Active-matrix (AM) displays are self-emissive devices in which each pixel of a light-emitting diode (LED) is independently controlled using thin-film transistor (TFT) circuits. Compared with conventional liquid-crystal displays, AM electroluminescent devices have a simplified structure, lower power consumption, wider view angle and higher contrast1. In AM displays, the screen refresh rate requires LEDs with a fast response time; the time from the onset of the pulse voltage until steady electroluminescence (EL) of LEDs (typically 90% of the final value) should be less than a millisecond when the refresh rate increases to 120 Hz or higher frequency2,3,4,5.

Metal halide perovskite LEDs (PeLEDs) are attractive for AM displays owing to their high colour purity, efficiency, growing operational stability and potential for cost-effective fabrication of large-area panels6,7,8,9. The peak external quantum efficiencies (EQEs) of prototype infrared-red, green and sky-blue emissive PeLEDs are approaching the theoretical limit and the operational stability has reached 520 h at 1,000 cd m−2 for green-emitting devices10,11,12,13,14,15,16,17. However, their EL response under pulsed operation has been largely overlooked. The response time of a PeLED has been decreased by reducing interfacial capacitance and resistance18. Nevertheless, the typical transient EL of PeLEDs shows two regions19,20,21,22,23,24: an initial fast rise (in the range of microseconds) upon application of the pulse voltage, followed by a slow rise (in the range of tens of milliseconds) before reaching the steady EL. The slow rise in the range of tens of milliseconds is a consequence of intrinsic ionic crystal structures of metal halide perovskites, which show ionic transport response under applied electric fields in the millisecond range25,26. These mobile ions change the internal electric field and thus slow the subsequent EL rise19. As a result, the transient EL response in typical PeLEDs is much slower than in other LED technologies, which exhibit EL response timed within or below microseconds27,28,29.

Among different perovskite emission layers (such as bulk films and nanocrystals), surface-passivated nanocrystals are promising candidates to eliminate the ionic transport-induced milliseconds EL rise in PeLEDs. Ions prefer to migrate through halide vacancies, and the defective surface and grain boundaries in particular act as essential ion migration channels30. Previous studies have shown that the energy barrier of ionic transport in a nanocrystal film is much higher than that in the bulk film, which is caused by the surrounding insulating ligands between the nanocrystals rather than the grain boundaries20,31. Therefore, for a perfect surface-passivated nanocrystal film, the energy barrier could be increased further if the vacancy sites in every single nanocrystal were removed (Fig. 1a). While the concept is known theoretically, it remains a challenge to create perfect surface-passivated perovskite nanocrystal films. Ligand loss is inevitable in perovskite nanocrystals during film deposition because of the weak ligand bonding energy32, which generates surface halogen vacancies. Inserting passivation layers covering the nanocrystal film cannot modify the surface vacancies of every single nanocrystal inside the film24,33.

Fig. 1: Individual-particle-treated CsPbBr3 nanocrystal film.
figure 1

a, Schematics of different perovskite films showing distinct energy barriers of ion migration. b, A cross-sectional high-angle annular dark-field image of the TBABF4-treated nanocrystal film and colour-mixed EDX mapping image presents the element distribution of Cs, Pb, Br and F. Scale bar, 10 nm. c, High-resolution XPS spectra of F 1s at different etch depths for the TBABF4-treated nanocrystal film. The untreated film with the F 1s signal being absent is also presented. d, HRTEM image of TBABF4-treated nanocrystal film in a LED configuration shows high crystallization. Scale bar, 10 nm. a.u., arbitrary units; DBSA, dodecylbenzenesulphonic acid; DDA+, didodecyldimethylammonium ions; ETL, electron transporting layer; HTL, hole transporting layer; OTAC, n-octanoic acid.

Source data

In this Article, we report the development of microsecond-response and efficient-driven AM PeLED displays using an individual-particle-passivated perovskite nanocrystal film as the emissive layer. The nanocrystal film exploits its high crystallinity and sufficiently anchored surface ligands to create a defect-free film with a discrete nanostructure that inhibits ion migration. The response time of the resulting AM PeLEDs is decreased from milliseconds to microseconds and the corresponding EQE is above 20.0% at a display brightness of 500–3,000 cd m−2. We generalize this concept of individual-particle passivation to halide-mixed perovskite nanocrystal films, creating blue PeLEDs with emission centred at 475 nm with an efficiency of 18.1%.

