Traditional light-emitting diodes (LEDs) on brittle and rigid substrates are not suitable to advanced applications such as wearable, bendable or stretchable display systems, which enable better man-machine interactions, provide wider fields of view, and bring more abundant visual experiences^{[1, 2]}. Nowadays, electronic products such as smart phones and watches equipped with curved screens are accessed into the markets. Nevertheless, these display screens are technically limited in mechanical robustness and flexibility to afford various elastic deformation required for advanced wearable displays. Furthermore, they fail to achieve full color displaying with high color purity. Therefore, it is of great necessity to develop flexible LEDs (FLEDs) to meet the demands of flexible display applications.

The first FLEDs were manufactured with organic emitters in 1992, simply replacing the rigid glass substrates with flexible polyethylene terephthalate (PET) substrates^[3]. Flexible organic LEDs (FOLEDs) have undergone rapid development and now are successfully commercialized, but the color gamut and purity are unsatisfactory^[4−8]. Apart from flexible organic LEDs, it has been demonstrated that flexible quantum dot LEDs (FQLEDs) have great potential for flexible displays since their first presence in 2009^[9]. At present, FQLEDs are facing the issues of toxicity, insufficient structural flexibility, and mass production techniques^[10]. Thus, it is an urgency to develop FLEDs featuring adequate flexibility, facile color tunability, high color purity as well as other excellent optoelectronic properties.

Recently, metal halide perovskites (MHPs) have attracted tremendous attention due to the prominent optoelectronic properties such as tunable bandgap, narrow full width at half maximum (FWHM), high charge mobility, high photoluminescence quantum yield (PLQY), and high defect tolerance^[11−16]. MHPs have been widely studied and employed to fabricate optoelectronic devices, such as solar cells (SCs), LEDs, and photodetectors^[17−28]. Right after the report about the first room temperature perovskite LEDs (PeLEDs) in 2014, the first flexible PeLEDs (FPeLEDs) were created by simply substituting the rigid glass substrate with flexible PET substrate, which paved the way for rapid development of FPeLEDs in the following decade^{[29, 30]}. Various study has been conducted to promote the development of FPeLEDs. For example, flexible polymeric substrates such as polyethylene terephthalate (PET), polyimide (PI), polyethylene naphthalate (PEN), polydimethylsiloxane (PDMS), polyurethane (PU) and Norland optical adhesive 63 (NOA 63) have been employed to sustain the elastic deformation of devices^{[29, 31−39]}. Low dimensional carbon materials, conductive polymers and metal nanowires have been developed for flexible and transparent electrodes to replace brittle indium tin oxide (ITO)^[39−43]. The optoelectronic and mechanical properties of perovskite films have been successfully enhanced by various strategies^{[31, 34, 44−46]}. Nowadays, the state-of-the-art FPeLEDs have achieved the record external quantum efficiency (EQE) of 24.5% and retained high efficiency over 90% of initial value after 1000 bending cycles at a curvature radius of 3 mm, demonstrating their great potential for next generation of display^[46].

In this paper, we review the recent efforts devoted to improving the optoelectronic and mechanical performance of FPeLEDs. Firstly, we give a brief introduction to perovskite materials, including the crystal structure, optoelectronic and mechanical properties. Subsequently, we summarize various strategies developed to enhance the device performance, among which we especially emphasize the significant roles of perovskite emitters and electrodes. Then we present applications of FPeLEDs in display and beyond. Finally, we summarize the development of FPeLEDs in the past and point out the existing challenges for future development.

Named after the discoverer of calcium titanate (CaTiO₃), perovskites now usually refer to a category of metal halide materials with similar lattice structure to CaTiO₃. With a chemical formula of ABX₃, the general crystal structure of three-dimensional (3D) perovskites is shown in Fig. 1(a), where A-, B- and X-sites are occupied by monovalent cations (e.g., cesium (Cs⁺), methylamine (MA⁺) or formamidine (FA⁺)), divalent metal cations (e.g., lead (Pb²⁺) or tin (Sn²⁺)), and halide anions (e.g., chlorine (Cl⁻), bromine (Br⁻) or iodine (I⁻)), respectively. Each B-site cation is coordinated with six X-site anions to shape a metal halide octahedron [BX₆]⁴⁻, where the B-site cation is in the center while the X-site anions located at the vertexes. Adjacent octahedrons are interconnected through corner-sharing B−X bonds to build the 3D octahedron network while the A-site cations are inserted into the interstices of the octahedron network. Treating all ions as rigid spheres, Goldschmidt’s tolerance factor t (Eq. 1) and octahedral factor μ (Eq. 2) have been established to estimate the crystal stability of 3D perovskites^[47],

Named after the discoverer of calcium titanate (CaTiO₃), perovskites now usually refer to a category of metal halide materials with similar lattice structure to CaTiO₃. With a chemical formula of ABX₃, the general crystal structure of three-dimensional (3D) perovskites is shown in Fig. 1(a), where A-, B- and X-sites are occupied by monovalent cations (e.g., cesium (Cs⁺), methylamine (MA⁺) or formamidine (FA⁺)), divalent metal cations (e.g., lead (Pb²⁺) or tin (Sn²⁺)), and halide anions (e.g., chlorine (Cl⁻), bromine (Br⁻) or iodine (I⁻)), respectively. Each B-site cation is coordinated with six X-site anions to shape a metal halide octahedron [BX₆]⁴⁻, where the B-site cation is in the center while the X-site anions located at the vertexes. Adjacent octahedrons are interconnected through corner-sharing B−X bonds to build the 3D octahedron network while the A-site cations are inserted into the interstices of the octahedron network. Treating all ions as rigid spheres, Goldschmidt’s tolerance factor t (Eq. 1) and octahedral factor μ (Eq. 2) have been established to estimate the crystal stability of 3D perovskites^[47],

