Surface and grain boundary carbon heterogeneity in CH₃NH₃PbI₃ perovskites and its impact on optoelectronic properties

Organic-inorganic hybrid perovskites have been regarded as the next-generation solar technology with over 25% efficiencies being achieved with single-layer perovskite solar cells (PSCs) in the laboratory.¹ PSCs benefit from a combination of desirable properties, including high absorption coefficients,^2,3 low exciton binding energies,^4,5 low carrier recombination velocities,^6,7 and ambipolar long-distance charge transport.^8,9 For a typical PSC, the nature of the surface of the perovskite absorber layer is of critical importance for affecting efficient charge transport and achieving high PSC efficiency. As a consequence of the polycrystallinity, the surfaces of perovskite films are generally populated with a high density of defects, including under-coordinated atoms,¹⁰ halide vacancies,¹¹ and chemical impurities.¹² Numerous efforts have focused on minimizing the energy loss from nonradiative recombination concentrated at the interface, through interfacial engineering of the active layers.^3,13–17 The surface properties are also dictated by the strain and grain boundary energies.^18–22 Further improvement in device efficiency and expansion into other energy-related applications²³ requires a quantitative understanding of the surface properties and their influence on the local and global charge carrier behavior.

Within the surface region, grain boundaries (GBs) have been given much attention, though there is disagreement as to how they impact photophysical processes. For example, some studies report that GBs do not impact photogenerated carrier (hole) transport,^24–27 while other studies demonstrated the detrimental impact of GBs on charge carrier dynamics under simulated working conditions.^28,29 Due to the surface-to-volume (GB to crystal volume) ratio, it is generally believed that larger crystal dimensions lead to better performance, since GBs are discontinuities in the crystal structure. The chemical and physical characteristics intrinsically rely on the formation of adjacent grains, giving rise to heterogeneities in perovskite films. By manipulating the surroundings of GBs, which can be done by surface modification, it is possible to critically address the role of GBs.

In this report, we focused on the surface properties of perovskite films by modifying their microstructure using solvent annealing (SA), which has proven to be an effective and repeatable way to improve the film quality, both for organic and perovskite systems.^30,31 Methylammonium (MA) lead triiodide perovskite films were imaged by x ray photoemission electron microscopy (X-PEEM) and grazing incidence hard x ray diffraction (GIXD) at the Advanced Light Source (ALS) at the Lawrence Berkeley National Laboratory (LBNL). As a powerful and highly sensitive spectromicroscopy, X-PEEM, with a spatial resolution of ∼30 nm, was used to probe the surface. By using an energy analyzer, chemically selective images were obtained with morphologic and chemical contrast. The local chemistry was found to depend on the preparation conditions, e.g., the GBs of the SA sample were rich in carbon, while the thermally annealed sample (the control) had GBs with carbon-related contaminants. Using static and in situ GIXD, the formation of a liquid-like intermediate phase during SA was found that influenced the homogeneity of the local chemistry. The homogeneity of the chemistry eliminated surface states, improving the local and overall photophysical responses.

We used X-PEEM to resolve detailed real-space information on perovskite surfaces. Figure 1(a) shows a schematic of X-PEEM, where electrons from core levels are excited into unoccupied states after absorbing x ray photons with energies characteristic of the elements present in the sample, leaving empty core states (process 1). These core holes predominantly decay through Auger electron emission (process 2), generating a cascade of low-energy secondary electrons (SEs) through inelastic scattering of Auger electrons and photoelectrons (process 3). Some of the SEs with energies above the work function of the material from the topmost layers can escape into the vacuum without being captured and are detected by the X-PEEM optics. Consequently, a spatial map of the absorption in the near-surface region of the sample is obtained, with a probing depth of at most a few nm.³² 