Individual-particle-treated perovskite nanocrystal film

We use an in situ modification strategy to uniformly treat individual particles by adding an appropriate amount of tetrabutylammonium tetrafluoroborate (TBABF4) dissolved in ethyl acetate (EA) as a passivant during nanocrystal film deposition (Fig. 1a, see Methods for details). Colour-pure CsPbBr3 nanocrystals with a full-width at half-maximum of 18 nm and photoluminescence quantum yield (PL QY) of 90 ± 2% in the solution (Supplementary Fig. 1) are used as a model system for the in situ treatment. The BF4 ions are chosen due to their strong electrostatic attraction with Pb2+ ions and similar ionic radius (218 pm) to Br ions (196 pm), and are therefore tightly coordinated to the surface of perovskite nanocrystals34,35. It is notable that perovskite nanocrystal films can be treated with an antisolvent, which differs from in situ film crystallization with an antisolvent. In comparison with modification when the nanocrystals are in solution, this postdeposition treatment strategy overcomes film deposition-induced surface defects.

Cross-sectional high-resolution transmission electron microscopy (HRTEM) and elemental mapping using energy-dispersive X-ray (EDX) spectroscopy (Fig. 1b) show uniformly distributed F atoms throughout the nanocrystal film. In-depth X-ray photoelectron spectroscopy (XPS) analysis reveals nearly identical F content following the etch depth until the underneath of the nanocrystal film is reached (Fig. 1c and Extended Data Fig. 1a–d). From the evenly distributed F content observed in EDX and in-depth XPS analysis, we conclude that the in situ TBABF4 modification enables uniform surface treatment of BF4on every nanocrystal inside the film. The monodispersed CsPbBr3 nanocrystals in solution exhibit excellent crystallinity (Supplementary Fig. 2a). The treated nanocrystal film maintains a perfect crystal structure and presents a discrete nanostructure composed of isolated particles (Fig. 1b,d and Supplementary Fig. 2b). The voids between the nanocrystals in the HRTEM can be filled with ligands. Notably, the TBABF4 treatment does not change the size of nanocrystals or change the morphology and compactness of the film in the device configuration (Supplementary Figs. 3 and 4).

Suppressed ionic transport through individual-particle passivation

Optical measurements combined with structural characterizations reveal that the BF4 ions effectively passivate the surface vacancies of the nanocrystals. The Fourier transform infrared (FTIR) spectrum shows a strong B–F stretching vibration peak at 1,053 cm−1 after washing of the TBABF4-treated nanocrystal film with EA (Extended Data Fig. 2a). Moreover, the treated nanocrystal film becomes more hydrophilic and can withstand washing of the original solvent by octane (Supplementary Figs. 5 and 6), suggesting that the BF4anion binds to the nanocrystal surface in replace of the original didodecyldimethylammonium bromide (DDAB) ligands lost during film deposition. The XPS core-level spectra of F 1s shifts towards higher binding energy by 0.5 eV for the TBABF4treated nanocrystal film compared with the TBABF4 film (Extended Data Fig. 1e). Although the Pb 4f and Br 3d remain unchanged for the untreated and treated nanocrystal films, they shift towards lower binding energy when comparing the mixed PbBr2–TBABF4 film with the pure PbBr2 film (Extended Data Fig. 1f,g). The XPS results reflect an electron cloud shift from the BF4 to the incomplete PbBrx octahedrons on the nanocrystal surface induced by the relatively low electron electronegativity of BF4 ions36, indicating a strong interaction between BF4 and Pb2+ besides the electrostatic interaction. The ultraviolet–visible absorbance of the nanocrystal film remains identical after TBABF4 treatment, whereas the PL displays a noticeably enhanced intensity (Fig. 2a). The transient PL spectra reveal an average exciton lifetime of 33 ns for the BF4-treated film, which is much longer than the value of 22 ns for the untreated film (Fig. 2b). The confocal PL lifetime mapping presents a narrowly distributed exciton lifetime of 31.7 ± 1.6 ns in an area of 20 × 20 μm2 (Supplementary Fig. 7a,b), confirming a uniform BF4 passivation of the film. The BF4-treated film exhibits a high PL QY of 80% at an excitation density of 0.5 mW cm−2 and 50% at 100 mW cm−2 (Fig. 2c and Supplementary Fig. 7c). It is known that the vacancies are closely associated with the trap states and influence the optical properties of perovskites37. The improvement in the optical properties validates the effective filling of vacancies with BF4, resulting in uniform and defect-less individual-particle-passivated perovskite nanocrystal films. In addition, the BF4-treated film shows improved PL stability in ambient conditions (Supplementary Fig. 7d), in accordance with effective surface vacancy passivation38.

