surpass 1073 Kelvin degrees, the content of nitrogen atoms is lower than 10 at%19. A spot of N could not provide sufficient pseudocapacitive binding sites. To date, many efforts including screening thermostable N-containing precursors, pre-crosslinking strategy, and reverse-defect-engineering strategy have been devoted to increasing the N-doped concentration to more than 20 at%20,21,22.

Inevitably, highly developed N atoms that are introduced into carbon lattices would destruct the short-range order, i.e., the C-sp2 conjugated structure. Thereby, the produced turbostratic structure would deteriorate the electron-transfer ability concomitantly, leading to active N bonding sites not being utilized efficiently23,24. It is full of challenging for balancing the tradeoff between C–N· active sites and electric conductivity. Currently, improving the conductivity of carbon in the short range is mainly based on adjusting process parameters (e.g., pyrolysis temperature) and adding mad accelerants (e.g., metal catalysts and hydrochloric acid). Increasing calcination temperature is somewhat beneficial to promote the formation of sp2-C conjugated structure, but at the expense of nitrogen sites17,25,26. Additionally, the graphitized structure can be improved by the addition of metal catalysts during thermolysis process, which requires the additional chemicals (e.g., strong acid and alkali) to remove residual metal27,28. To avoid the usage of metal catalysts, the protonation of HCl for g-C3N4 precursor is proved to be efficacious to lower the C-sp3/C-sp2 ratio and thus enhance the electron migration according to reverse-defect engineering, but the nitrogen content decreases to 14.7 at% from 22.3 at%22. Thus, rationally designing and constructing electron transfer-mediated nanoarchitectures to overcome this issue based on highly N-doped carbon indeed remains rare and challenging.

Herein, we develop a supramolecular self-assembly strategy to fabricate the high-level nitrogen-doped nanosheet-like carbon (NSLC) with enhanced electrical conductivity for electrochemical water desalination. The supramolecular crosslinking can produce abundant nanobubbles in boundary by introducing structural units of uric acid (UA) and melamine with differential thermal stability, thus tensionally inducing the formation of graphitic nanodomains among nitrogen-rich segments. The “order-in-disorder” nanosheet-like carbon obtained at 800 °C (O/D NSLC-800) discloses the application potentiality for brackish water (≈82 mgNaCl g−1 at a cell voltage of 1.6 V in a 1000 mg L−1 NaCl solution) and actual industrial circulating cooling water. The highway-mediated ordered/disordered nanoarchitecture facilitates to compensate the short-range electron transfer, resulting in a lowered resistance and activation energy of interfacial charge transfer, facilitating the full utilization of hosting sites of multiple nitrogen motifs. This work paves an avenue to fabricate highly heteroatom-doped carbon with high conductivity for advanced electrochemical applications based on precursor design at molecular level.

Results

Synthesis and characterizations of “order-in-disorder” carbons

In this work, a supramolecular self-assembly strategy was developed to construct “order-in-disorder” structure to promote the electron transfer. As illustrated in Fig. 1a, the NSLC by direct pyrolysis of UA at 800 °C is amorphous, whose disorder structure (i.e., abundant defect structure) blocks the efficacious charge transport. When adding melamine (MA) into UA aqueous dispersion, the MA–UA supermolecules are self-assembled under the Lewis pair interaction, which is confirmed by the variation of micromorphology, broadening of peaks for hydroxyl group recorded by Fourier transform infrared (FTIR) spectroscopy, and increasement of 1H chemical shift obtained through liquid 1H nuclear magnetic resonance (NMR) spectra (Supplementary Figs. 1–3). The hydrogen bonding coupled with π–π interaction promotes the evolution of supramolecular structure ultimately. Calcinating the MA–UA supramolecular under an inert atmosphere, the disordered defect structure is bridged by well-organized pseudographitic networks in obtained ordered/disordered nanosheet-like carbon (O/D NSLC), which dramatically favors the electron transfer in local nanodomains.

Fig. 1: Synthesis and morphological features of O/D NSLC-based materials.
figure 1

a Schematic illustration of the preparation process. TEM characterizations of b NSLC and O/D NSLC prepared at c 700 °C, d 800 °C, and e 900 °C (the insets: high-resolution images).

Full size image

The pseudographitic nanodomains coupling with disordered structure can be verified by transmission electron microscope (TEM). Obviously, the O/D NSLC-based samples show an ordered pseudographitic nanonetworks embedded in defect-rich segments, whereas NSLC is overall disordered amorphous structure (Fig. 1b–e). Accompanied by increasing the calcination temperature from 700 to 900 °C, the graphitization degree of pseudographitic nanodomains has been strengthened. The hierarchical structure of wrinkled, interconnected ultrathin nanosheets of O/D NSLC series are displayed (Supplementary Fig. 4). Atomic force microscopy (AFM) height profiles show a height of ca. 1.2 nm for the O/D NSLC-800 obtained at 800 °C for a given cross-section (Fig. 2a, b). Single-layer graphene oxide (GO) is typically found to be on the order of 0.6–1.2 nm observed by other AFM studies29,30,31. It can be rationally estimated that the thickness of O/D NSLC-800 was quantified to be mono- or bilayer GO. It has been reported that increasing carbonized temperature and precross-linking strategy of precursors are beneficial for horizontal growth of pseudographitic domains instead of vertical growth and reducing the number of stacked carbon layers32,33, which might be responsible for the no stacked structure.

Fig. 2: Thickness characterization of carbon nanosheets and evolution of thermal decomposition behavior of MA–UA precursor.
figure 2

a AFM image and b the corresponding height profiles of O/D NSLC-800. c TGA and DSC curves, and TGA–MS spectra of d UA (inset: enlarging of the shade section) and e UA–MA supermolecules. f Schematic diagram of pseudographitic boundary formation.