Here, we focused on the strong absorption at C K-edge of MAPbI₃ films and obtained images with morphologic and chemical contrast. The scanning parameters and the absorption edges were chosen to acquire reliable image contrast. The detailed experimental method is included in the Materials and Methods of the supplementary material. Figures 1(b)–1(e) show the morphologic images of the SA samples acquired at 284.7, 285.7, 287.2, and 289.2 eV, respectively. These energies were chosen based on absorption resonances shown in Fig. 1(f) and 1(g), which were acquired by integrating the intensity of images taken at discrete energies around C absorption edges. It is seen that the free surface of perovskite films has mainly three chemical states (labeled as C1, C2, and C3), located at 285.5 (285.6) eV, 287.2 (287.2) eV, and 288.8 (289.1) eV for SA (control) sample, respectively. These peaks are consistent with the reported C 1s spectrum obtained by x ray photoelectron spectroscopy (XPS), where C1 was attributed to intrinsic carbon species in perovskites, and C2 and C3 were related to the methyl and carbonyl groups of residual N,N-dimthylformamide (DMF) molecules, which may retard charge transfer at the interface by the introduction of non-radiative recombination sites.^33–35 This carbon contamination was verified by comparing with the spectrum of the control, in which the intensity ratios of both C2 and C3 to C1 were greatly increased. The grains and boundaries from the morphology images shown in Figs. 1(b)–1(e) are easily resolved, indicating sufficient resolution of X-PEEM to image spatial heterogeneities. Two phases (the grain and GB regions) are shown in all images with the boundary area being brighter, consistent with the dominant morphologic contrast.³⁶ Some grains show additional contrast, indicating an inhomogeneity in the crystal grains.²⁹ The similar contrast both below and above the C K-edge indicates that there is no significant carbon segregation when comparing GB and grain interior (GI) regions. Unfortunately, we could not obtain distinguishable morphologic contrast for the control (shown in supplementary material Figs. S1 and S2), due to surface charging, indicating a number of defect states were located at the sample surfaces, especially at the GBs, since they appear brighter. The blurry images are consistent with the increased signals from carbon-related impurities.

To obtain a contrast of chemical origin, the spatial distribution of chemical states, the images above the C K-edge [Figs. 1(c)–1(e)] were normalized to that below the edge [Fig. 1(b)]. Figures 2(a)–2(c) show these images with chemical contrast, and Figs. 2(d)–2(f) are the corresponding line profiles. It is evident that the C1 state is concentrated not only at the GIs but more markedly at the boundaries. This observation is quite surprising since the GBs are the discontinuities between crystal grains, where high concentrations of imperfections, disorders, and chemical contamination would be expected to concentrate. This should lead to a depletion of intrinsic elemental/chemical species in this area. For example, inhomogeneities in the concentrations of Pb and I were previously found due to the morphology.³⁷ The observed strong C-rich nature suggests an integrity and functionality of MA molecules at the GBs, which would assist in healing deep states and screening carriers.^38,39 On the other hand, the C2 and C3 impurity states were also located in the two regions, in keeping with the above arguments. Combining the distribution of the three detectable states, it is reasonable to argue that the GBs in the SA sample would not impede charge transfer, as previously reported,^24–27 while the defective ones in the control originated from the chemical contamination and non-homogeneity of the chemistry.³⁷ The observation of GBs with C-rich nature and defective GBs is in keeping with the arguments in this regard, suggesting a non-uniformity of local chemistry at film surfaces.

The X-PEEM images indicate that the SA process changes the local chemistry. This post-annealing process is associated with structural evolution, indicating that the variations in surface chemistries arise from grain growth dynamics. To confirm this, scanning electron microscopy (SEM) and GIXD measurements were performed. As shown in the SEM images [Figs. 3(a) and 3(b)], the SA film had larger grains (∼400 nm) in comparison to the control (∼200 nm), consistent with previous reports claiming that polar solvent vapor leads to a recrystallization process by wetting the surface layer, facilitating grain growth and coalescence.^31,40 Interestingly, the textured grain surfaces (the striations), which balance the strain energy in the film,⁴¹ significantly decreased after SA, indicating a surface relaxation. The smoothing of grain surfaces implies a lateral diffusion of material on the surface, which will modify the local chemistry. The atomic force microscopy (AFM) images (shown in supplementary material Fig. S3), also verified the variation of surface textures.

GIXD patterns of the perovskite thin films (thickness ∼370 nm) under different preparation conditions at different angles of incidence (penetration depths) are shown in the supplementary material (Fig. S4). From the 1D diffraction profiles, a tetragonal perovskite phase was evident for both films. The stronger and sharper diffraction peaks of the SA film indicate enhanced crystallinity and crystal size of the polycrystal grains (Fig. S5). The redistribution of crystal orientation was confirmed by the azimuthal dependent intensity at q = 1 Å⁻¹, the (100) plane. As the incidence angle was increased to 0.3°, corresponding to a hard x ray absorption depth of ∼105 nm, the peak at $ϕ$ = 90° for SA samples emerged. Further increases to the penetration depth led to no further changes in orientation distribution [Fig. 3(d)]. These changes are clearly indicative of a depth-dependent orientation of the crystals. The control sample, on the other hand, was laterally more homogenous regardless of the depth [Fig. 3(c)].

