Introduction

Perovskite solar cells (PSCs) have rapidly emerged as a high-efficiency photovoltaic technology, with their power conversion efficiency (PCE) soaring from 3.9% in 2009 to over 27% in just over a decade1,2,3,4,5. This remarkable progress is closely related to the interface engineering between charge transport layers and perovskite absorber layers, which profoundly affect interfacial energy and carrier dynamics6,7. SnO2 has emerged as a leading electron transport layer (ETL) in conventional (n-i-p) PSCs due to its high electron mobility, favorable conduction band alignment with perovskites, good optical transparency, and compatibility with low-temperature fabrication processes8,9. However, SnO2 thin films often suffer from intrinsic surface defects—primarily oxygen vacancies and adsorbed hydroxyl groups—that form during deposition and post-annealing processes10. These interfacial defects introduce deep trap states, disrupt energy-level alignment, and accelerate non-radiative recombination at the SnO2/perovskite interface, ultimately reducing device efficiency and stability11,12.

Recently, the interface dipole molecular passivation has been extensively explored as an effective strategy to modulate energy-level alignment, suppress defect-assisted recombination, and enhance charge extraction. Commonly employed interfacial modification materials in current research primarily include organic ammonium salts and carboxylic acid derivatives. Specifically, organic ammonium salts effectively passivate charged defects at buried interfaces through electrostatic interactions13,14,15, while carboxylic acid molecules optimize interface defects and energy level matching via chemical bonding with the SnO₂ surface16,17. Although these molecules have achieved notable success in interface engineering, the inherently low surface chemical activity of SnO₂ often results in the random and disordered distribution of interfacial dipolar molecules, leading to inhomogeneous passivation and ineffective interfacial dipole formation18,19,20,21,22. Therefore, constructing a robust bidirectional bridging interface with uniform molecular distribution and oriented growth at the buried interface remains a significant challenge.

Here, we report a bifunctional antisymmetric molecular trisodium N-(1-carboxyethyl)iminodiacetate (MGDA·3Na) with dual capabilities of directional alignment and stable anchoring to both SnO2 and perovskites. We found that the methyl side chain in the MGDA- anion induces intramolecular electronic polarization and dipole reorientation over the symmetric nitrilotriacetic acid trisodium salt (NTA·3Na) molecular, enabling oriented adsorption at the SnO2 surface and the buried interface of the perovskite layer. This allows the simultaneous efficient passivation of SnO2 surface defects and precise regulation of perovskite crystal growth, not only minimizing charge extraction barriers through energy level alignment but also templating the vertical growth of perovskite by regulating nucleation kinetics. As a result, we achieved a PCE of 26.43% in (0.05 cm2) small-area PSCs and 23.27% in 5 × 5 cm2 large-area modules, highlighting the strong potential for industrial-scale application. The device retains 96% of its initial efficiency after 2000 h in a nitrogen-filled glove box, 96% after 800 h at 55 °C/55% RH, and 99% after 800 h of continuous illumination at the maximum-power point.

Result and discussion

Mechanism of interface modification

Although common interface modifiers such as organic ammonium salts and carboxylic acid derivatives have achieved certain results in passivating defects and adjusting energy levels, they often struggle to achieve uniform distribution and ordered, directional arrangement of molecules at the SnO2 buried interface23. This disorder hinders the formation of a robust and uniform bidirectional bridging interface as well as the construction of effective interfacial dipoles. Therefore, it is crucial to design molecules with specific spatial structures to promote directional adsorption and uniform distribution. The symmetric NTA·3Na (Supplementary Fig. 1a), featuring a central nitrogen atom covalently bonded to three methylene-bridged carboxylate arms (-CH₂-COO-Na+), in which the three carboxylate groups of the NTA- ion are situated in the same plane. This structural configuration results in a dipole moment that is oriented perpendicularly to the molecular plane (Fig. 1a), with a magnitude of 2.1 Debye. This structural characteristic theoretically determines that the molecule tends to adopt a horizontally random aligned configuration at the interface. We employed Density Functional Theory (DFT) calculations to simulate the adsorption behavior of NTA- molecules on SnO2 and perovskite surfaces separately (Supplementary Fig. 2a–b). The results indicate that the three carbonyl groups of the NTA- ion are horizontally adsorbed on the SnO2 and perovskite surfaces through Sn-O and Pb-O bonds, respectively. Although this tridentate anchoring enables partial defect passivation, it concurrently occupies perovskite deposition sites, weakening the interfacial binding strength (Fig. 1b). Consequently, achieving ideal bidirectional molecular anchoring remains challenging. To overcome this limitation, we strategically introduced methyl groups to the side chain of NTA·3Na, designing the sterically structured MGDA·3Na (Supplementary Fig. 1b). This structural modification reconstructs the electron density distribution of the molecule, aligning the dipole moment parallel to the molecular backbone and increasing its magnitude to 4.2 Debye, which significantly enhances interfacial adsorption stability. Compared to NTA-, the electron-donating methyl group in MGDA- induces higher electron density in the directly connected carboxylate group (Fig. 1a), which preferentially binds to the SnO2 surface via Sn-O coordination. The remaining two carboxylates subsequently anchor the perovskite through Pb-O bonds, ultimately achieving a vertically bidirectional anchoring configuration (Fig. 1c, Supplementary Fig. 2c–d). Statistical analysis of adsorption energies for different molecular orientations (Fig. 1d) reveals that MGDA- predominantly adopts a vertically anchored configuration at the interface, in stark contrast to the planar adsorption tendency of NTA-, further confirming their distinct distribution patterns at the SnO2/perovskite interface. Moreover, the results demonstrate that MGDA- exhibits enhanced adsorption stability on both SnO2 and perovskite surfaces, which can be attributed to its larger dipole moment. Most importantly, conductive atomic force microscopy (c-AFM) measurements revealed that the average current on the SnO2 surface was 16.3 nA, which decreased to 13.4 nA for the SnO2/NTA- surface (Supplementary Fig. 3). In contrast, the MGDA--modified SnO2 exhibited a significantly higher average current of 109 nA (nearly 10 times greater) demonstrating its remarkable enhancement in conductivity24. This difference further confirms the distinct adsorption behaviors of NTA- and MGDA-—the random horizontal adsorption of NTA- ions impedes charge transport, while the vertical adsorption of MGDA- ions establishes an efficient pathway for charge migration.