Fig. 2: Effects of the individual-particle treatment.
figure 2

a, Absorption and PL spectra of the TBABF4-treated and untreated CsPbBr3 film on quartz substrates. The inset shows the normalized PL spectra. b, Time-resolved PL decays of the TBABF4-treated and untreated CsPbBr3 films show average lifetimes of 22 ns and 33 ns, respectively. The inset shows photographs of the films under ultraviolet illumination. c, The PL QYs of the nanocrystals in the solution and the corresponding PL QYs of films. The central line represents the average value, and the distributions of PL QYs are fitted by the Gaussian fit. d, Transient currents of the lateral devices at room temperature. e, Temperature-dependent ion decay rate (k) multiplied by temperature in untreated and TBABF4-treated CsPbBr3 film. The lines represent fitted curves by the Nernst–Einstein relationship, which gives the activation energy (Ea) of ion migration. f, Capacitance–frequency characterizations of the prototype PeLEDs based on untreated and TBABF4-treated CsPbBr3 emissive layers. The bias is 0 V with an alternating amplitude of 100 mV. g, Normalized transient EL of prototype LEDs based on TBABF4-treated CsPbBr3 films, untreated CsPbBr3 films and bulk FAPbBr3 films. The pulse voltage is 4.0 V with a duration of 50 ms and the time at the onset of EL is unified to zero. Transient EL also shows that the EL decay time is within 1 μs when the pulses are turned off. h, The pulse duration-dependent transient EL (TrEL) intensity. This figure corresponds to data with a 3.0 V pulse voltage. The transient EL intensities at the end time of each pulse are collected (the raw data are shown in Extended Data Fig. 5). Six values of PL QY are used to derive statistics for each condition. All the error bars are presented as average values ± standard deviation (s.d.).

Source data

For comparison, post-treatment of the perovskite nanocrystals in solution with various passivants, including the BF4 anion, does not produce excellent optical properties when these are made into nanocrystal films (Fig. 2c and Extended Data Fig. 3a). Typically, different amounts of TBABF4 are dissolved in EA to post-treat the nanocrystals in the solution during the purification process. Although the PL QYs of the nanocrystals in solution with post-treatment are greater than 90%, the corresponding films have PL QYs that are below 50% (Fig. 2c), this low value being caused by ligand loss-induced surface vacancies created during film deposition. High PL QYs of the films can only be achieved by in situ TBABF4 treatment during film deposition to fill halide vacancies with BF4 ions (Extended Data Fig. 3b). The above results demonstrate the superiority of in situ treatment during film deposition over post-treatment in solution.

Next, we investigate the ionic transport behaviour of uniform individual-particle-passivated perovskite nanocrystal films with few surface vacancies. Temperature-dependent transient current-response measurements were conducted to analyse the ionic current and obtain the activation energy for ionic transport. Lateral devices with interdigital gold electrodes deposited on glass/CsPbBr3 nanocrystal films (Extended Data Fig. 4) are used for characterizations. The BF4-treated nanocrystal film shows suppressed current decay compared with the untreated nanocrystal film at room temperature, indicating that fewer ions contribute to the current (Fig. 2d). The activation energy can be derived by fitting the temperature-dependent ionic current decay rate multiplied by the temperature using the Arrhenius equation39,40 (see Methods for details). The extracted activation energy for the untreated and BF4-treated nanocrystal films are 190 meV and 478 meV (Fig. 2e), respectively. The increased activation energy is consistent with the trend of theoretical calculation results (Fig. 1a and Extended Data Fig. 2b,c). The ionic transport behaviour was further investigated using capacitance measurements at low frequencies, which are sensitive to the motion of charged species41,42. In the capacitance–frequency characterization of PeLEDs, the ion migration can be indicated by the increase of capacitance following a decrease in frequency43. The prototype PeLEDs based on the individual-particle-passivated nanocrystal films show a much lower capacitance than the devices incorporating untreated films at the measured frequencies from 20 Hz to 106 Hz (Fig. 2f). This observation indicates that ion migration is effectively suppressed in the passivated nanocrystal films. Thus, we have verified that individual-particle-passivated nanocrystal films can effectively inhibit ion migration.

Suppressed ionic transport greatly decreases the EL rise time of PeLEDs from milliseconds to microseconds. The EL rise time of the prototype PeLEDs is characterized using a custom-built transient EL setup (Supplementary Fig. 8a). For PeLEDs, the fast electron/hole transport under an electrical field enables the EL to turn on within 2 μs of the pulse voltage (Supplementary Fig. 8b,c) whereas the subsequent EL rise usually continues for at least milliseconds (Fig. 2g)19,25. Therefore, the response time of PeLEDs is mainly determined by the time from the onset of EL to steady EL, that is, the EL rise time. When utilizing the individual-particle-passivated CsPbBr3 nanocrystal films as the emissive layer, the prototype LED exhibits a fast EL rise within 10 μs (Fig. 2g). In contrast, the EL rise time of the LEDs based on the untreated CsPbBr3 nanocrystal film or bulk FAPbBr3 film are longer than 20 ms and 50 ms, respectively. We note that the post-treatment of the nanocrystals in solution cannot tackle the slow EL rise (Supplementary Fig. 9). In addition, the transient EL of treated PeLEDs under a sustained pulse with 100 Hz in 1 h shows an identically fast response time, indicating inhibited ion migration under continuous electric field (Supplementary Fig. 10).

