We first conducted a precise molecule design to match the molecular anchor and perovskite lattice accurately. To enable the molecule to have interfacial anchoring properties, we attempted multi-site substitution of the molecule. We chose triphenylphosphine oxide (TPPO) as the basic molecular framework. The calculated interatomic distance of the O atoms from the P=O and the phenyl ring’s ortho, meta, and para positions was 3.1 Å, 4.4 Å and 5.3 Å, respectively. To study the effect of site spacing and nucleophilicity on QDs, we synthesised a series of multi-site anchoring molecules, including TMeOPPO-o, TMeOPPO-p, TFPPO, TClPPO, and TBrPPO (Fig. 1a). The site spacing of multi-site anchoring molecules in that order is 2.6 Å, 6.5 Å, 6.6 Å, 7.0 Å and 7.2 Å, respectively. The TMeOPPO-p could precisely match the lattice spacing of QDs, allowing it to attach to the surface of the perovskite lattice and provide adequate passivation (Fig. 1b)²⁶. Besides, the nucleophilicity of -OCH₃, F⁻, Cl⁻ and Br⁻ decreases in turn, according to the calculated electrostatic potential results (Supplementary Data 1). Additionally, the triple-attached nucleophilic groups significantly increase their probability of interacting with uncoordinated Pb²⁺.Fig. 1

a Schematic illustration of lattice-matched anchoring molecule design and the corresponding calculated electrostatic potential results. b Schematic illustration of the lattice-matched molecular anchoring effect. c PDOS of the perovskite QDs with surface defects. d PDOS of the perovskite QDs treated with the single-site molecule. e PDOS of the perovskite QDs with multi-site lattice-matched anchoring effect.

We synthesised the perovskite CsPbI₃ QDs by using a modified hot-injection method²⁶. Details of the synthesis and purification process are provided in the Methods. We first compared the average PLQYs of QD solutions treated with different molecules (concentration of 5 mg mL⁻¹ in ethyl acetate). The pristine QD and QDs treated with TPPO, TMeOPPO-o, TMeOPPO-p, TFPPO, TClPPO, and TBrPPO showed average PLQYs of 59%, 70%, 82%, 96%, 92%, 88%, and 87%, respectively (Fig. 2a and Supplementary Fig. 2). The QDs treated with TMeOPPO-p showed the highest PLQYs, which is consistent with the theoretical calculation results. TMeOPPO-p could offer multi-site anchoring interactions and eliminate defects in QDs (Fig. 2b); thus, it is designated as the representative molecule for further exploration.Fig. 2

a Average PLQYs of QD solution treated by different molecules. b Schematic illustration of anchoring effect on CsPbI₃ QDs. c STEM images of the pristine and target QDs. Scale bars, 50 nm (the left column) and 2 nm (the right column). d XRD patterns of the pristine and target QD films. e FTIR spectra of the QDs. f XPS spectra of Pb 4f of the QD films. g¹H NMR spectra of the pristine QDs, TMeOPPO-p, and target QDs. h³¹P NMR spectra of the pristine QDs, TMeOPPO-p, and target QDs.

Compared to the pristine ones, the target QDs showed enhanced and red-shifted photoluminescence (PL) emission with a full width at half maximum (FWHM) of 32 nm (Fig. 3a). The PLQY was 97%, much higher than the pristine QDs (65%). Time-resolved photoluminescence (TRPL) decay results (Fig. 3b) showed that the target QDs presented a longer average carrier lifetime (τ_avg) compared with the pristine ones, increasing from 8.22 ns to 19.17 ns (see Supplementary Table 1 for the fitting details). Besides, the trap-related fast decay component was reduced from 0.67 to 0.37, suggesting the suppressed trap states in the target QDs.Fig. 3

a PL spectra and the photographs of the CsPbI₃ QD solutions under the UV lamp. b TRPL curves of the QDs. c GSB kinetics at 1.83 eV extracted from transient TA measurements for pristine and target QDs under low fluence (2.14 eV, 0.61 μJ cm⁻²). d Energy diagram of the hole-only devices. e Current-voltage characteristics of the control hole-only device. f Current-voltage characteristics of the target hole-only device.