Full size image

To get deep insight into the evolution of pseudographitic nanodomains, the pyrolysis behaviors of pure UA, MA, and MA–UA supramolecular were tracked through thermogravimetric analysis–mass spectrometry (TGA–MS) equipped with differential scanning calorimetry (DSC). It can be found that pure UA experiences a prominent weight loss from 400 to 450 °C, corresponding to the melting and thermolysis of UA molecules (Fig. 2c). This process with a DSC peak centering at 430 °C was endothermic, releasing massive gases, which is further dissected by MS spectra (Fig. 2d). During the endothermic process, a mass of small nitrogen- and oxygen-containing gas molecules (e.g., CO2, H2O, NH3, NH2, CNH, CN, CH3NO, C4HNO) generate at about 430 °C, originating from the open-ring reactions of the pentagons and hexagons in the UA molecule. The MA undergoes a ca. 100% weight loss at the temperature range of 220–340 °C. Pure MA cannot be carbonized due to the significant sublimation and decomposition (Supplementary Fig. 5). Compared with pristine UA and MA, the MA–UA supramolecular firstly exhibits a slight weight loss at 220–340 °C, as demonstrated by the increased thermostability benefitting from intermolecular interaction. Subsequently, the MA–UA supermolecules experience the cross-linking process at temperatures ranging from 340 to 400 °C, which is verified by TGA–MS spectra (Fig. 2e). During this process, the H2O release peak occurs in the TGA–MS spectra of MA–UA supermolecules, while the corresponding peak is absent for pure UA or MA, which is speculated to be the imidization reaction (Supplementary Fig. 6). Intriguingly, the supramolecular self-assembly can eliminate the loss of CN species and sp2-conjugated C of C4HNO, benefitting from foregoing cross-linking interaction.

The difference of carbon structure obtained by UA and MA–UA supermolecules may lie in the impact of foregoing cross-linking process on the subsequent carbonization pathways. It has been reported that the tension on the surface of nanobubbles can induce an ordered alignment of carbon atoms, resulting in the formation of short-range graphitic domain34. In this regard, creating abundant nanobubbles is necessary to promote the production of pseudographitic domain. The mutual cross-linking process in this study reserves the elementary unit of melamine. When the UA molecules begin to decompose, massive gases produced by the decomposition of elementary unit of melamine would be generated in the boundary of UA molecules (Fig. 2f). After most of H atoms elimination at >600 °C, the carbon atoms on the nanobubble surface rearrange under the driving force of surface curvature, and the sp2-conjugated C atoms remain, resulting in the formation of highly graphitized carbon wall34. Thus, the short-range graphitic nanodomains are in situ built in disordered defect regions.

The material yields of O/D NSLC-700 acquired at 700 °C, O/D NSLC-800, and O/D NSLC-900 obtained at 900 °C were much lower than that of NSLC (Supplementary Table 1), indirectly demonstrating the generation of abundant decomposed gases by MA–UA supramolecular. The nitrogen contents of resulting NSLC-based materials decrease gradually accompanied by increasing calcination temperature (Supplementary Table 2), whose existing forms are C―N, C=N, and C≡N and N―H groups in the carbon framework (Supplementary Figs. 7–9).

Performance for electrochemical water desalination

Due to the promoted electronic conductivity and abundant nitrogen-rich sites at the same time, these O/D NSLC series could be favorably applied in diverse ionotronic applications such as CDI for electrochemical ion separation (see the CDI configuration in Supplementary Fig. 10). In standard 1000 mg L−1 NaCl with applying 1.2 V, there was no obvious loss of specific adsorption capacity (SAC) and specific current occurred during the first 4 cycles (Fig. 3a, Supplementary Fig. 11), demonstrating the decent reversible adsorption capacity for all samples. The symmetrical O/D NSLC-800 electrodes exhibited the highest SAC of 49.0 ± 1.4 mgNaCl g−1 within 30 min, while NSLC (26.5 ± 1.5 mgNaCl g−1) only reached half of that of O/D NSLC-800. The O/D NSLC-700 and O/D NSLC-900 achieved SAC of 42.4 ± 2.4, 33.9 ± 1.4 mgNaCl g−1, respectively, which was also far superior to the performance of NSLC. The breakthrough of hydrolysis voltage could significantly promote the desalination performance of CDI35. Thus, the SAC of prepared samples as a function of the applied voltage was conducted (Fig. 3b). When increasing voltage from 1.2 to 1.6 V, the SAC of all samples enhanced stepwise. The O/D NSLC-800 electrodes displayed the highest SAC at all applied voltages and owned an SAC as high as 81.9 ± 1.3 mgNaCl g−1 at 1.6 V. The improved SAC of NSLC, O/D NSLC-700, O/D NSLC-800, and O/D NSLC-900 at increased stride length of 0.2 V were ca. 10.2, ca. 7.7, ca. 16.5, and ca. 8.0 mgNaCl g−1, respectively, demonstrating highly sensitive response to voltage increase for adsorbing ions for O/D NSLC-800 electrodes. The charge efficiencies of O/D NSLC electrodes were much higher than that of NSLC electrode, demonstrating the superior energy utilization efficiency of O/D NSLC electrodes for ion capturing, which was also verified by the specific energy consumption (SEC) and energy-normalized adsorbed salt (ENAS) (Supplementary Fig. 12).

Fig. 3: Performance for electrochemical water desalination.
figure 3

a Specific adsorption capacity vs. time during the first 4 cycles (1.2 V, 1000 mg L−1 NaCl solution), b specific adsorption capacity at different voltages (1000 mg L−1 NaCl solution), and c the corresponding Kim-Yoon plots. d Wind rose of SAC comparison with reported state-of-the-art CDI electrodes at 1.2 V and high voltages. e Cycleability of O/D NSLC-800 electrodes (1.6 V, 1000 mg L−1 NaCl solution). f Schematic illustration of capacitive deionization system with multi-channel tandem. Concentration of different salt ions before and after treatment of g brackish water and h actual circulating cooling water at a cell voltage of 1.6 V. Error bars in (b) represent the standard deviation and were calculated on the basis of three experimental data points.

