Hollow metal halide perovskite nanocrystals with efficient blue emissions

See allHide authors and affiliations

Science Advances  24 Apr 2020:
Vol. 6, no. 17, eaaz5961
DOI: 10.1126/sciadv.aaz5961


Metal halide perovskite nanocrystals (NCs) have emerged as new-generation light-emitting materials with narrow emissions and high photoluminescence quantum efficiencies (PLQEs). Various types of perovskite NCs, e.g., platelets, wires, and cubes, have been discovered to exhibit tunable emissions across the whole visible spectrum. Despite remarkable advances in the field of perovskite NCs, many nanostructures in inorganic NCs have not yet been realized in metal halide perovskites, and producing highly efficient blue-emitting perovskite NCs remains challenging and of great interest. Here, we report the discovery of highly efficient blue-emitting cesium lead bromide (CsPbBr3) perovskite hollow NCs. By facile solution processing of CsPbBr3 precursor solution containing ethylenediammonium bromide and sodium bromide, in situ formation of hollow CsPbBr3 NCs with controlled particle and pore sizes is realized. Synthetic control of hollow nanostructures with quantum confinement effect results in color tuning of CsPbBr3 NCs from green to blue, with high PLQEs of up to 81%.


Metal halide perovskite nanocrystals (NCs) have received great attention as an emerging class of light-emitting materials for their narrow emissions, exceptional photoluminescence quantum efficiencies (PLQEs), and color tunability (14). Various approaches have recently been developed to achieve synthetic control of the composition, size, and shape of perovskite NCs to obtain emissions covering the whole visible spectrum, from deep blue to near-infrared (514). For instance, by manipulating the quantum size effect, the emission of CsPbBr3 NCs can be tuned from green for NCs with sizes larger than the exciton bohr radius (~7 nm) to deep blue for quantum dots, nanowires, and nanoplatelets with strong quantum confinement (1520). Several recent studies have demonstrated NC size control within the quantum confinement regime by introducing organic salts (21), controlling the acid-base environment (22), varying Pb-to-halide ratio (23), etc. (24). Although highly efficient perovskite NCs can be synthesized using colloidal methods, preserving the high PLQEs during transfer of colloidal NCs to thin films for device integration has been challenging (3). On the other hand, in situ formation of highly efficient perovskite NCs is not trivial either, because of the limited control on particle size, monodispersity, and surface passivation. This is even more relevant in blue-emitting perovskite NCs, which are generally in the size range of 4 to 6 nm for CsPbBr3 and thus have high surface area–to–volume ratios that readily lead to high surface defect density and low PLQEs.

Hollow NCs, nanostructures with pores, provide an additional degree of freedom to exploit the quantum size effect by controlling pore and grain sizes. Several classes of hollow NCs have previously been reported, e.g., metal chalcogenides (25, 26), metal oxides (27, 28), and metal nanoparticles (29), with applications in catalysis and energy storage. To the best of our knowledge, hollow metal halide perovskite NCs have not yet been discovered. Here, we report an approach to achieving highly efficient blue emissions from CsPbBr3 perovskite NCs by creating hollow nanostructures with controlled sizes. By spin-casting a mixture solution containing cesium bromide (CsBr), lead bromide (PbBr2), ethylenediammonium bromide [(H3N(CH2)2NH3)Br2], and sodium bromide (NaBr), in situ formation of hollow CsPbBr3 perovskite NCs was achieved. It was found that [NH3(CH2)2NH3]2+ (EDA2+) cations act as both surface passivation ligands and A-site cations that lead to the formation of randomly oriented Pb2+ vacancies. The presence of Na+ along with EDA2+ cations produces hollow nanostructures. By carefully controlling the EDABr2 and NaBr content, size tuning of pore and grain radii of hollow CsPbBr3 perovskite NCs was achieved, resulting in emission color tuning from green (525 nm) to blue (459 nm) with high PLQEs of up to 81%. Our work introduces a unique nanostructure to the family of metal halide perovskites, showing once again their exceptional structural versatility. The hollow perovskite NCs with tunable optical and electronic properties may have potential applications in various optoelectronic applications, for instance, blue light-emitting diodes.


