Report

A Eu3+-Eu2+ ion redox shuttle imparts operational durability to Pb-I perovskite solar cells

See allHide authors and affiliations

Science  18 Jan 2019:
Vol. 363, Issue 6424, pp. 265-270
DOI: 10.1126/science.aau5701

A redox road to recovery

Device longevity is a key issue for organic-inorganic perovskite solar cells. Encapsulation can limit degradation arising from reactions with oxygen and water, but light, electric-field, and thermal stresses can lead to metastable elemental lead and halide atom defects. Wang et al. show that for the lead-iodine system, the introduction of the rare earth europium ion pair Eu3+-Eu2+ can shuttle electrons and recover lead and iodine ions (Pb2+ and I). Devices incorporating this redox shuttle maintained more than 90% of their initial power conversion efficiencies under various aging conditions.

Science, this issue p. 265

Abstract

The components with soft nature in the metal halide perovskite absorber usually generate lead (Pb)0 and iodine (I)0 defects during device fabrication and operation. These defects serve as not only recombination centers to deteriorate device efficiency but also degradation initiators to hamper device lifetimes. We show that the europium ion pair Eu3+-Eu2+ acts as the “redox shuttle” that selectively oxidized Pb0 and reduced I0 defects simultaneously in a cyclical transition. The resultant device achieves a power conversion efficiency (PCE) of 21.52% (certified 20.52%) with substantially improved long-term durability. The devices retained 92% and 89% of the peak PCE under 1-sun continuous illumination or heating at 85°C for 1500 hours and 91% of the original stable PCE after maximum power point tracking for 500 hours, respectively.

Device lifetime and power conversion efficiency (PCE) are the key factors determining the final cost of the electricity that solar cells generate. The certified PCE of perovskite solar cells (PSCs) has rapidly reached 23.7% over the past few years (19), which is on par with that of polycrystalline silicon and Cu(In,Ga)Se2 solar cells, but poor device stability (1012) under operating conditions prevents the perovskite photovoltaics from occupying even a tiny market share (13, 14). Generally, commercial solar cells come with a warranty of a 20- to 25-year lifetime with a less than 10% drop of PCE, which corresponds to an average degradation rate of ~0.5% per year (15). Compared with those inorganic photovoltaic materials—e.g., silicon (IV group) and CIGS (I-III-VI group) (16)—the elements or components are mostly large and more polarized in organic-inorganic halide perovskite materials, such as I, methylammonium (MA+), and Pb2+. They construct a soft crystal lattice prone to deform (17) and vulnerable to various aging stresses such as oxygen, moisture (18, 19), and ultraviolet (UV) exposure (20, 21). By encapsulation (2224), interface modification (13, 2529), and UV filtration, the device lifetime can be prolonged by the temporary exclusion of these external environmental factors.

However, some aging stresses cannot be avoided during device operation, including light illumination, electric field, and thermal stress, upon which both I and Pb2+ in perovskites become chemically reactive to initiate the decomposition even if they are well encapsulated (30). Because of the soft nature of I, Pb2+ ions, and Pb-I bonding, intrinsic degradation would occur in perovskite materials upon various excitation stresses, which finally induce PCE deterioration. On one hand, I is easily oxidized to I0, which not only serve as carrier recombination centers but also initiate chemical chain reactions to accelerate the degradation in perovskite layers (31). On the other hand, Pb2+ is prone to be reduced to metallic Pb0 upon heating or illumination, which has been observed in Pb halide perovskite films (32, 33).

Pb0 is a primary deep defect state that severely degrades the performance of perovskite optoelectronic devices (34, 35), as well as their long-term durability (36). Furthermore, most soft inorganic semiconductors are suffering similar instability, such as PbS (37), PbI2 (38, 39), and AgBr (40), among others. Several attempts have been reported to eliminate either Pb0 or I0 defects, like optimizing film processing (41) and additive engineering (4244). To date, these additives are mostly sacrificial agents specific for one kind of defects, which diminish soon after they take effects. Long-term operational durability requires the simultaneous elimination of both Pb0 and I0 defects in perovskite materials in a sustainable manner.