Full size image

Figure 3c further provided the CDI Ragone plots for NSLC and O/D NSLC-800 samples, and the O/D NSLC-800 electrodes showed the higher SAC and time-average specific adsorption rate (ASAR) than those of NSLC simultaneously, especially at 1.6 V. Note that the maximum SAC value for O/D NSLC-800—49.0 mgNaCl g−1 at 1.2 V—are superior to that of almost reported advanced carbon-based CDI electrodes under the same applied voltage5,6,36,37,38,39,40, even more than or comparable to those of a part of Faradaic electrodes, such as CoAl-layered metal oxide (CoAl-LMO)41, Cu-Prussian blue analogues (Cu-PBA)42, Fe-N-C43, MnO244, Ti3C2Tx MXene45, Na(Cl)-FeOOH46, antimonene (Sb-ene)47 (Fig. 3d). Unexpectedly, the SAC of O/D NSLC-800 at 1.6 V far outperforms most of the reported common CDI electrode materials at the high applied voltage (1.4–2.0 V)48,49,50,51,52,53,54, highlighting its outstanding ion-adsorption capacity for electrochemical desalination.

Long-term adsorption―desorption cycling performance for O/D NSLC-800 electrodes was conducted in a 1000 mg L−1 NaCl saline solution at a cell voltage of 1.6 V over 60 cycles (Fig. 3e). The CDI configuration paired the O/D NSLC-800 electrodes exhibited stable performance without apparent decay of SAC and without obvious increase of SEC, indicating the excellent cycling stability. In comparison, the other samples experienced a performance degradation to different extents (Supplementary Fig. 13), which are also confirmed by specific current-time curves (Supplementary Fig. 14). The trace of pH, potential distribution of the anode and the cathode, and characterizations before and after 60 adsorption/desorption cycles were conducted to investigate possible Faradaic reactions at the carbon electrode surface. The results showed that pH decreased and increased as the charging and discharging steps (Supplementary Fig. 15), which could be due to the lower diffusivity of bulkier Cl− compared to that of lighter Na+, leading to the dissociation of water during charging/discharging steps55. During an adsorption/desorption cycle, the magnitude of pH fluctuation is ca. 0.4 in the neutral condition, implying no obvious signature of chlorine and hydrogen evolution reactions in that the pH would fluctuate significantly and become alkaline if chlorine and hydrogen evolution reactions occur. Galvanostatic charge/discharge (GCD) measurements through changing potential windows were carried out to determine the safe working potentials. The results showed that suitable working potentials of the anode and the cathode should be operated within 0.9 V and −1.2 V vs. Ag/AgCl, respectively (Supplementary Fig. 16). Additionally, we monitored potential distributions of the cathode and the anode at different cell voltages during charging process (Supplementary Fig. 17). The potentials of the anode and corresponding cathode are both located at safe working potentials, demonstrating that the cell working voltages ranging from 1.2 V to 1.6 V are suitable to apply.

It should be noted that there was a pH decay of ca. 0.7 after 60 adsorption/desorption cycles, which is ascribable to the oxidation reactions occurred at the electrode surface in different degrees. The possible Faradaic reactions are as follows:

$${{\rm{Cathode}}}:{{{\rm{O}}}}_{2}+2{{{\rm{H}}}}^{+}+2{{{\rm{e}}}}^{-}\to 2{{{\rm{H}}}}_{2}{{{\rm{O}}}}_{2}$$
(1)
$${{{\rm{O}}}}_{2}+4{{{\rm{H}}}}^{+}+4{{{\rm{e}}}}^{-}\to 2{{{\rm{H}}}}_{2}{{\rm{O}}}$$
(2)
$${{\rm{A}}}{{\rm{node}}}:{{\rm{C}}}+{{{\rm{H}}}}_{2}{{\rm{O}}}-2{{{\rm{e}}}}^{-}\to {{\rm{C}}}={{\rm{O}}}+2{{{\rm{H}}}}^{+}$$
(3)

The oxygen reduction reaction (ORR) consumes the protons (Eqs. 1, 2), thus leading to the increase of pH in cathode region. The carbon oxidation in the anode (by reaction with water) would release the protons (Eq. 3), decreasing pH value. The pH decline after long-term cycles indicates that the carbon oxidation reaction in the anode is more significant than ORR in the cathode. The resulting H2O2 in cathode region also oxidizes the cathode to some extent. According to X-ray photoelectron spectroscopy (XPS) analysis, ratios of carbon to oxygen for the cathode and the anode increase from pristine 5.9 to 8.2, 15.7 (Supplementary Fig. 18a), respectively, suggesting that the anode suffers from more significantly oxidative corrosion than the cathode. As shown by XPS C 1s and FTIR spectra, the emergence of peaks for C―O and C=O species clearly reveals that the electrodes undergo surface oxidation during 60 adsorption/desorption cycles (Supplementary Fig. 18b, c).

Although different degrees of oxidation occur in the anode and the cathode, there is no obvious decline for desalination performance. Thus, the electrochemical properties (e.g., cyclic voltammograms (CV), GCD) before and after long-term operation were further investigated (Supplementary Fig. 19). Apparently, there is no decay of electrochemical capacities for the cathode after 60 adsorption/desorption cycles. The retentions of specific capacities for the anode are ca. 90.6% and ca. 84.2 % measured by CV and GCD, respectively. The high retention of electrochemical capacities for the cathode and the anode contributes to the excellent cycling performance due to the partial oxidative corrosion to a low extent during 60 adsorption/desorption cycles. Additionally, considering the potential effect of N-doping on surface charge, possibly leading to inverted CDI, Zeta potential measurements were conducted. With increasing of nitrogen content, the surface charge is more negative (Supplementary Fig. 20). However, the performance of inverted CDI (O/D NSLC-700 with negative surface charge//amino functionalized NSLC with positive surface charge) was negligible.

