We next measured the operational lifetime of F- and FCT-PeLEDs with a high constant current density of 100 mA cm⁻². State-of-the-art NIR PeLEDs with the best stability, which showed a half operational lifetime (T₅₀) of 11,539 h at 5 mA cm⁻², demonstrated 120.3 h at such a harsh working condition of 100 mA cm⁻² for accelerating degradation³⁷. As shown in Fig. 1g, while the F-devices exhibit a short T₅₀ of 11 h, our FCT-counterparts deliver a drastically enhanced lifetime reaching 142 h. The excellent stability of our optimized devices is also indicated by the evolution of peak EQE upon voltage scanning cycles (0 ~ 6.0 V). As shown in Supplementary Fig. 3, the peak EQE of F-devices drop significantly after several cycles. In contrast, the peak EQE of FCT-devices can preserve half of the highest value until 52 cycles under the same scanning voltage span. The comparison of the stability of our PeLEDs with those of state-of-the-art is also summarized in Supplementary Table 1, where the much-improved lifetime at such a high electrical excitation represents a notable improvement compared with those of state-of-the-art PeLEDs.

Recombination kinetics in PeLEDs

To clarify the differences in device performance, particularly at high current densities, we performed a series of optical measurements to investigate the recombination kinetics of perovskites. We conducted transient absorption (TA) and fluence-dependent time-resolved-photoluminescence (TRPL) measurements. The characteristics and relevant 2D pseudo-colour images of TA measurements are shown in Supplementary Fig. 4 and Fig. 2a, b respectively. We observe two primary photobleaching bands situated at 450 nm and 780 nm in both cases, which can be ascribed to the typical sub-band of iodide-perovskites and radiative recombination respectively³⁸. Notably, no distinct charge carrier transfers in the different time scales and additional photobleaching features can be found, clearly suggestive of the absence of low-dimensional phases. These findings are in line with the X-ray diffraction (XRD) patterns and grazing-incidence wide-angle X-ray scattering (GIXWAS) measurements, where only the signatures of 3D perovskites were visible (Supplementary Fig. 5).

We then investigated the scaling of the initial PL intensity (at t = 0, I_PL0) with excitation density from TRPL measurements (Supplementary Fig. 6) to uncover the nature of the carrier recombination²⁴. As shown in Fig. 2c, the I_PL0 for both samples exhibit a quadratic dependence on carrier density, suggesting that the photogenerated carriers predominantly undergo bimolecular recombination, typical of free carrier processes. Additionally, we observe that the PL intensity gradually deviated from the quadratic relationship with the increase of the carrier concentration, indicating a growing prominence of the Auger process³⁹. A similar phenomenon is also visible in the evolution of relative PL quantum yields (PLQYs) as a function of carrier density for both cases (Fig. 2d). Notably, the relative PLQYs of FCT-films are much more tolerant to high carrier density compared to F-samples, suggesting a retarded Auger recombination^18,24. These observations are consistent with the device characteristics, where the peak EQE appeared at high current densities reaching hundreds of mA cm⁻². In addition, we note that PLQYs of FCT-films are also higher at low carrier densities (<~10¹⁵ cm⁻³), owing much to a mitigated trap-assisted non-radiative recombination. Such a difference eventually leads to a higher absolute PLQY of 81% for FCT-films compared to 62% of F-films at the best excitation conditions (Supplementary Fig. 7), in line with the higher peak EQE of FCT-devices than F-ones.

Having confirmed that bimolecular recombination is the dominant recombination pathway and the absence of low-dimensional phases, we are able to quantify the recombination kinetics with a combined use of TA and fluence-dependent TRPL by the following equation⁴⁰:

