Metallosupramolecular Polymer Precursor Design for Multi-Element Co-Doped Carbon Shells with Improved Oxygen Reduction Reaction Catalytic Activity

: Heteroatom-doped carbon materials have been extensively studied in the ﬁeld of electrochemical catalysis to solve the challenges of energy shortage. In particular, there is vigorous research activity in the design of multi-element co-doped carbon materials for the improvement of electrochemical performance. Herein, we developed a supramolecular approach to construct metallosupramolecular polymer hollow spheres, which could be used as precursors for the generation of carbon shells co-doped with B, N, F and Fe elements. The metallosupramolecular polymer hollow spheres were fabricated through a simple route based on the Kirkendall effect. The in situ reaction between the boronate polymer spheres and Fe 3+ could easily control the component and shell thickness of the precursors. The as-prepared multi-element co-doped carbon shells showed excellent catalytic activity in an oxygen reduction reaction, with onset potential (E onset ) 0.91 V and half-wave (E half-wave ) 0.82 V vs reversible hydrogen electrode (RHE). The ﬂuorine element in the carbon matrix was important for the improvement of oxygen reduction reaction (ORR) activity performance through designing the control experiment. This supramolecular approach may afford a new route to explore good activity and a low-cost catalyst for ORR. annular dark-ﬁeld scanning transmission electron microscopy (HAADF-STEM) images, elemental energy-dispersive X-ray spectroscopy (EDX) The


Introduction
During the commercialization process of hydrogen fuel cells, exploring electrocatalysts with high oxygen reduction reaction (ORR) activity and outstanding stability is the primary task [1][2][3]. At present, precious metals incorporating carbon materials, such as commercial Pt/C, have been successfully used as ORR catalysts [4,5]. However, their expensive cost and scarcity greatly limit their broader application [6][7][8]. Therefore, fabricating non-precious metal composite materials with low-cost, and a catalytic performance comparable or better than commercial Pt/C toward ORR, is of great significance to commercial applications [9,10].
Pure carbon materials have many advantages, including excellent electrical transport properties, a highly active surface area, chemical stability and superior thermal stability. They have thus become an ideal choice for electrochemical energy storage materials [11][12][13][14][15]. However, pure carbon materials have a highly hydrophobic surface and limited active sites, which bring many problems for the application have a highly hydrophobic surface and limited active sites, which bring many problems for the application in hydrogen fuel cells. Heteroatom-doping has been recognized as an effective approach to increase the electrochemical activity and surface wettability of carbon materials [16,17]. The chemical elements used for doping carbon materials include nitrogen, phosphorus, boron, fluorine and non-precious metals [18][19][20][21][22]. Notably, N, B co-doped carbon materials have shown excellent ORR performance, because they possess large amounts of defects and active sites [23,24]. Except for the heteroatom-doping materials, carbon materials incorporated with metal and N elements (M-N-C) have been considered as one of the promising candidates. Most of the M-N-C catalysts were prepared via heat-treating the carbon materials with an N-containing compound, as well as metal salts, or obtained from the simple pyrolysis of transition metal macrocyclic polymers [25][26][27]. Nevertheless, these approaches have shortcomings, such as inhomogeneity of the carbon materials, the aggregation of the active sites, complicated synthesis, as well as high-cost [28,29]. Hence, there exists a need to develop a facile and effective route to fabricate polymeric precursors, containing multiple elements like B, P, S, F, Co and Fe. This is of great importance to incorporate multi-elements into the carbon materials.
We have developed a supramolecular approach, in which the condensation reaction between boronic and catechol monomers is accompanied with the formation of B-N dative bonds, and can organize as formed boronate polymers into nanospheres with controllable sizes [30]. The catechol moiety has a high coordination efficiency with transition metal ions and the Kirkendall effect occurs during the reaction between boronate polymer nanospheres and transition metal ions, thus resulting in the formation of metallosupramolecular polymer hollow spheres [31]. In this work, we extended this approach through the design of the building blocks of the boronate polymers, and therefore metallosupramolecular polymer hollow sphere precursors, containing B, N, F and Fe elements, could be fabricated. Carbonation of the metallosupramolecular polymer hollow sphere precursor at 650 °C afforded carbon shells co-doped with B, N, F and Fe elements. We focused on the control over the thickness of the shell, the influence of the shell thickness and doping elements (B, N, F and Fe) on the ORR performance of the carbon materials.