Fig. 1: Molecule structure and interactions of NTA·3Na and MGDA·3Na with SnO2.
figure 1

a Chemical structure and electrostatic potential spectroscopy (ESP) images of NTA- and MGDA-. Molecular orientation of b, MGDA·3Na and c, NTA·3Na at the buried bottom interface. d Statistical analysis of DFT-calculated adsorption energies for NTA- and MGDA- molecules at the SnO₂/perovskite interface. e High-resolution XPS spectra of Sn 3 d core level of Ctrl-SnO2, NTA·3Na-SnO2, and MGDA·3Na-SnO2 films. f XPS spectra of Pb 4 f core level of buried-bottom perovskite interfaces with Control, NTA·3Na, and MGDA·3Na modifications. Source data are provided as a Source Data file.

Full size image

The surface chemical states of SnO2 were significantly investigated after modification with MGDA·3Na and NTA·3Na. The Sn 3 d signal of the pristine SnO2 film exhibit two characteristic peaks at 486.38 eV and 494.8 eV (Fig. 1e), corresponding to the Sn 3 d5/2 and Sn 3d3/2 of Sn4+, respectively25. Upon surface modification with MGDA·3Na and NTA·3Na, both Sn 3d peaks exhibit a slight shift to higher binding energies. Specifically, the Sn 3d5/2 peak shifts from 486.38 eV (Ctrl-SnO2) to 486.50 eV (NTA·3Na-SnO2) and 486.61 eV (MGDA·3Na-SnO2), indicating a more electron-deficient local environment around Sn atoms. This shift can be attributed to electron withdrawal from Sn via coordination with oxygen atoms of carboxyl groups26,27. Notably, the MGDA·3Na induces the most pronounced Sn 3d5/2 shift (0.3 eV), suggesting a stronger coordination interaction with the SnO2 surface. This is ascribed to the electron-donating effect of -CH3 substituent on the carboxyl group in MGDA·3Na, which enhances the Lewis basicity of the oxygen atoms and promotes stronger coordination with the surface Sn4+. This observation aligns perfectly with the DFT-calculated stronger adsorption energy of MGDA·3Na at the interface. In addition, the O 1 s spectra were analyzed to evaluate the density of oxygen-related defects. Asymmetric spectra profiles of O 1 s signal were observed in pristine SnO2, NTA·3Na-SnO2, and MGDA·3Na-SnO2 films (Supplementary Fig. 4), which can be deconvoluted into two peaks representing the lattice oxygen in SnO2 (OL) around 531 eV and the oxygen vacancies (OV) or surface-absorbed hydroxyl (OOH) at ~532 eV. Quantitative analysis revealed that MGDA·3Na modification significantly reduced the oxygen defect density, decreasing the (OV + OOH)/(OL + OV + OOH) ratio from 44.13% to 40.11%, while concomitantly increasing the lattice oxygen proportion from 55.87% to 59.89%28,29. The reduction of OV and OOH, along with the increase in the OL ratio, significantly suppresses oxygen vacancies, providing conclusive evidence of the surface passivation ability of MGDA·3Na30. Given the bidirectional coordination capability of MGDA·3Na and NTA·3Na to SnO2, their interaction with uncoordinated Pb²⁺ in the buried interface of the perovskite layer was evaluated. The perovskite layers were mechanically detached from the underlying SnO₂ layer and the Pb 4 f spectra exhibited two characteristic peaks located at 143.21 eV and 138.38 eV, corresponding to Pb 4f5/2 and Pb 4f7/2, respectively (Fig. 1f)31,32. These peaks shifted to higher values upon modification with NTA·3Na and MGDA·3Na, which can be attributed to the interaction between the carboxylate groups of the modifiers and uncoordinated Pb2+.

These interactions were further confirmed by Fourier-transform infrared (FTIR) spectroscopy (Supplementary Fig. 5), where a red shift in the C = O stretching vibration from 1572.5 to 1563.9 cm−1 was observed upon MGDA·3Na treatment of the SnO2 surface, indicating coordination between the carboxylate oxygen atoms and uncoordinated Sn4+. This coordination weakens the C=O bond through electron donation, consistent with the increased Sn 3 d binding energy observed in XPS analysis. Meanwhile, the Sn-O stretching vibration shifted from 1056.4 to 1045.0 cm−1 after MGDA·3Na treatment, further confirming the formation of Sn…O-C coordination bonds at the interface. For the perovskite layers, the Pb-I stretching vibration shifted from 1602.6 to 1598.3 cm−1 with the modification of MGDA·3Na, suggesting interaction between carboxyl groups of MGDA·3Na and uncoordinated Pb2+ at the perovskite surface through Pb…O coordination33,34 (Supplementary Fig. 6). This inference is corroborated by the positive binding energy shifts in the Pb 4 f XPS signals. These spectral changes provide molecular-level evidence for the bidirectional anchoring of MGDA·3Na at both SnO2 and perovskite interfaces.

