Anticorrelation between Local Photoluminescence and Photocurrent

Oct 17, 2016 - Dane W. deQuilettes , Sarthak Jariwala , Sven Burke , Mark E. Ziffer , Jacob T.-W. Wang , Henry J. Snaith , and David S. Ginger. ACS Na...
2 downloads 0 Views 10MB Size
Anticorrelation between Local Photoluminescence and Photocurrent Suggests Variability in Contact to Active Layer in Perovskite Solar Cells Giles E. Eperon, David Moerman, and David S. Ginger* Department of Chemistry, University of Washington, Seattle, Washington 98105, United States S Supporting Information *

ABSTRACT: We use high-resolution, spatially resolved, laser beam induced current, confocal photoluminescence, and photoconductive atomic force microscopy (pcAFM) measurements to correlate local solar cell performance with spatially heterogeneous local material properties in methylammonium lead triiodide (CH3NH3PbI3) perovskite solar cells. We find that, for this material and device architecture, the photocurrent heterogeneity measured via pcAFM on devices missing a top selective contact with traditional Aucoated tips is significantly larger than the photocurrent heterogeneity observed in full devices with both electronand hole-selective extraction layers, indicating that extraction barriers at the Au/perovskite interface are ameliorated by deposition of the organic charge extraction layer. Nevertheless, in completed, efficient device structures (PCE ≈ 16%) with state-of-the-art nickel oxide and [6,6]-phenylC61-butyric acid (PCBM) methyl ester contacts, we observe that the local photoluminescence (PL) is weakly anticorrelated with local photocurrent at both short-circuit and open-circuit conditions. We determine that the contact materials are fairly homogeneous; thus the heterogeneity stems from the perovskite itself. We suggest a cause for the anticorrelation as being related to local carrier extraction heterogeneity. However, we find that the contacts are still the dominating source of losses in these devices, which minimizes the impact of the material heterogeneity on device performance at present. These results suggest that further steps to prevent recombination losses at the interfaces are needed to help perovskite-based cells approach theoretical efficiency limits; only at this point will material heterogeneity become crucial. KEYWORDS: perovskite solar cells, laser beam induced current, photoluminescence microscopy, photoconductive atomic force microscopy, correlative microscopy, heterogeneity, contact-limited

P

Shockley−Queisser limit assumes only radiative recombination, and recent work has shown that there is a heterogeneous distribution of nonradiative recombination centers in thin films of organic−inorganic halide perovskites. Indeed, deQuilettes et al. found that thin films of the archetypal halide perovskite, methylammonium lead iodide, exhibited significant spatial heterogeneity in nonradiative recombination, with some grains, and most grain boundaries exhibiting large nonradiative losses.9 Importantly, regions of high nonradiative recombination will ultimately limit the performance of devices fabricated from these materials to below the Shockley−Queisser limit. This

erovskite solar cells have exhibited a dramatic rise in power conversion efficiencies over the last four years. This rise has been driven by an intense worldwide research effort, motivated by many beneficial properties of the low-temperature-processed halide perovskite materials: facile fabrication, strong absorption coefficients, long carrier lifetimes, high charge-carrier mobilities, and an apparent tolerance to processing-induced defects.1−3 The best perovskite photovoltaic devices are now over 22% efficient, rivaling established thin-film technologies.4 However, despite this rapid rise, the current power conversion efficiency (PCE) record is still only just over two-thirds the theoretical Shockley−Queisser limit of 31%5 for a cell with a 1.6 eV band gap.6 In contrast, the record PCE (28.8%) for GaAs cells is currently a much higher fraction (∼85%) of the Shockley−Queisser limit for GaAs.7,8 The © 2016 American Chemical Society

Received: August 29, 2016 Accepted: October 12, 2016 Published: October 17, 2016 10258

DOI: 10.1021/acsnano.6b05825 ACS Nano 2016, 10, 10258−10266

Article

www.acsnano.org

Article

ACS Nano

Figure 1. (a) Diagram of the perovskite cell architecture employed consisting of indium tin oxide (ITO)/nickel oxide/perovskite/[6,6]phenyl-C61-butyric acid methyl ester (PCBM)/bathocuproine (BCP)/Au. (b) Current−voltage characteristics of a representative CH3NH3PbI3 perovskite solar cell, measured under 100 mW cm−2 AM1.5G illumination at 0.38 V/s with no preconditioning, showing both short-circuit to forward bias scans (SC-FB) and vice versa (FB-SC). Performance parameters are recorded in the inset. (c) Diagram of the LBIC, LBIV, and PL mapping setup. (d) Diagram of pcAFM measurement setup.