The preferential orientation normal to the substrate was associated with the crystal formation starting from the bottom to the top during thermal annealing. The normal grain growth agrees with the grainy matrix of small units. When using solvent vapor, the surface is wetted and liquid-like so that growth into large units with a random orientation of grains occurs readily. To quantify the crystal orientation, orientation parameters (O.P.) were calculated as shown in Fig. 3(e). The O.P. of the control film was nearly a constant of 0.28, while SA film had a value of 0.32 in the surface region (∼60 nm) and 0.30 in the bulk part. According to the definition of O.P. (see the Materials and Methods in the supplementary material), a value of 0.33 indicates no preferential orientation. The variation of the O.P. indicates that the orientation decreases with penetration depth after SA treatment.

Lattice strain can be qualified by using a modified Williamson-Hall plot of the diffraction peaks broadening, as shown in Fig. 3(f). The reduction of microstrain was determined from the depth of the SA sample compared with the control, and was more pronounced (about 15% reduction from 5.5% to 4.7%) at the near-surface region. The strain gradient is consistent with recent observations.⁴² The stress relief confirmed the existence of a quasi-liquid surface previously described. As reported,¹⁸ strain increases with the elimination of boundaries during grain growth. Thus, the lattice strain cannot be relaxed if the solvent molecules only act as binders between adjacent grains. A liquid-like phase must be formed at the surface to allow mass flow to release the lattice strain²² and eventually change the chemical nature of the GBs. The existence of a liquid-like phase was further verified by the in situ GIXD measurements during post-SA where an intermediate non-perovskite phase was observed [Fig. 3(g)].^43,44 The static and in situ GIXD results confirmed the relationship between structural and chemical evolution during the post-annealing process and that the observed variation of surface chemical distribution was activated by generating a liquid-like surface allowing diffusion from the surface to the boundaries and stress relief, which resulted in a reduction of chemical contamination.¹⁸ 

Since a variation in the local chemistry was found, we probed the impact on the physical behaviors of perovskites. The X-PEEM results already showed that the C-rich nature of the surface prevented surface charging, indicating surface defect states can be significantly suppressed. This was confirmed by performing the incident direction-dependent photoluminescence (PL) to elucidate any contribution from defect traps that induce intraband energy levels.¹⁷ Figure S6 shows steady-state PL results by irradiating the sample from both the front (free surface) and back (ITO substrate) sides for SA and control samples, respectively. The excitation wavelength was 450 nm, with penetration of ∼50 nm. It showed that the PL emission from front excitation was redshifted in comparison to the back excitation (781 nm vs 777 nm). While for the SA sample, the PL spectra were nearly identical for both excitations with an emission at 782 nm. The relative intensity of PL emission was enhanced when solvent vapor was adopted. These PL measurements indicate that the SA sample was more defect-tolerant in comparison to the control that had considerable surface states.

To gain further insight on reduced surface states, Kelvin probe force microscopy (KPFM) measurements were performed using different intensities of light illumination. The corresponding contact potential differences (CPD) at different light intensities are shown in supplementary material [Fig. S7(a)–(f)]. The changes in the average CPD values for the two samples are shown in Fig. 4(a). The surface potential continuously increased with incident light intensity and tended to be saturated up to an illumination of 20 mW/cm², suggesting that the predominant surface states, if any, should act as electron traps, which resulted in free hole accumulation at the surface.⁴⁵ Compared with the control, the average CPD in the dark was significantly increased in the SA sample (>150 mV), meaning that the surface electron traps had been markedly eliminated, resulting in less hole accumulation due to surface states induced band bending, consistent with the observation from PL studies. As shown in Fig. 4(b), the difference of CPD between the two films (⁠$Δ$CPD) enlarged with increasing light intensity. One can clearly see that the difference between the CPD under illumination and in the dark was amplified for the SA sample, indicating photogenerated charge separation instead of a surface states related mechanism dominated the signal after SA treatment. It also implies that the SA sample had a stronger photovoltaic effect, which could help produce devices of a larger open-circle voltage.