To exploit more application scenarios, we conducted a multi-channel tandem CDI system with five processing units (Fig. 3f) to treat brackish water and real refining circulating cooling water. After the first run for brackish water, the concentration of Na+ and Cl− reduced from 382.0 mg L−1 and 589.7 mg L−1 to 160.0 mg L−1 and 246.9 mg L−1 (Fig. 3g), respectively, satisfying the drinking water standard (GB 5749-2022, China). The concentration of Na+ and Cl− further decreased to 25.2 mg L−1 and 38.9 mg L−1 after two runs, respectively, which can meet water demands for ultra-low salt concentration. Intriguingly, this coupled CDI system also displayed the excellent performance for mixed ion separation in actual refining circulating cooling water, that is, achieving a decrease of the concentration of Ca2+, Mg2+, Cl−, and SO42− from 292.9, 156.0, 658.5, and 1403.3 mg L−1 to 4.2, 3.0, 360.3, and 209.2 mg L−1 after three runs (Fig. 3h), respectively. The obtained treated circulating cooling water with ultra-low Ca2+/Mg2+ and low Cl−/SO42− concentrations can be recycled repeatedly benefitting from avoidance of scaling and pipeline corrosion.

Microstructure-to-performance response analysis

Compared with pure UA-derived carbon, the O/D NSLC-800 electrodes exhibited much higher SAC and better performance for real saline water. To get deep insight on microstructure–performance relationship, Raman spectra, XPS, and advanced electrochemical tests were performed. The Raman spectra can be deconvoluted into four subpeaks by Gaussian numerical simulation, including D1, D2, D3, and G band (Fig. 4a, b). The ratio of the D1 band area to the G band area (i.e., AD1/AG) of NSLC and O/D NSLC-800 are 1.18 and 1.27, respectively, representing the more exposed edged defects on the carbon layers for O/D NSLC-800. The D3 proportion manifests the distortion of C6 graphitic structure. The D3 proportion of O/D NSLC-800 is much lower than that of NSLC, demonstrating the maintenance of more complete graphite structure in inner region of carbon nanodomains for O/D NSLC-800. In addition, increasing calcination temperature could reduce D3 proportion and AD1/AG value (Supplementary Fig. 21). The high AD1/AG value with low D3 proportion is beneficial to promote the ion transport and/or electron transfer during electrochemical process34. The N 1s XPS analysis reveals that superstructure assembly can significantly reduce the graphitic-N proportion, and improve the ratios of pyrrolic and amino N (Fig. 4c, d). It has been demonstrated that reduced graphitic-N proportion and increased ratio of pyrrolic N to pyridinic N contribute to promoting cycling stability and electrochemical activity17,22,56. With increasing heating temperature, the graphitic-N content decreases and the pyrrolic N percentage increases dramatically (Supplementary Fig. 22).

Fig. 4: Microstructural and advanced electrochemical analysis.
figure 4

Raman spectra of a NSLC and b O/D NSLC-800. High-resolution N 1s spectra of c NSLC and d O/D NSLC-800. e EPR spectra. f Electrical conductivity vs. pressure measured by a four-point probe (inset: scheme of a four-point probe setup). g Chronoamperometry property at an overpotential of 50 mV (inset: scheme of ion adsorbed on electrode surface). h Activation energy of the charge transfer process. i Nyquist plots (inset: fitted equivalent circuit diagram). j N2 physisorption isotherms at 77 K (inset: pore size distribution). k RRDE voltammogram for the disk electrode coated with O/D NSLC-800 in a 1000 mg L−1 NaCl solution at 5 mV s−1 and 1600 rpm rotation rate. The inset: corresponding mean electrons transferred. l Schematic illustration of ion-capture mechanism.

Full size image

The electron paramagnetic resonance (EPR) spectra were further conducted to interpret the microstructure of defect-rich regions surrounded by the nanographitic network. The g values of all as-prepared carbon centers at 2.0031, manifesting the emergence of unpaired electrons on π-conjugated carbon skeletons due to nitrogen doping (Fig. 4e, Supplementary Fig. 23). The O/D NSLC-800 delivers a much lower Lorentzian linewidth (L.W.) than that of NSLC, indicating higher defect-rich edge-nitrogen-doped configurations in O/D NSLC-800, boosting ion accessibility as active adsorption sites11. Additionally, the results of CV and GCD reveal that the O/D NSLC-800 exhibits the highest electrochemical capacity and fast ion-diffusion kinetics (Supplementary Figs. 24, 25) among all samples, endowing O/D NSLC-800 with an extraordinary CDI performance. Moreover, the outer surface capacitance (Co) of O/D NSLC-800 accounts for up to 73.0% of the total amount of specific capacitance (Ct) (Supplementary Fig. 26), which is beneficial to enhance charge-storage kinetics.

The nitrogen contents of O/D NSLC-800 and NSLC are almost identical (Supplementary Table 2), their differences of defect structure and pseudographitic networks are also indicated by the results of four-point probe measurements (Fig. 4f). The electrical conductivity of O/D NSLC-800 is higher than that of NSLC, reflecting the promoted effect of embedded pseudographitic nanodomains on electron transfer. The results of electrochemical active surface area (ECSA) measured in non-Faradaic voltage window also manifest the contribution of EDLC for ion capture (Supplementary Fig. 27). Apart from distinctive electron-transfer ability, the coupling structure of carbon framework with a tradeoff between edge-nitrogen and electron-transport configurations can motivate higher charge carrier density (ND) bound for ion hosting, determined by Mott–Schottky plots (Supplementary Fig. 28). To better understand the unique microstructure coupling nitrogen doping with highly pseudographitic boundary, we conducted chronoamperometry (CA) and electrochemical impedance spectroscopy (EIS) at different temperature to comprehend ion-adsorption behavior at the interface between electrode and electrolyte. The CA results show that the absolute value of current response for O/D NSLC-800 is much higher than that of NSLC when positively or negatively charged (Fig. 4g), indicating more favorable ion-adsorption ability for O/D NSLC-800 electrode benefitting from the “order-in-disorder” structure. The enhancement of reaction kinetics was further explored by the activation energy (Ea,ct) through Arrhenius equation during charge transfer process, which was measured across a temperature range from 293.15 to 323.15 Kelvin degree (Fig. 4h, Supplementary Fig. 29). Obviously, the Ea,ct of O/D NSLC-800 (36.7 kJ mol−1) is much lower than that of NSLC (62.4 kJ mol−1), demonstrating the enhanced kinetics and lower ion-desolvation energy for O/D NSLC-800 during the charge transfer process57,58,59. The significant decreases in activation energy and resistance (Fig. 4i) of the charge transfer process for O/D NSLC-800 can be deduced as the formation of structure of highly graphitized boundary bridging edge-type nitrogen species, promoting the ion accessibility and electron-transfer ability synchronously.