LBIC on the same area of a device at high resolution. Hence, we are able to explore correlations between these properties on micrometer length scales.

initial report has been supported by a number of studies demonstrating that photoluminescence intensity can vary dramatically between grains in perovskite films prepared in many different laboratories.10−12 More recently, researchers have probed spatial variation in the photovoltaic properties of perovskite films to determine whether the spatial heterogeneity observed in nonradiative recombination translates to local variations in device performance. Notably, two studies using photoconductive atomic force microscopy (pcAFM) on devices missing one selective contact have appeared to confirm the existence of large local variations in photocurrent, with both groups showing that many regions of their films in fact generated little or no photocurrent under illumination.13,14 However, it should be pointed out that the current distribution established from the pcAFM measurements to date appears larger than might be expected based on the performance of the completed devices,15 suggesting that charge extraction at the pcAFM tip may complicate interpretation on perovskites studied without a top selective contact. Other researchers have used laser beam induced current (LBIC) to map photocurrent variation with much lower spatial resolution (∼50−1000 μm), in perovskites and other systems.16−18 For instance, Hinsch et al. found that LBIC could be used to resolve processing defects in the perovskite active layer and at the contacts, while Heben et al. investigated the large-scale evolution of LBIC uniformity as perovskite films degraded.19,20 However, the spatial resolution achieved in these studies has so far been too coarse to examine how the spatial heterogeneity in the perovskite itself is affecting device performance. Herein, we tackle this question using in situ photoluminescence, current and voltage mapping, and pcAFM, on full high-efficiency perovskite solar cells. We also investigate the variation in spatial heterogeneity as the device structure is built up. To map the photocurrent and local photovoltage, we utilize high-resolution laser beam induced current/voltage mapping (LBIC/LBIV). These measurements are normally carried out using conventional microscopy, with spatial resolution of tens of micrometers.18−21 Here, we employ this technique in a confocal microscope setup to achieve submicrometer resolution. Moreover, we measure photoluminescence (PL) and

RESULTS AND DISCUSSION We carried out in situ LBIC, LBIV, and PL mapping on solar cells based on methylammonium lead triiodide, with the device architecture shown in Figure 1a. We carried out pcAFM on devices missing the top (n-type) contact, to enable contacting of the perovskite surface with the AFM tip, as described in previous work.13,14,22,23 The perovskite deposition solution contains a slight stoichiometric excess of lead iodide, experimentally optimized and consistent with reports of the most efficient devices in the literature.24−26 We use state-of-theart contact materialsnickel oxide as the hole-selective contact and PCBM/BCP as the n-type contactas confirmed by the high open-circuit voltages of 1.06 V attained, close to the best reported in the literature.26,27 With a short-circuit current of 20.5 mA cm−2 and fill factor of 0.75, the best devices made with this approach attained 16.3% power conversion efficiency when measured under AM1.5 illumination, as shown in the current− voltage characteristics displayed in Figure 1b. Devices were fabricated and measured entirely under a dry nitrogen atmosphere, as moisture and oxygen are known to affect the properties of perovskite devices.28−30 Figure 1c shows the experimental setup for the in situ PL and electrical mapping. The device is mounted (in a nitrogen flow cell, not shown for clarity) on a piezoelectrically controlled confocal microscope stage, which raster scans the sample with nanometer precision, and is illuminated by a focused modulated 488 nm laser (square wave) from below. The laser spot size is ∼500 nm (full width at half-maximum) with an intensity of ∼22 000 W/m2 during each pulse (22 equiv suns). Current is collected using a lock-in amplifier, and a preamplifier applies a bias for the LBIV. PL is collected via a photomultiplier tube equipped with two 550 nm long-pass filters attached to the confocal microscope. For pcAFM, illumination is via a larger area LED, and an Au AFM tip is used to collect current under illumination at 0 V applied bias (short circuit), as we show in Figure 1d. In Figure 2 we compare pcAFM mapping on a device with no top contact with LBIC mapping on a full solar cell. As we 10259