Based on these high-quality perovskite QD films, we fabricated QLEDs with an ITO/ PEDOT: PSS: PFI/ PTAA/ QD film/ TPBi/ LiF/ Al architecture (Fig. 4a–c). The target QLED exhibited a lower turn-on voltage and a higher current density than the control device (Supplementary Fig. 6 and Fig. 4d), which can be attributed to the improved carrier transport characteristics. The radiance of the target QLED increased rapidly with the increase in voltage, yielding a maximum radiance of 119,037 mW Sr m⁻², a threefold improvement over the control device (Fig. 4e).Fig. 4

a Schematic illustration of the device structure. b The cross-sectional SEM image of the target QLED. Scale bar, 200 nm. c Energy diagram of the QLED. d Current density-voltage curves. e Radiance-voltage curves. f EQE curves. g The statistics of the distribution of max EQE of 30 devices for the pristine and target QLEDs. h EL spectra. i CIE diagram. The insert shows a photograph of the operating target device, illuminating the Tsinghua University logo.

Apart from enhancing exciton recombination (e.g. higher EQE and radiance), our strategy can also improve the stability of the devices. We first evaluated the operating stability of the target QLED. We individually tested the radiance decays of the control and target devices under varying initial radiance levels. Here, we estimated the T₅₀ through an empirical scaling law widely used for simulating LED degradation:1R0nT50=constant

Where n is the acceleration factor^34,35, and the value was obtained by fitting T₅₀ under different initial radiance. We calculated that the values of n are both 1.87 and the T₅₀ for the control and target devices could be estimated to be 1350 h and 23,420 h (Fig. 5a, b and Supplementary Fig. 8), respectively, at an initial radiance of 190 mW Sr⁻¹ m⁻² (corresponding to a luminance of 100 cd m⁻² for green perovskite LED with EL peak centred at 525 nm, see Supplementary Fig. 9 for details).Fig. 5

The T₅₀ lifetimes as a function of R₀ for the control and target QLEDs. The dashed line fits the T₅₀ data to the equation R₀ⁿT₅₀  =  Constant, where n is the acceleration factor. a the control device. b the target device. c Summary of the reported red perovskite LED characteristics based on maximum EQE, T₅₀ and roll-off. d–f EL performance of air-processed champion QLED. d Current density-voltage-radiance curve. e EQE curve. f EL spectra. g–i The storage stability of a typical air-processed QLED. g Current density-voltage curves of QLED after 0/1/2 weeks of storage. h Voltage-radiance curves of QLED after 0/1/2 weeks of storage. i EQE curves of QLED after 0/1/2 weeks of storage.

To represent the universal effect of phosphine oxide molecules and highlight the superiority of our lattice-matched multi-site anchoring strategy, here we compared the effects of other designed molecules, including TPPO, TMeOPPO-o, TFPPO, TClPPO and TBrPPO. Compared with the control one, the as-fabricated QLEDs presented improved performance of different magnitudes (Fig. 6 and Supplementary Fig. 12). For example, the devices based on the QDs treated with TPPO, TMeOPPO-o, TFPPO, TClPPO and TBrPPO achieved the highest radiance of 49,018 mW Sr m⁻², 54,821 mW Sr m⁻², 67,410 mW Sr m⁻², 85,859 mW Sr m⁻² and 69,587 mW Sr m⁻², respectively. The maximum EQEs in that order were 19.79%, 20.16%, 23.62%, 22.38% and 21.45%, respectively. For operating stability, we calculated that the values of n are 1.87, 1.87, 1.87, 1.86, and 1.86 for the devices based on the TPPO-, TMeOPPO-o-, TFPPO-, TClPPO- and TBrPPO-treated QDs, respectively. The T₅₀ of the devices in that order could be estimated to be 1944 h, 3127 h, 8307 h, 6139 h and 4767 h, respectively.Fig. 6