1t=RA+RX2(RB+RX),

2μ=RBRX,

where RA, RB, and RX stand for the ionic radii of A-, B- and X-site ions. When 0.813 < t < 1.107 and 0.442 < μ< 0.895, most 3D perovskites can be formed^[48]. In particular, the cubic perovskites are formed if 0.9 < t< 1.0 while other structural perovskites such as orthorhombic, tetragonal or hexagonal phase can be obtained if t < 0.9 or t > 1.0^{[49, 50]}.

The stability of 3D perovskites is susceptible to A-site cations^{[50, 51]}. When incorporating bulky organic cations such as PEA⁺ and PBA⁺ into 3D perovskites, which are too large to enter the clearances, the 3D octahedron network will be divided into various layered structures and separated by insulating organic spacer layers (Fig. 1(b)). As a result, two-dimensional/quasi-two-dimensional (2D/quasi-2D) perovskites (n = 1/n = 2, 3, 4, ···, p.s. n is the amount of octahedron layers) are formed on the base of 3D perovskites (n = ∞). Generally, 2D/quasi-2D perovskites show higher PLQY and more efficient energy cascade compared to 3D ones, resulting from the strong quantum and dielectric confinement effects due to the reduced dimensionality along the direction normal to spacing layers^[52].

Another type of perovskites adopting 3D structure is termed as perovskite quantum dots/nanocrystals (PeQDs/PeNCs). PeQDs/PeNCs can be synthesized by various methods including hot injection, ligand-assisted reprecipitation, anion exchange and chemical vapor deposition etc.^[53−55]. At present, PeQDs/PeNCs have been widely studied and exploited for LEDs and color conversion in display, mostly because of their excellent optoelectronic properties owing to strong quantum and dielectric confinement effects induced by the reduced size of crystals^{[54, 56−58]}.

As one kind of promising light-emitting materials, perovskites feature many outstanding optoelectronic characteristics including tunable bandgap^[59], narrow FWHM^[60], high PLQY^[12] and high defect tolerance^[61], which make them stand out of other organic or inorganic luminescent materials.

Previous study has reported that the bandgap of perovskites is primarily determined by metal halide octahedron [BX₆]^4-, but different A-site cations can cause the tilt and distortion of octahedron to affect the bandgap indirectly^[62−64]. In practical, the usual and effective way to achieve bandgap tuning is to adjust the ratio of X-site anions. With the increasing electronegativity of halogens in the sequence of I to Br to Cl, the bandgap gradually becomes wider and thus results in the blue shift in emission wavelength from infrared-red to blue-ultraviolet regions (Figs. 1(c) and 1(d))^[65]. The combination of red, green and blue perovskite emitters with narrow FWHM can achieve high color purity and broad color gamut covering more than 140% National Television System Committee (NTSC) standard (Fig. 1(e))^{[14, 66−68]}.

Compared to other inorganic or organic materials, perovskites exhibit bright and efficient photoluminescence. One of the major reasons is the high defect tolerance. Intrinsic point defects, dangling bonds or impurities usually act as mid-gap trap states in conventional semiconductor, resulting in severe degradation of luminescence, but they tend to form electronic states within or near conduction band (CB) and valence band (VB) in perovskites, which have negligible effects on efficiency (Fig. 1(f))^[61]. Besides, benefiting from the high exciton binding energy due to dielectric confinement and quantum confinement effects, the excitons in low-dimensional perovskites are more stable and effectively devoted to radiative recombination leading to high PLQY and brightness^{[69, 70]}. It is reported that the optimized perovskite QDs are able to achieve extremely high PLQY near 100%, indicating the practically complete elimination of intrinsic defects^[12].

In practical applications, especially in flexible devices, the mechanical properties of perovskites are as important as the optoelectronic properties. Previous study on the 3D single crystal perovskites has demonstrated that they have good flexibility and ductility for elastic deformation due to the low elastic modulus, which is determined by B−X bond and A-site cation, where the B−X bond leads a dominated role^{[75, 79−81]}. In general, the elastic modulus (E) along the parallel orientation increases with the enhanced strength of B−X bond as halogen changes from I to Br to Cl (Fig. 1(g))^{[74, 82]}. In terms of A-site cations, researches suggested that they were just filled into the gaps of octahedron network and offered positive charges to maintain the electrical balance^[81]. However, there is evidence that A-site cations also influence the elastic modulus^{[75, 83]}. As shown in Fig. 1(h), it is clear that MAPbBr₃ and FAPbBr₃ differ in elastic modulus. On one hand, the crystals composed of organic cations with different sizes form B−X bond with distinct length and strength leading to the discrepancy in elastic modulus; in another hand, it is observed that the distance between organic cations and anions is longer in FAPbBr₃ due to weaker hydrogen bond, implying the strength of formed hydrogen bond probably affect the mechanical properties as well^[75].