Scheme 1.
Synthetic process of the carbon spheres (CSs). Scheme 1. Synthetic process of the carbon spheres (CSs). With the assistance of the B-N coordination, a cross-linking reaction between DFC and TBB adopted a typical cooperative polymerization mechanism, and therefore, mono-dispersed boronate polymer nanospheres (BPN) could be formed in one-step (Figure 1a,b) [30]. The coordination reaction between catechol and Fe 3+ could evidently change the morphology of the boronate polymer nanospheres, which accorded well with a previous report [31]. The metallosupramolecular polymer could be easily carbonized to afford CSs with hollow structures. With the assistance of the B-N coordination, a cross-linking reaction between DFC and TBB adopted a typical cooperative polymerization mechanism, and therefore, mono-dispersed boronate polymer nanospheres (BPN) could be formed in one-step (Figure 1a,b) [30]. The coordination reaction between catechol and Fe 3+ could evidently change the morphology of the boronate polymer nanospheres, which accorded well with a previous report [31]. The metallosupramolecular polymer could be easily carbonized to afford CSs with hollow structures. Figures 1c-e show the TEM images of the CSs prepared from precursors derived from different reaction times. Small pores were created in the interior of the nanospheres within a 6 h reaction time ( Figure 1c). When the reaction time was 12 h, hollow nanospheres with shell thickness ~70 nm were obtained ( Figure 1d). However, too long of a reaction time was likely to destroy the nanospheres, because many of the obtained particles collapsed (Figures 1e). As demonstrated by the previous report, hollow structural metallosupramolecular polymer precursors were generated according to the Kirkendall effect during the reaction between boronate polymer nanospheres and Fe 3+ [31]. This process greatly relied on the removal of the boronic acid component. Too long of a reaction time between the boronate polymer nanospheres and Fe 3+ can result in excessive removal of the boronic acid component, therefore leading to the collapse of the particles. Dark-field TEM imaging of carbon particles obtained from a 12 h reaction time between the boronate polymer nanospheres and Fe 3+ confirmed the hollow structure ( Figure 1f). Energy-dispersive X-ray (EDX) mapping of representative hollow particles indicated the coexistence of C, N, B, F, O and Fe elements (Figures g-l). The carbonization temperature likely had no evident effect on the morphology of the carbon materials, as the TEM images of CS12-550 and CS12-750 ( Figure S1), display similar morphology to CS12-650. Figure 2a gives the Raman spectra of the CSs. The two prominent peaks at 1340 and 1571 cm −1 represent the D band of disordered graphitic structure and the G band of ordered carbon matrix, respectively. The calculated intensity ratio from the D band and the G band (ID/IG) was applied to evaluate the disorder degree of the carbon materials. The calculated ID/IG value of CS6-650, CS12-650 and CS18-650 were 1.22, 1.26 and 1.29, respectively. Probably, the content defect site in the carbon materials increased with the increasing reaction between BPN and Fe 3+ . The calculated ID/IG value of CS12-550 and CS12-750 were 0.94 and 1.19, respectively, which were lower than that of CS12-650.   Figure 1c). When the reaction time was 12 h, hollow nanospheres with shell thickness~70 nm were obtained ( Figure 1d). However, too long of a reaction time was likely to destroy the nanospheres, because many of the obtained particles collapsed ( Figure 1e). As demonstrated by the previous report, hollow structural metallosupramolecular polymer precursors were generated according to the Kirkendall effect during the reaction between boronate polymer nanospheres and Fe 3+ [31]. This process greatly relied on the removal of the boronic acid component. Too long of a reaction time between the boronate polymer nanospheres and Fe 3+ can result in excessive removal of the boronic acid component, therefore leading to the collapse of the particles. Dark-field TEM imaging of carbon particles obtained from a 12 h reaction time between the boronate polymer nanospheres and Fe 3+ confirmed the hollow structure ( Figure 1f). Energy-dispersive X-ray (EDX) mapping of representative hollow particles indicated the coexistence of C, N, B, F, O and Fe elements (Figure 1g-l). The carbonization temperature likely had no evident effect on the morphology of the carbon materials, as the TEM images of CS 12-550 and CS   (Figure S1), display similar morphology to CS 12-650 . Figure 2a gives the Raman spectra of the CSs. The two prominent peaks at 1340 and 1571 cm −1 represent the D band of disordered graphitic structure and the G band of ordered carbon matrix, respectively. The calculated intensity ratio from the D band and the G band (I D /I G ) was applied to evaluate the disorder degree of the carbon materials. The calculated I D /I G value of CS 6-650 , CS  and CS 18-650 were 1.22, 1.26 and 1.29, respectively. Probably, the content defect site in the carbon materials increased with the increasing reaction between BPN and Fe 3+ . The calculated I D /I G value of CS 12-550 and CS 12-750 were 0.94 and 1.19, respectively, which were lower than that of CS  .