More importantly, the continuous EL rise within tens of milliseconds would severely effect the brightness of high-refresh-rate AM displays, and the variable EL intensity would create difficulty in controlling the brightness44,45. As shown in Fig. 2h, the transient EL intensities of the LEDs based on individual-particle-passivated CsPbBr3 nanocrystal films remain nearly identical when the pulse duration is varied, in contrast with the PeLEDs based on the untreated nanocrystal films or bulk films. When a bulk-film-based PeLED operates at a display refresh frequency of 120 Hz with a duty cycle of 50%, the final EL intensity of each pulse reaches less than 50% of the stable brightness driven at constant operation (Extended Data Fig. 5). The above results highlight the critical importance of the microseconds EL rise to achieve high-refresh-rate AM displays.

Microsecond-response AM PeLEDs

The microsecond-response transient EL of the prototype PeLEDs based on individual-particle-passivated CsPbBr3 nanocrystal films encouraged us to integrate the LEDs with drive TFTs. A TFT backplane containing 16 × 16 pixel arrays in 1.36 × 1.36 cm2, which is 30 pixels per inch (PPI), is designed for fabricating a demonstration-of-concept AM PeLEDs display. Each pixel consists of a LED with an emissive area of 380 × 650 μm2 and three TFTs, which are the commonly used three-transistor-one-capacitor circuit46,47 (Extended Data Fig. 6). The cross-sectional structures of the drive TFT and PeLED are shown in Fig. 3a,b, respectively. Due to the thick passivation layer, our AM PeLEDs are fabricated by the sequential deposition of multiple layers on the TFT backplane without pixelation. The AM PeLED can be controlled independently by the TFT (Fig. 3c and Supplementary Video 1), exhibiting excellent efficiency in terms of display brightness. All 256 pixels exhibit uniform brightness and efficiency with a deviation of 5% (Supplementary Fig. 10a,b). The microphotograph displays the uniform EL of each pixel (inset of Fig. 3d). The typical current density–luminance–voltage characteristics of a LED in a single pixel are shown in Fig. 3d. Owing to the two to three nanocrystal monolayer of the emissive film and the high charge mobility of the transport layers (3.8 × 10−3 cm2 V−1 s−1 for poly[bis(4-phenyl)(2,4,6-trimethylphenyl) amine] (PTAA) and 4.4 × 10−3 cm2 V−1 s−1 for 2,4,6-tris[3-(diphenylphosphinyl)phenyl]-1,3,5-triazine (PO-T2T))48,49, the current density and luminance increase quickly once the LED turns on at a Vdd voltage (working voltage of devices integrated on the TFT backplanes) of 2.3 V, yielding a brightness of 1,000 cd m−2 at a voltage of 3.9 V. The peak EQE reaches 23.3% at a current density of 1.5 mA cm−2, corresponding to a brightness of 1,206 cd m−2, and the EQE remains above 20.0% at a brightness ranging from 500 to 3,000 cd m−2 (Fig. 3e). The maximum current efficiency reaches 81.2 cd A−1 at a brightness of ~1,200 cd m−2 (Fig. 3e). A histogram of EQEs at 4.0 V for 20 PeLEDs ranging from 22.5% to 23.5% (Supplementary Fig. 10c) suggests good reproducibility of the processing. The efficiency of the AM PeLED operated at display brightness is comparable to that of solution-processed AM organic or quantum-dot LEDs and represents a state-of-the-art electroluminescent device based on perovskite (Supplementary Table 1). In addition, the AM PeLED exhibits a half-lifetime (T50, defined as the time for the brightness to decrease to 50% of the initial value) of ~10 h at an initial brightness of 100 cd m−2 (Supplementary Fig. 10d). Notably, the overall performance of the AM PeLED shows a maximum EQE of 26.2% and an average value of 23.6% with a relative standard deviation of 7% (Extended Data Fig. 7), which is comparable to that of the prototype LEDs with identical structures and processing.