We demonstrated constant elimination of Pb0 and I0 simultaneously in PSCs over their life span, which leads to exceptional stability improvement and high PCE through incorporation of the ion pair of Eu3+ (f6) ↔ Eu2+ (f7) as the redox shuttle. In this cyclic redox transition, Pb0 defects could be oxidized by Eu3+, while I0 defects could be reduced by Eu2+ at same time. The Eu3+-Eu2+ pair is not consumed during device operation, probably because of its nonvolatility and the suitable redox potential in this cyclic transition. Thus, the champion PCE of the corresponding device was promoted to 21.52% (certified, 20.52%) with negligible current density-voltage (J-V) hysteresis. Devices with the Eu3+-Eu2+ ion pair exhibited excellent shelf lifetime and thermal and light stability, which suggests that this approach may provide a universal solution to the inevitable degradation issue during device operation.

The reaction between Pb0 and I0 is thermodynamically favored and has a standard molar Gibbs formation energy for PbI2(s) of −173.6 kJ/mol (45), which provides the driving force for eliminating both defects. However, simply mixing metallic Pb and I2 powder only led to limited formation of PbI2, which suggests the presence of kinetic barriers at room temperature. To enable elimination of Pb0 and I0 defects in PSCs simultaneously across device life span, we propose the “redox shuttle” to oxidize Pb0 and reduce I0 independently, wherein they can be regenerated during the complete circle. It requires selectively oxidizing Pb0 and reducing I0 defects without introducing additional deep-level defects. After finely screening many possible redox shuttle additives, the rare earth ion pair of Eu3+-Eu2+ was identified as the best candidate, mostly owing to their appropriate redox potentials. Eu3+ could easily be reduced to Eu2+ with the stable half-full f7 electron configuration to form the naturally associated ion pair. The redox shuttle can transfer electrons from Pb0 to I0 defects in a cyclical manner, wherein the Eu3+ oxidizes Pb0 to Pb2+ and the formed Eu2+ simultaneously reduces I0 to I (Fig. 1F). Thus, each ion in this pair is mutually replenished during defects elimination.

Fig. 1 Eu3+-Eu2+ ion pair promotes the conversion of Pb0 and I0 to Pb2+ and I in solution and perovskite film.

(A) I0 and Pb0 powder dispersed in mixed DMF/IPA solvent (volume ratio 1:10) with or without Eu3+ [Eu(acac)3], and the solutions were stirred at 100°C. (B) The UV-vis absorption spectra of the upper solution and (C) XRD patterns of the bottom precipitation from the sample and reference solutions (after 60 min) shown in (A). (D) The representative solution and the absorption spectra of bottom layer in which MAI mixed with Eu3+ or Fe3+ dissolved in water/chloroform. (E) EPR spectra of MAPbI3 film with or without Eu3+ incorporation and Eu2O3 sample, for which the value of proportionality factor (g-factor) is 2.0023. (F) Proposed mechanism diagram of cyclically elimination of Pb0 and I0 defects and regeneration of Eu3+-Eu2+ metal ion pair. a.u., arbitrary units; Ref, reference.

The proposed redox shuttle eliminates corresponding defects on the basis of the following two chemical reactions:2Eu3+ + Pb0 → 2Eu2+ + Pb2+(1)Eu2+ + I0 → Eu3+ + I(2)

We first explored the feasibility of the Eu3+-Eu2+ ion pair to promote electron transfer from Pb0 to I0 in solution (Fig. 1A) by dispersing I2 (25 mg) powder and metallic Pb powder (25 mg) in 2 ml of N,N-dimethylformamide (DMF) and isopropanol (IPA) that had a volume ratio of 1:10 as a reference solution. The Eu3+-Eu2+ ion pair was incorporated by further adding europium acetylacetonate [Eu(acac)3] (11 mg) into the 2-ml solution. Under continuous stirring at 100°C, the sample solution gradually turned from black to colorless with a large amount of yellow precipitates after 60 min, whereas the reference solution remained dark brown with little evidence of yellow precipitates.

UV-visible (UV-vis) spectra of the reference solution exhibited an absorption peak at ~370 nm (Fig. 1B), which we attributed to the presence of an I0 species (36) that was absent in the sample solution, which had an absorption peak at ~290 nm that we attributed to a PbIx species. Both the I0 and Pb0 species were effectively converted to I and Pb2+ upon Eu3+ addition. An x-ray diffraction (XRD) measurement on the precipitates revealed both PbI2 (12.7°, 25.9°, 39.5°) and metallic Pb (31.3°, 36.2°, 52.2°) species in both cases (Fig. 1C). In the sample, the characteristic peak intensity ratio of PbI2 to metallic Pb was larger than that of the reference. This result further confirmed that Eu3+ could accelerate the conversion of Pb0 and I0 to Pb2+ and I, respectively.

