Report

Impact of microstructure on local carrier lifetime in perovskite solar cells

See allHide authors and affiliations

Science  08 May 2015:
Vol. 348, Issue 6235, pp. 683-686
DOI: 10.1126/science.aaa5333

Going toward the grains

Great strides have been made in improving the efficiency of organic-inorganic perovskite solar cells. Further improvements are likely to depend on understanding the role of film morphology on charge-carrier dynamics. de Quilettes et al. correlated confocal fluorescence microscopy images with those from scanning electron microscopy to spatially resolve the photoluminescence and carrier decay dynamics from films of organic-inorganic perovskites. Carrier lifetimes varied widely even between grains, and chemical treatments could improve lifetimes

Science, this issue p. 683

Abstract

The remarkable performance of hybrid perovskite photovoltaics is attributed to their long carrier lifetimes and high photoluminescence (PL) efficiencies. High-quality films are associated with slower PL decays, and it has been claimed that grain boundaries have a negligible impact on performance. We used confocal fluorescence microscopy correlated with scanning electron microscopy to spatially resolve the PL decay dynamics from films of nonstoichiometric organic-inorganic perovskites, CH3NH3PbI3(Cl). The PL intensities and lifetimes varied between different grains in the same film, even for films that exhibited long bulk lifetimes. The grain boundaries were dimmer and exhibited faster nonradiative decay. Energy-dispersive x-ray spectroscopy showed a positive correlation between chlorine concentration and regions of brighter PL, whereas PL imaging revealed that chemical treatment with pyridine could activate previously dark grains.

As active layers in solar cells, organic-inorganic perovskites (1, 2) combine the promise of solution processing (3, 4) with the ability to tailor the band gap through chemical substitution (57), yielding solar cell power conversion efficiencies as high as 20.1% (8). Concomitant with their photovoltaic performance, perovskites also exhibit high fractions of radiative recombination, with apparent carrier lifetimes of 250 ns or longer (9, 10), and are challenging the dogma that solution-processed semiconductors inevitably possess high densities of performance-limiting defects. Ensuring that all recombination is radiative is critical for approaching the thermodynamic efficiency limits for solar cells and other optoelectronic devices (11).

Carrier recombination lifetimes measured by photoluminescence (PL) are commonly taken as a hallmark of perovskite film quality, with longer decay lifetimes used as indicators of better-performing materials (9, 10, 1214). Carrier recombination kinetics have been described as a combination of trap-assisted, monomolecular (first-order), and bimolecular (second-order) recombination (15). Although most studies agree that radiative bimolecular recombination dominates at high initial carrier densities (n0 > 1017 cm−3) (1518), reports of kinetics at lower excitation densities (and relevant to solar cell operation) (19) range from single-exponential (12, 20), to biexponential (13, 14), to stretched-exponential (6, 9) functions with varying levels of fidelity. These distributions have in turn been explained in terms of unintentional doping (21) or charge trapping (22). The perovskite growth conditions (3, 4, 10) and post-deposition treatments (12, 23) can greatly alter film morphology, carrier lifetime, and device performance, yet the underlying relations between these parameters are important open questions. For instance, perovskite films grown from nonstoichiometric mixed halide (Cl/I) precursor solutions have exhibited lifetimes of hundreds of nanoseconds, but PL lifetimes in films grown from chloride-free precursors are generally much shorter (9, 20).

Correlated confocal PL and scanning electron microscopy (SEM) have been a powerful tool to reveal structure/function relationships in biology (24). We applied similar techniques to study structure/function relationships in perovskite films. We found substantial local PL heterogeneity even for CH3NH3PbI3(Cl) films with average lifetimes of ~1 μs (comparable to the longest lifetimes reported) (9, 10), which suggests that considerable scope remains for reducing nonradiative recombination in these films. In addition to observing entire grains that appear dark, we also observed that grain boundaries are associated with PL quenching, indicating that they are not as benign as has been suggested previously (25, 26). We further used PL microscopy to show that post-deposition chemical treatments can activate previously dark regions in the film, and we correlated local energy-dispersive x-ray spectroscopy (EDS) with confocal fluorescence maps, finding that brighter grains with longer lifetimes were associated with local spikes in Cl concentration.

