*2.5. Characterization*

The transmission electrical microscopy (TEM) measurement was performed by FEI Tecnai G220 (Hillsboro, OR, USA) with an accelerating voltage of 200 kV. SEM characterization was conducted on a JEOL JSM-7500F field-emission scanning electrical microscope (Tokyo, Japan). The samples were coated with a layer of 10-nm-thick platinum film deposited by a JEOL JFC-1600 auto fine coater (Tokyo, Japan) before characterization. Wide-angle X-ray di ffraction (XRD: RIGAKU, Karlsruhe, Germany) was conducted at a scanning rate of 5◦ min−<sup>1</sup> with CuK α radiation (40 kV, 30 mA). Raman spectra were obtained by using a Confotec MR520 instrument (Graben, Germany)with an excitation laser wavelength of 532 nm, and Si wafers were applied as substrates. X-ray photoelectron spectroscopy (XPS) was conducted with a PHI Quantera SXM (ULVAC-PHI, Kanagawa, Japan) instrument with an AlK α X-ray source, and Ar ion etching was performed before measurement. All of the binding energies were calibrated by referencing to the C1's binding energy (285 eV). The specific surface area and the pore structure were measured by nitrogen sorption by using a JW-BK 112 physisorption analyzer (Beijing, China). The samples were degassed at 120 ◦C for 2 h before measurement. The specific surface area of samples was calculated by the Brunauer–Emmett–Teller and (BET) method. The pore size distributions were derived from the adsorption of isotherms using the Barrett–Joyner–Halenda (BJH) model.

#### *2.6. Electrical Measurements.*

The electrochemical measurements for all samples were characterized with a CHI 660E electrochemical workstation (Shanghai Chenhua, China) in a conventional three-electrode system, and a 6 M KOH aqueous solution was used as the electrolyte. The working electrode was prepared as follows: The active material (80 wt.%), acetylene black (10 wt.%), and polytetrafluoroethylene (PTFE) binder (10 wt.%) were mixed su fficiently with the help of ultrasonic machine. The mixture was pressed into a sheet with a piece of porous nickel net (diameter of 1 cm). The typical loading mass for each electrode is about 0.8–1.5 mg. Platinum and Ag/AgCl (3 M KCl) electrodes were used as the counter and the reference electrodes, respectively. The electrochemical properties of the working electrodes were measured by cyclic voltammetry (CV), galvanostatic charge-discharge (GCD), and electrochemical impedance spectroscopy (EIS). The voltage ranges for the CV and GCD tests were varied from −1 to 0 V. The current density for the galvanostatic measurement were varied from 1 to 20 A g<sup>−</sup>1. Electrochemical impedance spectroscopy (EIS) was measured in frequent ranges from 10−<sup>2</sup> Hz to 10<sup>5</sup> Hz at open circuit voltage with a current amplitude of 5 mV. Moreover, to characterize the cycling performance of the samples, galvanostatic measurement at a current density of 10 A g<sup>−</sup><sup>1</sup> was also carried out for 20,000 cycles. The capacitance of the electrode was calculated on the basis of the GCD curve according to the following equation:

$$\mathcal{C}\_{\mathfrak{m}} = \frac{I \times \Delta t}{m \times \Delta V}$$

where *Cm* (F g<sup>−</sup>1) is the specific capacitance, *I* (A) is the discharge current, Δ*t* (s) is the discharge time, m (g) is the mass of active material in the working electrode, and Δ *V* (V) is the potential window.

#### **3. Results and Discussion**

#### *3.1. Structure Design and Characterization*

The schematic illustration for the synthesis of *N*CNFs is shown in Figure 1. There are several appealing aspects for the use of linear PEI molecules for constructing the 3D structures: Firstly, as reported before, the L-PEI molecules can crystallize and self-assemble into diverse nano-structures under different crystallization conditions. Secondly, the amine groups on the main chains of PEI can catalyze the quick hydrolysis and condensation of silica source (e.g., TMOS), thus prompt deposition of SiO2 layers around PEI assembly; thirdly, as the protonated PEI (PEI-H+) is soluble in H2O, while the PEI was insoluble at room temperature, the crystallization of PEI can easily be fulfilled by neutralization of PEI-H<sup>+</sup> into PEI crystalline aggregates driven by the addition of alkali solution, which finally transformed into the cross-linked PEI@SiO2 nanotube networks once mixed with silica sources. Even more, the as-obtained PEI@SiO2 structures are also effective at prompting the polymerization of resorcinol and formaldehyde to form phenolic resins on the surface of PEI@SiO2 (Figure 1c), which can be easily turned into SiO2@C structures after carbonization. To further improve the porosity of 3D *N*CNFs, the inside SiO2 cores can be removed by HF to form cross-linked hollow nanotube structures (Figure 1d). The above solution-based processing also showed significant changes in powder colors with different compositions, which can be seen in the photograph of Figure 2.

