Scalable neutral H2O2 electrosynthesis by platinum diphosphide nanocrystals by regulating oxygen reduction reaction pathways

Despite progress in small scale electrocatalytic production of hydrogen peroxide (H2O2) using a rotating ring-disk electrode, further work is needed to develop a non-toxic, selective, and stable O2-to-H2O2 electrocatalyst for realizing continuous on-site production of neutral hydrogen peroxide. We report ultrasmall and monodisperse colloidal PtP2 nanocrystals that achieve H2O2 production at near zero-overpotential with near unity H2O2 selectivity at 0.27 V vs. RHE. Density functional theory calculations indicate that P promotes hydrogenation of OOH* to H2O2 by weakening the Pt-OOH* bond and suppressing the dissociative OOH* to O* pathway. Atomic layer deposition of Al2O3 prevents NC aggregation and enables application in a polymer electrolyte membrane fuel cell (PEMFC) with a maximum r(H2O2) of 2.26 mmol h−1 cm−2 and a current efficiency of 78.8% even at a high current density of 150 mA cm−2. Catalyst stability enables an accumulated neutral H2O2 concentration in 600 mL of 3.0 wt% (pH = 6.6).

H ydrogen peroxide (H 2 O 2 ) is a valuable chemical for a variety of industrial applications, as well as a potential energy carrier alternative to oil or hydrogen in fuel cells. H 2 O 2 is currently manufactured by a large-scale indirect anthraquinone process and the under-developed direct synthesis from a H 2 and O 2 mixture 1 . The anthraquinone process involves multiple redox reaction steps and requires expensive palladiumbased hydrogenation catalysts. Furthermore, energy-intensive distillation for obtaining high concentration H 2 O 2 is necessary to minimize transportation and storage costs. Direct synthesis via H 2 and O 2 is more straightforward but potentially explosive. Electrochemical H 2 O 2 production through the oxygen reduction reaction (ORR) in an electrolyzer or fuel cell is an attractive and cost-effective route due to its mild operation conditions, on-site production, and tunable concentration 2 . However, it is still a great challenge to develop efficient and stable electrocatalysts that are selective toward the two-electron ORR.
Incorporation of an efficient and stable catalyst into a protonexchange membrane electrolyzer or fuel cell is a promising route to commercialization. Many potential ORR electrocatalysts, mainly including carbon-based materials and noble metal-based materials, have been reported for H 2 O 2 production in alkaline or acidic electrolyte 3 . Carbon-based electrocatalysts typically perform well in alkaline solution but show low intrinsic activity and stability in acidic media 4 . Bimetallic noble metal alloys, such as Au-Pd, Pt-Hg, Pd-Hg, and Au-Pt-Ni, catalyze ORR through two-electron pathways with selectivity as high as 95% [5][6][7] . The goal of secondary metal incorporation is to change the electronic structure of the primary catalytic site and optimize the binding strength of reaction intermediates 8 . An ideal two-electron ORR electrocatalyst should possess a suitable binding strength for OOH* (not too strong or too weak) and suppress the O-O bond breakage in OOH* to O*. However, stability of the bimetallic alloys is a concern, particularly for medicinal or water treatment applications. Elemental leaching hinders the long-term ORR stability and the leached ingredient (particularly toxic Hg) devalues the H 2 O 2 product and increases the separation cost 9,10 . Apart from the metal alloying strategy, incorporation of nonmetal elements such as phosphorus, sulfur, and boron into metals to form multicomponent alloys has been demonstrated to be an attractive and effective way to improve the electrocatalytic activity of metal catalysts [11][12][13][14][15] . Particularly, our previous work found that electronegative phosphorus (P) was able to regulate the binding strength of intermediates in ORR and improve the four-electron ORR activity for cobalt phosphide 16 . Since Pt is the most widely used material for four-electron ORR, it is highly desirable to investigate if P alloying is able to alter the electronic structure of P-rich platinum phosphide and shift the reaction pathway from a four-electron to a two-electron pathway.
Ultrasmall nanoparticle catalysts not only have a high surface to volume ratio which reduces cost for precious metal catalysts by increasing the per mass surface area, but also lead to higher exposure of low-coordinated edge sites for improved electrocatalytic activity 17 . However, small nanoparticle electrocatalysts are prone to aggregation during long-term electrochemical operation, which lowers surface area, reduces available active sites and consequently causes electrocatalytic activity degradation 18 . Encapsulation of nanoparticles in a porous ultrathin shell of a stable metal oxide can largely preserve the catalytic activity while increasing the resistance against aggregation 19 . Atomic layer deposition (ALD), a thin-film deposition technique that allows growth of conformal coating through a self-limiting vapor growth process, has been employed to deposit ultrathin metal oxide layers to overcoat and stabilize nanoparticle catalysts for optimizing both activity and durability [20][21][22] . Therefore, ALD is of interest to prevent aggregation of ultrasmall platinum phosphide nanocrystals and preserve their ORR activity and selectivity during long-term electrocatalysis.