DOI: 10.1021/acsnano.6b05825 ACS Nano 2016, 10, 10258−10266

Article

ACS Nano

over this range the increased illumination intensity has little effect on the LBIC distribution, as we show in Figure S1. If the pcAFM current was a direct measure of local efficiency for perovskites, and we assume that the highest local currents correspond to 100% EQE (external quantum efficiency), then taking into account the lower current areas one would compute that the device EQE should be ∼28%, which would produce a current of around 7.5 mA cm−2.31 Given that our measured currents in solar cells are around 20 mA cm−2, we conclude that the spatial heterogeneity observed on perovskites with a highwork-function tip appears to be complicated by the lack of a top contact or by domain size issues. Therefore, we turn to LBIC measurements, which allow measurement of local current in a full device structure. Here, we still observe heterogeneity in the photocurrent, but the variation in current between regions of high and low current is relatively small; the lowest currents are still 95% of the highest currents. If the brightest regions gave 100% EQE, we would expect ∼26 mA cm−2, which is more than our devices generate, indicating that the brightest regions have EQEs closer to 80%, which is a reasonable value for a wellperforming device and in good agreement with our measured EQEs, as we show in Figure S2. Since the key difference between the two techniques, aside from spatial resolution, is the presence of the top contact, we conclude that this contact is a critical part of the device to allow effective operation and mitigate the heterogeneity in current. The heterogeneity observed in pcAFM does not imply that there is no heterogeneity in extraction via the AFM tip; however, it suggests such heterogeneity is not as relevant to completed devices as the LBIC measurements. The full devices do still show some heterogeneity, although of a smaller magnitude, and we now move to considering the origin of this heterogeneity as observed via LBIC. Figure 3 shows PL, LBIC, and LBIV maps measured at the same location for a typical sample. The PL, shown in Figure 3a, was measured at open circuit. This image, measured on a full device, appears similar to previous confocal PL images taken of perovskite active layers on glass, confirming that PL heterogeneity at submicrometer length scales persists in the devices studied herein.9 Here, the dark spots are approximately 40% darker than the bright spots. We point out that this difference and that observed for films on glass (see Figure 4) are smaller differences in PL between dark and bright regions compared to previous reports, which is promising; as our knowledge of film processing techniques advances alongside device efficiencies, we would expect film heterogeneity to decrease. Strikingly, the LBIC measured at short circuit, Figure 3b, also shows obvious heterogeneity on a similar scale to the PL. The current varies by a smaller relative amount than the PL, as we will discuss below. We also note that this LBIC heterogeneity is typical over larger areas of the device, excluding points where there are obvious processing defects, as we show in Figure S3. The LBIV map, Figure 3c, was generated by holding the device at the bulk open-circuit voltage and determining in which regions current was positive, negative, or negligible, to define three discrete regions as higher than, lower than, or equivalent to the open-circuit voltage, in a manner similar to that employed in previous work.13 We note that the open-circuit voltage of the device under illumination by the laser is 0.547 V, lower than the voltage attained when held under full-area AM1.5G illumination (1.06 V). This discrepancy occurs because the vast majority of the device is in the dark during LBIV (see SI Figure

Figure 2. (a) Photoconductive AFM measurement of current, illuminated with a 470 nm LED at ∼1000 W/m2 on an ITO/NiO/ perovskite structure showing severe heterogeneity. (b) Data from part (a) normalized and with filter applied to attain a similar length scale as LBIC. (c) LBIC measurement on the full device illuminated with a 22 000 W/m2 488 nm laser. (d) Normalized LBIC map. (e) Histogram showing counts from plots (b) and (d); red = pcAFM; green = LBIC.