The schematic diagram of a one-step spin-casting process for the preparation of metal halide perovskite thin films is shown in Fig. 1A. Briefly, metal halide perovskite precursors (CsBr and PbBr2) mixed with EDABr2 and NaBr at controlled ratios were spun cast on glass substrates, followed by low-temperature thermal annealing at around 100°C (see experimental details in Materials and Methods). By fixing the concentration of NaBr to 3 mole percent (mol %) with respect to the perovskite precursors and varying the molar ratio of EDABr2 (from 0 to 1.2 with respect to the perovskite precursors), uniform and smooth thin films were produced as evidenced by the atomic force microscopy (AFM) images (fig. S1, A to F). The thickness of these thin films was measured to be 42 ± 8 nm (fig. S2). The thin films were transparent under ambient light but displayed green and blue emissions under UV irradiation (Fig. 1B). The photophysical properties of the thin films were investigated via ultraviolet-visible (UV-vis) absorption, steady-state and time-resolved photoluminescence (TRPL) spectroscopy, and PLQE measurements. The absorption onset and the PL emission peaks were steadily shifted to lower wavelengths with the increase of the EDABr2 concentration, while keeping the NaBr content constant (Fig. 1C). This blue shift of emission spectra was also associated with a pronounced increase in the PL decay lifetimes (Fig. 1D and table S1), indicating the role of EDA2+ cations in surface defect passivation (30). The defect passivation is further supported by PLQE measurements, where samples containing increasing amounts of EDABr2 showed a steep increase (Fig. 1E), with a peak PLQE of 72% at 478 nm for the sample with 1:1 ratio of EDABr2 to the perovskite precursors.

Fig. 1 Processing and photophysical properties of CsPbBr3 thin films containing 3 mol % NaBr and varying concentrations of EDABr2.

(A) A schematic showing the one-step spin coating followed by low-temperature thermal annealing. (B) Thin films with various concentrations of EDABr2 under ambient light (top) and ultraviolet (UV) irradiation (bottom). (Photo credit: M.W., Materials Science and Engineering Program, Florida State University.) (C) Photoluminescence (PL) and absorption spectra, (D) time-resolved PL (TRPL) decay curves, and (E) PLQE and emission peak wavelength of thin films with EDABr2 molar ratio varying from 0 to 1.2 with respect to the perovskite precursors. a.u., arbitrary units.

To better understand the effects of NaBr and EDABr2 on the photophysical properties of perovskite thin films, we systemically adjusted the concentration of NaBr with and without the presence of EDABr2, and vice versa. Changing the NaBr content within the precursor solution (while keeping the concentration of EDABr2 constant at 1:1 ratio with respect to the perovskite precursors) also afforded emission color tuning, from 498 nm for the sample without NaBr to 459 nm for that containing 4 mol % of NaBr (fig. S3A). However, the PLQEs steadily declined with the increase of NaBr concentration (table S2). Controlling the content of NaBr without the presence of EDABr2, however, had little to no effect on the PL spectrum and PLQE (fig. S4A). These results suggested that NaBr did not play a passivation role in the thin films. Moreover, by controlling the EDABr2 content without the presence of NaBr, limited color tuning was achieved, with the blue shift saturating at around 498 nm (fig. S4B). To reveal the structure-property relationships and understand the roles of Na+ and EDA2+ cations in these systems, we investigated four representative samples in detail, whose compositions and photophysical properties are summarized in Table 1. The sample CsPbBr3 was prepared from the solution containing perovskite precursors without the addition of NaBr and EDABr2, the sample CsPbBr3-Na was prepared from the solution containing perovskite precursors and 3 mol % NaBr, the sample EDA5 was prepared from the solution containing perovskite precursors and EDABr2 at 1:1 ratio, and the sample EDA5-Na was prepared from the solution containing perovskite precursors and EDABr2 at 1:1 ratio and 3 mol % NaBr.