Electro-optical performance and stability of QLEDs by using various designed molecules, containing current density-radiance-voltage curve, EQE curve (the insert is the chemical structure) and operating stability. a TPPO. b TMeOPPO-o. c TFPPO. d TClPPO. e TBrPPO.

oleyl amine (OAm, 80~90% purity), oleic acid (OA, 80~90% purity), 1-octadecene (ODE, >80% purity), isopropanol (99.7% purity), ethyl acetate (99% purity) and toluene (≥99.5% purity) were supplied by Aladdin. Caesium stearate (CsSt, 98% purity) was supplied by J&K. N-octane (>98% purity) was supplied by TCI. Nafion perfluorinated ionomer (PFI, tetrafluoroethylene-perfluoro-3, 6-dioxa-4-methyl-7-octenesulfonic acid copolymer) was supplied by Sigma-Aldrich. Lead iodide (PbI₂, 99.99%), PEDOT: PSS (Baytron P VPAl 4083), TPBi and PTAA were supplied by Xi’an Yuri Solar Co., Ltd. Dichloromethane was supplied by Shanghai Titan Scientific Co., Ltd.

We synthesised the QD solution referencing our previous report²⁶. The caesium precursor was synthesised by mixing CsSt (2.5 g, 6 mmol), OA (2 mL) and ODE (40 mL). The mixture was then heated to 140 °C under continuous stirring until complete dissolution was achieved, as indicated by the formation of a homogeneous and transparent liquid. For pristine CsPbI₃ QDs, PbI₂ (0.25 g, 0.6 mmol), ODE (10.0 mL), OAm (1.5 mL) and OA (2.0 mL) were added into a 100 mL three-neck flask. The mixture was degassed at 120 °C for 5 min under nitrogen flow to remove oxygen and moisture. The temperature was raised to 180 °C to ensure the full solubilisation of PbI₂. Then, 1.0 mL of caesium precursor was quickly injected into the solution for a 15 s reaction. The flask was immediately immersed in an ice-water bath to quench the reaction and cool the solution to ambient temperature.

For the first-round purification process, the crude QD solution was transferred to centrifuge tubes, and then isopropanol with a 1:3 volume ratio was added, followed by centrifugation at 8000 × g for 1 min. The collected precipitate was dispersed in the toluene solution with a volume ratio of 1:1. For the second-round purification process, OAmI in toluene (0.05 mol L⁻¹) was added to a volume of 0.4 mL. Next, ethyl acetate was added to the solution in a 1:2 volume ratio, followed by centrifugation at 8000 × g for 1 min. The collected precipitate was then dispersed in n-octane. For treated QDs, different molecules (TPPO, TMeOPPO-o, TMeOPPO-p, TFPPO, TClPPO, and TBrPPO) in ethyl acetate (5 mg mL⁻¹) were added into the solution with a volume ratio of 1:2 during the second-round purification process. The remaining steps were the same.

General procedure: The substituted triphenylphosphine (1.0 eq.) was dissolved in the THF (3.0 mL). A hydrogen peroxide solution in H₂O (30%, 5.0 eq.) was added dropwise and stirred for one hour at ambient temperature. After the reaction was quenched by saturated sodium thiosulfate (Na₂S₂O₃) solution, the organic phase was extracted with DCM. The solvent was removed under reduced pressure and the residue was recrystallised twice in the mixed solvent of DCM and hexane to get the final product for further use.

Trisphenylphosphine oxide (TPPO) was isolated as a white solid (120 mg, yield: 75%). mp: 155–156 °C (lit.ref 155–157 °C). ¹H NMR (CDCl₃, 400 MHz): δ 7.55–7.42 (m, 6H), 6.97 (t, J = 7.5, 3H), 6.90 (dd, J = 5.3, 2.3 Hz, 3H), 3.56 (s, 9H). ¹³C NMR (CDCl₃, 100 MHz). δ 161.3, 134.0, 133.6, 121.2. ³¹P NMR (CDCl₃, 162 MHz): δ 25.8. HRMS (m/z): [M]⁺ calcd. for C₁₈H₁₅OP, 278.0861; found, 278.0868.