where k₁ is assigned to the trap-assisted nonradiative recombination constant, k₂ is the bimolecular recombination (band-to-band recombination) rate constant, k₃ is the trimolecular (Auger) recombination rate constant. Here, k₁ was experimentally determined by TRPL spectra at low fluences, while k₂ and k₃ are deduced from both TRPL and TA tests (Supplementary Figs. 4, 6, 8, and 9). We show the extracted ratios of the various recombination pathways upon carrier densities in Fig. 2e, f and summarized the average recombination constants in Fig. 2g and Supplementary Table 2. The TRPL-derived rates are in good agreement with that extracted from TA measurements. Compared to F-films, we observe largely retarded mono-molecular and Aguer recombination constants but close bimolecular recombination rates in FCT-samples. These contrasts thus lead to a much more dominant role of radiative recombination in FCT-films across a wide range of excitation fluence (Fig. 2e, f). The average k₃ of FCT-films extracted by TRPL and TA is (1.15 ± 0.64) × 10⁻²⁸ cm⁶s⁻¹ and (5.28 ± 0.43) × 10⁻²⁹ cm⁶s⁻¹, respectively. Notably, the best sample showed an ultra-slow Auger rate of 1.29 × 10⁻²⁹ cm⁶s⁻¹ (obtained by TA). These results are among the best so far for lead-halide perovskites and are only one order of magnitude faster than that of III-V emitters (e.g., GaInN, ~2 × 10^-30 cm⁶s⁻¹)^39,41,42. The notably suppressed Auger recombination in FCT-films is critical for the mitigated current-efficiency roll-off in the devices. More detailed discussions are provided in Supplementary Note 1.

Carrier injection kinetics in PeLEDs

Despite the slow-downed Auger recombination in FCT-films with optical excitation, we understand that Augur losses could happen at a low carrier density with electrical excitation, especially in the case of unbalanced charge injections. Given that both devices of interest employed identical charge transport layers, we focused our investigations on the perovskites. Ultraviolet photoelectron spectroscopy (UPS) was used to examine changes in the energy levels of the F- and FCT-films. The results reveal that the introduction of CsTFA causes a shift in the secondary electron cutoff from 17.25 eV to 17.43 eV (Supplementary Fig. 10a). Further calculations indicate that the valence band maximum (VBM) of the perovskite shifts from −5.46 eV to −5.26 eV. The resulting flat-band energy level diagrams are illustrated in Supplementary Fig. 10b, which display that the VBM of the perovskite becomes more closely aligned with the highest occupied molecular orbital (HOMO) level of TFB. Given that the electron mobility of the ZnO electron transport layer is typically much higher than that of TFB, the improved hole injection facilitates more balanced charge transport.

Another critical observation we noted is the significant difference in morphology between the F- and FCT-films. As illustrated in the SEM and atomic force microscopy (AFM) topographical images (Fig. 3a, b), the dense FCT-films exhibit a small root-mean-square (rms) roughness of 1.67 nm, in stark contrast to the F-films, which display prominent pinholes and an enhanced rms roughness of 5.42 nm. Achieving such a uniform thin film in the FCT-cases is very challenging for 3D perovskite thin-film emitters, considering the ultra-thin-film thickness typically required to enhance charge carrier confinement for efficient recombination.

We realize that the notable changes on film morphology may affect the electrical properties of devices⁴³. We thus conducted simulations of the electric potential distribution across simplified device architectures—one featuring densely packed emissive films and the other with discrete perovskite grains (Fig. 3c,d)⁴⁴. The simulations reveal that in both configurations, the electric potential gradient is concentrated near the hole transport layer (TFB)/anode interface, diminishing once moving away from the interfaces (region I and II in Fig. 3c and region III in Fig. 3d). However, it is worthy to mention that excessive holes accumulate at the HTL–ETL interface once increasing the bias, leading to additional charge transport pathways and hence potential electrical shunts (Fig. 3e). Figure 3f displays a further analysis regarding the quantification of charge carrier losses dependent on the surface coverage and applied bias. We find that the electrical shunt can be negligible at low bias even for those with low surface coverage. However, the loss of charge carriers significantly increases once the bias exceeds 4.0 V, suggesting an increasing current-efficiency roll-off over enhancement in bias.

To link the simulation and experimental results, we fabricated devices with a structure of ITO/PEIE-modified ZnO/TFB/MoO₃/Au and compared their J–V characteristics with those with perovskite emissive layers (Supplementary Fig. 11). Consistent with our simulation results, we note that the current densities of the devices without perovskites are more than one order of magnitude lower at a small bias (<2.0 V), suggesting that the charge injection barrier between TFB and ZnO is large enough to prevent the shunts at the TFB/ZnO interfaces for PeLED devices. In addition, the current density of the ZnO/TFB-only device quickly rises with increasing the bias and eventually reaches the same order of magnitude as the perovskite device at high voltages. These results align with our simulation results that the electrical shunts can be very severe at the high bias necessary to achieve high radiance. These findings demonstrate that dense film morphology enables minimized detrimental effects such as charge accumulation and electrical shunts, thereby mitigating the issues of current-efficiency roll-off^45,46.