Compared to 3D single crystal perovskites, 2D/quasi-2D Ruddlesen−Popper (RP) single crystal perovskites are more flexible due to the reduced octahedral layers, soft organic spacer layers, and weak van der Waals interaction between organic layers^[77]. Therefore, the mechanical properties 2D/quasi-2D RP single crystal perovskites are more complex, which are affected by the B−X bond, n value, and species of organic spacers. In general, the weaker B−X bond, smaller n value or the longer alkyl chains result in lower out-of-plane elastic modulus, while substituting the monoamine alkyl chains with rigid phenyl chains or diamine alkyl chains leads to higher mechanical stiffness due to the strong phenyl−phenyl interaction or covalent bond (Figs. 1(i)−1(l))^{[76, 77]}. In-plane elastic modulus is also studied, but its changing trend is not completely consistent with out-of-plane elastic modulus and their correlation remains further study^[84].

Compared to single crystal perovskites, polycrystalline perovskites are more common in the fabrication of FPeLEDs. Therefore, it is more meaningful to investigate the mechanical properties of polycrystalline perovskites, but most study is conducted on the base of single crystal ones. If we simply treat polycrystal as a collection of abundant crystals, the mechanical properties of polycrystal not only depend on those of each crystal but also the interaction between adjacent crystals. It proved that the existing inverse Hall−Petch effect on grain boundaries leads to enlarged yield regions as grain size increases, enabling extended deformation of perovskites films (Fig. 1(m))^[78]. In other words, the polycrystalline perovskites show better mechanical flexibility and ductility than single crystal ones due to extensive and continuous amorphization under tensile loading.

Generally, FPeLEDs and flexible perovskite solar cells (FPeSCs) employ the similar device structure. The essential components include substrates, charge transport layers (CTLs, comprising hole transport layers (HTLs) and electron transport layers (ETLs)), perovskite active layers and electrodes (anodes and cathodes) (Fig. 2(a)). Table 1 summarizes the performance of FPeLEDs reported previously. So far, the highest EQE of 24.5% has been achieved by green FPeLEDs, while the record efficiencies of red and blue counterparts are only 12.7% and 13.5%, respectively^{[45, 46, 85]}. To promote the development of FPeLEDs, we review the relevant improving strategies in the following sections.

The ideal substrate materials for flexible devices should meet the following requirements: excellent mechanical flexibility, high light transmittance, chemical and physical stability^{[1, 2]}. Theoretically, materials such as polymers, metal foils, ultrathin flexible glass and paper are suitable^[110]. In practical, transparent or colorless polymers like PET, PI, PU, PEN and PDMS etc. are more commonly selected for fabricating flexible substrates due to the advantages of light weight, low cost and roll-to-roll processibility^{[31−37, 110]}. Except for these traditional polymers, a kind of photopolymer, NOA 63, was also adopted due to its excellent transparency, flexibility and smooth flat surface^{[38, 39]}. Recently, Qin et al. reported the first FPeLEDs based on paper substrates, where the polymethyl methacrylate (PMMA) was adopted as a buffer layer to smooth the surface of paper and protect the devices from moisture^[111]. It is beneficial to further reduce the weight of FPeLEDs by using paper substrates, however, it should be noted that such devices must adopt a top emitting structure because the poor transmittance of paper.

In FPeLEDs, the performance of the devices primarily depends on the properties of perovskite emission layers (EMLs). Compared to rigid PeLEDs, perovskite EMLs in FPeLEDs are not only required for uniform morphology with less surface defects to boost the optoelectronic characteristics, but also excellent mechanical flexibility and stability to prevent devices performance from serve degradation during elastic deformation.

To boost the optoelectronic performance of FPeLEDs, the homogeneous and pinhole-free perovskite EMLs are indispensable. The comprehensive investigation on the crystallization dynamics during perovskites formation has verified that the chemical composition, additive, nature of underlying interface, fabrication techniques and conditions have significant effects on the morphology of as-prepared films^[112−119]. Therefore, it is rational to modulate the crystallization process and morphology of perovskite EMLs for FPeLEDs.

Incorporating polymer additives into perovskite precursor is an effective way to modulate the crystallization process to improve the morphology of perovskite EML. Hydrophilic polyethylene glycol (PEG) is commonly adopted to enhance the wettability of underlying surface and promote the formation of nucleation sites^[120]. Besides, the PEG can fill the grain boundaries and interact with perovskite grains to retard their growth and passivate the defects on the grains surface during crystallization process (Fig. 2(b))^{[86, 120−122]}. Combining the incorporation of PEABr, an ultrasmooth, continuous, and uniform perovskite films were obtained (Fig. 2(c)) and the corresponding FPeLEDs (514 nm) achieved maximum EQE and CE of 10.1% and 31.0 cd∙A⁻¹, respectively^[86]. Polyethylene oxide (PEO) is also widely used to modify the film morphology in traditional PeLEDs^[123−126]. It has proven that PEO matrix can lower the mobility of perovskite precursor at the solvent drying stage during spin-coating, which inhibits the growth and aggregation of perovskite crystals to enable smooth and uniform films with small grains size and better contact with overlying layer^{[90, 127]}. The ionic conducting property of PEO further offers enhanced conductivity of composite films^{[127, 128]}. In consequence, the FPeLEDs based on these composite films exhibited a low turn-on voltage (V_on) of 2.4 V and high luminance exceeding 15 000 cd∙m⁻²@8.5 V^[37]. Polyvinylpyrrolidone (PVP), rich in Lewis base pyrrolidone groups, is also exploited to modulate perovskite crystallization and coordinate with Pb²⁺ to suppress the formation of halide vacancies, resulting in homogeneous perovskite QDs films with superior PLQY near 100% (Figs. 2(d) and 2(e))^{[87, 129]}. The as-prepared FPeLEDs based on the perovskite QDs/PVP films showed electroluminescence (EL) wavelength and maximum luminance of 530 nm and 43 990 cd∙m⁻², respectively.