Based on the provided results, we can conclude that the homogeneous distribution of local chemistry is responsible for the suppression of trap-mediated processes, resulting in enhanced PL and CPD values. When surface chemistry is not uniformly distributed, defective states of different forms can be formed. It has been widely accepted that Frenkel defects are predominantly accountable for non-radiative recombination resulting from charged vacancies of ions and corresponding interstitials.^46,47 The Frenkel pair defects and the elemental defects like Pb, I, and MA vacancies derived from them act as recombination centers, leading to PL quenching and reduction of conductivity. What's more, chemical impurities or contaminations could be induced to the vacancies to reach energy minimum through bonding, forming impurity states. Consequently, trap states passivation should be closely linked to the homogeneity of chemistry, promoted during the solvent annealing process. Through recrystallization induced by the liquid-like intermediate phase, chemical species are more evenly distributed, mitigating the movement of ions and the formation of excess ionic vacancies, interstitials, undercoordinated atoms, and degradation products, which are considered as defect sites involved in non-radiative recombination and poor stabilities the perovskites are subjected to. The passivation also prohibits the invasion of impurities, which may inhibit grain growth and reduce grain size.

As the carbon primarily originates from the MA molecules in the perovskites, the benefit from the liquid-like orientation of the organic cations can be anticipated. It has been widely recognized that the existence of MA rotation is responsible for many advantages of perovskites, such as healing deep trap states,³⁸ screening hot carriers,³⁹ and generating high dielectric constant.⁴⁸ Given our results, it is expected that photogenerated charge separation could more readily occur around the GBs of C-rich nature compared with the defective ones. This hypothesis was confirmed by the KPFM line profiles. Figures 4(c) and 4(d) are zoomed-in topographic images where the lines are labeled for the SA and the control sample, respectively. Figures 4(e) and 4(f) show the corresponding line profiles. The CPD difference between grain and GB became smaller with incident light (from ∼0.09 V to ∼0.05 V) for the SA sample, meaning the boundary area gained more photocarriers and contributed to the increase in CPD, while the local difference in the counterpart became slightly larger. The local potential differences demonstrate that C-rich GBs can intrinsically act as channels to efficiently separate and accumulate carriers rather than nonradiative recombination centers.

It can be concluded that surface chemical engineering critically affects structural and physical properties. From the structural and optical characterizations, it is known that there are depth-dependent microstrain and bandgaps.⁴² The mismatch of associated bandgaps between the surface region and bulk part is detrimental as it could facilitate surface/interfacial nonradiative recombination when the surface has a narrower bandgap and form unfavorable band alignment at charge-selective interfaces. Conversely, the more homogenous the surface chemistry, the more united the energetic level within the perovskite absorber, which is more favorable for charge carrier transport, resulting in better performance. By fabricating complete devices, we can see the enhanced performance of devices with SA (supplementary material Fig. S8). All PV parameters show a noticeable improvement, verifying the benefit on both charge carrier separation and extraction.

In summary, GB regions were found to be C-rich by X-PEEM by using a post-annealing treatment. The film formation was altered by generating an intermediate liquid-like surface phase that ensured mass flow. The uniform chemical components and properties throughout sample surfaces were associated with a significant release of the lattice strain in the film. The local dielectric response was enhanced, the films more defect-tolerant, and energetic bandgaps became more uniform along the film thickness, suppressing the carrier recombination within the absorber and at the interface. This study described the correlation between the chemical nature and resulting performance, and provided an explanation of the photophysical performance within the boundary area. This study underscores the critical importance to engineer local chemistry toward high-performance and stable devices.

See the supplementary material for details of experiments and additional supplementary figures.

Y.L., Q.H., and P.W. contributed equally to this work.

This work was finally supported by the Young 1000 Talent Program of China, the National Natural Science Foundation of China (NFSC) (Nos. 61805138, 51973110, 21734009, and 21905102), Beijing National Laboratory for Molecular Sciences (BNLMS201902), and the Center of Hydrogen Science, Shanghai Jiao Tong University, China. Q.H. and T.P.R. were supported by the Office of Naval Research, Materials Division, under Contract No. N00014–151-2244. Portions of the research were carried out at beamline 7.3.3 and 11.0.1.1 at the Advanced Light Source, Molecular Foundry, Lawrence Berkeley National Laboratory (LBNL), which is supported by the DOE, Office of Science, and Office of Basic Energy Sciences.

There are no conflicts of interest to declare.

The data that support the findings of this study are available from the corresponding author upon reasonable request.