Introduction

In recent years, perovskite solar cells have reached power conversion efficiencies (PCE) exceeding 27%1 along with prospects to achieve years of operational stability2. Among other factors, the quality of the active material is most crucial in obtaining a high PCE. A key challenge in the preparation of perovskite thin films from precursor inks is to control the growth of the perovskite grains. During thin film deposition, several processes take place simultaneously, including solvent evaporation, progressing supersaturation, nucleation, and crystal growth3. Even minor deviations from an original processing protocol can result in substantial changes in the structure and properties of the deposited material4. Therefore, considerable effort has been devoted to control the deposition and crystallization of perovskites5,6. Commonly, precursor solvent mixtures, in which at least one solvent is able to form stable intermediate complex structures with the perovskite precursors, are used in combination with solvent extraction methods like anti-solvent-, gas- or vacuum-quenching. These deposition techniques are used to control the morphology of the perovskite layers to reduce the process variables and thereby to ultimately improve reproducibility7,8,9,10,11,12,13. Larger perovskite grains reduce the density of grain boundaries in the final perovskite layer leading to a reduction of deep trap states and non-radiative recombination—key factors influencing solar cell device performance14,15,16. While there are also other means to mitigate the impact of grain boundaries17, reduction of the number of grain boundaries by increasing the grain size is the most straightforward and highly popular approach. Aiming to reduce losses and increase perovskite solar cell performance, in addition to the aforementioned varieties of film deposition, various additives may be added to the perovskite precursor ink to impact the resulting perovskite layer. Although the impact of solvent and additive engineering on the final perovskite layer is well documented, the underlying mechanisms governing the crystallization process are still subject of a vigorous debate18. In the context of crystallization, the impact of additives is often attributed to a retardation of the nucleation process, caused by a suspected increase in Gibbs free energy due to additive coordination to the lead core19,20,21,22,23,24,25. A frequently cited hypothesis is a heterogeneous nucleation evolving from Pb2+-MA+-I− clusters that coordinate into [PbI6]4− octahedra and eventually rearrange into corner-sharing structures that grow in size and finally become seeds for the crystallization of the perovskite phase26,27,28,29. The existence of colloidal complex structures consisting of the lead component as an acceptor and the polar aprotic solvent as donor molecules, already present in highly diluted perovskite precursor inks, is widely accepted and frequently shown in the literature by multiple methods like UV-Vis, X-ray scattering, and nuclear magnetic resonance (NMR)13,18,30,31,32,33. Substantial evidence confirms the ability of common Lewis-base solvents to form such complex structures with perovskite precursors26,34,35,36. This concept implies a strong impact of complex coordination on the precursor nucleation and crystallization process, as the size and density of perovskite seeds would determine the size and shape of the grains that are formed in the film. The fact that the lead cation site acts as an electron acceptor (i.e. Lewis-acid) further matches well with several reports stating the beneficial impact of various Lewis-base additives on the crystallization process20,23,33,37,38,39. While those observations are often presented as a chain of causality, a clear systematic link between liquid (ink) phase, nucleation, and grain size is still lacking in the community. So far, nucleation theory is not able to provide solid, generalizable predictions, which is why the community currently relies heavily on intuition and heuristic approaches for process optimization. For process engineering beyond heuristics that will be key for tailored additive design and upscaling, in-depth knowledge of the root cause of perovskite grain morphology is critically needed40,41.

In this work, we provide a solid concept for the mechanism of how many popular crystallization agents mediate the grain formation in the final perovskite layer and explain why predictions based on nucleation theory are unreliable and prone to error. To this end, we present a multi-facet approach, in which we study the full route of the perovskite formation process, i.e. from probing the precursor ink over in situ grazing incidence X-ray scattering (GIWAXS) during thin film deposition to the final films and their implementation into perovskite solar cells (Fig. 1a). While we are able to confirm and complement reports in the literature on colloidal lead complexes in the perovskite precursor ink that grow in size upon increasing the concentration of the ink, we present compelling evidence that the key impact of several popular additives unfolds during the annealing step, when solvent removal, nucleation and initial perovskite crystallization have already taken place. We model the process of grain growth by phase-field simulations, using a coarsening growth process limited by the ion mobility of the perovskite constituents across the grain boundaries. The ion mobility in turn, is mediated by the presence of the additive at the grain boundaries of the film. With ultraviolet photoelectron spectroscopy (UPS), density functional theory (DFT) calculation, and Fourier transform infrared spectroscopy (FTIR), we present evidence that coordination of the additive to lead sites, along with the opening of interfacial defect states and/or ion shuttling during annealing, are the underlying mechanisms driving additive mediated grain growth. We validate the generalizability of our hypothesis by testing various perovskite compositions and additives, as well as showing its applicability to complementary post-processing methods. We confirm that the proposed mechanism is not only viable to explain the effect of additives, but also to link the additive mediated perovskite formation to post-processing approaches like thermal hot-pressing, where the mobility of the ions is increased by elevated temperature.