**Figure 1.** Schematic illustration for the synthesis of N-doped 3-dimensional carbon nanotube networks. (**a**) Linear polyethyleneimine (PEI) template, (**b**) deposition of SiO2 layer on PEI surface to form the PEI@SiO2 sample, (**c**) coating of RF onto the PEI@SiO2 structure to form the PEI@SiO2@RF sample, and (**d**) the formation of N-doped carbon nanonet flakes (*N*CNFs) and the demonstration of electron and ion transport pathway in carbon nanonet flakes (CNFs).

**Figure 2.** Photographs showing the evolution from linear PEI precursor to the CNFs during the synthesizing process. (**a**) Linear PEI, (**b**) PEI@SiO2, (**c**) PEI@SiO2@RF and (**d**) CNFs.

SEM and TEM characterizations were performed to investigate the morphology and structure of the as obtained materials, which are shown in Figure 3. We can see that SiO2 coated PEI powders (PEI@SiO2) showed flakes-like structure with cross-linked nanofiber networks. After deposition and carbonization of the carbon precursors and further HF etching, the morphologies were mostly retained, remaining in uniform contacts and does not collapse in microscale. The diameter of the nanofiber for PEI@SiO2 was about 20 nm with interconnected mesoscale pores of around 20–30 nm, which were expected to be beneficial for fast electrolyte ion transportation. After further integration with N-doped carbon precursors (melamine), most of the porous structures were still reserved, although the diameter of nanofibers increased by several nanometers. Figure 3e,f showed the TEM and high resolution TEM images for N-doped sample ( *N*CNFs-1) after HF etching. The mutually cross-linked nanotubes in the assembled flask structures can be clearly seen. The diameter of the nanotubes for the *N*CNFs-1 is about 15–20 nm with wall thickness of around 3–5 nm, and the length of each nanotube is about dozens to a few hundred nanometers. EDS mapping further confirmed that nitrogen was homogenously distributed around the carbon nanotube networks, which were expected to improve the electron conductivity and surface wettability with electrolytes. The large amounts of pores existing both in the inside of each nanotube and among the large interspaces of adjacent one, may provide enough ion pathways and e fficient surface double layer capacity that are required for high performance SC electrodes. Moreover, the special flake-like assembles, which were composed of long nanotubes that interconnected with each other and almost completely spread throughout every flake along the axial direction, could further prompt the formation of stable electron conductive networks. These special and relatively ordered structures are expected to be quite beneficial for high performance SC devices [37], which will be discussed in the following sections.

**Figure 3.** SEM images for PEI@SiO2 (**a**), SiO2@C, (**b**) an N-doped CNFs sample of *N*CNFs-1, (**c**) and *N*CNFs-2 (**d**). (**<sup>e</sup>**,**f**) Shows the TEM images for sample of *N*CNFs-1 with di fferent magnifications. (**g**–**i**) Shows the TEM dark field image and the corresponding energy dispersive spectrometer (EDS) element mapping results for *N*CNFs-1.

XRD and Raman spectroscopy were further performed to investigate the structure and compositions of CNFs and N-doped samples, which are shown in Figure 4. According to the XRD results, two broad peaks centered at around 23.4◦ and 43.0◦ were observed, which can be assigned to the (002) and (100) di ffractions for carbon. Raman spectra further showed the existence of the two typical modes of D (relating with sp<sup>3</sup> hybridization and in plane defects for carbon) and G (relating with sp2-bonded ordered graphitic carbon), located at around 1363 and 1585 cm<sup>−</sup>1, respectively, further indicating the formation of carbon in all samples. N2 adsorption/desorption isotherm measurements were performed to investigate the pore structure for CNFs and nitrogen doped samples, and the results are shown in Figure 4c. All of the N2 adsorption/desorption isotherm curves exhibited the type-IV profiles, with two typical steep uptakes (P/P0 < 0.01 and P/P0 > 0.98) and hysteresis loops (CNFs and *N*CNFs-1: 0.45 < P/P0 < 0.97; *N*CNFs-2: 0.82 < P/P0 < 0.98), demonstrating the coexistence of micropores (<2 nm) and mesopores (2–50 nm) in the samples. The pore size distributions of the samples were calculated by using the Barrett–Joyner–Halenda (BJH) method, which is shown in Figure 4d. As can be seen from Figure 4d, all samples demonstrated two distinct peaks in the ranges of 2 to 10 nm (mesopore) and 10 to 200 nm (meso- and macro-pore), respectively, which can be assigned to the inner side pores for each individual carbon nanotube and the void spaces among the interconnected adjacent CNFs, according to the TEM characterizations (Figure 4e,f). The pore contents and their distributions for *N*CNFs-1 were similar with that of CNFs in the range of 2 to 10 nm, although were reduced to some extent especially at the macropore range (>50 nm), which can be explained by the incorporation of additional thin N doped carbon layer on the CNFs. Such peculiar multi-level size distribution is expected to be beneficial for achieving high specific surface area and fast electrolyte ion transferring [38]. Moreover, the calculated specific surface areas were 860, 946, and 365 m<sup>2</sup> g<sup>−</sup>1, respectively, for CNFs, *N*CNFs-1, and *N*CNFs-2, which was quite attractive among reported 3D hierarchical carbon. The decrease of specific surface area for *N*CNFs-2 can be speculated to owe from the excessive coating of melamine, which suppress both of the mesopores and the macropores, as indicated by the significant decrease of pore volumes for *N*CNFs-2 in Figure 4d.