Composition and Structure Characterization
The crystalline structures of CSs were characterized through XRD. As shown in Figure 2b, an evident broad diffraction peak located at about 25 • could be attributed to the (002) plane of ordered graphitic structure. The sharp peak at about 45 • was the (110) plane of iron, reduced by the hydrogen-argon mixture gas. Two broad diffraction peaks at about 35 • and 43 • were derived from the (311) and (222) planes of Fe 3 O 4 [35]. With the elongation of reaction time between BPN and Fe 3+ , the characteristic peaks of both iron and Fe 3 O 4 were obviously enhanced. It is likely that penetration of Fe 3+ into BPN and formation of metallosupramolecular polymers was time dependent. The effect of carbonization temperature on the crystalline structure of CSs was studied by XRD. CSs obtained from different carbonization temperatures generally had similar peak positions. However, with the carbonization temperature increased from 550 • C to 750 • C, the shape, width and intensity of the peak at 25 • changed. Therefore, the carbonization temperature could affect the carbon matrix crystalline structure of CSs. It was observed that CS 12-650 prepared by the carbonization temperature of 650 • C had the best crystallinity. The crystalline structures of CSs were characterized through XRD. As shown in Figure 2b, an evident broad diffraction peak located at about 25° could be attributed to the (002) plane of ordered graphitic structure. The sharp peak at about 45° was the (110) plane of iron, reduced by the hydrogenargon mixture gas. Two broad diffraction peaks at about 35° and 43° were derived from the (311) and (222) planes of Fe3O4 [35]. With the elongation of reaction time between BPN and Fe 3+ , the characteristic peaks of both iron and Fe3O4 were obviously enhanced. It is likely that penetration of Fe 3+ into BPN and formation of metallosupramolecular polymers was time dependent. The effect of carbonization temperature on the crystalline structure of CSs was studied by XRD. CSs obtained from different carbonization temperatures generally had similar peak positions. However, with the carbonization temperature increased from 550 °C to 750 °C, the shape, width and intensity of the peak at 25° changed． Therefore, the carbonization temperature could affect the carbon matrix crystalline structure of CSs. It was observed that CS12-650 prepared by the carbonization temperature of 650 °C had the best crystallinity.