Energy level alignment and charge transport

The impact of interfacial modification on the energy level alignment and electronic structure of SnO2-based ETL was further investigated, as shown in Supplementary Fig. 7.The pristine SnO2 shows an EF of −4.51 eV and a CBM of −4.07 eV (Fig. 2a), resulting in a relatively deep conduction band edge compared to the CBM of perovskite (−3.88 eV). After surface treatment with NTA·3Na and MGDA·3Na, an upward shift was induced in both EF and CBM. Specifically, after NTA·3Na treatment, EF and CBM were altered to −4.48 eV and −4.01 eV, respectively. Following MGDA·3Na treatment, EF and CBM were further shifted to −4.34 eV and −3.95 eV, respectively. This shift narrows the conduction band offset between the SnO2 and perovskite, which potentially facilitates electron extraction and suppresses interfacial recombination (Supplementary Fig. 8). Notably, MGDA·3Na-treated SnO2 shows the most favorable alignment, with an EF of −4.34 eV and a CBM of −3.95 eV, closely matching that of the perovskite. Moreover, both SnO2/NTA·3Na and SnO2/MGDA·3Na exhibited markedly increased surface potentials (Supplementary Fig. 9), which indicates that the MGDA·3Na layer establishes a stronger dipole field and reduces the interfacial work function, suggesting a more n-type character and improved electron extraction capability, consistent with the above UPS results35. Especially, SnO2/MGDA·3Na demonstrated a more homogeneous potential field distribution (Fig. 2b), suggesting effective passivation and improved energy level alignment with the perovskite. DFT simulations further elucidate the interfacial charge distribution characteristics. As shown in the differential charge density maps and z-axis charge line profiles (Fig. 2c and Supplementary Fig. 10), both NTA- and MGDA- form strong electronic coupling with the SnO2 substrate through their carboxylate groups. Notably, compared to the sharp charge density gradient at the NTA-/SnO2 interface (Δρmax = 0.45 e/Å3), the MGDA-/SnO2 contact region exhibits a more gradual charge density gradient (Δρmax = 0.24 e/Å3), indicating effective suppression of interfacial charge accumulation. This characteristic promotes highly delocalized electrons at the MGDA--modified SnO2 interface, inducing favorable charge redistribution and establishing continuous transport pathways, thereby significantly enhancing charge transport efficiency36. Moreover, the results revealed that the electron mobilities of MGDA·3Na-SnO2 and NTA·3Na-SnO2 reached 2.12 × 10−3 cm2·V−1·s−1 and 1.12 × 10−3 cm2·V−1·s−1 (Supplementary Fig. 11a), respectively, nearly an order of magnitude higher than that of the pristine film (9.37 × 10−4 cm2·V−1·s−1). Additionally, the conductivity of the modified devices increased from the pristine film of 2.55 × 10−6 S·cm−1 to 4.89 × 10−6 S·cm−1 for NTA·3Na-SnO2 and 7.07 × 10−6 S·cm−1 for MGDA·3Na-SnO2 (Supplementary Fig. 11b). The observed enhancements in both mobility and conductivity of the MGDA·3Na-SnO2 film suggest that the bridging effect of MGDA·3Na facilitates charge extraction and transport while suppressing charge accumulation at the SnO2/perovskite interface37.

Fig. 2: Energy levels and transport properties of Ctrl-SnO2, NTA·3Na-SnO2, and NTA·3Na-SnO2 thin films.
figure 2

a Energy levels of the Ctrl-SnO2, NTA·3Na-SnO2, NTA·3Na-SnO2, and PVK thin films. b Statistical analysis of surface potential distribution for Ctrl-SnO₂, NTA·3Na-SnO₂, and MGDA·3Na-SnO₂ thin films. c Charge density difference of NTA·3Na and MGDA·3Na assembled on the SnO₂ (001) surface. Photoluminescence (PL) mapping images of perovskite films deposited on d Ctrl-SnO₂, NTA·3Na-SnO₂, and MGDA·3Na-SnO₂ substrates. Carrier density extracted from charge extraction (CE) measurements under variable light intensity (0.001-1 sun) for devices with e Ctrl-SnO₂, NTA·3Na-SnO₂, and MGDA·3Na-SnO₂ electron transport layers. Source data are provided as a Source Data file.