The porous structure that stores ions through capacitive behavior is crucial for the carbon electrode. All samples exhibit a typical type IV N2 adsorption–desorption isotherms (Fig. 4j, Supplementary Fig. 30), indicating a feature of mesopore-dominated hierarchical porous structure, which is also verified by pore size distribution. The Brunauer–Emmett–Teller (BET) surface areas of NSLC, O/D NSLC-700, O/D NSLC-800, and O/D NSLC-900 were calculated to be 67.2, 147.8, 245.1, and 153.2 m2 g−1, respectively, demonstrating the enhancement of porous network for O/D NSLC-800 at a calcination temperature of 800 °C. It is also confirmed by results of pore volume (0.42, 0.34, 0.52, and 0.49 cm3 g−1 for NSLC, O/D NSLC-700, O/D NSLC-800, and O/D NSLC-900, respectively). The pore size distribution manifests the coexistence of micro- and mesoporous structure for all samples, where dominates at 5.3, 6.9, 10.6, and 18.9 nm for mesopores. The enhanced specific surface area with multifarious mesopore distribution is conductive to promote the ion separation performance. Additionally, the ultrathin graphene-like structure of O/D NSLC-800 (a thickness of ca. 1.2 nm) facilitates the ion diffusion and exposure of hosting sites, which is also responsible for superior desalination performance. To roughly distinguish their respective contributions of surface area and nitrogen doping to the SAC, we prepared a kind of graphene (denoted as the G) that has a similar specific surface area and pore structure with that of O/D NSLC-800 (Supplementary Fig. 31). It can be found that the contribution of nitrogen doping to SAC is above 30% at different cell voltages (Supplementary Fig. 32).

ORR has a great impact on the performance of CDI electrodes due to unavoidable dissolved oxygen molecules in the solution. Hence, the behavior of ORR was explored by a rotating ring-disk electrode (RRDE) using linear sweep voltammetry (LSV). Figure 4k shows the calculated transferred electron numbers (ne−) based on the resulting currents of disk and ring electrodes. Since the ORR involves the two-electron (Eq. 1) and four-electron (Eq. 2) pathways, the mean ne− of the ORR is between 2 and 4. Our findings reveal that the ne− is near 3.14 at −0.42 V vs. Ag/AgCl, indicating the two-electron pathway becomes a comparable reaction to the four-electron pathway. When the potential of the cathode is more negative, the ne− increases to 3.71, manifesting the dominant process of the four-electron pathway. The calculated proportion of H2O2 produced (%H2O2) also confirms the results above (Supplementary Fig. 33). The H2O2 production reaches maximum value of ca. 42.7% at −0.42 V vs. Ag/AgCl, and gradually decreases to 14.5% with further negatively polarizing. The decreased %H2O2 at more negative potential might favor retarding oxidation of the cathode, but at expense of consuming more electric charges. Notably, the presence of cation-exchange membrane can limit the transport of dissolved oxygen from approaching the electrode surface, and suppress H2O2 production and hydrogen evolution reaction benefitting from the strong basicity near the cathode vicinity55.

To highlight the role of nitrogen doping, we fabricated N-deficient NSLC-800 by using H2 reduction to remove the lattice nitrogen atoms of O/D NSLC-800 at 800 °C. The nitrogen concentration of resulting N-deficient NSLC-800 decreased from 21.9 at% to 5.9 at% (Supplementary Table 3), but its specific surface area increased to 456.3 m2 g−1 (Supplementary Fig. 34a). Ultimately, the SAC of O/D NSLC-800 is about twice that of N-deficient NSLC-800 at different cell voltages (Supplementary Fig. 34b), highlighting the predominant role of high-level nitrogen species. The physical properties and corresponding desalination capacities of different-level nitrogen-doped carbons in this study are summarized in Supplementary Table 4. It is interesting to note that increasing nitrogen content can significantly improve the SAC with a certain range (Supplementary Fig. 35). Accordingly, it could be inferred that pyrrolic N, pyridine N, and amino N species are highly active pseudocapacitive sites for nitrogen-doped carbon, which play their roles through the N5Θ active sites60, C―N· radical sites61, and interaction of lone pair electrons for ion storage4, respectively (Fig. 4l). Additionally, the cyano groups (C≡N) occurring in O/D NSLC-800 might also play a role in improving working voltage and stability62,63. It is worth noting that the roles of all these diverse nitrogen configurations involve electron transfer-mediated surface Faradaic reactions that are limited by intrinsic electrical conductivity of carbon materials. The pseudographitic networks embedded in carbon framework provide the electron-transfer highways, which can boost the synergy of electric field-induced active nitrogen species to an extreme, promoting fast and high-capacity ion capture.