To confirm and generalize our findings, we investigated two additional types of devices with controlled surface coverages of perovskites. One approach was to modify the morphology by varying the quantity of excess FAI into FAPbI₃, with the molar ratios of FAI to PbI₂ ranging from 2.0:1 to 2.8:1. The other case is to introduce DMF vapour atmosphere during thermal annealing to control the grain growth, using a fixed FAI to PbI₂ ratio of 2.0:1⁴⁷. In brief, we find that the films with high surface coverage commonly give rise to enhanced performance at high current densities, thereby supporting above analysis that the dense film morphology is crucial. The corresponding device characteristics and film morphologies are detailed in Supplementary Figs. 12 and 13, and a more thorough description is available in Supplementary Note 2.

Roles of TFA⁻ addition

Having concluded that the decent performance of FCT-devices at high current densities mainly stems from the slowed-down Auger recombination rates of perovskites as well as balanced charge injection, the next question that arises is why TFA⁻ addition can achieve these effects. In addition, the observed mitigated trap-assisted recombination caused by TFA⁻ is possible to be correlated with the prolonged operational lifetime. To understand the underlying mechanisms, probing the TFA⁻ interaction with perovskites is crucial. We conducted X-ray photoelectron spectroscopy (XPS) and time-of-flight secondary ion mass spectrometry (ToF-SIMS) measurements, aiming to verify the presence of TFA⁻ within the perovskite films and investigate their distribution perpendicular to the substrate. Different from the F-films, the core-level spectra of C 1s and F 1s, as depicted in XPS images (Fig. 4a), clearly reveal the distinct features corresponding to −O−C=O and −CF₃ functional groups, confirming the presence of TFA⁻ on the surface of the perovskite films. In addition, ToF-SIMS measurements show that the TFA⁻ ions are evenly distributed throughout the perovskite films (Supplementary Fig. 14). These results not only confirm the incorporation of TFA⁻ into the perovskite films but also help to establish the connection to the modified recombination dynamics and film morphology.

Previous reports in II-IV semiconductor nanocrystals suggest that a decouple of wave-function overlap between electrons and holes usually gives rise to a decreased Auger process. Such an effect can be achieved by either constructing a type II core-shell heterojunction or a careful design of organic ligands on the surface^48,49,50. Given that TFA⁻ is a strong and well-known electron-withdrawing moiety, we used density functional theory (DFT) calculation to unveil the changes in electron cloud density of crystal surface with and without TFA adsorption. We examined the surface lattice structures of FAPbI₃ and its variant, where one Cs⁺ cation substitutes FA⁺ (denoted as FA_xCs_1-xPbI₃) for comparison (Fig. 4b). As expected, the strong electron-withdrawing capability of −CF₃ moiety leads to an electron polarization and thus results in the reduction of electron cloud density in the lattice. This redistribution results in a decrease in wave-function overlap, reducing Coulomb electron-hole interaction and thereby decreasing Auger recombination. Our analysis on temperature-dependent PL enabled us to determine the exciton binding energies (Supplementary Fig. 15), which are 44 meV for F-films and about 28 meV for FCT-films, respectively. The decrease in exciton binding energy often leads to a retarded Auger process due to the changes in Coulomb interactions within the materials. Additionally, the electron localization function (ELF) and the Bader charge analysis (Supplementary Fig. 16) of the samples suggests that the atoms of the FAPbI₃ loss electrons in the presence of TFA. This obviously changed electron distribution of lattices is attributed to large electronegativity of the dangling F atoms in the TFA.

Having revealed the underlying reasons behind the decreased Auger recombination rates, we proceeded to investigate the origins of morphological variation, which is commonly associated with a change of crystallization process. We then performed ¹H nuclear magnetic resonance (¹H NMR) and Fourier-transform infrared spectroscopy (FT-IR) measurements, aiming to gather more information about the chemical interactions between perovskite precursors and TFA⁻ anions, which could be signs of the changes in the formation of intermediates and thus varied crystallization process. In Fig. 4c, we present the changes in the chemical shifts of the active protons in FA⁺ upon the addition of CsTFA (highlighted by the red square). For both cases, the chemical shifts moved toward the low field after TFA⁻ addition in the respective systems (FAI/CsTFA and PbI₂/FAI/CsTFA), confirming the interactions between FA⁺ cations and TFA⁻ anions. This can be assigned to the ionic interactions between positively charged FA⁺ and negatively charged TFA⁻, or additional hydrogen bonding interactions. In addition, the FT-IR results showed that the C=O vibration of TFA⁻ shifts to a lower wavenumber by mixing with PbI₂, accompanied by the shift of the C–F vibration frequency to a higher wavenumber (Fig. 4d). All these observations demonstrated the interactions of TFA⁻ anions with both FA⁺ and Pb²⁺ cations³⁵.