**Figure 4.** XRD (**a**), Raman (**b**) and nitrogen adsorption/desorption isotherms; (**c**) measurements for CNFs and di fferent N-doped CNF samples. (**d**) Shows the calculated pore size distributions for corresponding samples.

To further investigate the composition of the hierarchical N-doped carbon nanonet flakes, XPS characterizations were performed which are shown in Figure 5. Wide range survey XPS spectra suggested the existence of nitrogen with a content of 1.88 wt.% in the powder after coating of melamine. The forms of nitrogen in carbon frameworks were studied by fitting the high-resolution XPS spectrum for N 1s. There were three significant peaks locating at 398.1 eV, 399.1 eV and 400.5 eV, respectively, for *N*CNFs-1 samples, which can be indexed to the pyridinic-N, pyrrolic-N and the graphitic-N (the configurations of doped N atoms in graphene layer is shown in the insert of Figure 5b). The formation of N doped structures can increase the electron density in carbon to achieve a high electron conductivity, and the defects formed by replacement of C with N can also introduce more active sites to improve the electrochemical performance, both of which are quite beneficial for SC devices.

**Figure 5.** (**a**) X-ray photoelectron spectroscopy (XPS) wide range survey spectrum and (**b**) high-resolution XPS spectrum of N 1s for a sample of *N*CNFs-1 (the inset shows the various configurations of N atoms doped in graphene layer).

## *3.2. Electrochemical Performance*

The electrochemical properties of the CNFs and N-doped CNFs were examined by a three-electrode configuration method, and a 6 M KOH solution was used as the electrolyte. The CV and galvanostatic charge-discharge curves were given in Figure 6. As shown in the CV curves (Figure 6a,c,d), under a fixed potential window of −1 to 0 V (versus Ag/AgCl), all samples indicated the typical response for electric double-layer electrodes, with quasi-rectangular shapes [39]. When a high voltage scan rate, e.g., 500 mV s<sup>−</sup>1, was applied, the rectangular-like curve was still able to be sustained, demonstrating a relatively fast electron and ion transportation during the charge and discharge process for these samples. Galvanostatic charge–discharge method was also applied to measure the charge and discharge capacitance for SC. The recorded specific capacitances were 461, 613 and 347.5 F g<sup>−</sup><sup>1</sup> (or 322, 351 and 205 F g<sup>−</sup>1) at a current density of 1 A g<sup>−</sup><sup>1</sup> (2 A g<sup>−</sup>1), respectively, for CNFs, *N*CNFs-1 and *N*CNFs-2. These values, especially for sample of *N*CNFs-1, are quite high among reports for carbon based EDLSs, which are usually lower than 300 F g<sup>−</sup>1. Recently, Wu et al. [40]. reported a similar capacitance of around 500 F g<sup>−</sup><sup>1</sup> based on hierarchical 3D carbon electrodes, which consisted of N-doped graphene quantum dots on carbonized MOF and carbon nanotubes hybrid structures. However, less controllability over the structural characteristics and the higher complexity of the synthesizing process can be expected from their reports. Moreover, as PEI refers to an interesting family of molecules, which can adopt a verity of 1D nanostructures to form different hierarchical nano- and micro-structures, and the benefits of feasible interaction chemistry for PEI monocular with SiO2 and the carbon precursors, the method reported here can provide an effective protocol for building a variety of interesting structures that are quite suitable for supercapacitors.

**Figure 6.** Cyclic voltammetry (CV) curves (**<sup>a</sup>**,**c**,**<sup>e</sup>**) at different scan rates and galvanostatic charge-discharge curves (**b**,**d**,**f**) at different current densities for CNFs, *N*CNFs-1 and *N*CNFs-2, respectively.