When we added Eu(acac)3 to the CH3NH3I solution of water/chloroform, we observed no I0 species absorption peak in the corresponding UV-vis spectrum (Fig. 1D), showing that Eu3+ selectively oxidizes Pb0 rather than I. The stronger oxidizing agent of Fe3+ oxidized I species, and the absorption peak of I0 was present. We verified that Eu3+ was reduced to paramagnetic Eu2+ in CH3NH3PbI3 (MAPbI3) perovskite films with 1% (Eu/Pb, molar ratio) Eu3+ incorporated, which showed a strong signal in electron paramagnetic resonance (EPR) measurements (Fig. 1E) that was absent in Eu2O3 and in the reference MAPbI3 film.

We compared the effect of Eu3+ by studying other ions, including redox-inert Y3+ and strong oxidizing Fe3+, by preparing film samples incorporated with 1% metal ions (M/Pb, molar ratio) and performed high-resolution x-ray photoelectron spectroscopy (XPS) analysis to elucidate the potential effects on both Pb0 and I0 defects. As shown in Fig. 2A, the binding energy (BE) at 142.8 and 137.9 eV were assigned to 4f5/2, 4f7/2 of divalent Pb2+, respectively, and the two shoulder peaks at 141.3 and 136.4 eV around lower BE were associated with metallic Pb0. We calculated the intensity ratio of Pb0/(Pb0 + Pb2+) for three metal-incorporated samples and the reference to observe a notable tendency (Fig. 2, A and D, and table S1). The Pb0 intensity ratio in reference reached 5.4%, which is comparable to that of Y3+-incorporated film. This ratio in the perovskite film with oxidative Eu3+ and Fe3+ additives was reduced to nearly 1.0%, indicating that metallic Pb0 was successfully oxidized.

Fig. 2 High-resolution XPS spectra of Pb 4f, I 3d, and the Eu 3d of perovskite films with the incorporation of 1% M/Pb different acetylacetonate metal salts [M(acac)3, M = Eu3+, Y3+, Fe3+].

(A) Pb 4f spectra, the insertions are the enlarged spectra of Pb0 4f. (B) I 3d spectra. (C) Eu 3d spectra. (D) Fitted results of the Pb0/(Pb0+Pb2+) ratio. (E) Fitted results of I/Pb ratio.

With respect to I0 species, it is difficult to obtain I0/(I0 + I) ratio by peak fitting accurately because I0 species are volatile during the annealing process of perovskite film preparation. Thus, we examined the ratio of I/Pb and BE shift to monitor the iodine evolution indirectly. As shown in Fig. 2, B and E, and table S1, we observed the similar I/Pb ratio in the reference and the Y3+-incorporated sample but a much lower ratio in the Fe3+ sample. Incorporation of Fe3+ likely generated I0 species that were released. A higher I/Pb ratio was observed in the Eu3+ sample compared with the reference, possibly indicating less volatile I0 species produced in the corresponding film. Furthermore, the BE of I 3d3/2 further confirmed the argument, wherein it shifted toward a higher value of 0.3 eV in Fe3+ sample but lower 0.2 eV in Eu3+ sample as compared with the reference. Given the lower BE of I, it clearly showed that I was well preserved in the Eu3+ sample. In addition, Eu2+ was 36% of the total Eu content, which further confirmed the Eu3+-Eu2+ ion pair working as a redox shuttle (Fig. 2C).

According to the charge conservation rule, the amount of I0 should be twice that of Pb0 involved in the entire redox reaction. Iodine species (HI and I2) are all volatile, which follows the 1:2 molar ratio (33). We checked the total change in the amount of iodine (ΔI) and lead (ΔPb0) in the film upon the addition of Eu(acac)3, wherein ΔI/ΔPb0 was calculated to be 3.5 (see table S1 and supplementary text). The change in the amount of iodine (ΔI) was about three times that of lead (ΔPb0) during the degradation process, indicating that the amount of I0 species preserved was twice that of Pb0 species consumed upon redox shuttle addition. In the context of a redox reaction, the standard electrode potential (Eθ) is often used as a reference point to rationally predict the occurrence of the reaction. According to the Eθ of each half reaction involved (which may deviate in solid materials) (table S2), Fe3+ is too oxidative and oxidizes Pb0 and I simultaneously. On the contrary, Eu3+ exhibited the suitable Eθ to selectively oxidize Pb0 without I oxidation, while the reduction product of Eu2+ reduced I0 to I at same time. Thus, the constant elimination of Pb0 and I0 defects still preserved the Eu3+-Eu2+ ion pair.