Generally, the ORR activity of an electrocatalyst is initially evaluated by the rotating-ring disk electrode (RRDE) technique, which provides only an upper limit to the activity and selectivity of two-electron ORR due to rapid transportation of H 2 O 2 from disk to ring 2 . In real devices, such as those based on a membrane electrode assembly (MEA) architecture, additional transport factors must be considered. These include the slow diffusion rate of H 2 O 2 from the catalyst and gas diffusion layers to the output stream and the chance for further H 2 O 2 reduction or chemical decomposition 23 . Yamanaka and Wilkinson's groups have reported MEA-based water electrolyzers and fuel cell reactors for production of small amounts of neutral H 2 O 2 by continuously feeding gaseous O 2 into the cathodic chamber [24][25][26][27] . However, the long-term stability for efficient O 2 -to-H 2 O 2 production at high current levels is still a large challenge due to the severe H 2 O 2 accumulation at the interface between ORR catalyst layer and proton-exchange membrane, particularly if there is no solvent flow to remove the concentrated product. To reach a high concentration at a large scale will require incorporation of an efficient two-electron ORR catalyst into a system with sufficient long-term stability in the presence of high H 2 O 2 content to allow for continuous recycling of the H 2 O 2 product to reach medical level concentrations.
Herein, monodisperse colloidal platinum diphosphide nanocrystals (PtP 2 NCs) with an uniform size of 3 nm are directly synthesized by a hot-injection method using platinum(II) 2,4pentanedionate and tris(trimethylsilyl)phosphine. Unlike ORR catalyzed by Pt NCs which follow a conventional four-electron pathway, the ultrasmall PtP 2 NCs proceed through a two-electron pathway with a nearly zero overpotential to initialize the O 2 -to-H 2 O 2 reaction and achieve a maximum selectivity of 98.5% at 0.27 V vs. RHE. DFT calculations reveal that changes in electron density and increased Pt atom separation due to P incorporation leads to a weaker adsorption of the key OOH* intermediate and inhibition of subsequent O-O breakage of OOH* to form the O* intermediate. The ultrasmall PtP 2 NCs are treated by ALD of an alumina overcoat and post-annealing to suppress aggregation and maintain electrocatalytic stability. The resulting catalyst is employed in a PEMFC to achieve a steady neutral H 2 O 2 formation rate of 2.26 mmol h −1 cm −2 and the accumulated H 2 O 2 concentration reaches 3 wt% in 65 h and as high as 1.21 M in 600 mL after continuous cycling for 120 h.

Results
Synthesis and characterization. To synthesize ultrasmall and highly monodisperse platinum phosphide NCs, platinum(II) 2,4pentanedionate and tris(trimethylsilyl)phosphine ((Me 3 Si) 3 P) are employed in a hot-injection synthesis to allow for separate control of the nucleation and growth processes. In a typical preparation, 0.3 mmol of platinum(II) 2,4-pentanedionate (0.118 g) was initially mixed with 8 mL oleylamine (OAm), 0.5 mL oleic acid (OA), and 8 mL octadecene (ODE). Oxygen and impurities were removed by placing the solution under vacuum at 120°C for 1 h. In a nitrogen atmosphere, the solution was then heated to 220°C at a rate of 10°C/min. Meanwhile, the P precursor was prepared by placing 1.2 mL (Me 3 Si) 3 P dissolved in hexane (10 wt%) and 1.0 mL ODE under vacuum to remove the hexane at room temperature. The (Me 3 Si) 3 P solution was quickly injected at 220°C and for 15 min the temperature was maintained. Inductively coupled plasma mass spectroscopy (ICP-MS) gives an atomic ratio of platinum to phosphorus of 0.498 (Supplementary Table 1), and energy-dispersive X-ray spectroscopy (EDS) of platinum phosphide NCs shows a similar Pt:P ratio of 0.509 ( Supplementary Fig. 1), confirming the formation of PtP 2 . The XRD pattern of PtP 2 NCs matches the cubic structure of bulk PtP 2 (ICDD PDF: 01-080-2220) ( Supplementary Fig. 2). The central Pt atoms are surrounded by 6 phosphorus atoms, which are situated at the corners of a slightly distorted octahedron. Similarly, the phosphorus atoms are surrounded by one phosphorus and three platinum neighbors ( Supplementary Fig. 3). The transmission electron microscopy (TEM) image of the assynthesized PtP 2 NCs shows a spherical morphology with average size of 3 ± 0.2 nm (Fig. 1a). A well-resolved lattice fringe with interplane distances of 0.33 nm is observed, corresponding to the (111) crystallographic plane of the cubic PtP 2 (Fig. 1b). The (111) plane, associated with other planes, such as (200), (211), and (222), are shown on the corresponding selected-area electron diffraction (SAED) image (inset of Fig. 1b), indicating good crystallinity of the as-synthesized PtP 2 NCs. The well-defined spherical geometry and high monodispersity of the PtP 2 NCs is clearly observed from the high-angle annular dark-field scanning TEM (HAADF-STEM) image (Fig. 1c). The elemental mapping of the PtP 2 NCs shows that the Pt and P elements are evenly distributed throughout the whole nanocrystal region ( Fig. 1d-f).