Fig. 1: Solvent chemistry—solvent type and concentration.

a Schematic illustration of the perovskite deposition process accompanied by the measurement approach. b UV-Vis absorbance measurements on MAPbI3 precursor inks with different concentrations in DMF. Note that reflection was neglected and absorbance was calculated as 1 − T, where T is transmittance. UV-Vis for dimethylsulfoxid (DMSO) and n-methyl-2-pyrrilidone (NMP) are shown in Fig. S4. c Illustration of the lead complex evolution in dependency of concentration, as also proposed by ref. 30. d 207Pb NMR studies of MAPbI3 precursor inks based on dimethyl formamide (DMF), DMSO, and NMP in dependency of precursor concentration. The dotted line represents the solid-state MAPbI3 signal derived from ref. 50. Chemical shifts are reported on a ppm scale relative to the standard Pb(CH3)4. e Conductance study comparing stoichiometric MAPbI3 precursor to MAI dissolved in DMF, NMP, or DMSO at different concentration. The bottom panel shows the relative difference of conductance (G) between the dissociated MAI and the colloidal MAPbI3 ink, referenced by MAPbI3, calculated as ({\Delta }_{{{\rm{EC}}}}=\frac{{G}_{{{\rm{MAI}}}}-{G}_{{{{\rm{MAPbI}}}}_{3}}}{{G}_{{{{\rm{MAPbI}}}}_{3}}}), following the apparent evolution in charge-to-volume ratio in the colloidal ink with increasing concentration. For details about the conductance measurement setup see the Methods section as well as Fig. S5.

Our findings provide a decisive piece that complements the nucleation theory for perovskite thin film fabrication and bridges the gap between the precursor phase and the final film. Thereby, we take a crucial step beyond heuristics and enable revised and more targeted additive and crystallization engineering.

Results and discussion

Before we focus on the properties of the perovskite precursor, we briefly address processing parameters that are frequently mentioned to affect the grain size during perovskite film deposition29,42,43. Firstly, we deliberately varied the speed of solvent removal (i.e., the evaporation rate, Fig. S1 and Fig. S2), by selecting different deposition and quenching techniques for thin film formation (one step, gas-, and anti-solvent-quenching) with dimethylformamide (DMF) and dimethyl sulfoxide (DMSO), that have different vapor pressures, as the solvents for the perovskite precursor. Secondly, we deposited perovskite films (without any additive) on both hydrophilic and hydrophobic substrates (Fig. S3)25,29,44,45. As a proxy for the speed of solvent removal, we recorded the timespan until the perovskite phase formation was visible in the deposited thin films by a distinct color change (Fig. S1). Details on the determination of the grain size in the resulting layers by scanning electron microscopy (SEM) can be found in the Supplementary Note 1. Interestingly, even if the macroscopic morphology may be affected by the drying process, we obtained very similar grain sizes independent of the chosen technique and substrate. We consider the results of this set of initial experiments as a strong indication that our findings for the development of the final perovskite grain size will be transferable to both hydrophilic and hydrophobic substrate types, slow and fast deposition techniques and different popular perovskite solvent systems.

Probing colloidal lead complexes in the precursor ink

So far, most investigations of lead complexes in perovskite precursors have been conducted on diluted precursor inks using optical absorption techniques or X-ray scattering methods. Due to the highly absorptive nature of the lead-based precursors, optical measurements at high concentrations are experimentally challenging. Our absorption experiments reached a limit at around 0.3–0.5 M (M = mol l−1) (see Fig. 1b and Fig. S4). As visualized in Fig. 1c, absorbance allows for the identification of polynuclear lead complexes up to trimers in the precursor solution30, but in our case proved unpractical for typical perovskite precursor ink concentrations of 1 M or higher.

To probe precursor solutions at higher concentrations, we therefore employed 207Pb NMR and electrical conductance (EC) measurement techniques, which are not constrained by the above limitations. NMR is informative about the electronic environment of the lead core, and therefore its complex formation, while EC can be used to probe changes in the volume-to-charge ratio of charged species, which is an indicator for the growth of colloidal aggregates in the solution.