We examined the effectiveness of Eu3+-Eu2+ redox shuttle in the film. Metallic Pb0 is the major accumulated defect in aged perovskite films because of its nonvolatility (33). The content of Pb0 is a measure of the extent of decomposition in the perovskite film. When the sample was subjected to 1 sun illumination or 85C aging condition for more than 1000 hours, the Pb0/(Pb0 + Pb2+) ratio in films with redox shuttle were 2.5% or 2.7%, compared with 7.4% or 11.3% in the reference film, respectively, as shown in fig. S1 and table S3. The redox shuttle can preserve the I/Pb ratio in the aged film. Meanwhile, the corresponding I/Pb ratio in Eu3+-incorporated film was 2.68 or 2.57 as compared with that of reference 2.30 or 2.13, indicating the perovskite film was well preserved.

We also examined the crystallographic and optoelectronic properties perovskite films with the redox shuttle. According to XRD results, the phase structure was retained in the perovskite films with improved crystallinity upon Eu3+ addition (figs. S2 to S4). No residual acetylacetonate anion was detected by XPS and Fourier transform infrared spectroscopy measurement (figs. S5 and S6). The Eu3+-Eu2+ ions were concentrated near the film surface, wherein the detected Eu/Pb ratio was much higher than the precursor ratio (table S1). When the Eu(acac)3 was introduced from 0.15 to 4.8%, we observed neither extra diffraction peaks nor an obvious shift of diffraction peaks in the XRD patterns (figs. S2 to S4), which indicates that Eu3+-Eu2+ ions may not necessarily accommodate in the crystal lattice.

Given the similar radius of Eu2+ [117 pm (46)] and Pb2+ (119 pm), however, we cannot confidently rule out the possibility that Eu2+ replaces Pb2+ at B site, wherein direct evidence is expected. In addition, europium-iodine–based organic-inorganic perovskite (47) and lanthanide ions doped CsPbX3 perovskite nanocrystals were found in previous reports (48). The morphology and grain size of the perovskite film with the tiny amount redox shuttle remained similar to the reference (Fig. 3A and fig. S7). Also, we did not observe obvious orientation variation by synchrotron grazing-incidence wide-angle x-ray scattering (GIWAXS) analysis (Fig. 3B and fig. S8).

Fig. 3 Influence of morphology, orientation, electronic structure, carrier behaviors of Eu3+-incorporated perovskite film, and results of DFT calculations.

The characterization of reference and 0.15% Eu3+-incorporated perovskite film: (A) scanning electron microscopy images; (B) GIWAXS data; (C) time-resolved photoluminescence spectra; (D) J-V characteristics of devices (ITO/perovskite/Au), used for estimating the SCLC defects concentration (Ndefects = 2εε0VTFL/eL2, ε and ε0 are the dielectric constants of perovskite and vacuum permittivity, L is the thickness of the perovskite film, and e is the elementary charge). (E) The interface ultrathin Eu clustering-layer-incorporated structural model. (F) Left: half-reaction potential barriers; right: overall redox charge-transfer reaction barrier for Eu incorporated at the interface. (G) The summary of ΔG between MAPbI3 and MAPbI3 incorporated with Eu at the interface.

In addition, the optical bandgap of the perovskite film upon Eu3+ addition was calculated to be 1.55 eV, similar to that of the reference (fig. S9). The photoluminescence (PL) intensity (fig. S10) and carrier lifetime (Fig. 3C) increased in the perovskite film with the incorporation of Eu3+, indicating the decrease of nonradiative recombination centers from defects elimination. The improvement of the morphology and grain size could also lead to the increased PL lifetime, so the defects reduction should be further confirmed by other methods. We used the space charge–limited current (SCLC) measurement to quantify the defect density Ndefects of 5.1 × 1015 and 1.5 × 1016 cm−3 for Eu3+-incorporated samples and the reference, respectively (Fig. 3D).