The surface nature of underlying layer is a key factor to modulate the morphology of deposited perovskite EMLs. For most p−i−n type devices, conducting polymer poly (3,4-ethylenedioxy-thiophene): poly (styrene sulfonate) (PEDOT:PSS) is chose as HTL material and perovskite films are deposited on the underlying PEDOT:PSS films. Ethanolamine (ETA) additive can improve the hydrophilicity of PEDOT:PSS films and then induce the morphology transition of perovskite films from formless grains to uniform square-like ones (Fig. 2(f)), leading to enhanced photoluminescence (PL) performance^{[46, 121]}. Given that the acidic PEDOT:PSS may corrode the underlying layer and then bring detrimental effect on device performance, a kind of water-soluble conjugated polyelectrolytes with neutral pH value, termed as TB(MA), was adopted to replace PEDOT:PSS films and function as HTL^[35]. These novel films not only prevented the corrosion of underlying layer from acidity, but also showed strengthened hydrophilicity to improve the cover rate of perovskite films. Thus, the EQE and CE of prepared FPeLEDs (488 nm) were boosted to 8.3% and 14.7 cd∙A⁻¹, respectively. A passivation potassium iodide (KI) layer was introduced to enhance the affinity between HTL and EML, where the strong ionic interaction between K⁺ and I⁻ induced the uniform and compact arrangement of perovskite NCs and suppressed the formation of I-vacancies (Fig. 2(g))^[85]. As a result, the optimized devices (687 nm) achieved the record EQE of 12.7% among red emissive FPeLEDs.

The growing process engineering is also an effective way to improve the perovskite EML morphology. An encapsulation growth method was employed to modulate the crystallization process and phase distribution of quasi-2D perovskite films (Fig. 2(h)). perovskite crystals were inclined to transform into randomly oriented large-n phases induced by space-confined growth effect, enabling highly efficient charge and energy transport due to sufficient contact between individuals (Fig. 2(i))^[88]. Instead of traditional anti-solvent treatment, N₂ gas blowing was adopted to modulate the perovskite crystallization process (Fig. 2(j))^[89]. The nucleation rate generally depends on the spin rate and liquid thickness, so the nitrogen (N₂) gas blowing can help accelerate the solvent evaporation and increase nucleation sites to form high quality perovskite films during spin-coating^[89]. Compared to conventional thermal annealing, the flash light annealing (FLA) method can trigger ultrafast recrystallization of perovskites to form a smooth film with dense and small grains, leading to significant improvement of current efficiency (CE) (Figs. 2(k) and 2(l))^[31]. It is noteworthy that this ultrafast recrystallization activated by high temperature of 250 °C is completed within 660 μs without damage to the underlying heat-sensitive flexible substrate, indicating FLA method is compatible to commercialized mass production of FPeLEDs.

To boost the efficiency of photons harvesting and converting, perovskite active films in FPeSCs are usually about one order of magnitude thicker than those in FPeLEDs. The thin films in FPeLEDs allow the devices to bend or stretch more easily. Nonetheless, perovskite films are still the most fragile section in FPeLEDs due to its relatively high elastic modulus compared to other components^[94]. The mechanical properties of perovskite films do not receive enough attention and the poor mechanical stability usually results in severe degradation of luminance and efficiency during repetitive elastic deformation, let alone long terms operation. Therefore, it is of great necessity to enhance the mechanical properties of perovskite films.

The mechanical stability and stretchability of perovskite films can be strengthened by incorporating PEO, where the perovskite crystals are embedded into PEO matrix^{[36, 37, 44, 134]}. The PEO matrix acts as an elastic connecter cross-linking dispersed crystals to impart excellent stretchability and absorb tensile stress during stretching, so the crystals in perovskite/PEO composite films nearly remain intact while those in pristine films are broken and decomposed after stretching (Fig. 3(a))^{[37, 44, 90]}. Accordingly, the PL intensity of perovskite/PEO composite films slightly decreased under low tensile strain and maintained 60% of initial value when strain increased to 70% (Fig. 3(b)). Besides, the excellent mechanical stability was reflected by slight PL degradation without significant change in wavelength and FWHM after 1000 stretch-release cycles under 50% tensile strain. Similarly, a low-cost and green biomass ethyl cellulose (EC) was blended with CsPbI₃ NCs to serve as an elastic cross-linker between neighboring isolated NCs through hydrogen bonds and Pb−O bond^[130]. The optimized films showed an apparent enhancement of mechanical stability compared to pristine ones, and the corresponding FPeLEDs (686 nm) not only achieved a high EQE of 12.1% but also exhibited excellent mechanical flexibility and stability that the devices maintained high brightness as the bending radius decreases to 1 mm (Fig. 3(c)).