Ion-capture mechanism

Quantum chemical calculations based on density functional theory (DFT) were performed to further underscore the significance of modulating atomic/lattice arrangements. Figure 5a–c shows the electrostatic potential (ESP) diagrams of different lattice defect structures, that are, highly nitrogen-doped carbon, highly nitrogen-doped carbon with pseudographitic networks, and highly graphitized carbon, respectively. The negative and positive regions represent the active sites of electrosorption for cationic and anionic ions, respectively. In marked contrast with highly nitrogen-doped and highly graphitized carbons, the ESP distribution of highly nitrogen-doped carbon with graphitized carbon walls is much more inhomogeneous, indicating the greater propensity for capturing more ions. Simulations of molecular orbital interactions also highlight the structure of embedded pseudographitic networks (Fig. 5d–f, Supplementary Fig. 36). Accordingly, the adsorption energies (Eads) of a Na ion enriched onto highly nitrogen-doped carbon, highly nitrogen-doped carbon with pseudographitic networks, and highly graphitized carbon were calculated to be −0.45, −0.66, and −0.15 eV, respectively. A more negative Eads suggests the stronger binding affinity and thus higher adsorption capacity. Additionally, the highly nitrogen-doped carbon with graphitized carbon walls (Q = 0.89 e−) exhibits a greater potential for electron transfer than that of highly nitrogen-doped carbon (Q = 0.12 e−) and highly graphitized carbon (Q = 0.54 e−).

Fig. 5: Quantum chemical calculations and ion-capture mechanism.
figure 5

Distributions of ESP for a highly nitrogen-doped carbon, b highly nitrogen-doped carbon with pseudographitic networks, c highly graphitized carbon. Simulations of molecular orbital interactions of d highly nitrogen-doped carbon, e highly nitrogen-doped carbon with pseudographitic networks, f highly graphitized carbon with a Na ion, respectively, and corresponding calculated Eads and electron transfer number (Q). Top (top) and side (bottom) views of charge density differences of g pyridinic nitrogen, h pyrrolic nitrogen, i amino group, and j diverse nitrogen configurations (pyridinic, pyrrolic, amino, and cyano N) doped in the carbons for the interaction with a Na atom. k In situ ATR–FTIR spectra of O/D NSLC-800 electrode.

Full size image

In order to get insightful perspective of the function of different nitrogen configurations, the differential charge density maps for pyridinic, pyrrolic, amino, and diverse nitrogen functionalities were further performed (Fig. 5g–j). Obviously, the electron-rich regions (yellow) prefer to accumulate around the N atoms, facilitating to capture the Na+ ion. The Eads of Na atom on pure pyridinic, pyrrolic, and amino N were calculated to be only −0.73, −1.07, and −0.35 eV, respectively, reflecting a relatively weak adsorption. The Eads could further increase to as high as −2.74 eV after introducing pyridinic, pyrrolic, amino, and cyano N into carbon framework, implying the stronger interaction of diverse nitrogen speciation with Na. The abundant N configurations synergistically modulate the spatial charge redistribution and increase the active hosting sites, boosting the electrosorption capability significantly. Additionally, the results of simulations for a Na atom diffusion demonstrate that various N functionalities can effectually lower the diffusion barrier (Edif) (Supplementary Fig. 37), improving the ion separation kinetics. In situ attenuated total reflection (ATR)-FTIR microscopy was employed to further evaluate the role of different nitrogen configurations for the adsorption of sodium ions (Fig. 5k). It is worth noting that with increasing the cell voltage from 0 V to 1.6 V, the peak intensities of C―N, C=N, and ―NH2 groups decrease gradually, which could be inferred that these nitrogen species are electrified highly active pseudocapacitive sites for ion hosting64,65.

Discussion

We have demonstrated the ultrathin carbon nanostructure with N-rich defect surrounded by pseudographitic domains for fast, high-capacity ion capture. Such carbon electrodes (symmetric O/D NSLC-800) exhibited a high SAC of ca. 82 mgNaCl g−1 (at a cell voltage of 1.6 V in a 1000 mg L−1 NaCl solution) and a decent reuse potentiality for brackish water and real refining circulating cooling water. The desalination performance of O/D NSLC-800 electrodes outperforms most of the reported state-of-the-art CDI electrode materials. The extraordinary CDI performance is mainly due to the construction of pseudographitic domains bridging edge-type nitrogen species, which can accelerate the local electron transfer, thus boosting the electron transfer-mediated surface-capacitive behaviors of diverse nitrogen functionalities. Meanwhile, the enhanced specific surface area with multifarious mesopore distribution is also responsible for enhanced performance of ion separation. The first-principles calculations and in situ ATR–FTIR spectra manifest that highly doped nitrogen motifs with pseudographitic networks are highly active surface-capacitive sites for nitrogen-doped carbon materials. The electrically conductive nanoarchitectures demonstrated here provide an attractive strategy to fabricate highly heteroatom-doped materials for other competing electrochemical applications such as energy storage and electrocatalysis.

Methods

Synthesis of O/D NSLC

Typically, 6 mmol UA (99%, Macklin Co., Ltd., China) was dispersed in 100 mL deionized water. After stirring for 1 h, 2 mmol melamine (≥99.0%, Macklin Co., Ltd., China) was added into the above solution. The suspension was vigorously stirred for another 4 h before being transferred into an oven at 60 °C to age for 12 h. After self-assembly finished, the white powder (UA–MA supermolecules) was attained through filtration and then dried at 60 °C overnight. The UA–MA supermolecules were calcinated at a certain temperature (700 °C, 800 °C, 900 °C) with a ramping rate of 5 °C min−1 for 2 h under the Ar atmosphere. According to the calcination temperature, the final obtained thin carbon nanosheets (O/D NSLC) were denoted as O/D NSLC-700, O/D NSLC-800, and O/D NSLC-900, respectively. For comparison, pure UA (6 mmol) without adding melamine was conducted to heat at 800 °C for 2 h with a ramping rate of 5 °C min−1 in the Ar atmosphere, which labeled as NSLC. The graphene was obtained by thermal reduction of GO (250 mg, XF-205, XFNANO Co., Ltd., Jiangsu Province, China) at 800 °C for 2 h with a ramping rate of 5 °C min−1. The N-doped graphene (N-G) was prepared through calcination of mixture of GO (250 mg) and melamine (500 mg) at 800 °C for 2 h with a ramping rate of 5 °C min−1. The N-deficient NSLC-800 was fabricated by using H2 (a flow of 80 mL min−1) reduction to remove the lattice nitrogen atoms of O/D NSLC-800 (240 mg) at 800 °C with a ramping rate of 5 °C min−1 for 2 h.