We studied CH3NH3PbI3(Cl) films prepared on glass slides by spin-coating a nonstoichiometric mixed halide precursor solution composed of CH3NH3I and PbCl2 (3:1) in N,N-dimethylformamide (19, 27). Films prepared under identical conditions and incorporated into standard solar cell device architectures (fig. S1) (19) exhibit power conversion efficiencies (η) up to 14.5% (Fig. 1A), which is comparable to efficiencies in other reports using this architecture (2, 28). Figure 1B shows that our PL lifetimes are as long as those reported for films used in the best devices to date (10). These films exhibited average carrier lifetimes >1000 ns when excited at low intensity (30 nJ/cm2, n0 ~ 1015 cm−3). At short times (Fig. 1B, inset), the PL decay could appear nearly single-exponential, but at longer times, the decay deviated from a single-exponential decay (6, 9, 15). We fit the decay in Fig. 1B with a stretched-exponential function of characteristic lifetime τc = 431 ns and distribution parameter β = 0.57, which we interpret as arising from a superposition of exponential relaxation functions (see below) with an average lifetime of <τ> = 1005 ns (19, 29).

Fig. 1 Solar cell device measurement, bulk PL lifetime measurement, and correlated images from (SEM) and fluorescence microscopy experiments.

(A) Light current-voltage (J-V) characteristics of a high-performing mixed halide perovskite solar cell. (B) Bulk time-resolved PL decay trace of CH3NH3PbI3(Cl) perovskite film on glass after excitation at 470 nm, 125 kHz, 30 nJ/cm2 (n0~1015cm−3) and fitted to a stretched-exponential function with <τ> = 1005 ns, (τc = 431 ns, β = 0.57), with nearly single-exponential dynamics at short times (inset). (C) Correlated SEM micrograph, (D) fluorescence image, and (E) composite image showing significant variations in PL intensity across different grains and grain boundaries.

Green and co-workers recently examined microscopic PL quenching of discontinuous perovskite islands with n- and p-type capping layers (30). Here, we used fluorescence microscopy to probe the inherent decay properties of neat semiconducting films. Figure 1 shows a correlated SEM micrograph (Fig. 1C), confocal PL image (Fig. 1D), and an overlaid SEM/PL microscopy image (Fig. 1E) of a high-performing perovskite film on a glass substrate. Although this film appears contiguous (Fig. 1C) and exhibits <τ> = 1005 ns, Fig. 1D shows a large distribution in local PL intensity across the film. We observed these large distributions in films prepared in different research labs (fig. S2D) (19), and we exclude variations in film thickness (fig. S2) (19) and photodegradation during imaging (fig. S3) (19) as primary causes. The PL intensity not only varied from grain to grain, with roughly 30% of grains imaged in Fig. 1C consisting of dark grains (19), but we also observed ~65% lower PL intensity at grain boundaries (fig. S4, A to C) (19), after deconvolution of the microscope point-spread function (fig. S5) (19). These results are surprising because, through considerations of detailed balance (11, 31), one expects high-performance films to have minimal nonradiative decay.