The effect of bulky organo-ammonium halide additives on mechanical properties of perovskite films was studied by introducing butylammonium iodide (BAI), dodecylammonium iodide (DDAI), benzylammonium iodide (PMAI), phenethylammonium iodide (PEAI) and 4-fluorobenzylammonium iodide (p-FPMAI), respectively^[34]. It found that the first four additives indeed contributed to the strengthened cohesion energy (G_c), a parameter indicating the mechanical robustness or ability to resist cracks propagation, but there existed a trade-off between G_c and PLQY with the increment of chains length (Fig. 3(d))^{[34, 135]}. In contrast, the p-FPMAI additive enabled increment both in G_c and PLQY of perovskite films compared to those added with PMAI, which was mainly attributed to the presence of fluorine at the phenyl group enhancing the binding energy and resulting in better coverage of additives around crystals. As a result, the p-FPMAI-modified FPeLEDs exhibited outstanding mechanical flexibility and stability, which maintained unchanged efficiency after 10 000 bending cycles at a curvature radius of 2 mm and 80% of initial values at 1 mm (Fig. 3(e)).

It was reported that FPeSCs and photodetectors adopted a kind of films with self-healing structure to improve the mechanical flexibility and stability, which can be applied to FPeLEDs as well^[136−138]. Qian et al. designed a kind of self-healing FPeLEDs by employing an elastomer diphenylmethane diisocyanate polyurethane (MDI-PU) to modulate the perovskites crystallization, where the MDI-PU played a role of connector between adjacent crystals and the repairable hydrogen bonds formed in soft segment of polymer chains introduced a great self-healing ability (Figs. 3(f) and 3(g))^[45]. Another self-healing FPeLEDs were achieved by incorporating (3,3,3-trifluoropropyl) trimethoxysilane (TFPTMS) into perovskite precursor, which also exhibited exceptional mechanical robustness and flexibility as well as self-healing capacity^[139]. However, the self-healing ability of both MDI-PU and TFPTMS-doped perovskite films is thermally activated, which limits their applications in flexible devices working at room temperature. The self-healable perovskite films at room temperature were successfully obtained by introducing a fluoroelastomer into the perovskite precursor, which was mainly attributed to strong dipole−dipole interaction between -CF₃ groups (Fig. 3(h))^[131]. The cracks in the modified films underwent a self-healing process at room temperature and almost disappeared after 24 h (Fig. 3(i)). Furthermore, two individual parts cut from different perovskite films surprisingly merged into a novel film and still maintained outstanding resistance to stretching, indicating the superior self-healing ability and mechanical stability. Besides, these modified perovskite films exhibited exceptional mechanical stability (Fig. 3(j)), implying their great potential for FPeLEDs.

Recently, a novel kind of light-emitting layer−perovskite/polymer composite nanofiber membrane has popped up and rapidly become a research hotpot^{[133, 140−143]}. The perovskite NCs with poor stability are embedded into the elastic polymer nanofibers through electrospinning (Fig. 3(k)), which endows great ambient stability and mechanical ductility^[132]. Under the protection from polymer matrix, the bright light emission achieved even under 170% tensile strain (Fig. 3(l)), indicating such elastic composite nanofibers can be applied to prepare FPeLEDs with outstanding performance^[133].

Electrodes with remarkable flexibility, ductility, conductivity and optical transmittance are essential to obtain efficient FPeLEDs. However, ITO is not suitable to flexible devices due to its brittle nature despite the high conductivity and optical transmittance. To maximize the performance of FPeLEDs, the selection of suitable electrodes should be concerned. Currently, materials such as carbon nanotubes (CNTs), graphene, conductive polymers and metal nanowires are emerging as promising candidates for transparent and flexible electrodes^[144−147]. The comparison of these materials is depicted in Fig. 4.

Low dimensional carbon materials like CNTs and graphene have been developed for flexible electrodes. Both CNTs and graphene feature excellent mechanical flexibility, great optical transmittance and high chemical stability, but CNTs show poorer conductivity than graphene (Fig. 4).

For CNTs, their conductivity is limited by undesired barrier between individual CNTs^[149]. In spite of the superior conductivity (20 000 S∙cm⁻¹) of individual CNTs, the high junction resistance results in significant degradation in average conductivity of the whole CNTs film (6600 S∙cm⁻¹), which is probably attributed to the diverse length, diameter and chirality of individual CNTs^{[145, 147, 148, 150]}. The precursor containing a number of impurities and diverse structures also leads to the rough surface of CNTs films inducing poor contacts with overlying layer resulting in inferior device performance^{[145, 147, 149]}. Therefore, the early reported fully printed FPeLEDs based on CNTs anodes only reached the maximum luminance and EQE of 360 cd∙m⁻² and 0.14%, respectively, as well as minimum bending radius of 5 mm (Fig. 5(a))^[40].