In 207Pb NMR, a high (more positive) chemical shift (referred to as “downfield” shift) indicates a diminishing electron density (de-shielding) around the Pb center while lower chemical shifts (referred to as “upfield” shift) indicate a higher electron density around the lead center. We observed only a single resonance peak from 207Pb in all inks, because the processes of lead complexation and complex dissociation are fast on the timescale of the 207Pb NMR measurement and thus the averaged electron density at the lead core is probed46,47.

To warrant generalizability, we used the three most common perovskite solvents DMF, DMSO, and NMP, all of which are Lewis-bases. DMSO has a slightly higher donor number—a common measure for Lewis basicity—of 29.8 kcal mol−1 in comparison to DMF and NMP with 26.6 kcal mol−1 and 27.3 kcal mol−1, respectively48,49. As evident by UV-Vis and NMR spectroscopy (Fig. 1b–d, Fig. S4), all three solvents formed complexes with the lead cation that evolved with precursor ink concentration, in line with earlier findings for diluted precursor inks13,30,32. As expected, due to its higher donor number, DMSO caused an overall higher electron density around the lead nuclei compared to NMP and DMF, as evident by NMR in Fig. 1d. Interestingly, when the concentration was increased, the chemical shifts of DMF and NMP on the one side and the DMSO shift on the other side approached each other on a seemingly asymptotic path. This hypothetical asymptote can be approximated at around 1400 – 1450 ppm and has a striking resemblance with the chemical shifts reported for solid-state MAPbI3 perovskite (i.e., 1423 ppm)50. Considering the averaging effect of 207Pb NMR, this trend can indicate the shift of the equilibrium of all lead complex species towards the formation of an increasing amount of poly-nuclear, perovskite-like species in the precursor ink, which have been predicted in previous works13,30.

Similarly, we find indications of growing lead-based complexes by EC measurements, where a comparison of solutions of the MAI salt with the perovskite precursor ink at similar concentrations revealed a reduction in conductance, which we can interpret as a decrease in charge-to-volume ratio indicating larger and/or less charged species in the ink (Fig. 1e, top). An overall higher conductance for precursors based on DMF in comparison to those based on DMSO can be attributed to different viscosities of the solvents, which directly impact the ion mobility51,52,53. We finally calculated the relative difference between the ink conductances employing either MAI and MAPbI3 in relation to the conductance of the MAPbI3 precursor ink ({\Delta }_{{{\rm{EC}}}}=\frac{{G}_{{{\rm{MAI}}}}-{G}_{{{{\rm{MAPbI}}}}_{3}}}{{G}_{{{{\rm{MAPbI}}}}_{3}}}) (Fig. 1e, bottom). In line with our interpretation of UV-Vis and NMR, the rising ΔEC with increasing concentration follows the reduction of charge-to-volume ratio, that is consequential from the growth of colloidal structures in the ink. Furthermore, a comparison of DMF and DMSO at lower concentrations, where the ΔEC differs more than at higher concentrations, agrees well with the asymptotic behaviour of the respective 207Pb NMR measurements. We speculate that for lower concentrations, the higher donor number of DMSO allows for a stronger dissociation of the PbI2 than DMF, leading to a ΔEC below zero, meaning that the conductance actually increases upon addition of PbI2. For higher concentrations, in all cases, colloid formation dominates the electronic precursor properties in both solvents. We also found a similar behavior for MAPbI3 in NMP as we show in Fig. S6.