Table 1 Composition and photophysical properties of representative thin-film samples.

λem is the wavelength at the emission maxima, ϕ is the PLQE, and τavg is the average PL lifetime. FWHM, full width at half maximum.

View this table:

The structural properties of the four samples were investigated by selected-area electron diffraction (SAED) and x-ray powder diffraction (XRD) (Fig. 2, A and B). The sample CsPbBr3 was identified to have an orthorhombic phase (space group Pnma) and the diffraction peaks were indexed accordingly. Apart from small peak shifts (Fig. 2, C and D), the other three samples displayed the same diffraction patterns without any sign of low-dimensional or nonperovskite phases, indicating the formation of a three-dimensional (3D) perovskite structure. The shifts of the diffraction peaks can be ascribed to unit cell contraction or expansion. For the sample CsPbBr3-Na, slight shifts to higher diffraction angles were observed [Δ2θ ≈ 0.022° for the (121) and ≈ 0.054° for the (202) planes], which was expected as the incorporation of small Na+ cations (ionic radius of 116 pm) at the A-site could lead to unit cell contraction (31). This lattice contraction was further confirmed by XPS, which showed a shift of the Pb 4f and Br 3d photoelectron spectra to higher binding energies in comparison to the control sample (Fig. 2, E and F). This indicates an increased Pb-Br interaction as a result of reduced bond length. In contrast, the diffraction peaks for the samples EDA5 and EDA5-Na displayed small shifts to lower 2θ as a result of unit cell expansion. This shift of diffraction peaks could be well explained to result from the incorporation of large EDA2+ cations [ionic radius, 333 pm (32)] within the 3D perovskite structure, which has previously been shown in other halide perovskites (3236). Previous reports had proposed lattice strain caused by Pb-X (X = Cl, Br, I) bond elongation (35) and the formation of a hollow 3D structure (32) as mechanisms to explain lattice expansion upon the incorporation of large organic cations. Moreover, because the valence band maximum of ABX3-type 3D halide perovskites is an antibonding hybrid state of the metal s and halide p orbitals, reduced interaction of the B2+ and X ions, due to either bond elongation or discontinuities, would lead to energy bandgap increase (37).

Fig. 2 Structural and surface characterization of thin films.

(A) SAED patterns of thin films. (B) XRD patterns of thin films. (C) Pseudo-Voight peak fitting of the diffraction peaks for the (121) and (D) (202) crystallographic planes of thin films. Photoelectron spectra from x-ray photoelectron spectroscopy (XPS) and Gaussian peak fitting for (E) Br 3d and (F) Pb 4f core levels.

XPS was used to probe changes of the Pb-Br interactions in the samples EDA5 and EDA5-Na. Lattice expansion caused by Pb─Br bond elongation would shift the Pb 4f and Br 3d photoelectron spectra to lower binding energies (38, 39) without affecting Br/Pb ratio. In contrast, a hollow structure with Pb2+ vacancies should display higher Br/Pb ratio but little to no change in binding energies of the core levels. Pb 4f and Br 3d photoelectron spectra for EDA5 show no shift compared to those from CsPbBr3 (Fig. 2, E and F). Moreover, quantitative analysis of the XPS spectra showed that Br/Pb ratio increased substantially with the incorporation of EDA2+ (~3.97 for the sample EDA5) compared to the sample CsPbBr3 (~2.27), suggesting the formation of Pb2+ vacancies and a hollow 3D structure. We attribute the below-nominal Br/Pb ratio in pure CsPbBr3 to the presence of high surface defect density as corroborated by the short average PL decay lifetime. Energy-dispersive x-ray fluorescence (EDXRF) was also conducted to probe the relative concentrations of Pb2+ and Br in the studied samples. Br Kα1 and Br Kβ1 as well as Pb Lα1 and Pb Lβ1 lines from the four samples are shown in fig. S5A. To quantify the molar ratios between the different components, the intensity versus concentration scatterplots for the standards (fit using a linear equation) were used as calibration curves (fig. S5, B to D). The Br/Pb ratio in EDA5 obtained through EDXRF displayed a similar trend as what was observed from the XPS quantitative analysis. This confirms the formation of Pb2+ vacancies. The blue shift of emissions for the CsPbBr3 thin films upon the addition of EDABr2 is also consistent with what was observed in hollow 3D MASnI3, MAPbI3, and FASnI3 perovskites (3234, 40).