The promising graphene films are commonly fabricated by mechanical exfoliation, liquid phase exfoliation, reduction of graphene oxide (rGO), chemical vapor deposition (CVD) and so on^[151−153]. In spite of their smooth surface, graphene films are limited in conductivity due to their low charge carrier mobility, which can be improved by fabricating pure and large graphene sheets with few boundaries or doping strong acids, metal halide, metal oxide and alkali metal carbonates^{[145, 148]}. It was reported the best graphene films with low sheet resistance (30 Ω∙sq⁻¹) and high optical transmittance (90%) were obtained through CVD method and p-doping of hydrogen nitrate (HNO₃)^[154]. Hence, Seo et al. first employed the graphene films as anodes to design the ITO-free FPeLEDs, and these devices (542 nm) exhibited a high luminance about 13 000 cd∙m⁻² and maximum CE and EQE of 16.1 cd∙A⁻¹ and 3.8%, respectively^[43]. The results of bending tests demonstrated that the graphene-based devices showed significantly enhanced mechanical flexibility and stability compared to those based on ITO anodes (Fig. 5(b)).

Transparent conductive polymers represented by PEDOT:PSS are promising candidates for electrodes in flexible devices due to their excellent mechanical flexibility, great optical transmittance, low surface roughness and good conductivity (Fig. 4). Besides, their solution processability makes them compatible to low-cost roll-to-roll production^[148].

PEDOT:PSS, as a common material of HTL, has been widely studied to unlock its potential for flexible electrodes. On one hand, the excess insulated PSS chains in pristine PEDOT:PSS films result in extremely poor conductivity (below 1.0 S∙cm⁻¹)^[155−157]. Thus, many dopants have been exploited to remove the excess PSS to enhance the electrical conductivity, such as polar solvents, ionic liquids and strong acids, and thereby the conductivity of optimal PEDOT:PSS films can reach more than 4000 S∙cm^{−1[147, 149, 156, 158]}. On the other hand, the poor mechanical stability of pristine PEDOT:PSS films leads to the sharp increase in sheet resistance after bending or stretching in spite of their great flexibility^{[37, 159]}. Elastomer PEO was usually adopted to enhance the mechanical stability, so the all-inkjet-printed FPeLEDs based on PEDOT:PSS/PEO composite electrodes with maximum luminance of 10227 cd∙m⁻² and CE of 2.01 cd∙A⁻¹ displayed highly stable EL characteristics after 5000 bending cycles at a curvature radius of 2.5 mm (Fig. 5(c))^{[36, 37, 44, 100, 134]}.

To overcome the poor conductivity and mechanical stability meanwhile, the PEDOT:PSS films were blended with co-dopants dimethyl sulfoxide (DMSO) and Zonyl-FS300 (Zonyl)^{[160, 161]}. Despite the stable sheet resistance of optimized PEDOT:PSS films during 1000 bending cycles at a curvature radius of 1 mm, the severe EL quenching of as-prepared FPeLEDs occurred early before 80 cycles mainly due to the mechanical failure of EML (Fig. 5(d))^[94]. Recently, the outstanding FPeLEDs (513 nm) based on modified PEDOT:PSS electrodes have been fabricated and achieved maximum luminance, CE and EQE of 25972 cd∙m⁻², 25.1 cd∙A⁻¹ and 8.0%, respectively (Fig. 5(e))^[159]. In this work, a photopolymerizable additive hexa-2,4-diyne-1,6-diol (HDD) was introduced into PEDOT:PSS films and the cross-linked polymer network was formed induced by interaction of PEDOT:PSS and HDD during photopatterning, resulting in the significantly enhanced conductivity and mechanical stability.

Notably, the biggest advantage of PEDOT:PSS films is that they can work as bifunctional layers to simplify the device structure. After incorporating tetrafluoroethylene-perfluoro-3,6-dioxa-4-methyl-7-octenesulphonic acid (PFSA) into PEDOT:PSS precursor solution, the AnoHTL films were formed, where the highly conductive PEDOT:PSS chains with low surface energy contributed to the formation of bottom anode while PFSA chains with high surface energy accumulated at upper part to gradually increase the work function and narrow the hole injection barrier (Fig. 5(f))^[162].

As shown in Fig. 4, metal nanowires exhibit the most balanced performance compared to other materials, among which silver nanowires (AgNWs) are widely used as flexible electrodes. AgNWs have exceptional mechanical flexibility, conductivity and optical transmittance, but their rough surface and dramatically increased contact resistance under deformation lead to unsatisfactory performance^{[1, 148]}.

One of the effective solutions is to prepare polymer/AgNWs composite electrodes. Through a three-step process, the AgNWs were embedded into the polymer matrix to prepare composite electrodes with a smooth surface (<1 nm), low sheet resistance (10 Ω∙q⁻¹) and high transmittance (86% at 550 nm)^[91]. Similarly, the PI/AgNWs composite electrodes were developed, which had an ultrasmooth surface and outstanding mechanical stability (Fig. 5(g))^{[34, 106]}. Furthermore, the minimum bending radius of FPeLEDs based on PI/AgNWs anodes and optimized perovskite films could be reduced to 0.25 mm without failure (Fig. 3(e) (right))^[34]. It was also reported that the ultra-stable FPeLEDs were achieved by using AgNWs/sulfuric acid (H₂SO₄)-treated PEDOT:PSS composite anodes, which could withstand 1000 bending cycles at a curvature radius of 2.5 mm without significant degradation and showed a minimum bending radius of 1.5 mm (Figs. 5(h) and 5(i))^[39].