We found that the additional interactions of TFA⁻ with perovskite precursors not only facilitate the perovskite formation but also homogenize the grain growth. The former is evident from the presence of 3D perovskites within the precursor films (before thermal annealing), as their diffraction (Fig. 4e) and absorption features can be clearly identified by XRD and UV-Vis absorption measurements (Supplementary Fig. 17). In comparison, films without TFA⁻ displayed no prominent perovskite signals, despite the presence of PbI₂ signal. In the insets of Fig. 4e, the digital images visually capture the differences between F- and FCT-precursor films, highlighting the effect of TFA⁻ addition on facilitating perovskite nucleation. We attribute these discrepancies to the reduced formation of lead-Lewis base adducts with TFA⁻ addition, that is, Pb• DMSO in the current case; Pb• DMSO can slow down perovskite crystallization as they require extra energy to release the coordinating solvents³⁵. This speculation can be further evident from FT-IR spectra that the S=O stretching vibration of FAI•PbI₂•DMSO complexes shifts to a higher wavenumber once incorporating TFA⁻ anions, confirming a reduced formation of intermediate FAI•PbI₂•DMSO adducts (Supplementary Fig. 18). Additionally, our dynamic light scattering (DLS) analysis revealed that TFA⁻ addition reduces the size of the halide plumbate colloids, from an average size of 164 nm down to 105 nm (Supplementary Fig. 19) in precursor solutions. This suggests a more uniform nucleation process and a lower total free energy during perovskite phase formation. Taken together, these observations consistently point to a faster and more uniform crystallization process with TFA addition, which is usually beneficial for formation of dense and smooth perovskite films.

Interestingly, although fast crystallization of perovskites commonly leads to high densities of defects³⁵, we still observed reduced trap-assisted non-radiative recombination rates in FCT-films (Fig. 2g). This contrast suggests an effective capability of TFA⁻ in defect passivation^35,36. We thus investigated the adsorption geometry of TFA⁻ on vacancy-involved crystal lattice by DFT calculations. Our calculations indicate that TFA⁻ anions tend to adsorb on the surface of lattice instead of being in the crystals (Supplementary Fig. 20), as a result of their large ionic radius (2.38 Å). Notably, the adsorption energy (E_ads) of TFA⁻ anion on perovskite crystals (exposed lead cations) is as high as 4.55 eV, much larger than that of the iodide anions (~3.11 eV). This difference indicates a strong tendency of TFA⁻ to work with defect involved perovskite surfaces, and hence the passivation effect can be rationalized.

One consensus is that the reduced formation of vacancy-type defects contributes to the stabilization of ions within perovskite structures. To establish the correlation between ion movement and device stability, we calculated the desorption energy (E_d) of surface iodides (I⁻) in different local environment (Fig. 4f). A lower E_d indicates facilitated ion migration and thus accelerated device degradation, and vice versa. For pristine crystals and those with V_I, the E_d is 1.01 eV and 1.35 eV, respectively, suggesting that iodide anions adjacent to the defects are more likely to migrate out of the lattice. In contrast, the E_d values significantly increase up to 2.28 eV for the crystal lattice healed by TFA⁻, indicative of a greatly enhanced difficulty for neighboring iodides to become mobile²⁹.

To experimentally confirm the role of TFA⁻ in suppressing ion migration, we investigated the ion distribution in fresh and aged devices by TOF-SIMS measurements. The aged devices were measured at a fixed current density of 100 mA cm⁻² for 10 h ahead of tests. The comprehensive TOF-SIMS results for the devices of interest are shown in Supplementary Fig. 21. Figure 4g displays the evolution of iodide distribution in F- (upper panel) and FCT- devices (bottom panel). The former exhibits a distinct accumulation of iodides at the interface between HTL and the contacts, whereas the FCT-devices don’t show a substantial variation in iodide distribution after aging. These results are consistent with the previous report that halide migration toward the anode is the predominant factor limiting the operational lifetime of PeLEDs⁵¹.