The pore character of CSs was tested by physisorption of nitrogen at 77 K. As shown in Figure  3, CSs comprise both microporous and mesoporous structures. Detailed Brunauer-Emmett-Teller (BET) data is also listed in Table 1. The surface areas of CS6-650, CS12-650 and CS18-650 are 361.62, 439.47 and 451.13 m 2 g −1 , with relative pore volumes of 0.34, 0.40 and 0.36 cm 3 g −1 , respectively. Obviously, their surface areas mainly resulted from the microporous-and mesoporous-pore structures. The specific surface area of the CSs increased gradually with the increase of reaction for the generation of the precursor. The mesoporous-pore volumes of CS6-650, CS12-650 and CS18-650 were 212.08, 258.68 and 190.00 m 2 g −1 , respectively. Apparently, CS12-650 had the maximum mesoporous-pore compared with CS6-650 and CS18-650. The mesoporous-pore is important and beneficial for ORR. The surface areas of CS12-550 and CS12-750 were 394.38 and 445.20 m 2 g −1 , with relative pore volumes of 0.34 and 0.35 cm 3 g −1 (Table 1), respectively, which were lower than that of CS12-650 and CS18-650. This result indicated that the precursor might be not completely carbonized at 550 °C, and a carbonization temperature of 750 °C was not helpful for the development of the pore structure. The pore character of CSs was tested by physisorption of nitrogen at 77 K. As shown in Figure 3, CSs comprise both microporous and mesoporous structures. Detailed Brunauer-Emmett-Teller (BET) data is also listed in Table 1. The surface areas of CS 6-650 , CS 12-650 and CS 18-650 are 361.62, 439.47 and 451.13 m 2 g −1 , with relative pore volumes of 0.34, 0.40 and 0.36 cm 3 g −1 , respectively. Obviously, their surface areas mainly resulted from the microporous-and mesoporous-pore structures. The specific surface area of the CSs increased gradually with the increase of reaction for the generation of the precursor. The mesoporous-pore volumes of CS 6-650 , CS 12-650 and CS 18-650 were 212.08, 258.68 and 190.00 m 2 g −1 , respectively. Apparently, CS 12-650 had the maximum mesoporous-pore compared with CS 6-650 and CS  . The mesoporous-pore is important and beneficial for ORR. The surface areas of CS 12-550 and CS 12-750 were 394.38 and 445.20 m 2 g −1 , with relative pore volumes of 0.34 and 0.35 cm 3 g −1 (Table 1), respectively, which were lower than that of CS 12-650 and CS  . This result indicated that the precursor might be not completely carbonized at 550 • C, and a carbonization temperature of 750 • C was not helpful for the development of the pore structure. The crystalline structures of CSs were characterized through XRD. As shown in Figure 2b, an evident broad diffraction peak located at about 25° could be attributed to the (002) plane of ordered graphitic structure. The sharp peak at about 45° was the (110) plane of iron, reduced by the hydrogenargon mixture gas. Two broad diffraction peaks at about 35° and 43° were derived from the (311) and (222) planes of Fe3O4 [35]. With the elongation of reaction time between BPN and Fe 3+ , the characteristic peaks of both iron and Fe3O4 were obviously enhanced. It is likely that penetration of Fe 3+ into BPN and formation of metallosupramolecular polymers was time dependent. The effect of carbonization temperature on the crystalline structure of CSs was studied by XRD. CSs obtained from different carbonization temperatures generally had similar peak positions. However, with the carbonization temperature increased from 550 °C to 750 °C, the shape, width and intensity of the peak at 25° changed． Therefore, the carbonization temperature could affect the carbon matrix crystalline structure of CSs. It was observed that CS12-650 prepared by the carbonization temperature of 650 °C had the best crystallinity.