Except for polymers, the AgNWs-based composite electrodes could be obtained by depositing the sol–gel-derived zinc oxide (ZnO) films upon AgNWs network, where the ZnO could fill the voids between AgNWs network to smooth the surface and enhance the chemical stability and mechanical flexibility^[46]. It is noteworthy that the resulted FPeLEDs achieved the record EQE and CE of 24.5% and 75.0 cd∙A⁻¹. Interestingly, the novel silver-nickel NWs (Ag-Ni NWs) composite electrodes with a core-shell structure (Fig. 5(j)) were developed to improve the performance of FPeLEDs, where the shell (Ni) was electroplated on the core (AgNW) to enhance the conductivity, increase the work function and improve mechanical stability, resulting in high CE and EQE of 44.01 cd∙A⁻¹ and 9.67%, respectively^[98].

Recently, MXenes have attracted increasing attention due to its exceptional metallic conductivity, excellent mechanical flexibility, smooth and hydrophilic surface as well as other appealing advantages^[163−166]. Thus, a kind of composite electrodes with smooth and uniform surface and dramatically decreased sheet resistance (Fig. 5(k)) were developed through the sequent deposition of silver nanoparticles (AgNPs), PEDOT:PSS and MXenes on AgNWs films^[97]. These novel composite electrodes also showed a stable resistance during repetitive bending and thereby the corresponding FPeLEDs achieved excellent EL characteristics and mechanical stability (Fig. 5(l)).

Although (F)PeLEDs and (F)PeSCs commonly share the similar device structure, they differ in the energy band distribution and charge carrier dynamics. For (F)PeSCs, when the photons are absorbed by perovskite films, the electrons will be formed in CB quickly and prone to transport along the direction of decreasing energy levels while the holes generated in VB move in the opposite direction^[167]. Differently, in (F)PeLEDs, the electrons and holes are usually injected into conduction band (CB) and valance band (VB) at high energy levels and transported to perovskite emitting layer through ETL and HTL, respectively, where the energy levels of CB and VB are lower. However, the large carrier/energy injection barrier between CTLs and EML is one of the main reasons for poor performance of traditional PeLEDs and FPeLEDs, especially between HTL and EML^{[29, 39]}. It is necessary to tailor energy band structure of devices to achieve efficient performance.

Sequent deposition of multi-HTLs with the different highest occupied molecular orbital (HOMO) level can construct a step gradient energy band structure to accelerate the hole injection and smooth energy transport^{[168, 169]}. In this way, the bi-HTLs FPeLEDs (521 nm) with step gradient energy band structure were obtained and achieved maximum luminance and CE of 1000 cd∙m⁻² and 10.4 cd∙A⁻¹ (Fig. 6(a))^[91]. However, deposition of multi-HTLs is time costly and makes the device more complicated. A self-organized gradient buffer HTL (Buf-HTL) composed of PEDOT:PSS and perfluorinated ionomer (PFI) has a gradually increased work function from 5.2 to 5.95 eV and enables better energy band alignment and narrow hole injection barrier, leading to more efficient and balanced carrier injection^{[8, 29, 170]}. Therefore, the graphene-based (work function: 4.4 eV) FPeLEDs adopted such Buf-HTL to minimize the hole injection barrier and achieved high luminance and CE of ~13 000 cd∙m⁻² and 16.1 cd∙A⁻¹ (Fig. 6(b))^[43].

Aforementioned in Section 3.2.2, the co-dopants of DMSO and Zonyl-modified not only can enhance the conductivity and mechanical stability of PEDOT:PSS films, but also raise the HOMO level to 5.45 eV, significantly lowering the hole injection barrier from 0.76 to 0.35 eV (Fig. 6(c))^[39]. The optimized FPeLEDs showed a low V_on of 3.0 V and maximum luminance and CE of 1060 cd∙m⁻² and 17.9 cd∙A⁻¹, respectively. Similarly, poly(sodium 4-styrenesulfonate) (PSS-Na) was doped into PEDOT:PSS precursor to enhance HOMO level to boost the hole injection, and thereby the maximum CE and EQE of the prepared FPeLEDs increased from 6.64 cd∙A⁻¹ and 1.57% to 25.13 cd∙A⁻¹ and 5.91%, respectively (Fig. 6(d))^[99].

In addition to modifying the HTL, it is feasible to engineer the perovskite films to balance the carrier injection and improve the devices efficiency. A dendritic perovskite film has been developed to confine the holes diffusion and increase the electrons injection to enable efficient radiative recombination, where only one side of crystals contacts with HTL while other sides are exposed to ETL (Fig. 6(e))^[171]. The corresponding FPeLEDs achieved the highest EQE of 3.2% while that of the pristine devices is 1.5%. Besides, it was reported that the perovskite films doped by carbon dots had a lower HOMO level (Fig. 6(f)), so the carrier injection barrier between HTL and EML could be narrowed^[172]. The FPeLEDs showed a maximum luminance of 2259 cd∙m⁻² with a high current density of 474 mA∙cm⁻² due to the strengthened balance of carrier injection.

At present, the performance of FPeLEDs is still far away from the implementation in display applications. However, there are still studies working towards the real display application. In making a display, the FPeLEDs should be patterned or pixelated. A simple way of patterning the device was developed by Liu et al. who reported a method to pattern the electrodes, instead of the whole device pattern, providing a potential way to fabricate patterned or pixelated FPeLEDs (Fig. 7(a))^[159].