Preparation of CDI electrodes

To prepare a CDI electrode, 70 wt% active materials, 20 wt% acetylene black (≥99.9% metals basis, Sinopharm Chemical Reagent Co., Ltd., China), and 10 wt% polyvinylidene fluoride (Mw~400,000, Macklin Co., Ltd., China) were blended homogenously by magnetic stirring. Then, an appropriate amount of N-methyl-2-pyrrolidone (NMP, 99.5%, Aladdin Chemistry Co., Ltd., Shanghai, China), as a kind of solvent, was added into the mixture above. The obtained solid–liquid mixture was stirred magnetically for at least 12 h to produce a slurry. The as-prepared slurry was uniformly coated on a titanium sheet by filming coating machine and then dried at 60 °C overnight. The obtained electrode, thickness of ~50 μm, had a coating surface area of ~4.5 × 5 cm2 with a mass loading of ~15 mg.

Capacitive deionization tests

The CDI performance was conducted in a continuous circulation system containing a power supply, data recording systems, a peristaltic pump, a conductivity meter, a CDI device, and the feed container. The CDI unit cell was washed by deionization water overnight before the CDI process. A certain concentration (1000 mg L−1) of the feed solution was pumped into the CDI unit cell. The volume and flow rate of the solution were kept at 20 mL and 20 mL min−1, respectively. Before desalination, the electrodes were rinsed by the feed solution without an external bias to achieve physisorption equilibrium and stabilize the solution conductivity. Constant voltage mode at different voltages (1.2, 1.4, 1.6 V) was applied for the charging and discharging processes. The desalination experiments were conducted in an environmental chamber, and the cycling experiment was carried out at an environmental temperature (ca. 25 °C). The SAC (mgNaCl g−1) was calculated by the following formula:

$${{SAC}}=\frac{({C}_{0}-{C}_{{{\rm{e}}}})\times V}{m}$$
(4)

where C0, Ce, V, and m signify the initial and final NaCl concentrations (mg L−1), the volume of NaCl solution (L), and the total mass of active material of two electrodes in CDI cell.

The ASAR (mgNaCl g−1 min−1) was evaluated by the following equation:

$${ASAR}=\frac{{\rm{SAC}}}{t}$$
(5)

where t is the charging time (min).

The ENAS (mgNaCl J−1) was calculated according to the following equation:

$${{ENAS}}=\frac{({C}_{0}-{C}_{e})\times V}{{E}_{{\rm{in}}}}=\frac{({C}_{0}-{C}_{e})\times V}{U{\int }_{0}^{t}Idt}$$
(6)

where Ein, U, t, and I are energy input during charging (J), applied voltage (V), charging time (s), and current (A), respectively.

The SEC (kWh kgNaCl−1) including charging and discharging processes was obtained based on the following formula:

$${{SEC}}({{{{\rm{kWh \, kg}}}}_{{NaCl}}}^{-1})=\frac{U{\int }_{0}^{2t}Idt}{3600\times ({C}_{0}-{C}_{e})\times V}$$
(7)

The charge efficiency (CE) was determined by the following equation:

$${{CE}}=\frac{({C}_{0}-{C}_{{{\rm{e}}}})\times V\times F}{58.5\times 1000\times {\int }_{0}^{t}Idt}$$
(8)

where F denotes Faraday’s constant, 96,485 C mol−1.

Physicochemical characterization

X-ray diffraction experiments were conducted on specimens using an X-ray diffractometer (Bruker D8 Advance, Bruker AXS) operating at 40 kV and 40 mA. The FTIR spectra of the prepared sample were acquired by using a Tensor 27 FTIR spectrometer (Bruker Optics, Inc.) over a scan range of 400–4000 cm−1. A BELSORP instrument (BEL, Japan, Inc.) was operated to confirm the specific surface area and pore size distribution of the products according to the adsorption/desorption isotherms of N2 obtained at 77 K by BET and Barrett–Joyner–Halenda methods. The morphology of samples was obtained using field-emission SEM (JSM-6330F) and TEM (JEOL-2010F). The Raman spectroscopy was recorded through LabRAMHR, HORIBA in the range of 800 to 2000 cm−1. XPS analysis was carried out with a Kratos Axis Ultra DLD spectrometer with an Al Kα X-ray beam. The Zeta potential was measured by a Zetasizer Nano ZS 90. The EPR spectra were characterized by the Bruker EMX Plus at room temperature. Liquid 1H NMR spectra were recorded on a Bruker Avance III HD500 instrument at room temperature. AFM was carried out using a Digital Nanoscope III in the tapping mode, and an etched silicon tip was used as a probe for imaging the samples. TGA–MS measurement was carried out on TA Instruments coupled with a SDT 650-Discovery MS by using argon as the protective atmosphere (20 sccm) under a heating ramping rate of 5 °C min−1. The electrical conductivities of powder samples were measured by a four-point probe (ST-2258C).