The pore character of CSs was tested by physisorption of nitrogen at 77 K. As shown in Figure  3, CSs comprise both microporous and mesoporous structures. Detailed Brunauer-Emmett-Teller (BET) data is also listed in Table 1. The surface areas of CS6-650, CS12-650 and CS18-650 are 361.62, 439.47 and 451.13 m 2 g −1 , with relative pore volumes of 0.34, 0.40 and 0.36 cm 3 g −1 , respectively. Obviously, their surface areas mainly resulted from the microporous-and mesoporous-pore structures. The specific surface area of the CSs increased gradually with the increase of reaction for the generation of the precursor. The mesoporous-pore volumes of CS6-650, CS12-650 and CS18-650 were 212.08, 258.68 and 190.00 m 2 g −1 , respectively. Apparently, CS12-650 had the maximum mesoporous-pore compared with CS6-650 and CS18-650. The mesoporous-pore is important and beneficial for ORR. The surface areas of CS12-550 and CS12-750 were 394.38 and 445.20 m 2 g −1 , with relative pore volumes of 0.34 and 0.35 cm 3 g −1 (Table 1), respectively, which were lower than that of CS12-650 and CS18-650. This result indicated that the precursor might be not completely carbonized at 550 °C, and a carbonization temperature of 750 °C was not helpful for the development of the pore structure.   The X-ray photoelectron spectroscopy (XPS) survey spectra of CS 6-650 , CS 12-650 and CS 18-650 are shown in Figure 4a. In addition, the high-resolution XPS spectra of C 1s, N 1s, B 1s, F 1s and Fe 2p of CS 12-650 were characterized (Figure 4b-f). The C 1s signal can be split into four representative peaks at about 288.  [36]. The F 1s signal is displayed in Figure 4e. This indicated that fluorine remains in the carbon matrix. The weak peak at 685.7 eV was assigned to the F − anions. The existence of the F element had the ability to accelerate the oxygen reduction reaction process of carbon catalysts [37]. For the Fe 2p spectrum shown in Figure 4f, the peak at 723.2 eV was assigned to the binding energy of Fe 2+ , and the peak for Fe 3+ was detected at 725.4 eV for the 2p 1/2 band. Another two peaks at 714.4 and 710.5 eV could be respectively attributed to the binding energies of the 2p 3/2 orbitals of Fe 3+ and Fe 2+ species [38]. The Fe 2p 3/2 peak at approximately 703.5 eV corresponded to Fe 0 . The last peak at 719.6 eV was a satellite peak. The XPS results, in combination with the XRD results, clearly confirmed that only a small amount of iron element was transformed into Fe 3 O 4 , and most of iron element was changed into zero-valent iron, which may be of potential in improving the electrical conductivity of the carbon materials.  The X-ray photoelectron spectroscopy (XPS) survey spectra of CS6-650, CS12-650 and CS18-650 are shown in Figure 4a. In addition, the high-resolution XPS spectra of C 1s, N 1s, B 1s, F 1s and Fe 2p of CS12-650 were characterized (Figures 4b-f). The C 1s signal can be split into four representative peaks at about 288.  [36]. The F 1s signal is displayed in Figure 4e. This indicated that fluorine remains in the carbon matrix. The weak peak at 685.7 eV was assigned to the F − anions. The existence of the F element had the ability to accelerate the oxygen reduction reaction process of carbon catalysts [37]. For the Fe 2p spectrum shown in Figure 4f, the peak at 723.2 eV was assigned to the binding energy of Fe 2+ , and the peak for Fe 3+ was detected at 725.4 eV for the 2p1/2 band. Another two peaks at 714.4 and 710.5 eV could be respectively attributed to the binding energies of the 2p3/2 orbitals of Fe 3+ and Fe 2+ species [38]. The Fe 2p3/2 peak at approximately 703.5 eV corresponded to Fe 0 . The last peak at 719.6 eV was a satellite peak. The XPS results, in combination with the XRD results, clearly confirmed that only a small amount of iron element was transformed into Fe3O4, and most of iron element was changed into zero-valent iron, which may be of potential in improving the electrical conductivity of the carbon materials.  The XPS survey spectra of CS 12-550 and CS  are shown in Figure S2. CS 12-550 obtained at 550 • C had F content of 0.94%. However, at carbonization temperature of 750 • C, the prepared CS  comprised no fluorine element (Table S1). Because the C-F bond in the precursor accorded to a regular fracture, the temperature of 650 • C was the highest temperature that could retain the fluorine element in the carbon matrix, as well as endow the materials with adequate carbonization.

ORR Performances
We then tested the electrochemical performance of the CSs to evaluate their potential as ORR electrocatalysts. The cyclic voltammetry (CV) curves of CS 6-650 , CS 12-650 and CS 18- The XPS survey spectra of CS12-550 and CS12-750 are shown in Figure S2. CS12-550 obtained at 550 °C had F content of 0.94%. However, at carbonization temperature of 750 °C, the prepared CS12-750 comprised no fluorine element (Table S1). Because the C-F bond in the precursor accorded to a regular fracture, the temperature of 650 °C was the highest temperature that could retain the fluorine element in the carbon matrix, as well as endow the materials with adequate carbonization.