We studied the influence of the Eu3+-Eu2+ ion pair on the formation energies of redox reaction, lattice stability, and energy band structure by density functional theory (DFT) calculations. To construct the model, a small fraction of metal ions (Eu3+) was intercalated into two adjacent lattices (Fig. 3E), given the observation that Eu was concentrated at surfaces and grain boundaries. The formation energies for defects elimination (Eqs. 1 and 2) were calculated (Fig. 3F). For both reference and Eu3+-incorporated systems, the half reactions related to Pb0 elimination required a substantially high potential energy as the main barrier, whereas the I0 elimination half reactions were comparably favorable. However, after introducing Eu species at the interface, the barrier in Pb0 elimination half reactions was greatly decreased, but the barrier for I0 elimination half reactions decreased only slightly. With the assistance of Eu species at the interface, the overall redox potential energy has been much lowered, representing an energetical stabilization trend for the charge-transfer reaction (Fig. 3F).

We also compared the thermodynamic properties for reference and Eu-incorporated systems. Figure 3G shows that the MAPbI3 with Eu incorporation has a steeper slope in change of free energy ΔG than in that of reference, meaning that Eu-incorporated MAPbI3 shows a more energetically favorable physicochemical trend than pure MAPbI3 does. Additionally, it reveals Eu incorporation in MAPbI3 materials did not bring in obvious electronic disorders as extra traps (fig. S11).

We incorporated the perovskite absorber equipped with the redox shuttle in two device configurations. One is based on ITO/TiO2/perovskite/spiro-OMeTAD/Au, wherein spiro-OMeTAD refers to 2,2′,7,7′-tetrakis-(N,N-di-p-methoxyphenylamine)-9,9′-spirobifluorene, with MAPbI3(Cl). The other is based on ITO/SnO2/perovskite/spiro-OMeTAD (modified)/Au for higher PCE and stability, with (FA,MA,Cs)Pb(I,Br)3(Cl), in which FA is formamidinium. Both perovskites were deposited by means of a traditional two-step method, during which Eu(acac)3 or other additives were added in PbI2/DMF precursor solution. The two devices showed similar trends (Fig. 4A and fig. S12). The Eu3+-incorporated devices exhibited the best PCE, whereas the Fe3+-incorporated devices suffered from the markedly decreased PCE. The average PCE increased from 18.5 to 20.7% in the mixed perovskite upon Eu3+ addition (Fig. 4A), which is attributed to the effective defects elimination. We attributed the decreased PCE in Fe3+-incorporated devices to the additional I0 defects introduced by oxidation.

Fig. 4 Long-term stability and original performance evolution of PSCs.

(A) Original performance evolution based on (FA,MA,Cs)Pb(I,Br)3(Cl) perovskite with the incorporation of 0.15% different M(acac)3 (M = Eu3+, Y3+, Fe3+). (B) The J-V curve, stable output (measured at 0.97 V), and parameters of 0.15% Eu3+-incorporated champion devices. (C) Long-term stability of PSCs based on MAPbI3(Cl) perovskite absorber with the incorporation of 0.15% different [M(acac)3 (M = Eu3+, Y3+, Fe3+)], stored in inert condition. The PCE evolution of Eu3+-Eu2+-incorporated and reference devices under 1 sun illumination or 85°C aging condition: (D) half PSCs (original PCE: 0.15% Eu3+ incorporated PSCs, 19.21 ± 0.54%; reference PSCs, 18.05 ± 0.38%) and (E) full PSCs (original PCE: 0.15% Eu3+ incorporated PSCs, 19.17 ± 0.42%; reference PSCs, 17.82 ± 0.30%). Scanning speed is 20 mV/s. (F) The MPP tracking of 0.15% Eu3+-incorporated device, measured at 0.97 V and 1-sun illumination.