Full size image

The average PL lifetime of MGDA·3Na-pero was significantly shortened to 68.43 ns compared to 114.29 ns and 127.87 ns for the NTA·3Na and unmodified samples. The shortened lifetime indicates that electrons can be effectively extracted at the SnO₂/perovskite interface and trap-assisted non-recombination pathways are well suppressed (Supplementary Fig. 12). The reduced carrier lifetime aligns well with the steady-state PL intensity (Supplementary Fig. 13), where the MGDA·3Na-pero exhibited the most pronounced PL quenching. Furthermore, PL and Time-resolved photoluminescence (TRPL) mapping results were highly consistent with the steady-state PL and TRPL measurements, revealing a more uniform PL distribution with reduced intensity in the modified samples (Fig. 2d and Supplementary Fig. 14). These results collectively demonstrate that MGDA·3Na, as an effective interfacial modifier not only lowers the energy barrier for interfacial charge transfer but also improves the quality of perovskite films by defect suppression. The ultrafast carrier extraction process from perovskite to the SnO2 ETL was investigated using transient absorption (TA) spectroscopy. In the two-dimensional TA contour plots, the Ctrl-pero, NTA·3Na-pero, and MGDA·3Na-pero thin films all exhibit a ground-state bleaching (GSB) peak at approximately 810 nm (Supplementary Fig. 15). The GSB signal is proportional to the photogenerated carrier density. For the SnO2/perovskite bilayer structure, the decay of GSB intensity directly reflects the number of carriers remaining in the conduction band and valence band. Notably, at the same delay time, MGDA·3Na-pero shows a significantly weaker GSB signal compared to Ctrl-pero and NTA·3Na-pero, which indicates a faster carrier extraction rate from the perovskite to the SnO2 ETL.

To gain in-depth insights into the performance enhancement mechanism through interface modification, we systematically investigated the charge dynamics of the processed devices using multiple characterization techniques. The voltage decay time of the control device was 70.95 μs, while the modified MGDA·3Na-PSCs and NTA·3Na-PSCs exhibited significantly prolonged decay times of 117.89 μs and 107.83 μs, respectively, indicating effective suppression of non-radiative recombination channels (Supplementary Fig. 16). Moreover, the pristine device had a decay lifetime of 2.23 μs, while MGDA·3Na-PSCs and NTA·3Na-PSCs exhibited faster decays with lifetimes of 1.10 μs and 1.54 μs, respectively (Supplementary Fig. 17), confirming enhanced charge carrier extraction efficiency. Furthermore, the modified devices consistently demonstrated higher charge collection efficiency across various illumination intensities, indicating reduced trap density and suppressed non-radiative recombination in MGDA·3Na-SnO2 devices38. Additional charge extraction (CE) studies were performed to evaluate carrier extraction from Ctrl-SnO2 devices, NTA·3Na-SnO2 devices and MGDA·3Na-SnO2 devices under pulsed light of different intensities (Fig. 2e). The MGDA·3Na-SnO2 devices consistently extracted more carriers than their Ctrl-SnO2 devices under all tested light intensities. Notably, the MGDA·3Na-SnO2 devices maintained high-efficiency carrier collection even under weak illumination conditions ( < 3.16% of 1-sun intensity). This observation provides direct evidence for the improved energy level alignment between the MGDA·3Na interlayer and perovskite active layer, which substantially enhances interfacial charge transport efficiency.

Crystallization of perovskite thin films

To elucidate the regulatory mechanism of MGDA·3Na on the buried perovskite interface through its bifunctional anchoring effect, we systematically investigated its induced modulation on perovskite crystallization behavior. The control perovskite thin film (Ctrl-pero) exhibited obvious pinholes and unreacted PbI2 residues at the buried interface (Fig. 3a). In contrast, the MGDA·3Na-treated sample (MGDA·3Na-pero) displayed enlarged and more uniformly distributed bottom grains, attributed to the dual-anchoring between carbonyl groups in MGDA·3Na and both the underlying SnO2 and overlying PbI2, which effectively guided the growth of high-quality perovskite films. Moreover, the Ctrl-pero exhibited randomly oriented columnar grains (Supplementary Fig. 18), while the MGDA·3Na-pero demonstrated conformal single-crystal growth throughout the entire film thickness. The improved crystalline quality is further evidenced by the uniform surface potential distribution in the bottom perovskite layer. The modified perovskite maintained higher surface potential (Supplementary Fig. 19), suggesting improved morphological and compositional homogeneity of the buried interface with more ordered crystal structure, thereby facilitating photogenerated carrier transport to the ETL39. Moreover, contact angle measurements revealed that the MGDA·3Na-modified SnO2 substrate exhibited significantly enhanced hydrophilicity, with the contact angle decreasing from 27.9° to 11.4°. This effectively suppressed nanobubble nucleation and pore formation during precursor solution deposition (Supplementary Fig. 20). This improvement in buried interface quality reduced defect density and minimized non-radiative recombination, as further evidenced by the enhanced PL intensity at the buried interface of MGDA·3Na-pero (Supplementary Fig. 21). Subsequent grazing-incidence wide-angle X-ray scattering (GIWAXS) measurements at an incident angle of 0.7° revealed strong diffraction rings at the scattering vector q ≈ 0.99 Å−1, corresponding to the (100) crystallographic plane, for all samples (Ctrl-pero, NTA·3Na-pero, and MGDA·3Na-pero) (Fig. 3b). One-dimensional integration analysis of the GIWAXS data demonstrated that the (100) diffraction peak intensity of MGDA·3Na-pero increased by two-fold compared to the control, accompanied by a reduction in full width at half maximum (FWHM) from 12° to 9° (Supplementary Fig. 22). This result is consistent with the X-ray Diffraction (XRD) test results (Supplementary Fig. 23), confirming the significant improvement in crystallinity and orientation ordering of the MGDA·3Na-treated perovskite, thereby validating the vertically preferential growth induced by MGDA·3Na. It is noteworthy that buried interface defects not only induce lattice mismatch but also introduce residual stress in perovskite films, leading to the accumulation of lattice strain (ε)40. Through variable-incidence-angle GIWAXS analysis, we quantitatively characterized the strain evolution at buried interfaces (Supplementary Fig. 24). When the incidence angle was increased from 0.1° to 0.7°, the control sample exhibited a pronounced 0.32° shift in the (001) diffraction peak, corresponding to a substantial lattice strain of ε = 0.64%. In marked contrast, the NTA·3Na-pero and MGDA·3Na-pero samples demonstrated significantly reduced peak shifts of merely 0.04° (ε = 0.08%) and 0.01° (ε = 0.02%), respectively. These striking differences unambiguously demonstrate that the bidirectional anchoring effect of MGDA·3Na effectively alleviates interfacial residual stress, thereby establishing an optimal structural framework for highly efficient charge transport.