Increasing the EDA2+ content also led to decreasing grain sizes as evidenced by the XRD peak broadening (fig. S6). The NC formation and surface defect passivation by EDA2+ cations resulted in considerably increased PLQEs, with the highest value recorded for EDA5 at ~81%. Overall, EDA2+ cations can form small crystalline domains of CsPbBr3, passivate their surfaces, and penetrate the lattice to form hollow 3D structures. Further lattice expansion was observed for the sample EDA5-Na, which showed strongly blue-shifted emission as compared to the other samples. We speculate that Na+ cations occupy interstitial sites within the perovskite structure as has been reported for alkali ions in organo-metal halide perovskites (41), because Na+ cations are too small to create more Pb2+ vacancies as seen with EDA2+. A slight shift to higher binding energies was also observed for Br 3d and Pb 4f core orbitals of EDA5-Na. Because of the increased lattice parameter for EDA5-Na, improved orbital overlap between Pb2+ and Br ions cannot explain the observed peak shift in the photoelectron spectra. However, the presence of interstitial defects has previously been shown to result in increased lattice parameters and photoelectron spectra shift to higher binding energies (42). Thus, we ascribe the observed lattice expansion and photoelectron spectra shift to the presence of Na+ interstitials. Hence, EDA5-Na samples could be composed of hollow 3D structures with additional Na+ interstitials.

To directly visualize the effects of the different ions on the microstructural properties of these samples, we used transmission electron microscopy (TEM) and high-resolution TEM (HRTEM) to characterize the thin films. TEM images of CsPbBr3-Na, EDA5, and EDA5-Na indicate the formation of perovskite NCs (Fig. 3, A to C). Particle size distribution analysis revealed median sizes of around 40, 12, and 17 nm for CsPbBr3-Na, EDA5, and EDA5-Na, respectively. As the median particle size of EDA5-Na is still much larger than the exciton bohr radius of CsPbBr3 (~7 nm), blue emission from small quantum-confined CsPbBr3 NCs could not be the case here. With HRTEM (Fig. 3, D to F), large pores at the center of CsPbBr3 NCs with a median size of 4.25 nm were observed in EDA5-Na. The NCs with pores, or hollow NCs, could be observed throughout the whole film (fig. S7, A to D). These hollow NCs were not present in CsPbBr3-Na or EDA5 (Fig. 3, D to E and fig. S8). The median shell thickness of these hollow CsPbBr3 NCs was found to be 6.5 nm (inset of fig. S7C). This shell thickness is slightly larger than the grain size needed for CsPbBr3 to have an emission energy of ~2.6 eV (~477 nm), i.e., ~6 nm (5). However, aside from the quantum confinement effect, the hollow 3D crystal structure of EDA5-Na will also contribute to the energy bandgap widening, as discussed above. Thus, we ascribe the observed optical bandgap widening and emission peak shift to the combined effect of quantum confinement and a hollow 3D crystal structure. These hollow nanostructures answer how the optical bandgap of CsPbBr3 NCs could be tuned while the size of the NC grains was still well above the exciton bohr radius of ~7 nm.

Fig. 3 TEM characterization of thin films.

Transmission electron microscopy (TEM) images of (A) CsPbBr3-Na, (B) EDA5, and (C) EDA5-Na thin films. Insets: particle and pore size distribution of NCs. High-resolution TEM (HRTEM) images of (D) CsPbBr3-Na, (E) EDA5, and (F) EDA5-Na thin films. Inset: Simulated HRTEM images showing lattice fringes for the (101) planes.