show in Figure 2a, we find that when measuring photocurrent via pcAFM, we obtain a similar distribution of current to that found in previous work, with a large fraction of the film generating insignificant currents compared to the brightest parts.13,14 The highest currents are greater than expected under this illumination if each point measured was acting individually, suggesting that instead grains act as conjoined aggregates, as previously observed;14 a grain of approximate area 0.1 μm2 would be necessary to produce the highest currents measured, which agrees well with the size of apparent grains. We show the LBIC measurement in Figure 2c, for which the relative variation in current is much smaller. To directly compare the two images, accounting for the difference in resolution, we applied a lowpass filter to the pcAFM data and normalized both images as shown in Figure 2b and d. We also show the distributions of the normalized data in Figure 2e. Even after filtering the pcAFM to account for the lower resolution of the LBIC, the current measured from the full device via pcAFM is significantly more heterogeneous, with a wider distribution centered around a lower fraction of the maximum than the LBIC. We note that 10260

DOI: 10.1021/acsnano.6b05825 ACS Nano 2016, 10, 10258−10266

Article

ACS Nano

Figure 3. (a) Photoluminescence map; (b) LBIC map; (c) relative voltage map measured in the same area for a representative area of a perovskite solar cell device using 22 000 W/m−2 488 nm laser illumination modulated at 1 kHz; (d) binned PL map showing areas with medium and high PL intensity more clearly; (e) effective PCE map attained by determining regions of high current and voltage; (f) maps from (d) overlaid on (e) showing regions of high PL mainly in regions of low effective PCE; (g) binned current plotted against PL intensity showing significant anticorrelation; (h) open-circuit voltage plotted against PL intensity showing anticorrelation; (i) current plotted against binned voltage showing correlation. In all floating column plots, the central line represents the mean, and the top and bottom of the bar represent the standard deviation.

Figure 3g, h, and i verify this anticorrelation between LBIC and local PCE by plotting PL intensity as a function of binned current (Figure 3g) and voltage (Figure 3h) and current as a function of voltage (Figure 3i). In these plots, the central line of the bar represents the mean, and the top and bottom of the bars give the standard deviation. While the total spatial variation in current is relatively small, there is a notable trend that regions of higher PL have lower current (Figure 3g), a weaker trend to open-circuit voltage (Figure 3h), and that current and voltage have a strong positive correlation (Figure 3i). This result confirms the visual analysis that regions of lower PL are generally more efficient. We note that these trends were observed multiple times, across separate samples fabricated at different times (Figure S5). Determination of the Pearson coefficient from the raw data also confirms this anticorrelation (Figure S15). This conclusion leads us to further investigate why these bright regions are “worse”. First, we note that these regions are not much worse. To quantify how severely the PCE is affected,

S4). However, the qualitatively high and low voltage regions remain indicative of the relative voltage generated when those regions are illuminated. In a PV device we are ultimately interested in the PCE. Thus, we define an effective PCE, being assigned as “high” in regions with both high current and voltage, “low” when they are both low, and “medium” otherwise. We note that this estimate does not take into account the fill factor (FF), which full device PCE would do, but we assume the FF to vary little over the area for the sake of analysis (see SI Figure S6, as discussed later). The effective PCE map is shown in Figure 3e. We also classify the PL into regions of high, medium, and low, as shown in Figure 3d. We can now establish visually whether there is a relation between the PL by overlaying these binned maps on top of each other, as we do in Figure 3f. It is immediately apparent that, in these devices, regions of higher PL (red) tend to lie on top of regions of lower PCE, and regions of higher PCE (blue) appear to correlate to regions of lower PL. 10261