X-ray absorption spectroscopy was carried out to reveal the local geometric and electronic structures of PtP 2 and Pt NCs. Figure 1g shows the Pt L 3 -edge X-ray absorption near-edge structure (XANES) spectra of PtP 2 NCs, Pt NCs, and Pt foil. The rising edge of PtP 2 NCs shows a positive shift compared with that of Pt NCs and Pt foil due to the increased valence oxidation state after incorporation of P into Pt. This result can be ascribed to an electron density shift from metallic Pt to local P atoms with high electronegativity. The donor-acceptor nature of electron density distribution in PtP 2 NCs is further confirmed by the X-ray photoelectron spectroscopy (XPS) of Pt 4f ( Supplementary Fig. 5). The binding energy of Pt 4f 7/2 for PtP 2 NCs is positively shifted (0.9 eV) from that of Pt 4f 7/2 for Pt NCs. This shift is larger than that of conventional metal alloying, suggesting that the stronger electron delocalization in PtP 2 NCs may have greater influence on the intrinsic electronic properties and reaction intermediates  adsorption/desorption behavior 28 . The intensity of the Pt L 3 white line is a qualitative indicator of electron vacancies in the 5d orbitals of Pt atoms. The white line intensity for PtP 2 NCs is higher than that for Pt NCs and Pt foil, which is attributed to an electron density shift from Pt to P which creates electron vacancies in Pt and increases probability for electron transition from 2p to the unoccupied 5d orbital 29 . The Fourier transforms of the k 3 -weighted extended Pt L 3 -edge X-ray absorption fine structure (EXAFS) spectra were shown in Fig. 1h. The EXAFS fitting results are summarized in Supplementary Fig. 6 and  (Fig. 2a). A remarkable ring current in the potential range of 0.1-0.708 V vs. RHE is observed on PtP 2 NCs, compared to a negligible current measured for the Pt NCs control. The slightly higher onset potential of the PtP 2 NCs than the thermodynamic limit of 0.70 V vs. RHE could potentially be ascribed to a Nernstrelated potential shift. The PtP 2 NCs achieve a maximum H 2 O 2 selectivity of 98.5% at 0.27 V vs. RHE, and the corresponding electron transfer number (n) is 2.03 ( Supplementary Fig. 8), while the Pt NCs follow a typical four-electron pathway during ORR process. This indicates that phosphorus incorporation can regulate the electronic structure of surface Pt active sites and therefore alter the ORR pathway. Figure 2b summarizes the mass activity of different electrocatalysts for O 2 -to-H 2 O 2 conversion, and further details of these calculations are available in Supplementary Table 3. The remarkable mass activity and low overpotential of PtP 2 is superior to most reported electrocatalysts. While it is inferior to the state-of-the-art Pt-Hg and Pd-Hg alloys, replacement of Hg with minimal loss in efficiency is promising since the potential for Hg to leach directly into a medicinal product is a considerable barrier for practical application.
To directly compare stability, highly monodispersed Pt-Hg nanocrystals were also synthesized ( Supplementary Fig. 9). The PtP 2 NCs show very similar mass activity to the-state-of-art Pt-Hg NCs ( Supplementary Fig. 10), but the leaching of heavy metals is greatly reduced as determined by ICP-MS. The concentration of Hg leached from Pt-Hg NCs is 74.6 × 10 3 ppb after ORR for 6 h in 0.1 M HClO 4 , which is three orders of magnitude higher than the Pt and P leached from PtP 2 NCs (Supplementary Table 4). This severe leaching phenomenon limits the wide application of H 2 O 2 produced by the Pt-Hg nanoparticle electrocatalyst, particularly for medicine or water purification. In general, the chemical stability of bimetallic Pt-Hg alloy catalysts is highly dependent on their synthesis routes and surface properties and further studies need to be done to systematically establish Hg leaching discrepancy between the discussed Pt-Hg NCs and the reported electrodeposited Pt-Hg nanoparticles 6 .