Fig. 3: Performance of AM PeLED.
figure 3

a, Cross-sectional scanning electron microscope image of the TFT backplane. PV1, passivation layer 1; PV2, passivation layer 2; GI, gate insulator. Scale bar, 1 µm. b, Cross-sectional HRTEM image of an AM PeLED, showing the structure of indium tin oxide (ITO), PEDOT:PSS (25 nm), PTAA (35 nm), CsPbBr3 nanocrystal (NC) film (25 nm), TPBi (5 nm), PO-T2T (25 nm), LiF (1 nm) and aluminium (Al, 100 nm). Scale bar, 50 nm. c, Digital photographs of an AM PeLED operated at the switch state. d,e, Typical current density–luminance–voltage curves (d) and corresponding EQE and current efficiency (CE) (e) of a LED in a single pixel. The inset is a microphotograph showing uniform pixels. f, Transient EL intensity as a function of the pulse voltage. The pulse duration is 5 ms. g, Pulse-duration-dependent transient EL intensity. The pulse voltage is 4.0 V. The time at the onset of EL is unified to zero in f and g.

Source data

Notably, the response time of AM PeLEDs satisfies the requirement for displays with high refresh rates of up to 120 Hz. When the AM PeLED drives under the Vdd voltage from 3.0 V to 8.0 V, the transient EL shows an identical steep EL rise of tens of microseconds and remains steady on the time scale of milliseconds (Fig. 3f). When the pulse duration is varied from 50 μs to 50 ms, all transient ELs reach a steady EL intensity within tens of microseconds (Fig. 3g). The results indicate that the response time of the AM PeLED is independent of the pulse duration and voltage. Considering that the response of ions is strongly dependent on the pulse duration and electric field, the short rise time to steady EL at all pulse frequencies and voltages verifies the mostly suppressed ion migration. In contrast, for the AM PeLED based on the untreated CsPbBr3 nanocrystal film or bulk FAPbBr3 film, the transient ELs show quick responses at high frequencies, while exhibiting a continuous rise at pulse widths of milliseconds (Extended Data Fig. 8).

Finally, we demonstrate microsecond-response red, green and blue emissive AM PeLEDs with a resolution of 64 × 64 in 1.8 × 1.8 cm2, that is, 90 PPI, utilizing the individual-particle-passivated nanocrystal films as the emissive layers. CsPbClxBr3−x and CsPbBrxI3−x nanocrystals with emission peaks at 475 nm and 635 nm are used as the blue and red emitters, respectively, which both have high optical performance and crystal quality in solution (Supplementary Fig. 12). The prototype PeLEDs show substantially improved efficiency when integrating the individual-particle-passivated nanocrystal films as the emissive layer, reaching peak EQEs of 18.1% and 22.7% for the blue and red devices (Extended Data Fig. 9), respectively. Note that this blue LED represents the most efficient device with an emission below 480 nm. The 1-inch AM PeLEDs present clear cartoon pictures with uniform brightness (Fig. 4a–c and Supplementary Video 2), with a resolution comparable to that of a commercial liquid-crystal display50. The peaks of the EL spectra are nearly identical when the driving voltage Vdd is increased for all colour AM PeLEDs (Fig. 4d–f). For the blue and red AM PeLEDs, their peak EQEs reach 17.9% and 21.1%, respectively, with little efficiency loss compared with their prototype LEDs (Supplementary Fig. 13). The AM PeLEDs exhibit a fast response time within tens of microseconds and maintain a steady EL when the pulse duration is varied from 50 μs to 50 ms (Fig. 4g–i). Under a high Vdd voltage of 7.0 V, the AM PeLEDs have identical transient EL behaviour, for which the performance is noticeably opposed to that of the AM PeLEDs based on the untreated nanocrystal films (Extended Data Fig. 10). The above results indicate that individual-particle passivation can be extended to halide-mixed perovskite nanocrystal films to suppress ion migration.

Fig. 4: One-inch AM PeLEDs with a resolution of 90 PPI (pixel size 270 μm × 270 μm).
figure 4

ac, Digital photographs of the red (a), green (b) and blue (c) emissive AM PeLEDs showing cartoon pictures. df, EL spectra of the red (d), green (e) and blue (f) emissive AM PeLEDs at various Vdd driving voltages, displaying a single peak with narrow full-width at half-maximum (FWHM). gi, Transient EL intensities of red (g), green (h) and blue (i) emissive AM PeLEDs under various pulse durations. The pulse Vdd voltage is 4.0 V unless specifically labelled. The times at the onset of EL are unified to zero. The increased response speed at high frequency compared with the data in the 16 × 16 display originates from the reduced RC time constant.

Source data


We have reported microsecond-response and efficient-driven AM perovskite displays based on individual-particle-passivated perovskite nanocrystal films. We developed a passivant that is introduced during perovskite nanocrystal film deposition to compensate for the deposition-induced ligand loss. The individual-particle-passivated nanocrystal film has excellent crystallization and a uniformly passivated surface, leading to a discrete nanostructure with few defects and thus mostly suppressed ionic transport. The treated nanocrystal films can be used to fabricate AM PeLEDs with EL rise times reduced to microseconds, which is comparable to AM LEDs based on organic dyes and quantum dots as emitters. The individual-particle passivation strategy could also be combined with other types of solution processing, such as inject printing and transfer printing, to produce fast-response, large-area and full-colour AM PeLEDs for high-refresh-rate displays.