ORR Performances
We then tested the electrochemical performance of the CSs to evaluate their potential as ORR electrocatalysts. The cyclic voltammetry (CV) curves of CS6-650, CS12-650 and CS18-650 were tested in Aror O2-saturated 0.    The evidently improved electrocatalytic performance of CS 12-650 could be explained by the following two reasons. First, CS 12-650 had relatively higher specific surface area compared with CS 6-650 as proved by BET results, thus leading to the exposure of more active sites. CS 12-650 also had a higher content of mesoporous-pore than CS 6-650 and CS  , and the mesoporous-pore was beneficial for ORR. Second, CS 12-650 had relatively thin shell thickness and a regular hollow morphology, which was beneficial for the transfer of water and oxygen during the catalytic reaction. As illustrated by the LSV results, the second platform of the LSV curves of CS 12-650 became very flat in comparison with CS  , indicating the equilibrium of the catalytic reaction. As the morphology of CS 18-650 was completely damaged and collapsed, the irregular structure could have increased the specific surface area of CS 18-650 to some extent, but also could have prevented the generation and loading of active sites during carbonization. The activity of CS 18-650 was thus reduced. Therefore, we consider that a synergistic effect between surface area, shell thickness, morphology and enough available active sites, directly led to the optimization of ORR catalytic activity of CS 12-650 .
The LSV curves at different rotating rates were tested to study the reaction kinetics of the ORR catalyzed by CSs. The current density of CS 6-650 , CS 12-650 and CS 18-650 increased when increasing the rotating rate from 400 to 2500 rpm, as shown in Figure 5b-d. This could be attributed to the decrease of the diffusion distance. The Kouteckye-Levich (K-L) plots were calculated from the above LSV curves, which revealed a clear linear relationship and quite similar slopes at the corresponding potential, ranging from 0.3 V to 0.5 V (Figure S3a-c). Obviously, the ORR catalyzed by CSs accorded classic first-order reaction kinetics. By calculating underlying scope shown in Figure 5f, the electron transfer numbers (n) derived from the corresponding slopes of Kouteckye-Levich (K-L) plots were approximately 3.94 for CS 6-650 , 3.96 for CS 12-650 and 3.88 for CS  , verifying an atypical four-electron transfer process. This is the same as the many reported B and N elements co-doped carbon materials [23,24,39].
The CV curves of CS 12-550 and CS  were also tested in Ar-or O 2 -saturated 0.1 M KOH solution (Figure 6a), and the corresponding LSV curves of CS 12-550 and CS  are shown in Figure 6b,c. The onset potential (E onset ) of CS 12-550 and CS 12-750 were 0.85 V and 0.86 V. The half-wave (E half-wave ) potentials of CS 12-550 and CS 12-750 were 0.74 V and 0.77 V. In comparison, the E onset and E half-wave of CS 12-650 were 0.91 V and 0.82 V, respectively, which were much better than the CS 12-550 and CS  . The electron transfer numbers (n) derived from the corresponding slopes of Kouteckye-Levich (K-L) plots ( Figure S3d-f) were approximately 3.26 for CS 12-550 and 3.56 for CS  , verifying a typical four-electron transfer process.
Since they were derived from the same precursor, we considered that the carbonization temperature could evidently affect the ORR performance of the CSs. First, CS 12-650 had higher I D /I G value than CS 12-550 and CS  , implying more defect sites in the carbon matrix. Second, CS 12-650 had relatively higher specific surface area relative to CS 12-550 , as proved by BET results, and thus lead to the exposure of more active sites. The specific surface area of CS 12-650 and CS 12-750 was nearly the same, so CS 12-650 and CS 12-750 had the same utilization towards the active site in the same condition. Third, comparing CS 12-650 and CS  , the existence of the fluorine element in CS 12-650 was important for the improved ORR performance of CS  . Therefore, we considered that the fluorine element in the carbon matrix could promote the generation of defect sites, which was beneficial and important for the ORR activity.