Influence of additives on grain growth and complex formation

As the variety of additives that have been reported to act as so-called “crystallization agents” is very large, we selected one additive to conduct an in-depth case study and later on verify the general validity of our insights for other additives and perovskite compositions. Initially, we chose the Lewis-base thiourea, whose effect on perovskite crystallization is well reported and often related to its ability to interact with lead20,37,38,54,55, e.g. via the formation of hydrogen bonds56. Our decision for thiourea as a first case study was based on the fact that it has no secondary interference with the applied measurement techniques. Many other popular crystallization additives like MACl, Pb(SCN)2 or HI impair the complex measurements, as they shift the precursor stoichiometry (Pb(SCN)2, MACl), or add charged species by dissociation (MACl, HI). Altering the precursor stoichiometry generally impacts 207Pb NMR measurements, independent of the source of the shift (e.g., addition of excess MA, see Fig. S7), while additional ions impact conductance, which makes disentanglement of primary and secondary effects close to impossible in both cases. Thiourea dissolved in DMF, DMSO, or NMP proves to be non-conductive and does not impact the stoichiometry of the perovskite solutions. As we show in Fig. 2a, b, the addition of 0.1 M thiourea to the precursor ink resulted in a drastic increase of the grain size and degree of orientation of perovskite thin films (MAPbI3 in this particular example). We found the impact of thiourea to be largely independent of the solvent-system (Fig. S8) and the perovskite deposition technique (i.e. drying speed, Fig. S9) and therefore chose the DMF:NMP solvent system (7:3 volumetric ratio) and the gas-quenching protocol as a reference deposition procedure which is known for its low process variation9. To underline the relevance for applications, we fabricated perovskite solar cells and confirmed a performance improvement due to thiourea addition (Fig. 2c). Further data on the solar cells are summarized in Fig. S10 and Fig. S11. As solar cells without the additive also showed reduced operation stability, similar as described in earlier work57, we conducted fast hysteresis measurements (Fig. S12) revealing increased ionic losses in devices without thiourea addition, which are likely related to recombination at the higher number of grain boundaries16,58. We found similarly beneficial effects of the thiourea additive for perovskite solar cells with a FA0.94Cs0.06PbI3 composition (Fig. S13) as well as mixed halide perovskites, which we published earlier59,60.

Fig. 2: Impact of thiourea additive.

a Grain size distribution and scanning electron microscopy (SEM) images of MAPbI3 layers desposited from an DMF:NMP precursor ink by gas-quenching with a concentration of 1.0 M with and without (w/o) 0.1 M of thiourea additive. Topography recorded by atomic force microscopy and perovskite layers deposited from other solvent systems can be found in Fig. S8. b XRD patterns of respective layers with and without 0.1 M thiourea. c Solar cell characteristics employing an ITO/PTAA/MAPbI3/PCBM/AZO/Ag device stack using MAPbI3 with and w/o thiourea additive as active layers. Here, measurements in the stabilized state (after light soaking) are shown. Statistics, other solar cell characteristics and solar cells based on a FA0.94Cs0.06PbI3 active system with and w/o 0.1 M thiourea are shown in Fig. S9, Fig. S10, Fig. S11, and Fig. S12. d 207Pb NMR spectra showing the chemical shift of a 1.0 M MAPbI3 ink with and w/o addition of 0.1 M thiourea in common solvents DMF, DMSO, and NMP, as well as our reference solvent system DMF:NMP. e Electrical conductance (EC) measurements of concentration series of the same perovskite and solvent systems as probed by 207Pb NMR.

Despite such dramatic changes in the perovskite layers upon addition of thiourea, we could only detect minimal variations in the electronic environment of the lead nuclei by NMR spectroscopy, which are within the error margin of the highly sensitive 207Pb NMR measurement (Fig. 2d). Likewise, the impact of thiourea on the conductance of the respective inks proved to be negligible (Fig. 2e). Taken together, we conclude that if thiourea were to influence the complex and colloid formation in the precursor ink, it would be below the detection limits of our measurements. This finding is unexpected, because thiourea, as a sulfur-donor, has been predicted to possess a higher coordination affinity towards lead compared to oxygen donors such as DMF, NMP and DMSO34,61,62. Based on our combined data, we suspect that the decisive impact of the crystallization agent on the perovskite film morphology may occur later in the deposition process.

In situ GIWAXS with and without additive

In an effort to access the later stages of the perovskite deposition process (supersaturation, crystallization, and annealing), we constructed a setup that enables us to monitor the perovskite formation during spin-coating and different quenching processes by recording in situ GIWAXS data using synchrotron X-ray radiation (Fig. 3a)63,64. A comparison of perovskite layers annealed via infrared annealing or on a hotplate can be found in Fig. S14. Besides the apparent increase of grain size when employing the thiourea additive, also the texture of the perovskite film was strongly affected (Fig. S15). While for perovskite layers processed without the thiourea additive we found an essentially random orientation of the crystalline domains, layers processed with thiourea showed a distinct orientational order with the < 110 > direction perpendicular to the substrate surface. For coherently scattering crystallite domains with more than a few hundred nanometers in size, precise dimensions cannot be extracted reliably from the peak width as this is limited by resolution effects[65](https://www.nature.com/articles/s41467-025-65484-7#ref-CR65 “Scherrer, P. Bestimmung der Größe und der inneren Struktur von Kolloidteilchen mittels Röntgenstrahlen. Nachrichten von der Gesellschaft der Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse 98–100 http://resolver.sub.uni-goettingen.de/purl?GDZPPN002505045