The TEM and HRTEM images were further analyzed to confirm that the “lighter” regions were not results of intergrain voids, sample preparation artifacts, or electron-beam damage. Fast Fourier transform of TEM and scanning TEM images of monograin hollow NCs show single sets of spots, indicating the single crystalline nature of these NCs (fig. S9). Moreover, the absence of spots from multiple grains conclusively proves that the pores are not a result of intergrain voids. To exclude the TEM sample preparation as the source of the observed unique nanostructures, we conducted the TEM characterization using an alternative sample preparation technique (fig. S10). The collected TEM images were consistent with what was observed using the “scratch and sonicate” TEM sample preparation technique (experimental details in Materials and Methods), confirming that these nanostructures were not sample preparation artifacts. Moreover, the low-contrast regions in these grains do not resemble the products reported to arise from electron beam damage (Pb-rich nanoparticles) (43). In addition, TEM images taken 10 s apart for a total of 60 s show that there is little change in the microstructure of the sample, indicating that the pores are not results of electron beam damage (fig. S11).

Previous studies on hollow NCs have attributed their formation to nanoscale Kirkendall effect (25, 44), Ostwald ripening (28), galvanic replacement (27), and cation exchange (26). To gain a better understanding on the roles played by the different components during the growth of these hollow NCs, we used a combination of surface and bulk-sensitive techniques (XPS and EDX). It was found that electrons from Cs 3d5/2, C 1 s of C─N bond, and N 1 s orbitals were detectable in XPS (fig. S12), indicating the presence of Cs+ and EDA2+ ions on or near the surface. While electrons from Na 1 s orbital were readily discernable in XPS for CsPbBr3-Na, they were not observed in EDA5-Na (fig. S12). In contrast, EDX spectrum of EDA5-Na confirmed the presence of Na+ ions within the bulk (fig. S13). These results suggest that Na+ cations were located deep in the shell or within the core of the hollow NCs. On the basis of the changes observed in the lattice, surface chemistry, and microstructure of the hollow NCs as compared to the regular NCs, we propose a formation mechanism (Fig. 4A), in which the diffusion of Na+ ions from the surface to the core selectively draws Br anions from the perovskite lattice to the surface (45), reducing the coordination number of Pb2+ cations. Because undercoordinated ions generally display higher free energy compared to those that are fully coordinated, Pb2+ ions will likely diffuse to lattice sites with higher coordination number driven by the free-energy landscape. Although the diffusion of Pb2+ ions has been reported to have a high diffusion barrier of ~2.3 eV, the presence of Pb2+ vacancies due to the incorporation EDA2+ could lower this energy barrier (46). The diffusion of Pb2+ cations will be associated with vacancy migration in the opposite direction resulting in vacancy coalescence and void formation. To test the proposed mechanism, we introduced NaBr before thin-film deposition in the form of a 20-nm thin film on glass substrate. The EDA5 precursor was then deposited on top of the NaBr layer. The obtained thin film displayed a broad PL spectrum that can be deconvoluted to the spectra of EDA5 and EDA5 containing 2 mol % NaBr (fig. S14C). This result indicates that Na+ ions can diffuse into the EDA5 thin film to form hollow NCs. Overall, the effects of individual additives (NaBr and EDABr2) and their combination on the morphological and optical properties of CsPbBr3 can be summarized in Fig. 4B: the addition of only Na+ has little to no effect on the crystalline lattice, the addition of only EDA2+ creates hollow 3D structures, and the addition of both Na+ and EDA2+ leads to the formation of hollow perovskite NCs.

Fig. 4 Proposed formation mechanism and crystal structures.

(A) Proposed formation mechanism of hollow CsPbBr3 NCs beginning from Na+ and Br diffusion from the surface to the core and vice versa, respectively, followed by Pb2+ diffusion via Pb2+ vacancies culminating in vacancy condensation (B) changes of the crystal structures of CsPbBr3 upon the additions of NaBr, EDABr2, and both NaBr and EDABr2.