Caesium carbonate (Cs2CO3, 99.9%), lead bromide (PbBr2, 99.999%), lead iodide (PbI2, 99.999%), 5-aminovaleric acid (98%), triphenylphosphine oxide (TPPO, 98%), chlorobenzene (99.5%) and dodecylbenzenesulphonic acid (≥95%) were purchased from Sigma Aldrich. 1-Octadecene (ODE, 90%), oleic acid (OA, 90%), N,N-dimethylformamide (DMF, 99.8%) and octane (97%) were purchased from Alfa Aesar. EA (99%), tetraoctylammonium bromide (TOAB, 98%), DDAB (98%), didodecyldimethylammonium chloride (98%), TBABF4 (98%), tetrabutylammonium hexafluorophosphate (98%), oleylamine (80–90%) and formamidine acetate (99%) were purchased from Shanghai Macklin Biochemical Co. Ltd. n-Octanoic acid (99%) was purchased from Aladdin Industrial Corporation. Formamidine hydrobromide, 2,2′,2′-(1,3,5-benzinetriyl)-tris(1-phenyl-1-H-benzimidazole) (TPBi), PO-T2T, PTAA and poly(ethylenedioxythiophene):poly(styrenesulfonate) (PEDOT:PSS) were purchased from Xi’an Polymer Light Technology Corp. Poly(9,9-dioctylfluorene-co-N-(4-butylphenyl)diphenylamine) was purchased from LinkZill Technology Co. Ltd. All chemicals were used directly without further purification.

Synthesis of nanocrystals

For CsPbBr3 nanocrystals, 0.1 g of dodecylbenzenesulphonic acid (diluted in 100 μl toluene) was first added into 2.5 ml of a Pb2+ precursor (0.1 mol l−1 PbBr2 and 0.2 mol l−1 TOAB in toluene). After vigorous stirring for 1 min, 250 μl of Cs and FA precursor (0.1 mol l−1 Cs2CO3 and formamidine acetate in n-octanoic acid,the volume ratio of Cs and FA precursor VCs:VFA = 0.85:0.15) was added. Then 750 μl of DDAB solution (0.025 mol l−1 in toluene) was added after a reaction time of 2 min. The reaction was conducted at room temperature (25 °C) in a nitrogen-filled glove box. After 2 min, the crude solution was transferred to air and centrifuged at 7,104g for 2 min. EA was then added to the supernatant at a ratio of 2:1. The precipitate was collected after centrifugation and dispersed in octane. The purification process was repeated two to three times to obtain a clean nanocrystal solution for device fabrication.

The post-treatment of nanocrystals in solution with various passivants was conducted by using different concentrations of TBABF4 (0.25, 0.5, 0.75 and 1 mg ml−1 in EA) or tetrabutylammonium hexafluorophosphate (0.5 mg ml−1) to purify the nanocrystals. For DDAB or TPPO post-treatment, different amounts of DDAB (10 mg ml−1 in toluene) or TPPO (10 mg ml−1 in toluene) were added to the purified nanocrystal solution.

CsPbClxBr3−x nanocrystals were synthesized following the same procedure as for the CsPbBr3 nanocrystals. The 2.5 ml Pb2+ precursor was composed of 0.2 mmol PbBr2, 0.05 mmol PbCl2, 0.37 mmol TOAB and 0.16 mmol didodecyldimethylammonium chloride.

CsPbBrxI3−x nanocrystals were synthesized using a typical hot injection method51. Typically, Cs2CO3 (0.307 mmol), OA (0.4 ml) and ODE (3.5 ml) were loaded into a 25 ml three-necked flask. The mixture was degassed for 30 min at 80 °C, and then heated to 120 °C for another 30 min under continuous nitrogen flow. PbI2 (0.190 mmol), ZnI2 (1 mmol), ZnBr2 (0.6 mmol), OA (2 ml), oleylamine (2.4 ml) and ODE (5 ml) were added to a 25 ml three-necked flask. The mixture was degassed at room temperature for 20 min, and then heated to 170 °C for another 30 min under continuous nitrogen flow. Cs precursor (0.4 ml) was swiftly injected into the flask. After 5 s, the reaction mixture was cooled down using ice water. For the purification, toluene (10 ml) was added to the crude solution and centrifuged at 1,000g for 3 min. The precipitate was discarded and methyl acetate (16 ml) was added to the upper solution. After centrifugation at 8,991g for 5 min, the precipitate was dispersed in the mixture of toluene and methyl acetate (volume ratio of toluene:methyl acetate = 1:3) for further purification. Finally, the supernatant was discarded and the purified CsPbBrxI3−x nanocrystals were dispersed in octane. The upper solution was filtered with a 0.22 µm PTFE for further use after being centrifuged for 3 min at 1,000g.