Fig. 3: Characterizations of the perovskite films.
figure 3

a Cross-sectional SEM images of the buried interface in surface-modified perovskite films. b GIWAXS patterns of perovskite thin films with modified interfaces. c XRD intensity profiles of interface-modified perovskite thin films measured at six selected areas on 5 × 5 cm substrates (see inset in Fig. S32). Source data are provided as a Source Data file.

Full size image

To quantitatively evaluate the regulatory effect of MGDA·3Na bifunctional anchoring on defect states in perovskite films, we fabricated electron-only devices [FTO/SnO2 (20 nm)/perovskite (750 nm)/PCBM (30 nm)/Au (100 nm)] (Supplementary Fig. 25). Compared to the Ctrl-SnO2 devices (VTFL = 0.410 V), NTA·3Na-SnO2 and MGDA·3Na-SnO2 devices exhibited significantly reduced VTFL values (0.277 V and 0.137 V, respectively). The calculated trap density (Nt) revealed that modified samples showed Nt reduction from 4.33 × 1015 cm−3 (Ctrl-SnO2 devices) to 2.297 × 1015 cm−3 (NTA·3Na-SnO2 devices) and 1.37 × 1015 cm−3 (MGDA·3Na-SnO2 devices). Notably, the MGDA·3Na-modified sample displayed more pronounced Nt reduction, indicating stronger passivation effect at the SnO2/perovskite buried interface, consistent with previous morphological characterization results. Further mechanistic investigation through DFT calculations revealed that in the unmodified FAPbI3 lattice the formation energy of a perfect crystal is −223.48 eV, whereas Pb- and I-vacancy defects stabilize to markedly lower energies of −239.72 eV and −236.70 eV (Supplementary Fig. 26a–c), respectively. This thermodynamic preference for defect formation undermines film quality. By contrast, the formation energy of MGDA- modified perfect FAPbI3 lattice was −248.48 eV, significantly lower than that of defect structures containing Pb vacancies (−243.27 eV) and I vacancies (−242.19 eV) (Supplementary Fig. 26d–f). This energy difference clearly demonstrates that MGDA- modification effectively increases the formation energy barrier of defects, thermodynamically suppressing intrinsic defect formation and providing theoretical foundation for high-quality perovskite film fabrication. It is noteworthy that beyond the coordination passivation by MGDA- anions, DFT calculations indicate the free Na+ cations in MGDA·3Na molecules can effectively fill Pb vacancy defects at the buried interface due to their small atomic radius (Supplementary Fig. 27)41. This is further supported by XPS depth profiling (Supplementary Fig. 28). After MGDA·3Na treatment, a pronounced Na 1 s signal is detected at the buried perovskite interface, and this peak shifts to higher binding energy compared with that of pristine MGDA·3Na, directly confirming Na+-mediated passivation. This finding is highly consistent with the observed trap density reduction in SCLC measurements, providing atomic-scale insight into MGDA·3Na’s multi-functional interfacial modification mechanism.

Subsequently, we fabricated 5 × 5 cm2-scale perovskite films to investigate the potential of MGDA·3Na modification for large-area PSCs manufacturing. MGDA·3Na-pero exhibited consistent diffraction peak intensities across all regions without significant peak shifts (Fig. 3c). Comprehensive characterization through normalized XRD intensity distribution (Supplementary Fig. 29a) and full width at half maximum (FWHM) analysis (Supplementary Fig. 29b) demonstrated that MGDA·3Na-pero showed more uniform and narrower FWHM distribution (0.07–0.08) compared to Ctrl-pero (0.08–0.12) and NTA·3Na-pero (0.08–0.11), indicating improved crystalline orientation stability and grain size uniformity. This enhanced structural homogeneity can be attributed to the synergistic coordination between amino and carbonyl groups in MGDA·3Na molecules, which enables directional anchoring control of perovskite nuclei. Such molecular-level regulation effectively suppresses random nucleation during solution processing, promoting ordered crystal growth along preferential orientations and thereby overcoming the size limitations inherent to conventional perovskite films. Furthermore, PL mapping (Supplementary Fig. 30 and 31) confirmed enhanced spatial uniformity in MGDA·3Na-pero, exhibiting minimal PL intensity variation compared to Ctrl-pero, which demonstrates the effectiveness of MGDA·3Na modification for producing high-quality large-area perovskite films. At the module level, the 5 × 5 cm2 MGDA·3Na-device displays the most uniform emission without local quenching or dark shunting spots, whereas both Ctrl-device and NTA·3Na-device show noticeable brightness heterogeneity (Supplementary Fig. 32). This direct comparison confirms that MGDA·3Na enables homogeneous interfacial molecular coverage, thereby supporting the reliability of large-area perovskite photovoltaic modules.