With the understanding of the roles of the different ions, we demonstrated rational control of the particle and pore sizes of CsPbBr3 hollow NCs and, subsequently, their optical properties, by controlling the Na+ and EDA2+ contents. The increase in Na+ content (in the presence of EDA2+ 1:1 with the perovskite precursors) had no substantial effect on grain size but generally led to larger pore sizes from a median pore size of 3.2 nm for samples with 1 mol % NaBr to 4.8 nm for those containing 4 mol % of NaBr (fig. S15). This pore size increase, in turn, resulted in a narrower shell thickness and a corresponding optical bandgap widening. Conversely, increasing the EDA2+ concentration (in the presence of Na+) had no obvious influence on the pore size but led to the reduction of the particle sizes. The median particle size decreased from 19 to 17 nm for the samples prepared with of 0.4 and 1.2 molar ratios of EDABr2, while keeping the NaBr content constant at 3 mol % (fig. S16).


Obtaining a monograin-thick thin film could lead to further insight into the role of each component by, for instance, allowing an accurate elemental mapping, which can conclusively show the distribution of each component. However, in situ formation of monograin-thick nanocrystalline thin film was found to be a nontrivial task. Thus, we believe that future effort in this area should be directed toward colloidal synthesis of these multicomponent hollow NCs. Freestanding NCs would not only lend a major insight into the morphology of these hollow structures but also provide a path to investigate their formation through in situ microscopy methods (47). Moreover, the increased degrees of freedom in colloidal synthesis of NCs may also lead to the discovery of other exotic structures.

In summary, by carefully choosing appropriate organic and alkali metal salts, light-emitting thin films containing hollow perovskite NCs can be prepared via facile solution processing. The organic cations not only penetrate the perovskite lattice to create vacancies, but also passivate the surfaces of the perovskite NCs. The small alkali metal cations can diffuse throughout the perovskite lattice to assist the formation of hollow structures with the presence of large organic cations. As a result, quantum confinement can be realized in hollow perovskite NCs with controlled pore sizes and shell thicknesses, which exhibit highly tunable emission with high PLQEs. Our work represents a groundbreaking advance in the field of metal halide perovskite NCs, from well-developed nanostructures with positive curvatures to unique hollow structures, which will stimulate exploration of other nanostructures with remarkable and unique properties. Applications of hollow perovskite NCs in various types of optoelectronic devices are envisioned in the near future.



PbBr2 (99.999%), CsBr (99%), ethylenediamine (99%), hydrobromic acid [48 weight % (wt %) in H2O], NaBr (99%), and dimethyl sulfoxide (99.9%) were purchased from Sigma-Aldrich. All reagents and solvents were used without further purification unless otherwise stated.

Synthesis of C2 N2H10Br2 (EDABr2)

Ethylenediamine (1 eq) and 48 wt % hydrobromic acid (2.1 eq) were combined in ethanol at 0°C. After 30 min, white precipitate was extracted through evaporation of the solvent using a rotary evaporator. The white precipitate was washed with diethyl ether twice and was left to dry in a desiccator overnight before use. The salt was stored in a glove box.

Thin-film processing

CsBr (0.2 mmol) and PbBr2 (0.2 mmol) were dissolved in 1 ml of dimethyl sulfoxide to make the perovskite precursor solution. EDABr2 was added to the perovskite precursor solution in varying concentrations from 0 to 1.2 molar ratio. The Na+ content in the perovskite precursor solution was controlled by varying the amount of NaBr from 0 to 4 mol %. Indium tin oxide substrates were purchased from thin-film devices (any glass substrate that does not contain sodium would work). Before use, they were sequentially cleaned by sonicating in detergent (Hellmanex III) water, deionized water, acetone, and isopropanol for 15 min and were dried in an oven overnight. The substrates (bare side) were then cleaned under UV-ozone treatment and transported into a glove box for one-step spin coating of the precursor solution at 3000 rpm for 60 s. Subsequently, the films were annealed at 100°C for 20 min.

Atomic force microscopy

AFM images were taken on a Bruker Icon scanning probe microscope in tapping mode.