DOI: 10.1021/acsnano.6b05825 ACS Nano 2016, 10, 10258−10266

Article

ACS Nano

Figure 4. Photoluminescence maps measured using 22 000 W/m−2 488 nm pulsed laser illumination at 1 kHz for typical samples of (a) perovskite film only, (b) perovskite/PCBM, (c) ITO/NiO/perovskite, (d) ITO/NiO/perovskite/PCBM, and (e) ITO/NiO/perovskite/ PCBM/Au. Samples were fabricated on glass or ITO-coated glass and illuminated from below (glass side). The value σ/m refers to the standard deviation divided by the mean for the whole area images. (f) Line scans extracted from the PL maps in a−e, taken from x = 0, y = 5 μm to x = 10 μm, y = 5 μm in all cases. σ/m values represent the coefficient of variation for each map.

observed previously in low-resolution measurements on perovskite devices; heterogeneity is mainly ascribed to electrode heterogeneity arising from the deposition method or degradation of the active layer in these cases.12,19,20 Anticorrelation between local PL and local photocurrent has also been observed in other systems including amorphous silicon, organic photovoltaics, and copper indium gallium selenide.16,17,36,37 Notably, these systems all tend to be less mature, with contacts less optimized than polycrystalline silicon PV. Therefore, it seems likely that the anticorrelation we observe between photoluminescence and photocurrent is due to variation in quality of the contact between the perovskite and the extraction layers. This is something we now seek to explore in detail below. To determine the role of the contact layers in this anticorrelation, we took confocal PL maps of perovskite active layers in various stages of being contacted by the extraction layers and electrodes. Figure 4 shows PL maps of samples as the device stack is built up, from bare perovskite (Figure 4a) through perovkite/PCBM (Figure 4b), NiO/perovskite (Figure 4c), NiO/perovskite/PCBM (Figure 4d) to the full device, NiO/perovskite/PCBM/Au (Figure 4e). All maps were taken with the same laser intensity, so the absolute counts can be directly compared (note the different z-scale bars). Figure 4f shows line scans bisecting each image in order to facilitate comparison of absolute PL intensity more easily. We also calculate the value of the standard deviation divided by the mean (σ/m, or coefficient of variation) for each map, as a quantitative measure of the PL heterogeneity. Notably, Figure 4 shows that while contact deposition depresses overall PL intensity, the PL heterogeneity remains similar after adding either or both contact layers. The PL is quenched upon

we took LBIC maps at several applied biases. These maps allowed us to reconstruct I−V curves at these locations and thus attain a more exact relative efficiency than the estimation made in Figure 3. We show the data in the SI (Figure S6); the worse regions give an effective PCE that is ∼95% of that from the best regions. Approximating the impact this would have on a device, by considering the distribution of bright and dark regions, allows us to estimate that bringing all regions to the same performance level as the best grains would result in a ∼2.5% relative increase in PCE, bringing the 16.3% efficient device to 16.7%. Clearly these “worse” grains are not crippling the device, rather having a more minor impact. Nevertheless, their origin is interesting. In a photovoltaic architecture where the device is limited only by the fundamental properties of the active layer, we would expect that regions that have higher photoluminescence at open circuit also have superior photovoltaic behavior. These regions would have reduced nonradiative recombination rates and hence provide both a higher open-circuit voltage (governed logarithmically with PL by the relations detailed by Rau et al.)32−34 and a higher short-circuit current. Such a correlation has been observed previously for mature technologies such as silicon solar cells, where the regions of higher PL and PV performance are assumed to be those with a lower density of defects.35,36 We do not observe this situation here, however. Once the system is contacted by selective contacts or electrodes, more causes for possible heterogeneity are present. If the effectiveness of carrier extraction by the contacts is heterogeneous, then regions where excited carriers are less effectively extracted would have higher PL, but lower current, under short-circuit conditions, since carriers are confined to the active layer. In such a case, an anticorrelation between PL and LBIC would be observed. Such an anticorrelation has been 10262