In situ attenuated total reflection infrared spectroscopy (ATR-IR) was carried out to investigate adsorbed oxygen intermediates on the supported PtP 2 catalyst during ORR in O 2 -saturated 0.1 M HClO 4 (Fig. 2c). As detailed below, the in situ observation of OOH ad and HOOH ad during ORR confirms the associative two-electron pathway for PtP 2 . In accordance with previous literature, the bands at 1435 and 1330 cm −1 are assigned to the functional groups of the carbon support, and the bands at 1126 and 1052 cm −1 are ascribed to the ClO 4 − species 30 . The absorption band at 1488 cm −1 , present at all applied potentials, is assigned to the O-O stretching mode of adsorbed molecular oxygen (O 2,ad ). This assignment agrees with previous report that the band of adsorbed O 2 is observed around 1468 cm −1 for Pt/C 31 . The band intensity shows a continuous decrease with increase of overpotential due to O 2 consumption as well as substitution by the newly formed oxygen species ( Supplementary Fig. 11a). This band disappears in N 2saturated solution and shifts to 1406 cm −1 in the 18 O 2 isotope spectrum ( Supplementary Fig. 12). The band at 1264 cm −1 can be assigned to the O-O stretching mode of adsorbed superoxide (OOH ad ), which is consistent with that reported for the Au electrode 32 . It appears at 0.7 V vs. RHE and increases with increasingly negative potential ( Supplementary Fig. 11b), which is consistent with the trend of the H 2 O 2 current in the linear sweep voltammograms of PtP 2 NCs. The OOH ad band shows a significant shift to 1193 cm −1 in 18 O 2 ( Supplementary Fig. 10) , while remaining the same in the deuterated medium (Supplementary Fig. 13). As for Pt NCs, the OOH ad band is initially observed at 0.9 V vs. RHE, consistent with its onset potential of four-electron ORR. Lastly, the band at 1396 cm −1 is attributed to the OOH bending mode of adsorbed hydroperoxide (HOOH ad ) 33 . A slight shift is observed for HOOH ad in 18 Supplementary  Fig. 12), while the band is not detected in the deuterated medium presumably due to a large shift to lower wavenumber ( Supplementary Fig. 13). The HOOH ad band first appears at 0.7 V vs. RHE for PtP 2 NCs, in accordance with the onset in H 2 O 2 production. Figure 2d shows the normalized in situ Pt L 3 -XANES spectra of PtP 2 NCs recorded at various ORR potentials in 0.1 M HClO 4 . Commercial PtO 2 , PtCl 2 , and Pt foil were used as references and a linear combination of XANES spectra was fitted to the in situ Pt L 3 -edge spectra (Supplementary Figs. 14 and 15). The calculated oxidation states of platinum species in PtP 2 under various potentials are shown in Fig. 2e and summarized in Supplementary Table 5. For the as-prepared PtP 2 NCs, the initial oxidation state of platinum species is calculated to be +3.26, which is attributed to the electron density shift from Pt to P. Under opencircuit potential (OCP) and 0.9 V, a slight change of oxidation state is observed. This is presumably attributed to the formation and dissolution of minor surface oxide based on the small change of white line intensity. At a potential of 0.7 V a negative shift in the absorption edge is observed the average oxidation state decreases to +2.72 since the surface platinum active sites in PtP 2 start to be involved in O 2 activation and H 2 O 2 generation. The oxidation state further decreases to +2.25 by 0.5 V and remains the same to 0.3 V. It should be noted that the oxidation state of +2.25 keeps stable at 0.3 V for 6 h of measurement (Supplementary Fig. 16), indicating that the high P content can stabilize Pt +2.25 in the PtP 2 for sustaining rapid and stable O 2 -to-H 2 O 2 conversion.
The inset of Fig. 2d illustrates the changes in the Pt L 3 -edge XANES spectra with the in situ potential. Here, Δμ is the value obtained by subtracting the Pt L 3 -edge XANES spectrum at 0.54 V from that collected at elevated potentials. The |Δμ| of Pt NCs increase monotonically with potential, while for PtP 2 |Δμ| increases sharply from 0.54-0.7 V but remains unchanged to 0.9 V. The value of |Δμ| is expected to increase with increasing amounts of oxygen species adsorption 34 . The relatively high |Δμ| for PtP 2 within 0.54-0.7 V indicates a high coverage of oxygen intermediates in this region of intermediate activity, potentially due a rate limiting step beyond generation of the initial oxygen intermediates such as *OOH adsorption.