Instead, the spatial variations in PL intensity in the polycrystalline perovskite films are suggestive of variations in local nonradiative decay rates. By taking local steady-state and time-resolved PL data, we confirmed that darker regions have greater nonradiative loss. Figure 2 shows a confocal PL image (Fig. 2A) along with local PL spectra (Fig. 2B) and lifetime data (Fig. 2, C to E) from a film with a long average bulk lifetime (<τ> = 1010 ns). Figure 2B shows the steady-state spectra of a bright (red square) and dark (blue circle) region. The PL spectrum collected at the dark region is both red-shifted (~2 nm) and slightly broader than the bright region (fig. S6) (19). These trends suggest a less sharp band edge (32), probably caused by the presence of defect states or shallow trapping levels in the darker regions. In Fig. 2, C to E, we show local PL decays of the indicated dark and bright regions at low (1 μJ/cm2), medium (2.1 μJ/cm2), and high (3.4 μJ/cm2) excitation fluences. Several studies have reported a transition from trap-assisted monomolecular recombination to free-carrier bimolecular recombination over this fluence range (15, 18). Consistent with the picture that bright regions have fewer nonradiative pathways, bright regions show a slower decay, a transition to bimolecular recombination–dominated kinetics at a lower excitation fluence, and more efficient PL quenching when contacted by fullerene (fig. S7) (19) in comparison to dim regions. We modeled the PL dynamics (black lines in Fig. 2, C to E) as a combination of trapping, monomolecular, and bimolecular recombination (19). We report a higher deep trap-state density in the dark region (4 × 1016 cm−3) as compared to the bright region (<1 × 1015 cm−3). In addition, we extracted the trapping, monomolecular, and bimolecular decay rates in both regions to be 1 × 10−8 cm3 s−1, 1 × 106 s−1, and 2.3 × 10−11 cm3 s−1 to 7.8 × 10−11 cm3 s−1, respectively. We also report consistent ratios of PL intensity measured across the steady state and integrated time-resolved PL measurements (fig. S8) for bright and dark regions (19).

Fig. 2 Fluorescence microscopy of CH3NH3PbI3(Cl) film and local PL measurements.

(A) A 3 μm–by–3 μm fluorescence image of the perovskite film with bulk lifetime <τ> = 1010 ns (τc = 433 ns, β = 0.57). (B) Relative steady state PL spectra of bright (red square) and dark (blue circle) regions. (C) Time-resolved PL decay curves of bright (red square) and dark (blue circle) regions after excitation at 470 nm, 125 kHz, ϕ = 1 μJ/cm2 (n0 ~ 5 × 1016 cm−3), (D) ϕ = 2.1 μJ/cm2 (n0 ~ 1 × 1017cm−3), and (E) bright region measured at ϕ = 2.1 μJ/cm2 versus dark region measured at ϕ = 3.4 μJ/cm2 (n0 ~ 1.6 × 1017cm−3), showing that dark regions require higher initial carrier densities to exhibit kinetics dominated by bimolecular recombination. Black traces are simulations to the data (19).

The local PL lifetimes are also shorter at grain boundaries (fig. S4E) (19). Grain boundaries frequently serve as nonradiative recombination centers in polycrystalline semiconductor films (33). Studies have suggested that grain boundaries in perovskites are less detrimental than in other semiconductors (25, 26), or even beneficial (34). Other results suggest that single-crystal perovskites exhibit even higher performance (3537), and some describe improvements in carrier lifetime and device performance from post-growth treatments, such as exposure of the film to pyridine (C5H5N), in the context of surface passivation (12, 23).

In this context, we next show that pyridine vapor exposure can brighten dark domains. In Fig. 3, A and B, we show the PL from a CH3NH3PbI3(Cl) film before and after exposure to pyridine vapor. The entire film was both brighter (~8× integrated over the entire image) and more uniform after pyridine exposure (no enhancement without pyridine, fig. S9) (19). For instance, the relative increase of a dark domain (squares in Fig. 3, A and B), was 180% larger than the relative increase of a bright domain (circles). The PL emission also blue-shifted (by ~3 nm) and narrowed slightly (Fig. 3C) after pyridine exposure, which could be caused by a reduction in shallow trap density. Finally, Fig. 3D shows that the grain boundary brightness, relative to the surrounding grains, increased by 11% and the width decreased by 25% after pyridine exposure (see fig. S10 for other examples) (19). Both trends are consistent with passivation, albeit incomplete, of defects at grain boundaries. Although pyridine treatment can also result in some restructuring of film morphology (fig. S10) (19), these data suggest that pyridine was indeed remediating nonradiative defects in the perovskite film.

Fig. 3 Fluorescence microscopy of CH3NH3PbI3(Cl) film with pyridine vapor treatment.