Device performance

We prepared PSCs with a planar structure of FTO/Ctrl-SnO2, NTA·3Na-SnO2 and MGDA·3Na-SnO2 (20 nm)/perovskite ( ~ 750 nm)/spiro-OMeTAD (80 nm)/MoO3 (5 nm)/Au (100 nm). The champion PCE reached for the MGDA·3Na-SnO2 device was 26.43% with a VOC of 1.187 V, a short-circuit current density (JSC) of 26.06 mA/cm2, and a fill factor (FF) of 85.47% (Fig. 4a). This performance represents a significant enhancement compared to the control device with 24.20% efficiency and the NTA·3Na-SnO2-based device with 25.54% efficiency. The JSC values calculated by EQE integral are 25.58 mA/cm2 (MGDA·3Na-PSCs), 25.03 mA/cm2 (NTA·3Na-PSCs), and 24.57 mA/cm2 (Ctrl-PSCs), closely matching the J-V curve results (Supplementary Fig. 33). Statistical analysis of the three devices shows that the MGDA·3Na-SnO2 device outperforms the NTA·3Na-SnO2 and Ctrl-SnO2 devices in all photovoltaic parameters (Supplementary Fig. 34). In addition, the MGDA·3Na-treated devices exhibit a hysteresis effect index (HEI) of merely 3.8% (Supplementary Fig. 35), markedly lower than that of the control devices (10.4%) and NTA·3Na-SnO2 devices (6.4%). This enhanced performance stems from the directional dual-anchoring of MGDA·3Na, which simultaneously passivates SnO2 oxygen vacancies and perovskite iodine vacancies, suppresses ion migration, and minimizes interfacial charge accumulation—collectively leading to a pronounced reduction in J-V hysteresis. This is due to the enhanced electron mobility and optimized energy level matching provided by the MGDA·3Na modification layer. Subsequently, we fabricated a perovskite solar module (PSM) comprising six subcells on a 5 × 5 cm2 substrate (active area: 12.6 cm2) (Supplementary Fig. 36). The MGDA·3Na-SnO2 device achieved a champion PCE of 23.27% (certified 21.22%, Supplementary Fig. 37), compared to 21.96% for the control module and 22.34% for the NTA·3Na-SnO2 module (Fig. 4c). Consistent with the trend observed in small-area devices, the statistical analysis of 12 independent 5 × 5 cm2 modules for each of the three systems further verified the reproducibility of the MGDA·3Na modification strategy at the module scale (Supplementary Fig. 38).

Fig. 4: Device performance.
figure 4

a Current density-voltage characteristics of the best performance of the Ctrl-SnO2, NTA·3Na-SnO2, and MGDA·3Na-SnO2 devices. b Statistical of PCE of the Ctrl-SnO2, NTA·3Na-SnO2, and MGDA·3Na-SnO2 devices. c Champion current density-voltage (J-V) characteristics of 5 × 5 cm Ctrl-SnO₂ and MGDA·3Na-SnO₂ devices. d Stability of unencapsulated Ctrl-SnO₂, NTA·3Na-SnO2, and MGDA·3Na-SnO₂ devices stored in a N2-filled glove box in the dark. e Thermal and moisture stability (55 °C and 55% R.H.) of encapsulated Ctrl-SnO₂, NTA·3Na-SnO2, and MGDA·3Na-SnO₂ devices under a white LED lamp (AM 1.5 G illumination). f Operational stability of unencapsulated devices for Ctrl-SnO₂, NTA·3Na-SnO2, and MGDA·3Na-SnO₂ devices under a white LED lamp (AM 1.5 G illumination) in a N2-filled glovebox. Source data are provided as a Source Data file.

Full size image

The improvements in VOC and FF can be attributed to favorable charge extraction/transport behavior and suppressed non-radiative recombination. We determined the built-in potential (Vbi) through Mott-Schottky analysis, where a higher Vbi value facilitates efficient extraction of photogenerated carriers by enhancing the built-in electric field (Supplementary Fig. 39). The elevated VOC in MGDA·3Na-SnO₂ device is further corroborated by the increased Vbi value of 1.01 V observed in these devices, compared to 0.82 V recorded for Ctrl-SnO2 devices. Notably, light-intensity-dependent VOC measurements revealed that the ideality factor (nid) of MGDA·3Na-SnO2 devices decreased to 1.41 from 1.75 in Ctrl-PSCs, indicating effective mitigation of non-radiative recombination losses upon MGDA·3Na treatment (Supplementary Fig. 40). This improvement is further reflected in the reduced leakage current observed in MGDA·3Na-SnO2 devices (Supplementary Fig. 41). Additionally, MGDA·3Na-SnO2 devices exhibited lower series resistance (Rs) and higher recombination resistance (Rrec), demonstrating that the MGDA·3Na-modified SnO2 enhances charge transport while suppressing non-radiative recombination (Supplementary Fig. 42).

We performed stability tests to assess the interface treatment durability under humidity, heat and light. Remarkably, the MGDA·3Na devices retained 96% of their initial power conversion efficiency (PCE) when stored in a nitrogen atmosphere (25 °C, RH < 5%) for 2000 h, whereas the control and NTA·3Na devices degraded to 76% and 83% of their initial performance, respectively. (Fig. 4d). Under accelerated damp-heat testing conditions (55 °C, RH 55%), the encapsulated MGDA·3Na devices maintained 96% of their initial PCE after 800 h, demonstrating significantly enhanced stability compared to the control group (68%) and the NTA·3Na devices (75%) (Fig. 4e). Continuous maximum-power-point (MPP) tracking tests revealed that after 800 h the control device dropped by 41%, the NTA·3Na device by 20%, while the MGDA·3Na device maintained 99% of its initial efficiency (Fig. 4f). This stability likely stems from the high-quality buried interface and interface passivation effect of the directed anchoring strategy of MGDA·3Na, enhancing the overall performance and reliability of n-i-p PSCs.