Theoretical DFT calculation for ORR. Two-electron ORR generally follows two hydrogenation steps for H 2 O 2 production, i.e., O 2 → OOH* → H 2 O 2 , while four-electron ORR proceeds with four successive hydrogenation steps for H 2 O generation, i.e., (Fig. 3a). The optimized geometries of OOH*, O*, and OH* intermediates on PtP 2 (111) are shown in Supplementary Fig. 17. Recent theoretical studies have pointed out that the ORR activity for H 2 O 2 production is strongly related to the binding ability of OOH* 6 . Specifically, for an ideal catalyst for H 2 O 2 production, the adsorption of OOH* should be neither too strong nor too weak.
The adsorption configuration of OOH* is determined by the orientation type of the adsorbed oxygen molecule. There are usually two different adsorption configurations, end-on adsorption and side-on adsorption, for O 2 on the catalytic surface of Pt and PtP 2 ( Supplementary Fig. 18). For pure Pt, the optimized adsorption energy (E ads ) of the end-on configuration is more negative than that of side-on configuration, indicating that the former orientation is more favorable and subsequently that associative hydrogenation prevails over the dissociative reaction pathway. For PtP 2 , the adsorption energy of the end-on    than that in Pt-OOH* (2.08 Å) (Fig. 3a). This effect is attributed to electron delocalization in phosphorus-rich PtP 2 (Fig. 3c). For the bridge site of PtP 2 (111), the average Pt-O bond length is increased because the Pt-Pt distance is larger due to the high degree of P incorporation (Fig. 3b). A lower overlap among the binding states between Pt 5d and O 2p is observed when OOH* is adsorbed on PtP 2 (111) compared with Pt (111) (Fig. 3d), leading to a weaker binding strength of OOH* over PtP 2 (111) surface. Figure 3f shows the potentialdependent free-energy diagram for two-electron and fourelectron ORR pathways on the PtP 2 (111) surface. At zero electrode potential (U = 0 V) all elementary steps of the ORR are exothermic. At the equilibrium potential (U = 1.23 V), based on the free energies of the intermediates species the first two steps are predicted to be rate determining for the four-electron ORR. At an electrode potential of 0.7 V, the free-energy difference between OOH* to H 2 O 2 is 0.106 eV compared to 0.180 eV for that of OOH* to O*, consistent with the experimental observation that the adsorbed OOH* preferentially undergoes hydrogenation to form H 2 O 2 rather than continuing the dissociation path under a small overpotential. This result can be understood not only by the weakening of the Pt-OOH* bond, but also by the tightening of the O-O bond in OOH* adsorbed to PtP 2 relative to *OOH adsorbed on Pt. For both the top and bridge sites on PtP 2 , the O-O bond length of adsorbed OOH* is    shorter than that on Pt, which is beneficial to suppress the OOH*-to-O* dissociation tendency.
ALD overcoat for stabilization of ultrasmall PtP 2 NCs. Although the ultrasmall PtP 2 NCs gave efficient and selective O 2to-H 2 O 2 conversion, both the disk and ring current quickly decayed and around 60% activity was lost within 60 h at an applied potential of 0.4 V vs. RHE (Fig. 4a). Considering that the elemental composition, crystalline structure, and surface electronic structure of PtP 2 all remain the same following the stability test ( Supplementary Fig. 19), the degradation is ascribed to size instability, which leads to nanocrystal aggregation (Supplementary Fig. 20) and a significant decrease in the electrochemical active surface area (EASA, Supplementary Fig. 21). The left shift of Pt L 3 -rising edge of XANES is also attributed to small size nanocrystal aggregation (Fig. 4l). Ultrathin metal oxides by ALD  40 ). ALD of a thin Al 2 O 3 layer was carried out to stabilize the ultrasmall PtP 2 NCs on a commercial carbon support by alternately exposing the sample to cycles of trimethylaluminum (TMA) and water at 175°C (Fig. 4b). Figure 4c shows that the average size of PtP 2 NCs is increased to 5.2 ± 0.4 nm due to minor thermal aggregation during the mild-ALD process. The Al 2 O 3 overcoat thickness is 1.8 ± 0.2 nm after 42 cycles (Fig. 4d, Supplementary Fig. 22a). This overcoat thickness is smaller than that expected for the typical ALD growth rate for Al 2 O 3 of 1.19 Å cycle −1 , ascribed to the site blocking from any hydrophobic organic ligands remaining on the PtP 2 NCs surface ( Supplementary Fig. 23). We determined that 42 cycles of ALD Al 2 O 3 is required to stabilize PtP 2 NCs and maintain ORR durability, but without further processing the ring current shows a 32.5% decrease compared to the uncoated sample (Supplementary Fig. 24). An observed decrease in EASA is consistent with an expected blocking of active sites by the ALD layer (Supplementary Fig. 25). A balance between increasing the stability and maintaining high activity is achieved by annealing the sample following ALD at 600°C in N 2 gas for 2 h. The size dispersity and element distribution are well maintained (Fig. 4e, g-i), while the overcoat thickness is diminished (~0.54 nm) (Fig. 4f, Supplementary Fig. 22b) and the EASA and Brunauer-Emmett-Teller (BET) surface area is almost restored (Supplementary Fig. 25).