One of the optimized devices achieved the PCE of 21.52% (reverse 21.89%, forward 21.15%) (Fig. 4B) with negligible hysteresis (certified reverse 20.73%, forward 20.30%, average 20.52%, certificate attached in fig. S13). The measured stable output at maximum point (0.97 V) was 20.9%. Integrating the overlap of the incident-photon-to-current-efficiency spectrum of Eu3+-incorporated PSCs under the AM 1.5-G solar photon flux generated the current density of 23.2 mA·cm−2 (fig. S14). The stabilized J-V performance of PSCs was evaluated as follows (49): parameters are measured under a 13-point IV sweep configuration wherein the bias voltage (current for open circuit voltage VOC determination) is held constant until the measured current (voltage for VOC) was determined to be unchanging at the 0.05% level. The original, stabilized, and poststabilized efficiency of Eu3+-incorporated PSCs tested by third-party certification institution were similar, which indicates the stable characteristics of the devices (fig. S15).

The shelf lifetime of the corresponding devices was investigated, wherein the PCE evolution was descripted for solar cells stored in an inert environment (Fig. 4C). With the Eu3+-Eu2+ redox shuttle incorporated, the devices maintained 90% of the original PCE even after 8000 hours storage because of improved long-term VOC, short-circuit current density (JSC) and fill factor (FF) stability (fig. S16). Although the stability of Y3+-incorporated PSCs was comparable to the reference, Fe3+-incorporated PSC showed severely deteriorated stability, which lost the photoelectric conversion capability completely after merely 2000 hours of storage.

To estimate the stability of Eu3+-incorporated PSCs under operational conditions, half solar cells were subjected to either continuous 1 sun illumination or 85°C aging condition, respectively (Fig. 4D), in which the top charge-transfer materials and electrode were deposited after aging test. Improved long-term VOC and FF stability (fig. S17) allowed the devices, after 1000 hours, to retain 93% of the original PCE continuous 1 sun illumination or 91% after heating at 85°C. Several previous studies showed that small-molecule spiro-OMeTAD would crystallize under thermal stress and create pathways that allow for an interaction of the perovskite and the metal electrode (50, 51). By modifying the hole-transport materials (spiro-OMeTAD) with conductive polymer poly(triarylamine), the full devices incorporated with the Eu3+-Eu2+ ion pair maintained 92% and 89% of the original PCE because of obvious long-term VOC and FF stability improvement (fig. S18) under the same light or thermal stress for 1500 hours, respectively (Fig. 4E). Furthermore, the Eu3+-incorporated full devices could maintain 91% of the original stable PCE tracked at maximum power point (MPP) for 500 hours (Fig. 4F).

Supplementary Materials

www.sciencemag.org/content/363/6424/265/suppl/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S18

Tables S1 to S3

References (5257)

References and Notes

Acknowledgments: The manuscript was improved by the insightful reviews of anonymous reviewers. We thank Y. Yang (University of California, Los Angeles), Y. Li (Beijing Institute of Technology), H. Xie (Central South University), Q. Bao (East China Normal University), and J. Xiao (Beijing Institute of Technology) for insightful data analysis and valuable discussion. We also thank the third certification institutions National Institute of Metrology (China) and Newport Technology and Application Center PV Lab (USA) for authentication tests; beamline BL14B1 (Shanghai Synchrotron Radiation Facility, SSRF) for providing beam time and help during the experiments; and Enli technology Co., Ltd. for help with PV efficiency and EQE measurement. Funding: This work was supported by National Natural Science Foundation of China (nos. 91733301, 51672008, 51722201, 21425101, 21331001, and 21621061), MOST of China (2014CB643800), National Key Research and Development Program of China (grant nos. 2017YFA0206701 and 2017YFA0205101), Beijing Natural Science Foundation (4182026), National Key Research and Development Program of China (grant no. 2016YFB0700700), National Natural Science Foundation of China (51673025), Beijing Municipal Science and Technology Project (no. Z181100005118002), and Young Talent Thousand Program. Author contributions: L.W. and H.Z. conceived the idea and designed the project. H.Z., C.-H.Y., and L.-D.S. directed and supervised the research. L.W. fabricated and characterized devices. Y.H., Y.C., L.L., Z.X., and N.L. also contributed to device fabrication. L.W. performed the SEM, PL, UPS, UV-vis, XPS, and XRD measurements. GIWAXS was performed and analyzed by G.Z., supported by BL14B1 beamline of SSRF. B.D. and Z.L. performed EPR. M.S. and B.H. carried out DFT calculation. L.W. drafted the manuscript; Q.C. and H.Z. revised and finalized the manuscript. Competing interests: The authors have no competing interests. Data and materials availability: All data are available in the main text or the supplementary materials.
View Abstract

Stay Connected to Science

Navigate This Article