To further examine the electrochemical performance, the specific capacitance plots and the long-time cycling measurements for all samples were given, which were shown in Figure 7. We can see that, all samples reached relatively stable capacitance at a high current density range of 5 to 20 A g<sup>−</sup>1, and more competitively, the *N*CNF s-1 sample showed a high value of 242 F g<sup>−</sup><sup>1</sup> at 20 A g<sup>−</sup>1, which was quite high among reports for EDLSs [41–43]. The relatively large drops of the specific capacitance especially at the low current density range (from 1 A g<sup>−</sup><sup>1</sup> to 4 A g<sup>−</sup>1) could be understood by the suppressed contributions for small pores to the specific capacitance at increased current density as reported by Teng [44] and Daraghmeh [45]. Surface functional groups (see Figure S1 in the Supporting Information) and the related pseudocapacitance [46], as indicated by the slight tailing in the discharge curves in Figure 6, can also cause attenuations of the specific capacitance. However, because of the large amounts of exposed macropores in the hierarchical 3D structures and the excellent electron conductivity derived from the interconnected carbon nanotubes, a relatively high specific capacitance at high current density can still be obtained. Moreover, all samples obtained from PEI templates also showed good cycling stability. As shown in Figure 7b, after 20,000 cycles at a current density of 10 A g<sup>−</sup>1, specific capacitances of 213 F g<sup>−</sup><sup>1</sup> and 232 F g<sup>−</sup><sup>1</sup> can be obtained for CNFs and *N*CNFs-1, respectively, with high retention rate of both 95%. Additionally, to investigate the practical benefits of the as-synthesized *N*CNFs-1 sample, a symmetric two-electrode soft pack device was also constructed (see Figure S2 in the supporting information). The device demonstrated excellent electrochemical

performance with high specific capacitances of up to 313.6 F g<sup>−</sup><sup>1</sup> at 0.5 A g<sup>−</sup><sup>1</sup> and 262.4 F g<sup>−</sup><sup>1</sup> at 2 A g<sup>−</sup>1, which were quite attractive among the reported two-electrode supercapacitor devices. Schematic illustration of the local electron and ion pathway for the 3D hierarchical *N*CNFs is shown in inset of Figure 7b. The excellent electrochemical performance can be attributed to the following reasons: Firstly, the vast specific surface area as measured by BET can allow large amounts of charge accumulations on the surface or interface between electrodes and electrolyte; secondly, there were a grea<sup>t</sup> many stable interconnected and penetrable mesoscale pores (as shown in the SEM and TEM images), which can effectively facilitate the electron and ion transportation for fast electrode dynamics; moreover, the incorporation of nitrogen dopant in carbon nanotubes can significantly enhance the electric conductivity and electrolyte solution wettability.

**Figure 7.** (**a**) The specific capacitance calculated by galvanostatic charge-discharge curves at different current densities ranging from1Ag−<sup>1</sup> to 20 A g<sup>−</sup>1, (**b**) cycling stability at 10 A g<sup>−</sup><sup>1</sup> for *N*CNFs-1, CNFs and *N*CNFs-2 (the inset in (**b**) shows the schematic illustration for electron and ion pathway of the 3D hierarchical nitrogen doped carbon nanonet flakes). (**c**) Nyquist plots. (Z': real impedance, Z": imaginary impedance. And the inset shows a partial enlarged view in high frequency range). (**d**) Bode plots of phase angle relate to frequency in the low frequency range (the inset shows the normal bode plot).

To further explorer the influence of hierarchical pores and the interconnected flake-like structures on the properties of the working electrodes, electrochemical impedance spectroscopy (EIS) was conducted, and the results are shown in Figure 7c,d. As can be seen from Figure 7c, the Nyquist plots, for CNFs, *N*CNFs-1 and *N*CNFs-2, constituted a semicircle in the high frequency region and a straight line in the low frequency region. The nearly vertical profile in the low frequency region for all samples indicated the desired electrical double-layer behavior. As reported before, the semicircle at a high frequency region is related with the electronic resistance between the electrode materials and the electrolyte [47,48]. The resistance for *N*CNFs-1 is 0.73 Ω, which is slightly lower than that for the CNFs and *N*CNFs-2 (details can be seen in the enlarged view in Figure 7c), indicating a higher electron conductivity for *N*CNFs-1. Bode plots of phase angle versus the applied frequency were further performed, which are shown in Figure 7d. We can see that the phase angles of all samples were

located around 82◦ to 85◦ at the low frequency of zero, agreeing with ideal capacitive behavior for EDLS devices [49]. Moreover, the calculated characteristic frequencies *f* 0, defined as the frequency at phase angle of <sup>−</sup>45◦, were 3.2 Hz, 3.2 Hz and 1.1 Hz for *N*CNFs-1, CNFs and *N*CNFs-2, respectively, corresponding to a time constant τ0, defined as τ0 = *f* 0 −1, of 0.31 s, 0.31 s and 0.91 s. The high phase angle and short time constant indicated a faster frequency response and enhanced ionic transport rate for *N*CNFs-1 than that of CNFs and *N*CNFs-2 samples, which were quite beneficial for ion transport dynamic and high performance EDLS devices.