The disk and ring current of the annealed sample (Al 2 O 3 / PtP 2 -600) remain the same before and after 60 h of ORR testing in 0.1 M HClO 4 (Fig. 4a), indicating that the Al 2 O 3 overcoat and activation can stabilize the PtP 2 NCs, while maintaining open pathways to the NC surface ( Supplementary Fig. 26). It should be noted that the O 2 -to-H 2 O 2 selectivity is nearly the same after the ALD surface modification, indicating that the ultrathin Al 2 O 3 overcoating has a negligible geometric effect on the electrocatalytic selectivity of PtP 2 electrode (Supplementary Fig. 27). The Pt L 3 -edge XANES and corresponding EXAFS of PtP 2 , Al 2 O 3 / PtP 2 , and Al 2 O 3 /PtP 2 -600 were compared in Supplementary  Fig. 28. No change in the absorption edge (E 0 ) and Pt-P bond distance was observed, which suggests that the intrinsic electronic structures of PtP 2 were well maintained after ALD coating and annealing treatment. This could be further supported by the Pt 4f XPS results ( Supplementary Fig. 29). The slight increase for the white line intensity of XANES attributes to the formation of small amount of platinum oxide during ALD process. To further evaluate the effect of the activated coating on the electrochemical properties, CO desorption (Fig. 4k) was studied by electrochemical CO stripping. Prior to the activation step, the shift in the oxidation peak is lost after Al 2 O 3 overcoating and the current decreases suggesting the active surface is blocked. We also note that incorporation of P into Pt facilitates CO desorption, as proven by the negative shift of CO oxidation peak (0.11 V) in the CO stripping test (Fig. 4k).  Figure 5a shows a schematic diagram of a PEMFC using commercial Pt/C anode for H 2 oxidation, Nafion 117 membrane for proton transportation and elimination of gas crossover, and our Al 2 O 3 /PtP 2 -600 cathode for two-electron ORR. The details of the expeirmental setup and operation are shown in Supplementary Fig. 31. The Al 2 O 3 / PtP 2 -600 catalyst ink was spray coated onto the carbon gasdiffuse layer (GDL) and hot pressed with the Nafion membrane to form the MEA. The overall MEA is compactly connected and no voids are observed between the interfaces (Fig. 5b, c). The Pt, P, and Al elements are well distributed across the MEA (Fig. 5d-f). The optimized conditions (e.g., mass loading, catalyst support, water flow rate, and operation temperature) for H 2 O 2 production in PEMFC are summarized in Supplementary Table 6.
The optimized mass loading of Al 2 O 3 /PtP 2 -600 is determined to be 0.8 mg cm −2 ( Supplementary Fig. 32), as lower loading leads to kinetics loss and higher loading causes high O 2 gas mass transport reisistance 41 . The current efficiency (CE%) and H 2 O 2 production rate (r(H 2 O 2 )) at 50 mA cm −2 is 60.8% and 0.57 mmol h −1 cm −2 , respectively. A high CE is achieved despite the more challenging kinetics for H 2 O 2 production at neutral conditions compared to the commonly used acidic conditions. At high current density, cathode flooding is typically observed in carbon-based GDE due to the loss of hydrophobicity during PEMFC operation 42 . For the PEMFC with 20 wt% teflon-treated GDL, the operating current density can be increased up to 125 mA cm −2 with significantly improved r (H 2 O 2 ) (1.51 mmol h −1 cm −2 ) compared to that with the nonteflon-treated GDE ( Supplementary Fig. 33). Both the highest CE% and r(H 2 O 2 ) are achieved at a water flow rate of 10 mL min −1 (Supplementary Fig. 34). Increasing the water flow rate within 2-10 mL min −1 is beneficial to remove the generated H 2 O 2 and minimize its thermochemical decomposition and/or further electroreduction 27 (Fig. 5g). The difference in CE for of O 2 -to-H 2 O 2 in RRDE and MEA system is attributed by the transport efficiency of H 2 O 2 away from the electrode surface 2 .