Methods

Materials

SnO2 colloid precursor (tin (iv) oxide, 15% in H2O colloidal dispersion) was purchased from Alfa Aesar. PbI2 (Lead iodide); Spiro-OMeTAD (2,2’,7,7’-tetrakis(N,N-di-p-methoxy-phenylamine)-9,9’-spirobifuorene) were purchased from Advanced Election Technology Company in China. FAI (formamidinium iodide), MACl (mthylammonium chloride) was purchased from Great Cell. DMSO (dimethyl sulfoxide), DMF (N,N-dimethylformamide), t-BP (4-tert-butyl pyridine), CB (chlorobenzene), OAI (n-Octylammonium Iodide), LiTFSI (lithium bis(trifluoromethanesulfonyl)imide), ACN (acetonitrile) and IPA (isopropyl alcohol) were purchased from Sigma-Aldrich. NTA·3Na (Nitrilotriacetic acid (trisodium salt)), MGDA·3Na (Trisodium N-(1-Carboxylatoethyl)iminodiacetate Hydrate) were purchased from Aladdin.

Device fabrication

FTO substrates were ultrasonic cleaned sequentially in deionized water and isopropanol for 30 min respectively and dried by a compressed nitrogen gun. After 30 min UV-Ozone surface treatment, we deposited SnO2 ETL by spin coating a 1:3 diluted SnO2 nanoparticle water solution at 4000 rpm for 30 s, followed by annealing at 150 °C for 20 min in air. Afterwards, a thin dipole layer (NTA·3Na or NTA·3Na) was deposited by spin coating 2 mg ml-1 deionized water solution at 4000 rpm for 30 s and annealed at 100 °C for 10 min. The perovskite FAPbI3 active layer precursor was prepared by mixing 1.81 M PbI2, 1.8 M FAI, 0.42 M MACl, and dissolved in anhydrous DMF and DMSO with the volume ratio of 8:1 mixed solution. For the recipe with 1 mg mL-1 Potassium Iodide (KI, ≥99%, Sigma-Aldrich) was added into the perovskite precursors as additive. The perovskite solution was then deposited via spin coating at 5000 rpm for 30 sec (2,000 rpm ramp). Within 10 s of the 5000-rpm spinning, 1 mL of diethyl ether as the antisolvent was deposited onto the film. After spin-coating, the perovskite film was then annealed at 120 °C for 40 min in ambient air. Then the 5 mg mL−1 OAI solution in IPA was spin-coated onto the perovskite surface at 5000 rpm and annealed at 100 °C for 1 min. Later, 72.3 mg/mL spiro-OMeTAD in 1 mL CB and doped with 30 mL t-BP and 25 μL LiTFSI salt in acetonitrile (520 mg mL-1) was deposited at 3000 rpm for 30 s. Metal electrode of 80 nm Au was deposited on hole transporting layer through thermal evaporation method under a vacuum degree higher than 5 × 10−4 Torr to accomplish the solar cell fabrication. A 0.05 cm2 shadow mask was used to define the effective working area of the solar cells.

Perovskite solar mini-module fabrication: 5 cm×5 cm

The modules consisted of six solar cells connected in series using interconnects P1, P2, and P3 laser structuring, respectively (Fig. S31a). The P1 line is etched on the FTO electrode to define individual cell units. The P2 channel, serving as the interconnection for series connection of cells, is etched after the Spiro-OMeTAD oxidation is completed, ensuring effective electrical connections between each cell unit. The P3 scribe is then performed after all preceding steps are completed, further dividing the device to form the final independent modules. It was then cleaned and deposited with an SnO2 layer as described above. The large-area perovskite thin films were prepared by the blade coating of the perovskite precursors and subjected to a vacuum-flash process and subsequent annealing. A 20 ~ 30 μL droplet of perovskite precursor was added into the gap ( ~ 150 μm) between the blade and substrate, and then the blade was moved on with a speed of ~50 mm s-1. Other procedures were consistent with small-size perovskite solar cell.

Film characterization

Fourier Transform Infrared (FTIR) spectroscopy was characterized by FT/IR-6100 (Jasco).

X-ray photoelectron spectroscopy (XPS) were conducted with a Thermo Fisher Scientific K-Alpha by Al Kα X-ray source.

Ultraviolet photoelectron spectroscopy (UPS) was measured by Thermo ESCALAB 250 with a non-monochromated He Iα photon source (hν = 21.22 eV).

Scanning electron microscopy (SEM) analysis was performed on a SU8010 electron microscope. SEM images were captured using a 5-kV acceleration voltage and an aperture size of 20 μm.

Grazing-incidence wide-angle X-ray scattering (GIWAXS) was performed at the BL14BL beamline of the Shanghai Synchrotron Radiation Facility (SSRF) using X-ray with a wavelength of 1.24 Å.

Space-charge limited current (SCLC) measurements were used to determine the hole mobility of pristine and doped ETL films on an elec-only device (FTO/Ctrl-SnO2, NTA·3Na-SnO2 and MGDA·3Na-SnO2/Spiro-OMeTAD/MoO3/Au). The MoO3 and Au layers were all deposited by thermal evaporation.