(1918).“),66. Thus, for the thiourea additive we chose to use the distinct contrast in crystallite orientation as an additional indicator to monitor the dynamics of the grain formation process which is established simultaneously with the size of crystallites and, ultimately, grains. Along with the radial profiles of the measured reciprocal space maps (integration along the azimuthal angle φ, plot along the radial direction Qr, Fig. 3b), we plot the azimuthal profiles of the MAPbI3 < 110 > signal at Qr = 1 Å−1 (plot along φ, Fig. 3c)67. The corresponding time-resolved GIWAXS data is shown in Fig. 3d-i. The full reciprocal space maps at distinct points in time during perovskite formation can be found in Fig. S16 and S17. We want to note that a dynamic system of perovskite-like colloidal structures, where complexes continuously form and dissociate, differs substantially from a nanoparticle dispersion and does not imply crystallinity, which is in line with our observation by GIWAXS, showing no detectable crystalline features in the precursor ink (Fig. 3e, h, before the spin-coating is started). We speculate that these dynamic processes are also the reason why dynamic light scattering techniques, which use models of dispersed particles, are unreliable when probing perovskite precursors68,69.

Fig. 3: In situ investigations of the deposition process.

a Illustration of the measurement setup for in situ grazing incidence wide angle X-ray scattering (GIWAXS) characterization of the perovskite formation process. Exemplary plots for b radial and c azimuthal profiles that are plotted versus time in (e, f), and (h, i), respectively. Reciprocal space maps of MAPbI3 perovskite thin films deposited from precursors without (d) and with 0.1 M (g) of thiourea additive. In d, also the plot direction for the time-dependent plots is highlighted. Radial Intensities (azimuthally integrated) of reciprocal space maps obtained by in situ GIWAXS as a function of time (e, without thiourea, h, with thiourea). The decline of the intermediate (cross) and formation of the perovskite phase (circles) in both cases is visible. The orange asterisk marks PbI2. f, i Azimuthal angular intensity of the perovskite <110> signal (around Qr = 1 Å−1) as a function of time, representing orientational order (f, without thiourea, i, with thiourea). A contrast is visible only after the annealing step has started. Vertical lines in all panels indicate the start of spinning (dotted), gas-quenching (GQ, dot-dashed) and annealing (heat, dashed). For all samples, a solvent mixture of 7:3 DMF:NMP was used.

The first crystalline feature can be seen briefly after the gas-quenching started in both samples with and without thiourea. The signal at Qr = 0.6 Å−1 (crosses, Fig. 3e, h) indicates the PbI2 ⋅ NMP solvent complex structure, which has been reported to occur as an intermediate phase in perovskite formation (Fig. S18)70,71. The lower vapor pressure of NMP and its greater coordination affinity to the lead core compared to DMF34 renders it more likely to remain in the film during spin-coating and gas-quenching. Interestingly, as indicated by the emergence of the signal around Qr = 1 Å−1 (circles, Fig. 3e, h), a substantial conversion of this intermediate structure into a MAPbI3 perovskite phase takes place already during the gas-quenching process67. Azimuthal profiles reveal random orientation of the perovskite phase crystallites formed during the spin-coating and gas-quenching, regardless of whether thiourea was added or not (Fig. 3f, i; before dashed line). Upon annealing (after the dashed line), the situation changes completely, as the perovskite layer with thiourea undergoes a drastic reorientation, resulting in a thin film with highly oriented crystallites as the end product, while its counterpart without thiourea remains randomly oriented.

To illustrate the process, we plotted the intensity of the intermediate and the perovskite phase along with an indicator for the degree of order of the perovskite, represented by the quotient of scattering intensities from the perovskite phase at 90° and 45° azimuthal angle from Fig. 3f, i. Thereby, a representation of the sequence of processes during perovskite layer formation can be drawn, as shown in Fig. 4. On the one hand, the development of the intermediate and perovskite phase (top panels of Fig. 4b) clearly shows similarity in the initial crystallization kinetics with and without thiourea throughout the supersaturation, nucleation, intermediate phase conversion and the perovskite formation process. On the other hand, the degree of orientational order (bottom) reveals that the thiourea additive reorients the perovskite phase upon annealing. In line with est