The FAPbBr3 precursor was prepared by dissolving formamidine hydrobromide (0.24 mmol), PbBr2 (0.1 mmol) and 5-aminovaleric acid (0.075 mmol) in DMF (1 ml) followed by stirring for 8 h.

In situ nanocrystal film treatment and fabrication of AM PeLEDs

TFT substrates were sonicated in ethanol for 10 min and subsequently treated with oxygen plasma for 10 min. Afterwards, PEDOT:PSS solution was spin coated at 6,000 rpm for 50 s and baked at 150 °C for 15 min. The PEDOT:PSS-coated substrates were transferred to a nitrogen-filled glove box for subsequent film deposition. PTAA (in chlorobenzene, 8 mg ml−1) was spin coated layer-by-layer at 2,000 rpm for 50 s. The PTAA layer was baked at 120 °C for 20 min before the deposition of the phenethylammonium bromide. The perovskite nanocrystals (in octane, ~ 25 mg ml−1) were deposited by spin coating at 2,000 rpm for 60 s, during which the TBABF4 (in EA, typically 1 mg ml−1) was rapidly added into the wet film at 25 s. The critical addition time of the EA solution at 25 s was optimized to maintain the smooth morphology of the nanocrystal film. Then the samples were transferred to a high-vacuum evaporation chamber with ~ 2×10−4 Pa for sequent deposition of the TPBi (5 nm), PO-T2T (25 nm), lithium fluoride (LiF, 1 nm) and Al (100 nm) without a shadow mask. Prototype devices were fabricated using the aforementioned process.

The devices based on bulk FAPbBr3 film were fabricated by spin coating the ZnMgO precursor, perovskite precursor and poly(9,9-dioctylfluorene-co-N-(4-butylphenyl)diphenylamine) (in chlorobenzene, 12 mg ml−1) at 4,000 rpm for 50 s, 4.000 rpm for 40 s and 2,000 rpm for 50 s, respectively. The ZnMgO films were spin coated and annealed at 160 °C for 30 min in an air atmosphere. The perovskite films were annealed at 70 °C for 4 min in the glove box. Finally, MoOx (8 nm) and Al (100 nm) were deposited by high-vacuum thermal evaporation.

Optical and film characterizations

The ultraviolet–visible absorption spectra of the nanocrystals and films were measured using an Agilent Cary 5000 spectrophotometer. The PL spectra were obtained using an Agilent Cary Eclipse instrument. Time-corrected single-photon counting measurements were performed using an Edinburgh Instruments FLS-920 with a 405 nm pulsed laser diode (EPL405, 58.6 ps pulse width). The absolute PL QYs were measured by applying a three-step method52,53 using a 405 nm LED as an excitation light source and a home-designed integrating sphere coupled with a QE Pro spectrometer. For power-density-dependent PL QY measurements54, the PL intensity was detected using a photodetector (PDA100A, Thorlabs) connected to a lock-in amplifier (SR830, Standard Research System). The excitation light source was a 405 nm continuous laser, which was modulated by a lock-in amplifier. FTIR spectra were obtained using a Thermo Fisher IS-50 spectrometer. The PL intensity and lifetime mapping were performed using a home-built epifluorescence microscope. The PL of one pixel (1 × 1 μm2), excited by a 405 nm pulsed laser (LDH, PicoQuant), was collected by a 60× oil immersion objective (numerical aperture = 1.49) and recorded with an EMCCD (IXon, Andor Technology) and single-photon counting (Time Tagger Ultra). The spot size was 1 × 1 μm2 with a power density of 45 nW cm−2 and a scanning step of 1 μm.

The in-depth XPS spectra were obtained on a Thermo Fisher ESCALAB Xi+ with a monochromatic Al (Kα) X-ray source providing photons with 1,486.6 eV and argon ion gun with an etching voltage of 3,000 eV. Atomic force microscopy measurements were conducted using an Oxford Cipher S instrument. TEM observations of the nanocrystals were conducted using a Hitachi HT-7700 microscope operated at 100 kV. HRTEM and EDX analyses of the cross-sections of the AM PeLEDs were carried out using a Thermo Scientific TalosF200XG2 device. Cross-sectional samples were prepared using focused ion beam equipment (Helios G5 UX, Thermo Fisher Scientific). Cross-sectional scanning electron microscope images were obtained using Thermo Scientific Helios G5 UX and Hitachi SU-70 devices.