PL spectroscopy

Steady-state PL was carried out using an Edinburgh FS5 steady-state spectrometer with a 150-W xenon lamp at an excitation wavelength of 365 nm. TRPL was collected using time-correlated single photon counting for 10,000 counts. Excitation was provided by an Edinburgh EPL-360 ps pulsed diode laser. The PL decay was fit using a biexponential decay curve for CsPbBr3 CsPbBr3-Na and EDA5 and triexponential decay function for EDA5-Na. The average lifetime was obtained from the bi- or triexponential decays according to Eq. 1.τave=αiτi2/αiτi, i=1,2,3(1)where τi represents the decay time and αi represents the amplitude of each component.

Absorption spectrum measurements

Absorption spectra were obtained using an Agilent Technologies Cary 5000 UV-vis near-infrared spectrophotometer.

PL quantum efficiency

PLQE measurements were performed in accordance to what has been reported previously (48). Briefly, a Hamamatsu Quantaurus-QY Spectrometer (Model C11347-11) was equipped with a xenon lamp, an integrating sphere sample chamber, and a charge-coupled device detector. The PLQEs were calculated by the equation: η𝑄𝐸 = IsEREs, in which IS represents the luminescence emission spectrum of the sample, ER is the spectrum of the excitation light for the reference (empty substrate), and ES is the excitation spectrum for exciting the sample. The PLQE measurements were done consecutively, and a density of 0.3 mW/cm2 at an excitation wavelength of 365 nm was used for all samples.

Powder x-ray diffraction

XRD patterns were obtained using a Siemens D500 powder diffractometer equipped with a Cu Kα x-ray source. Diffraction patterns were recorded from 5° to 40° 2θ, with a step size of 0.05° under a tube current of 30 mA and a tube voltage of 40 kV.

TEM images

Microstructural characterization was performed using TEM, on a JEOL JEM-ARM200cF at 200 kV. TEM samples were prepared by scraping thin films from substrate and dispersing the powder in a nonpolar solvent (hexane). The dispersion was sonicated for 10 min to ensure uniform distribution before dropping it on a 200-mesh Cu grid. Alternatively, Cu grids were pressed facedown on thin films prepared on polydimethylsiloxane substrates, followed by applying a single drop of 2-propanol on the Cu grid. The Cu grids were then picked up using fine tweezers before the solvent completely evaporated. TEM image analysis was performed in Gatan microscopy software suite.

X-ray photoelectron spectroscopy

XPS was conducted using a PHI 5000 series XPS equipped with a dual anode x-ray source. For our purpose, Al Kα radiation with a photon energy of 1486.6 eV at a take-off angle of 45° and a pass energy of 35.75 eV were used. Charge compensation was performed using adventitious C 1 s peak (284.6 eV). Spectra background was fit and subtracted using an integrated Shirley function. XPS curves were deconvoluted using a Voight peak function for metal core electron spectra and Gaussian peak functions for the rest.

Energy-dispersive x-ray fluorescence

EDXRF was performed using a Panalytical Epsilon 3 spectrometer equipped with a rhodium x-ray tube and a silicon drift detector.


Supplementary material for this article is available at

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: We thank E. Lochner for help with XPS measurements and helpful discussions. Funding: We acknowledge the support from the National Science Foundation (DMR-1709116 and ECCS-1912911), the Air Force Office of Scientific Research (AFOSR) (17RT0906), and the FSU Office of Research. TEM work was performed at the National High Magnetic Field Laboratory, which is supported by National Science Foundation Cooperative agreement no. DMR-1644779 and the State of Florida. Author contributions: M.W., Y.T., and B.M. conceived the experiments. M.W. and B.M. analyzed and interpreted the data. M.W. performed the measurements with help from C.Z., H.L., M.C., L.-j.X., Q.H., D.B., Y.Z., and X.L.; Y.-f.S and Y.X. performed TEM and SAED measurements. The manuscript was written by M.W. and B.M. The project was planned, directed, and supervised by B.M. All authors discussed the results and commented on the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.
View Abstract

Stay Connected to Science Advances

Navigate This Article