In addition, patterning the perovskite EML represents another approach. A developed double-layer transfer printing method was reported to prepare perovskite NCs pixels, which can precisely control the size and position of perovskite NCs pixels and prevent the internal cracking in pixelated perovskite films during transfer printing process (Fig. 7(b))^[107]. Furthermore, the ultrathin attachable displays based on pixelated FPeLEDs arrays were successfully obtained. Such attachable displays showed great adaptability on different objects, such as human skin, leaf and the edge of blade, and exhibit excellent water resistance and deformation stability (Figs. 7(c) and 7(d)).

Given that aforementioned FPeLEDs arrays with specific patterns are limited in practical display applications, a kind of writable and wipeable FPeLEDs was created to display various patterns through writing or printing on the top surface (Fig. 7(e))^[109]. The initial perovskite films showed a large bandgap due to the crystals with low crystallinity randomly oriented arrangement, so the devices could not illuminate under normal bias voltage. After heat pen writing or stamp printing, the heat-treated areas lit up owing to the much narrower bandgap resulting from heat-induced crystallization and oriented arrangement while other areas maintain darkness. The present patterns could be wiped by applying increasing voltage until the whole areas of devices emitted bright light (Fig. 7(f))^[109]. However, the electrical heat also induced the crystallization of perovskites in other areas, so the cyclic use of such writable and wipeable FPeLEDs still remains challenging requiring reversible crystallization process.

In addition to display, FPeLEDs also have great potential in other applications. Using spacers to separate the top electrodes and EMLs, the FPeLEDs can be developed for light-emitting touch-responsive devices (LETDs)^{[36, 90]}. Such LETDs can be activated by instantaneous contact between top electrodes and EMLs when applying pressure on the top electrodes (Fig. 7(g)), which are promising for human-machine interaction, fingerprint identification and motion detection^[90]. Considering that the near-infrared (NIR) light can penetrate deep into human body rather than be greatly absorbed or scattered by human skin, the large-area NIR emissive FPeLEDs were applied to non-invasive deep-tissue imaging on human body^[173]. Combing with silicon photodiodes, these devices could realize real-time tracking subcutaneous blood flow variation to monitor heart rate. Inspired by the concept of visualized signals, the multi-color FPeLEDs were adopted in wearable electrocardiogram (ECG) monitors to detect health condition according to the emission color evolution (Fig. 7(h)), indicating the probability of FPeLEDs in health care fields as well^[122].

In this review, we summarize recent advances in development of FPeLEDs. Especially, we emphasize the crucial roles of perovskite films and flexible electrodes among all components in FPeLEDs. It is noted that the mechanical properties are as important as the optoelectronic characteristics, whereas most research underestimate the effect of the former. Besides, ideal electrodes possessing great conductivity, transmittance, flexibility and work function are essential to promote the performance. We also review current research on FPeLEDs for display applications and beyond, such as the combination of the ECG signals and wearable multi-color FPeLEDs to achieve health monitoring, which is meaningful in health-care fields^[122].

Now the state-of-the-art FPeLEDs can reach maximum EQE of 24.5% and withstand more than 10 000 bending cycles at a curvature radius of 2 mm without severe degradation^{[34, 46]}. Nevertheless, the performance of FPeLEDs still lags far behind other FLEDs such as FOLEDs, and there exist huge challenges on the way to the commercialization process of FPeLEDs yet. The biggest challenge at present is to achieve the balance between optoelectronic and mechanical properties of devices. Previous research has systematically studied the PL and EL characteristics of MHPs and provided various effective methods to improve the performance of devices. Nevertheless, it remains room to explore and upgrade their mechanical properties. First, as the most important parts among all layers, perovskite film emitters are extremely fragile and verified as the weakest components in the FPeLEDs^[94]. An effective solution is to mix perovskite precursor with elastic polymers, which help absorb and relieve strain and protect the perovskite crystals from damage during deformation^{[36, 45, 130]}. Second, electrodes with poor mechanical flexibility and stability usually result in severe EL quenching due to the increased resistance after bending or stretching. It is needed to develop novel types of electrodes to address these problems, such as Maxine-based composite electrodes^[97]. Third, suitable CTLs can boost the transport efficiency of charge carriers, but there is little research about how the mechanical properties of CTLs affect the device performance.

To fulfill the commercialization of FPeLEDs, full color display, high color purity and resolution are required. However, the big gap between the green FPeLEDs and red and blue counterparts impedes the achievement of full color display. And the pixelated FPeLEDs is needed further study as well. In addition, other drawbacks such as device lifetime, air, moisture and thermal stability and toxicity of lead etc. should be considered^[2]. The device lifetime is still too short to support the long-term operation of displays. Because perovskite films are inherently sensitive to air, water and temperature, it is easy to cause device degradation when FPeLEDs are exposed and unprotected^[174]. Suitable encapsulation can improve the device stability in ambient environment. Given that the toxicity of heavy metal, the lead-based FPeLEDs are environment-unfriendly and possibly lead to the damage to people’s health system. Lead-free FPeLEDs are very significant for the commercialization of FPeLEDs. Although there exist many difficulties to boost the advances in FPeLEDs, it is undoubted that FPeLEDs are of great potential for next generation of display technology. It is convinced that the development of FPeLEDs will make a breakthrough soon.