In the RRDE, the small amount of generated H 2 O 2 molecules are rapidly transported away from the disk electrode and oxidized at the ring electrode, leading to a low steady-state surface H 2 O 2 concentration. By contrast, the H 2 O 2 molecules produced at the catalyst/membrane interface in the MEA system need a longer time to diffuse though the catalyst layer and gas diffusion layer before removal into the water stream, which causes a higher local concentration of H 2 O 2 at the vicinity of catalyst surface and increases the possibility of further H 2 O 2 reduction 27 . Moreover, nearly no degradation of the H 2 O 2 concentration (14.7 mmol L −1 , pH = 6.8) is observed at a constant potential of 0.4 V in 120 h (Fig. 5h). The compact interfaces for both cathode/electrolyte and anode/electrolyte are well maintained after 120 h of measurement (Supplementary Fig. 35). For comparison, a Pt-Hg NC-based MEA was fabricated in the same way and run in the PEMFC under the same conditions. The concentration of leached Hg is 5.9 × 10 3 ppb in neutral H 2 O 2 catholyte solution (Supplementary Table 7), and approximately 38% degradation for H 2 O 2 concentration was observed after 6 h of PEMFC operation without recycling of the product (Supplementary Fig. 36
Synthesis of platinum diphosphide (PtP 2 ) NCs. A solution of platinum(II) 2,4pentanedionate (0.3 mmol), OAm (8 mL), OA (0.5 mL), and ODE (8 mL) was placed into a round bottom flask with a stir bar and degassed at 120°C under vacuum for 1 h. To prepare for injection of the P source, the solution was then heated to 220°C under nitrogen. Meanwhile, the P precursor was prepared by placing 1.2 mL (Me 3 Si) 3 P dissolved in hexane (10 wt%) and 1.0 mL ODE under vacuum to remove the hexane at room temperature. The (Me 3 Si) 3 P solution was quickly injected at 220°C and for 15 min the temperature was maintained. To facilitate Ostwald ripening, during cooling the system was kept at 120°C for 10 min prior to cooling to room temperature. To purify, the precipitate was collected following centrifugation at 6000 rpm for 6 min and redispersed with hexane. To further purify the NCs, aggregation was induced with a 1:6:1 (v:v:v) hexanes:acetone:methanol solution followed by centrifuged and resuspension with hexane and this procedure was then repeated for a total of four times. The final suspension yield a solution with NC density by weight of 2 mg/mL. For comparison, Pt NCs were synthesized in a similar co-heating way without (Me 3 Si) 3 where i d and i r are the disk and ring currents, respectively. N is the Pt ring current collection efficiency. The N value in our system was calibrated in 0.1 M HClO 4 with a 10 mM K 3 Fe(CN) 6 electrolyte and is approximately 0.25 ( Supplementary Fig. 38).
For the CO striping test, the electrodes were initially immersed in the CO-saturated 0.1 M HClO 4 by purging with 10 wt% CO in N 2 gas for 30 min and then set to 0.10 V vs. RHE for 15 min to form a CO adsorption layer on the catalyst surface. Then the electrolyte was purged by N 2 gas for 10 min to remove the remaining CO in solution. The CO stripping CVs were obtained in a potential range of 0-1.2 V with a scan rate of 20 mV/s. Electrochemical impedance spectroscopy was performed on a Reference 600 (Gamry Instrument Inc.) with the working electrode biased at OCP, while sweeping the frequency from 10 5 to 0.1 Hz with a 10 mV AC dither. To determine the electrochemical capacitance, CV scans in the non-faradaic potential region were conducted and the capacitive current was obtained at the middle potential value for each scan rate. CV was carried out in a nitrogen-purged 5 mM K 3 Fe(CN) 6 /0.1 M HClO 4 solution with platinum foil as the counter electrode. EASA values were calculated using the Randles-Sevcik equation 43 : where I p is peak current (A), n = 1, D = 4.34 × 10 −6 cm 2 s −1 , A is the EASA (cm 2 ), C is the concentration of potassium ferricyanide (5 × 10 −6 mol cm −2 ), and υ is the scan rate (5 mV s −1 ). Conversion from vs. Ag/AgCl to vs. RHE was done by adding 0.197 + 0.059 × pH. For in situ ATR-FTIR measurements, a diamond-like carbon was coated onto a Si wafer (5 × 8 × 1 mm 3 ) to prepare the internal reflection element (IRE). The coated IRE was ultrasonicated for 2 min with 30% concentrated H 2 SO 4 followed by rinsing with DI water before experiments. A 50 µL of 2 mg mL −1 catalyst ink (no Nafion binder) was dropcast on the IRE and dried under air at room temperature. A glassy carbon paper was placed on top of the catalyst layer for good electrical contact. Glassy carbon rod connected to the IRE, Pt gauze, and Ag/AgCl in 3 M KCl were used as the working electrode, counter electrode, and reference electrode, respectively. An FTIR spectrometer with a MCT detector was used for the in situ ATR-FTIR measurements. Solutions were saturated either with O 2 for ORR or with Ar as a control. Gamry Reference 600 potentiostat during recording of the IR spectra.