Electrochemical measurements

Electrochemical measurements were conducted in a three-electrode system using a CHI 760E electrochemical workstation (Shanghai CH Instruments Co., China) in a single chamber. A glassy carbon electrode coated with the as-prepared samples was used as the working electrode. A platinum sheet and Ag/AgCl were adopted as the counter and reference electrodes, respectively. A slurry was prepared by adding 70 wt% active materials, 20 wt% acetylene black, and 10 wt% polyvinylidene fluoride in NMP, and was stirred for 12 h. A certain amount of as-prepared slurry was coated on a glassy carbon by drop coating method. The diameter of coated materials on glassy carbon electrode was 3 mm, and the corresponding area was calculated to be ca. 0.071 cm2. The mass loading was approximately 3.5 mg cm−2. CV, GCD, EIS, Mott–Schottky curves, CA, and LSV were performed on an electrochemical workstation. The sweep speeds of CV were conducted at 1, 2, 4, 6, 8, 10, 20, and 50 mV s−1 at the voltage window of −1 ~ 0.8 V. The scan rates for measuring ECSA were 0.5, 1, 2, 3, 4, 5, and 6 mV s−1 at the voltage window of 0.15 ~ 0.25 V. The GCD tests were performed at the specific current of 100 mA g−1 and 500 mA g−1. The frequency of EIS analysis was used in the range of 0.1 Hz to 100 kHz upon a potential amplitude of 5 mV at an open-circuit potential. The frequency of Mott–Schottky test was fixed at 0.5 Hz at a potential amplitude of 5 mV. The CA measurements were recorded at an overpotential of 50 mV. The scan rate of LSV was set at 5 mV s−1.

The gravimetric specific capacitance based on CV curves (Cg, F g−1) was calculated through the following equation:

$${C}_{{{\rm{g}}}}=\frac{1}{2mv\Delta U}{\int }_{{U}_{1}}^{{U}_{2}}idV$$
(9)

where m, v, U, and i are the mass of electrode (g), the potential scan rate (V s−1), the potential window (V), and the current (A), respectively.

The capacitive contribution based on CV curves at different sweep speeds was calculated according to the following equation:

$$i(U)={k}_{1}v+{k}_{2}{v}^{1/2}$$
(10)

where i(U) and v represent the total current at a given potential (A), the scan rate (V s−1), respectively. k1 and k2 are the constants. Of which, k1v and k2v1/2 denote the capacitive process and diffusion-controlled process, respectively.

The relationship between ND and Mott–Schottky plot was depicted in the following equation:

$$\frac{1}{{C}^{2}}=\frac{2}{\varepsilon {\varepsilon }_{0}e{N}_{{\rm{D}}}}(U-{U}_{{FB}}-\frac{{k}_{{B}}T}{e})$$
(11)

where ε, ε0, e, ND, U, UFB, kB, and T denote the dielectric constant, vacuum permittivity constant, charge of electron, charge carrier density, applied potential, flat-band potential, Boltzmann’s constant, and temperature, respectively. Accordingly, the charge carrier density is inversely proportional to the slope of Mott–Schottky curve.

To obtain the activation energy (Ea) for the charge transfer process, the EIS profiles were measured at different temperatures (20, 30, 40, and 50 °C). Then the charge transfer resistance (Rct) was obtained by using ZView software to fit the EIS data. After that, 1/Rct vs. 1000/T curve was plotted, and the Ea was calculated according to the following Arrhenius equation:

$${{R}_{{{\rm{ct}}}}}^{-1}=A{e}^{-{E}_{a}/RT}$$
(12)

where A is the pre-exponential constant, R is the gas constant, and T is the Kelvin temperature.

RRDE voltammetry was performed at a rotation rate of 1600 rpm on a RRDE-3A (BAS Inc., Japan). The number of electrons transferred was calculated by using the following relationship:

$${n}_{{e}^{-}}=\frac{4{i}_{{disk}}}{{i}_{{disk}}+\frac{{i}_{{ring}}}{N}}$$
(13)

where idisk and iring are the currents at the disk and ring electrodes, respectively, and N is the collection efficiency of the Pt ring (0.37).

The %H2O2 produced during ORR was quantified through the following formula:

$$\%{{\rm{H}}}_{2}{{\rm{O}}}_{2}=\frac{100\times \left(\frac{2{i}_{{ring}}}{N}\right)}{{i}_{{disk}}+\frac{{i}_{{ring}}}{N}}$$
(14)

Quantum chemical calculations based on DFT

The DFT calculations of Fig. 4a–f and Supplementary Fig. 36 were carried out with the Gaussian 09 software. The B3LYP functional was adopted for all calculations in combination with the D3BJ dispersion correction. For geometry optimization and frequency calculations of molecule, the 6-311 G(d) was used. The electron energy of all molecules is obtained by using Def2-tzvp. To simulate the real situation as closely as possible, we took solvent circumstance and the charge of Na into consideration. The charge number of Na was set to be +1, reflecting the form of sodium ion. Additionally, the modeling of solvation model density was adopted to investigate the influence of water solvent, where the parameter of dielectric constant was set as a value of 78.3. The DFT calculations of Fig. 4g–j and Supplementary Fig. 37 were performed by using the Vienna Ab Initio Simulation Package (VASP, version number 6.1) via the projector-augmented wave (PAW) method. Generalized gradient approximation (GGA) was realized by the Perdew–Burke–Ernzerhof (PBE) function with a plane-wave cutoff energy of 500 eV. The structural models with different N configurations were built based on a 7 × 7 graphene supercell, and a 15 Å vacuum layer was used to separate the slabs. Monkhorst-Pack 2 × 2 × 1 k-point mesh was adopted for the integration of Brillouin zone and Gaussian smearing scheme with a smearing parameter of 0.05 eV was used. The convergence criterion for Hellman–Feynman force and total energy were set to 0.02 eV Å−1 and 10−5 eV.

In situ ATR–FTIR measurements

In situ, ATR–FTIR was performed on a Bruker INVENIO S instrument equipped with a mercury–cadmium–telluride (MCT) detector cooled by liquid nitrogen. The electrode was prepared according to a similar method above. A certain amount of as-prepared slurry was carefully dropped on the surface of the gold film, and used as the cathode and the anode. The cation-exchange membrane was used to separate the cathode and anode. Then, a 1 mol L−1 NaCl solution was employed as the electrolyte. The cell voltage was provided by a CHI 760E electrochemical workstation, and all the spectra were collected at the resolution of 4 cm−1. Prior to the test, the initial state with an open circuit was scanned for background correction.