Tris(2-methoxyphenyl)phosphine oxide (TMeOPPO-o) was isolated as a white solid (115 mg, yield: 80%). mp: 203–205 °C (lit.ref 205–206 °C). ¹H NMR (CDCl₃, 400 MHz): δ 7.55–7.42 (m, 6H), 6.97 (t, J = 7.5, 3H), 6.90 (dd, J = 5.3, 2.3 Hz, 3H), 3.56 (s, 9H). ¹³C NMR (CDCl₃, 100 MHz). δ 161.3, 134.0, 133.6, 121.2. ³¹P NMR (CDCl₃, 162 MHz): δ 25.8. HRMS (m/z): [M]⁺ calcd. for C₂₁H₂₁O₄P, 368.1177; found, 368.1182.

Tris(4-methoxyphenyl)phosphine oxide (TMeOPPO-p) was isolated as a white solid (106 mg, yield: 94%); mp: 143–145 °C (lit.ref 147–149 °C). ¹H NMR (CDCl₃, 400 MHz): δ 7.56 (dd, J = 8.8 Hz, J = 11.2 Hz, 6H), 6.95–6.93 (m, 6H), 3.81 (s, 9H). ¹³C NMR (CDCl₃, 100 MHz): δ 162.2, 133.7, 124.2, 114.0, 55.3. ³¹P NMR (CDCl₃, 162 MHz): δ 29.23. HRMS (m/z): [M]⁺ calcd. for C₂₁H₂₁O₄P, 368.1177; found, 368.1183.

Tris(4-fluorophenyl)phosphine oxide (TFPPO) was isolated as a white solid (120 mg, yield: 91%). mp: 119–121 °C (lit.ref 120–121 °C). ¹H NMR (CDCl₃, 400 MHz): δ 7.67–7.60 (m, 6H), 7.19–7.15 (m, 6H). ¹³C NMR (CDCl₃, 100 MHz): δ 165.2, 134.5, 128.2, 116.1. ³¹P NMR (CDCl₃, 162 MHz): δ 26.9. ¹⁹F NMR (CDCl₃, 376 MHz): δ−106.0. HRMS (m/z): [M]⁺ calcd. for C₁₈H₁₂F₃OP, 332.0578; found, 332.0582.

Tris(4-chlorophenyl)phosphine oxide (TClPPO) was isolated as a white solid (108 mg, yield: 89%). mp: 171–172 °C (lit.ref 172–174 °C). ¹H NMR (CDCl₃, 400 MHz): δ 7.60–7.55 (m, 6H), 7.48–7.44 (m, 6H). ¹³C NMR (CDCl₃, 100 MHz). δ 139.4, 133.7–133.6, 129.4. ³¹P NMR (CDCl₃, 162 MHz): δ 26.9. HRMS (m/z): [M]⁺ calcd. for C₁₈H₁₂Cl₃OP, 379.9691; found, 379.9692.

Tris(4-bromophenyl)phosphine oxide (TBrPPO) was isolated as a white solid (130 mg, yield: 85%). mp: 178–180 °C (lit.ref 179–180 °C). ¹H NMR (CDCl₃, 400 MHz): δ 7.65–7.62 (m, 6H), 7.52–7.49 (m, 6H). ¹³C NMR (CDCl₃, 100 MHz). δ 133.3, 132.0, 130.6, 127.7. ³¹P NMR (CDCl₃, 162 MHz): δ 27.8. HRMS (m/z): [M]⁺ calcd. for C₁₈H₁₂Br₃OP, 513.8155; found, 513.8158.