The durability of CS  was tested at 1600 rpm in O 2 -saturated 0.1 M KOH aqueous solution after 1000 cycles of CV curves (Figure 7a). The onset and half-wave potentials of CS 12-650 were kept at 0.90 V and 0.80 V. No evident current decrease in the onset and half-wave potentials was observed. The attenuation of electrocatalytic activity was less than 5%, which was much better than the commercial 20 wt% Pt/C catalysts (as shown in Figure S4). Thus, CSs had preferable durability for ORR in the alkaline condition. The relevant crossover effects test was also carried out by taking CS  as an example through a chronoamperometric experiment. After adding 3.0 M methanol, the methanol oxidation reaction made the current density of the commercial Pt/C decrease immediately (Figure 7b). However, the current density of CS 12-650 did not display an evident change. These results directly confirmed that CS 12-650 has good catalytic selectivity for the oxygen reduction reaction. [23,24,39].
The CV curves of CS12-550 and CS12-750 were also tested in Ar-or O2-saturated 0.1 M KOH solution (Figure 6a), and the corresponding LSV curves of CS12-550 and CS12-750 are shown in Figures 6b,c. The onset potential (Eonset) of CS12-550 and CS12-750 were 0.85 V and 0.86 V. The half-wave (Ehalf-wave) potentials of CS12-550 and CS12-750 were 0.74 V and 0.77 V. In comparison, the Eonset and Ehalf-wave of CS12-650 were 0.91 V and 0.82 V, respectively, which were much better than the CS12-550 and CS12-750. The electron transfer numbers (n) derived from the corresponding slopes of Kouteckye-Levich (K-L) plots ( Figure S3d-f) were approximately 3.26 for CS12-550 and 3.56 for CS12-750, verifying a typical four-electron transfer process.  Since they were derived from the same precursor, we considered that the carbonization temperature could evidently affect the ORR performance of the CSs. First, CS12-650 had higher ID/IG value than CS12-550 and CS12-750, implying more defect sites in the carbon matrix. Second, CS12-650 had relatively higher specific surface area relative to CS12-550, as proved by BET results, and thus lead to the exposure of more active sites. The specific surface area of CS12-650 and CS12-750 was nearly the same, so CS12-650 and CS12-750 had the same utilization towards the active site in the same condition. Third, comparing CS12-650 and CS12-750, the existence of the fluorine element in CS12-650 was important for the improved ORR performance of CS12-650. Therefore, we considered that the fluorine element in the carbon matrix could promote the generation of defect sites, which was beneficial and important for the ORR activity.
The durability of CS12-650 was tested at 1600 rpm in O2-saturated 0.1 M KOH aqueous solution after 1000 cycles of CV curves (Figure 7a). The onset and half-wave potentials of CS12-650 were kept at 0.90 V and 0.80 V. No evident current decrease in the onset and half-wave potentials was observed. The attenuation of electrocatalytic activity was less than 5%, which was much better than the commercial 20 wt% Pt/C catalysts (as shown in Figure S4). Thus, CSs had preferable durability for ORR in the alkaline condition. The relevant crossover effects test was also carried out by taking CS12-650 as an example through a chronoamperometric experiment. After adding 3.0 M methanol, the methanol oxidation reaction made the current density of the commercial Pt/C decrease immediately (Figure 7b). However, the current density of CS12-650 did not display an evident change. These results directly confirmed that CS12-650 has good catalytic selectivity for the oxygen reduction reaction.

Catalyst Preparation
DFC and TBB were dissolved in methanol to afford a concentration of 1.0 mg/mL. TBB solution (6 mL, 0.087 mmol) was added into DFC solution (10 mL, 0.132 mmol) dropwise under N2 atmosphere with vigorous stirring. The mixture solution became deep orange and a suspension of boronate

Catalyst Preparation
DFC and TBB were dissolved in methanol to afford a concentration of 1.0 mg/mL. TBB solution (6 mL, 0.087 mmol) was added into DFC solution (10 mL, 0.132 mmol) dropwise under N 2 atmosphere with vigorous stirring. The mixture solution became deep orange and a suspension of boronate polymer nanospheres (BPN) formed. BPN powder was obtained by centrifugation and washed with anhydrous methanol three times.