(A) Fluorescence image before and (B) after treatment showing activation of the CH3NH3PbI3(Cl) film. (C) Bulk steady-state PL spectra showing the relative PL intensities before (blue circle) and after (red square) treatment (inset) and normalized spectra showing a slight blue shift and narrowing of full width at half maximum after treatment. (D) Grain boundary PL line scan before [blue line in (A)] and after [red line in (B)] treatment, showing slight relative reduction in PL quenching across the grain boundary after treatment.

Finally, we explored the role of Cl in perovskite films by comparing SEM/EDS elemental composition traces with local PL intensity traces. Figure 4A shows an overlaid SEM/PL microscopy image of a CH3NH3PbI3(Cl) film on glass. We tracked the changes in PL intensity with Cl content (Fig. 4B) and showed that bright regions correlate with areas of higher relative Cl content (Cl/Cl+I) (trace Cl in CH3NH3PbI3 control films, fig. S11) (19). Although films prepared from Cl/I mixed halide precursors stoichiometrically resemble triiodides (38), there is evidence for residual Cl at levels of ~2 weight % or less (13, 39). The lifetime enhancement in the presence of Cl is consistent with recent findings that Cl-rich nucleation sites lead to better crystal coalescence (40, 41) and helps explain why films grown in the presence of Cl have slower recombination rates (9, 17). We hypothesize that Cl could be present at the surface or within the crystals, interstitially or substitutionally, or simply at the substrate surface as a residual but unincorporated component left over from the seeding of low-defect crystallites. We performed time-of-flight secondary ion mass spectrometry (ToF-SIMS) (fig. S11F) (19) and found higher Cl content in CH3NH3PbI3(Cl) films than in CH3NH3PbI3 films without Cl. This technique probes the top 2 nm of the film.

Fig. 4

Correlated images and line scans of CH3NH3PbI3(Cl) film using fluorescence microscopy, SEM, and EDS. (A) SEM micrograph overlaid on fluorescence image and (B) EDS line scan showing that the local elemental weight ratio of Cl/(Cl+I) tracks areas of higher integrated PL intensity, indicating that Cl is associated with better-performing grains.

Although perovskite solar cells have better radiative efficiencies than many other types, such as dye-sensitized, organic, or even cadmium telluride solar cells, they still suffer from greater nonradiative losses than inorganic materials such as gallium arsenide and are only at present approaching the radiative efficiencies of copper indium gallium selenide (CIGS) (31). Our results identify a subpopulation of dark grains and grain boundaries as specific targets for perovskite growth and passivation studies, and show that local fluorescence lifetime imaging provides a route by which changes in film processing can be evaluated to assess their influence on carrier recombination in films. By removing these nonradiative pathways to obtain uniform brightness with high emissivity across all grains, it is likely that we will see the performance of perovskite devices approach the thermodynamic limits for solar cells and other light-emitting devices.

Supplementary Materials

www.sciencemag.org/content/348/6235/683/suppl/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S11

References and Notes

  1. See the supplementary materials on Science Online.
  2. Acknowledgments: This material is based in part on work supported by the State of Washington through the University of Washington Clean Energy Institute. D.W.D. acknowledges support from an NSF Graduate Research Fellowship (DGE-1256082). S.M.V. acknowledges support from a National Defense Science and Engineering Graduate Fellowship. The research leading to these results has received funding from the European Union Seventh Framework Program (FP7/2007-2013) under Grant Agreement No. 604032 of the MESO project. G.E.E. is supported by the Engineering and Physical Sciences Research Council and Oxford Photovoltaics through a Nanotechnology Knowledge Transfer Network Collaborative Award in Science and Engineering. The authors gratefully acknowledge funding from the National Institute for Biomedical Imaging and Bioengineering (NIH grant EB-002027) supporting the National ESCA and Surface Analysis Center for Biomedical Problems and ToF-SIMS instrumentation. D.W.D. thanks I. Braly, S. Braswell, D. Moerman, and B. Miller for valuable assistance. S.M.V. gratefully acknowledges D. Graham for assistance with ToF-SIMS. Additional data, including materials, methods, and key controls, are available online as supplementary materials (19).
View Abstract

Navigate This Article