First, the PEDOT: PSS solution was mixed with the PFI solutions at a mass ratio of 1:1 and then filtered for further use. Then, the mixture was spin-coated onto the substrates at 4000 × g for 45 s in the atmosphere, followed by annealing at 150 °C for 30 min. Second, PTAA (5 mg mL⁻¹ in chlorobenzene) was spin-coated onto the substrate at 4000 × g for 45 s, followed by annealing at 120 °C for 25 min in a nitrogen-filled glovebox. The emitting layer was fabricated by spin-coating QD solutions at 2000 × g for 45 s. Finally, TPBi (40 nm), LiF (100 nm) and Al (1 nm) were evaporated in sequence under a vacuum of ~2 × 10⁻⁴ Pa. The emitting area was 0.15 cm × 0.2 cm. For air-processed QLEDs, we spun PEDOT: PSS: PFI, PTAA and QD solution in the atmosphere at ~25 °C and a relative humidity of 20~30%. The remaining vacuum evaporation steps were the same.

The DFT calculations were carried out using the PWmat code^38,39. The optimised Vanderbilt norm-conserving (OVNC) pseudopotential NCPP-SG15-PBE and PBE functional were adopted under the generalised gradient approximation (GGA)^{40, 41–42}. The k-mesh and encut were set to 2 × 2 × 2 and 50 Ry for structural optimisation, self-consistent field and projected density of state (PDOS) calculations, respectively. The vacuum level was set to be larger than 15 Å to avoid the interaction between different layers. The Fermi level was set to 0 in all PDOS results.

For STEM analysis, the diluted QD solution was dropped onto a carbon-coated Cu grid. STEM was performed applying an aberration-corrected FEI (Themis Z) at 300 kV with a beam current of ~2 pA. PL spectra were collected applying a Varian Cary Eclipse spectrometer (F-7000 FL Spectrophotometer; Excitation light source: xenon lamp; Excitation light wavelength: 405 nm; Scan speed: 240 nm min⁻¹). PL decay measurements were conducted applying a time-correlated single-photon counting (TCSPC) spectrofluorometer (FLS920, Edinburg Instrument, UK). The decay curves were fitted by applying the bi-exponential equation: I=A1exp−tτ1+A2exp−tτ2. The average PL lifetime was calculated by: τavg=A1τ12+A2τ22A1τ1+A2τ2. XRD patterns were obtained by applying a Bruker D8 Advance X-ray diffractometer with Cu Kα radiation (λ = 1.5406 Å). AFM images were obtained and analysed by applying an Atomic Force Microscope (Cypher ES, Oxford Instruments Asylum Research Inc.). NMR spectra were measured by using a Brucker AVANCE Ⅲ spectrometer in deuterated chloroform with tetramethyl silane as the internal standard. All the reactions were carried out using Schlenk techniques under nitrogen conditions. For transient absorption spectroscopy, the pump-probe experiments were conducted by using a transient absorption/transmission spectrometer (Helios Fire, Ultrafast Systems). A Ti: Sapphire amplifier (Coherent Inc.) operating at a 1 kHz repetition rate generated infrared pulses (800 nm, 35 fs), which were divided into two beams. The first beam was directed into an optical parametric amplifier (OPA) to produce tunable pump pulses for resonant or near-resonant excitation of the sample. The second beam generated white-light continuum probe pulses through a sapphire plate focusing, with temporal delay controlled by a motorised delay line. Both beams were spatially overlapped in the sample, exhibiting spot diameters of 200 μm (pump) and 6 μm (probe). The transmitted probe signals were detected by a spectrometer equipped with a back-thinned CCD array. Differential transmission (ΔT/T) was calculated as (T_on−T_off)/T_off, where T_on and T_off correspond to pump-activated and baseline probe intensities. The EL spectra, L-J-V characteristics and EQE were all measured in a nitrogen-filled glovebox at room temperature through an integrated LED testing equipment. The equipment is composed of a Keithley 2400 source, a fibre integration sphere, and a PMA-12 spectrometer, which is designed by Ocean Optics Co., Ltd.