Steady-state photoluminescence (PL) spectra and Time-resolved photoluminescence (TRPL) spectra were obtained by FLS1000. The excitation wavelength was set as 520 nm. The TRPL decay data were modeled by a biexponential formula:

$$Y={A}_{1}\exp \left({{{\rm{\hbox{--}}}}}t/{\tau }_{1}\right)+{A}_{2}\exp \left({{{\rm{\hbox{--}}}}}t/{\tau }_{2}\right)$$
(1)

Device characterizations

Charge extraction (CE) was performed using PAIOS system under different illumination intensity (1– 100 mW cm−2) with an illumination duration of 100 us. Integrating the extraction current over time yields the extracted charge. The extracted charge carrier density nCE is then calculated according to the following equation:

$${nCE}=\frac{1}{{L}_{q}}\left(\int j\left(t\right)d\left(t\right)-\left({V}_{a}-{V}_{e}\right){C}_{{{{\rm{geom}}}}}\right)$$
(2)

where L is the perovskite thickness, q is the unit charge, t is the extraction time, j(t) is the transient current density, Cgeom is the geometric capacitance, Va is the voltage applied prior extraction (in most cases Voc), and Ve is the extraction voltage. The charge on the capacitance needs to be subtracted17 as only the charge carrier density inside the bulk is of interest.

Transient photocurrent (TPC) measurement was performed with a system excited by a 532 nm (1000 Hz, 3.2 ns) pulse laser.

Transient photovoltage (TPV) measurement was performed with the same system excited by a 405 nm (50 Hz, 20 ms) pulse laser. A digital oscilloscope (Tektronix, D4105) was used to record the photocurrent or photovoltage decay process with a sampling resistor of 50 Ω or 1 MΩ, respectively.

Photo current-voltage (J-V) curves were measured by using 2400 Series Source Meter (Keithley Instruments) under a SS-F5-3A solar simulator (AM 1.5 G, 100 mW cm-2) (Enlitech) calibrated by a NREL standard Si cell, and no additional UV filter equipment was used. The measurements were carried out with the devices inside the glove box ( < 0.1 ppm O2 and H2O). To ensure the accuracy of the Jsc measured from J-V scans, a mask with an aperture area of 0.05 cm2 was covered during the measurement. The J-V curves were scanned by reverse (forward bias (1.3 V) → short circuit (−0.2 V)) or forward (short circuit (−0.2 V) → forward bias (1.3 V)) scan with scan rate of 50 mVs−1, and 1 ms delay time.

Stability tests

In order to test the stability of the device, we placed the device without any encapsulation in a N2-filled glovebox or for more than 2800 h. The encapsulation of PSCs devices is using EVA as the encapsulant and using glass as the cover. For hygrothermal stability studies, the devices are stored in a temperature and controlled-humidity cabinet with a transparent glass door. The internal relative temperature and humidity were controlled at the required 55 ± 1 °C and 55 ± 3% humidity. The J-V curves were recorded with a Keithley 2400 source meter and under one sun illumination (AM 1.5 G, 100 mW cm−2) equipped with 425 W collimated Xenon lamp (EnliTech SS-F5-3A), in which the light intensity was calibrated by the NREL certified silicon solar cell. The operational stability tests were carried out at the MPPT for the unencapsulated devices under AM1.5 illumination (100 mW cm−2) in atmosphere N2 at 30 ± 5 °C. The voltage at the MPPT was automatically applied, and the power output of the devices was tracked.

DFT calculations

Quantum chemistry calculations were carried out using the Gauss View 6.0 and Gaussian 16 computational software package. For the initial geometric structures of MGDA- and NTA-, the B3LYP-D3 functional combined with the 6-311 + G(d,p) basis set was selected for geometry optimization. And frequency calculation verification was carried out at the same level.

The First-principles calculation related to adsorption energy was carried out using the VASP(Vienna Ab-initio Simulation Package) 5.4.4 calculation software function package42. The generalized gradient approximation is achieved by using the Perdew-Burke-Ernzerhof (PBE) functional of the projection augmentation wave method43 and the plane wave energy cutoff of 450 eV. The SIGMA parameter is set to 0.1 and the maximum number of electron steps is set to 200. In the iterative solution of the Kohn-Sham equation, the energy criterion for structural optimization is set at 10–5 eV, and the energy criterion for the static self-consistent calculation of SCF is set at 10 − 6 eV, the step size is set to 0.5, the maximum ion step number is set to 500, and the maximum force convergence tolerance is set to 0.05 eV/Å. On this basis, considering the influence of van der Waals forces, the exchange correlation term was modified to 11. The K-point setting for bulk phase structure optimization adopts Gamma Scheme grid selection, which is set to 3 × 3 × 3. The structure optimization of the adsorption model is set to 3 × 2 × 1. To avoid periodic influences, a vacuum layer thickness greater than 15 Å is set. The adsorption energy is calculated according to the following formula14:

$${E}_{{ads}}={E}_{{ionslab}}-{E}_{{ion}}-{E}_{{slab}}$$
(3)

Defect formation can be calculated according to the following formula44:

$${E}_{{ads}}={E}_{{defect}-{slab}}-{E}_{{perfect}-{slab}}+{\sum }_{i}{n}_{i}{\mu }_{i}$$
(4)

Here, Edefect-slab represents the total energy of the defective surface, Eperfect-slab represents the total energy of the intact surface, ni represents the number of defect atoms, and μi represents the chemical potential of the defect atoms.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.