ALD Al 2 O 3 overcoat for stabilization. The ALD Al 2 O 3 overcoat for supported PtP 2 was grown in a GEMSTAR-6 atomic layer deposition (ALD) system using trimethylaluminum (TMA) and distilled water (H 2 O) at 175°C. The precursors were kept in the chamber for 2.2 s and a 28 s purge was used. To ensure the deposition occurred with typical growth per cycle, a silicon wafer with native oxide was included alongside the sample as a control and the Al 2 O 3 thickness on the wafer was determined by the X-ray reflectivity. After 42 cycles, the overcoated PtP 2 was activated at 600°C for 2 h in tube furnace under N 2 gas flow (Al 2 O 3 /PtP 2 -600). For comparison, the pure Al 2 O 3 thin film with around 2 nm thickness on silicon was fabricated by ALD for 20 cycles.
Scalable H 2 O 2 production in fuel cell. The MEA for testing the activity in a H 2 -O 2 fuel cell was prepared using Al 2 O 3 /PtP 2 -600 catalyst on GDL as cathode, Pt/ C catalyst on GDL as anode, and Nafion 117 membrane. To prepare the cathode, a catalyst ink composed of Al 2 O 3 /PtP 2 -600 dispersed in a water-ethanol mixture with ionomer (Nafion solution, 5 wt%) was sprayed on Teflon-treated or non-Teflon-treated GDL. Anode was prepared with commercial Pt/C (20 wt%) catalyst in the same manner as the cathode. The catalyst loading amount for cathode and anode is 0.8 mg PtP2 cm −2 and 0.3 mg Pt cm −2 , respectively. A hot press (120°C and 40 MPa for duration of 5 min) was used to press the components together with a Nafion 117 membrane. The MEA was then assembled in a single fuel cell consisting 4 cm 2 serpentine flow fields. Humidification of the MEA was performed for 60 min by flowing N 2 with 100% relative humidity at a cell temperature of 80°C. The flow rates for H 2 and O 2 gases are 150 and 200 mL min −1 , respectively. The flow rate of external neutral water was controlled by a peristaltic pump. The pure water flow through the cathode chamber is beneficial for the removal of generated H 2 O 2 molecules and to decrease the thermochemical decomposition and/or further electroreduction of H 2 O 2 . In the product recycling mode of operation, initially the system is run for 1 h run to accumulate 600 mL H 2 O 2 solution which is then continuously cycled back through the system without any separation of the H 2 O 2 . All the conditions, such as mass loading, catalyst support, water flow rate, and operation temperature, were optimized for H 2 O 2 production in the PEMFC. For quantitative analysis of H 2 O 2 concentration, the interaction of the H 2 O 2 with a modified iodate solution was monitored with the UV-vis spectrscopy method 44 . Briefly, solution A for the I 3 − method consisted of 33 g of KI, 1 g of NaOH, and 0.1 g of ammonium molybdate tetrahydrate diluted to 500 mL with water. The solution was stirred for~10 min to dissolve the molybdate. Solution A was kept in the dark to inhibit the oxidation of I − . Solution B, an aqueous buffer, contained 10 g of KHP per 500 mL. The pH was measured using a pH meter. Equal weights of A and B was subsequently mixed, followed by addition of the H 2 O 2 solution. The absorbance of the resulting solution was measured at a maximum wavelength of 351 nm.
Based on the obtained concentration of flow H 2 O 2 solution, the CE of H 2 O 2 production can be calculated by the following equation: where F is Faraday's constant (96485 C mol −1 ), Q is the water flow rate (L s −1 ), C is the H 2 O 2 concentration (mol L −1 ), and I is the current (A). The corresponding H 2 O 2 production rate can be expressed as following equation: where A is the MEA area (4 cm 2 ). Full details of experimental procedures can be found in the Supplementary Information.

Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.
Received: 14 October 2019; Accepted: 6 July 2020; ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-17584-9 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/.