BPN was redispersed in 16 mL of methanol to get 2.0 mg/mL of particle suspension. Then 2.37 mL of methanol solution of FeCl 3 (10 mg/mL, 0.037 mmol/mL) was added to the suspension of BPN. After different reaction times, such as 6, 12 and 18 h, the resultant hollow particles were collected by centrifugation, washed by anhydrous methanol, and then dried in vacuum oven at 50 • C overnight. The hollow particles were firstly carbonized at 650 • C for 2 h in an argon atmosphere with a fixed heating rate of 5 • C /min. The second carbonation was performed at 650 • C for 1 h in a mixture gas containing 5% hydrogen and 95% argon with a fixed heating rate of 10 • C /min to reduce Fe 3+ into Fe, thus forming B, N, F and Fe co-doped hollow carbon nanospheres (denoted as CSs). CS 6-650 , CS 12-650 and CS 18-650 represent carbon shells prepared from 6, 12 and 18 h reaction times between BPN and FeCl 3 , respectively. To evaluate the carbonization temperature on the electrochemical properties of carbon shells, control experiments were performed at pyrolysis temperatures of 550 and 750 • C, and the prepared samples were denoted as CS 12-550 and CS 12-750 .

Characterization
The scanning electron microscopy (SEM) were characterized using an SU-70 microscope (HITACHI, Tokyo, Japan). The morphology of the samples was tested by transmission electron microscopy (JEM-2100) (JEOL, Tokyo, Japan). The high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images, elemental energy-dispersive X-ray spectroscopy (EDX) mapping and line scanning analyses were acquired on a FEI TECNAI F20 microscope (Hillsboro, OR, USA) tested at 200 kV. The Raman spectra were tested on a Labram HR800 Evolution (Horiba, Lille, France). The powder X-ray diffraction (XRD) patterns were obtained through a desktop X-ray Diffractometer using Cu (600 W) Kα radiation (Rigaku, Tokyo, Japan) to characterize the crystallographic structure of the samples. The Brunauer-Emmett-Teller (BET) surface area and pore volume of the samples were tested through an ASAP 2460 system (Norcross, GA, USA). Before measurement, all of the samples were uniformly degassed at 120 • C for 12 h under vacuum before the test. The X-ray photoelectron spectroscopy (XPS) was tested by PHI Quantum-2000 photoelectron spectrometer (Physical Electronics, Inc., Chanhassen, MN, USA) through a monochromatic Al X-ray source of Kα radiation (1486.6 eV), and all of the spectra had been calibrated with the C 1s peak at 284.6 eV as an internal standard.

Electrochemical Measurements
The electrochemical properties of the samples were measured on an electrochemical workstation (CHI 760E), through a conventional three-electrode system. Before testing, 5.0 mg of CSs was dispersed in 1.0 mL of a mixed solvent containing anhydrous ethanol (500 µL), H 2 O (450 µL) and 5 wt% Nafion (50 µL). The above slurry (4.5 µL) was dropped onto a glassy carbon electrode and used as working electrode. The quality of the commercial 20 wt% Pt/C was half of the CSs catalyst. ORR performance was tested in freshly made KOH aqueous solution (0.1 M) at room temperature. Pt foil and an Ag/AgCl (KCl saturation) electrode were used as the counter electrode and the reference electrode, respectively. The potential in this article was relative to the Ag/AgCl electrode.

Conclusions
In summary, a new type of carbon shell co-doped with multi-element including B, N, F and Fe was designed and synthesized. The morphology and the content of the Fe element of the carbon shell could be easily tuned by changing the reaction time between the boronate polymer and Fe 3+ . The CS 12-650 catalyst showed excellent catalytic activity toward ORR (E onset = 0.91 V, E half-wave = 0.82 V vs. RHE), which were comparable to commercial Pt/C in an alkaline system. We verified that the existence of the fluorine element in the carbon matrix was important for the improvement of ORR performance. From the prospective of methodology, we consider that the Kirkendall effect-based route to metallosupramolecular polymer hollow spheres may be of great interest in the design of precursors incorporated with transition metal elements, thereby generating high-performance doped carbon materials.