Device characterizations

Current density–luminance–voltage characteristics were measured using a Keithley 2400 electrometer and an integration sphere coupled with a QE Pro spectrometer55. For the AM PeLEDs, we randomly chose a single pixel to conduct the measurements. The on–off state of each pixel can be controlled using a custom-built operating system that modulates the scan and data signals to control the address TFT (TFT1). The on state of TFT1 can activate the drive TFT (TFT2), inputting a power signal (Vdd) to drive the LED. A Keithley 2400 source meter was used to provide the Vdd voltage. The half-lifetime T50 was measured using a commercial aging system (Guangzhou New Vision Optoelectronic Technology Co. Ltd). The AM PeLEDs were operated in the all-on state during aging.

The transient EL was obtained using a home-built setup consisting of an oscilloscope (TDS 2024 C, Tektronix), a pulse generator (SDG 2122X, Siglent) and a Si avalanche photodetector (APD 130A2/M, Thorlabs). For the transient EL measurements, the AM PeLEDs were operated in an all-on state, and the on states of TFT1 and TFT2 were maintained to exclude interference. A pulse generator was used to provide the Vdd voltage. The resistor–capacity (RC) time constant of our 16 × 16 TFT backplane reaches 13.5 µs, which is the same order of magnitude as the response time of optimized PeLEDs. Therefore, the response of the TFT backplane will not limit the EL response of the AM PeLED.

Capacitance–frequency characterizations were performed using a TH2838 Precision LCR Meter. The temperature-dependent temporal responses were measured using a Keithley 2635B electrometer. The temperature of the device was controlled using a semiconductor refrigeration system. A temperature sensor was placed on the surface of the device.

The activation energy of ion migration measurements

The ionic conduction in perovskite films is mainly caused by the vacancy mechanism, which neglects the formation of vacancy complexes or superlattices39,40. The kinetics of ionic movement can be reflected by the decay in the temperature-dependent temporal response. The temperature-dependent ionic conductivity (\(\sigma T\)) was addressed as40:

$$\sigma{{T}}=\left(\frac{{{{Z}}}_{{\rm{i}}}{{{e}}}^{2}{{C}}_{{\rm{v}}}^{\;0}{{{D}}}_{{\rm{v}}}^{0}}{{{{k}}}_{\rm{B}}}\right)\exp \left(-\frac{\Delta {{{H}}}_{{\rm{s}}}}{5{{{k}}}_{{\rm{B}}}{{T}}}-\frac{{{{E}}}_{{\rm{a}}}\,}{{{{k}}}_{{\rm{B}}}{{T}}}\right)={{{\sigma }}}_{0}\exp \left(-\frac{{{{E}}}_{{\rm{a}}}^{\,{\rm{eff}}}}{{{{k}}}_{{\rm{B}}}{{T}}}\right)$$

where Zi is the number of the ionic charge, kB is Boltzmann constant, Cv is the concentration of vacancies, Dv is the diffusion coefficient of vacancies, ΔHs is the formation energy of ionic vacancies, Ea is the diffusion barrier for ion migration and \({{{E}}}_{{\rm{a}}}^{{\rm{eff}}}\) is the effective activation energy for ion transport.

The decay rate (k = τ−1) of the current decay represents the ionic transport dynamics and is proportional to the ionic conductivity. Therefore, the decay rate can be used to obtain the thermal activation energy for ion transport. The time constant (τ) could be obtained by the current decay with an exponential function I(t) = I0 + A1exp[−(t − t0)/τe] + A2exp[−(t − t0)/τ], where τe relates to equipment response (RC delay, which is supported by the temperature independent τe) and τ represents the time constant of ion migration. The measured results are plotted by using the Arrhenius equation \({\mathrm{ln}}(kT\;) = C^-{{{E}}}_{{\rm{a}}}^{{\rm{eff}}}/k_{\mathrm{B}} T\). Here, \({{{E}}}_{{\rm{a}}}^{{\rm{eff}}}\) can be derived from the slope of the ln(kT) versus 1/T plot.

Calculation of energy barrier for ionic transport

First-principles calculations were performed using density functional theory with the generalized gradient approximation of Perdew–Burke–Ernzerhof implemented in the Vienna Ab-Initio Simulation Package56. The valence electronic states were expanded based on plane waves, with the core–valence interaction represented using the projector augmented plane wave57 approach and a cutoff of 400 eV. A periodic slab with a (3 × 3) surface unit cell was considered to model the CsPbBr3 (001) surface, which contains seven atomic layers and a vacuum space of 30 Å. Owing to the large supercell dimensions, the k-point sampling was restricted to the Γ point. All the structures were fully relaxed until the maximum force on each atom was less than 0.05 eV Å−1. The climbing image nudged-elastic band algorithm was employed to identify transition state structures for Br migrations58.