DOI: 10.1021/acsnano.6b05825 ACS Nano 2016, 10, 10258−10266

Article

ACS Nano

the image before and after contacting with PCBM suggests that the bright regions in the final device, those which generate lower photocurrent, originate in the perovskite film, rather than the deposition of the contact layer introducing the dominating heterogeneity. We note that we have also shown the holeselective contact to have homogeneous properties, as we show in the SI (Figure S9). Thus, we conclude that the heterogeneity in contact quality stems from the perovskite itself. Furthermore, now that we know that the bright regions remain bright upon application of the contact layers, we can gain additional insight. Using a simple model, we find that we can explain the negligible change in σ/m if the brighter regions are those that are quenched less effectively (see Figure S8). This situation is slightly different from those studied previously in the literature.16,17,36,37 Previously, contact heterogeneity has been responsible for heterogeneous performance, but it has been introduced by inherent heterogeneity in the extraction layers rather than the perovskite itself. We thus move toward determining the underlying cause of this heterogeneity in the perovskite film. We carried out scanning Kelvin probe microscopy on the perovskite film to test for work function offsets at the interface, which may suggest local fluctuations in doping density, but we observed negligible heterogeneity (Figure S10). Thus, we reasoned that a more likely hypothesis is that a physical barrier, perhaps a very thin insulating layer, may be present at these grains. As previously mentioned, our perovskite films contain a small excess of lead iodide.38,39 Lead iodide is a wide gap semiconductor, with some discrepancy in reported conduction and valence band positions; depending on the literature reference, it could either be a passivating insulator for the purposes of charge extraction from MAPbI3 or in fact facilitate charge transfer at the electron-selective interface.38−40 It is however agreed upon that it can affect charge extraction, either positively or negatively. Heterogeneity in lead iodide distribution could result in heterogeneity in charge transfer, with more efficient regions having either an excess or deficit of lead iodide. Jsc is related to Voc by the relation Voc = (kT/q) ln(1 + Jsc/J0), where k, T, q, and J0 are the Boltzmann constant, temperature, electronic charge, and the dark saturation current, respectively.43 Therefore, the observation that voltage as well as current is also lower in the bright regions is not unexpected; if Jsc is affected more than J0, then Voc would drop if Jsc did, although it would be less affected due to the logarithmic relation. J0 relates to the shape of the dark current−voltage curve, and it may not be as affected as Jsc by a thin surface layer, especially in cases where the voltage attained in a full device is

application of either contact layer and quenched further when both are present. While expected, this quenching indicates that the contacts introduce additional nonradiative recombination pathways to the film. Importantly, neither the relative magnitude, as measured by the coefficient of variation, nor the characteristic length scales (see FFT analysis in Figure S7) of the PL heterogeneity undergo a statistically significant change after addition of the contacts. The fact that the coefficient of variation remains similar (within normal film-to-film variation; see Table ST1) as the PL is quenched by the application of contacts is noteworthy. If the contacts were to uniformly quench the films (i.e., quenching rate into the contact is the same in all regions), we would expect a significant reduction in σ/m, as we discuss in the SI (Figure S8) and has been observed for some films.9 The fact that σ/m does not decrease in these samples suggests that the contacts are not quenching PL uniformly. We expect quenching efficiency to be correlated to how wellcontacted the region is; so this result is consistent with the proposed cause of the PL/photocurrent anticorrelation being nonuniform contact quality between the perovskite and the contacts. In order to test this hypothesis further, we directly imaged PL in the same location before and after coating with the electron-selective contact (PCBM). We show these PL maps in Figure 5. The negligible change in the spatial heterogeneity,

Figure 5. Photoluminescence maps measured on a NiO/perovskite film in the same region and at the same laser fluence before (a) and after (b) coating with PCBM. σ/m values represent the coefficient of variation for each map.

quantified by σ/m (rather than a marked decrease), is consistent with the results above, suggesting heterogeneity in the contact to the active layer. However, the same general features are observable in Figure 5a and b, despite a ∼50% reduction in overall PL. The presence of the same structure in

Figure 6. Photoluminescence map (a) and LBIC map (b, e) measured in the same area using 22 000 W/m−2 488 nm modulated laser illumination at 1 kHz for devices fabricated with a stoichiometric 1:1 MAI:PbI2 perovskite. (c) PL intensity as a function of binned current for this device. 10263

DOI: 10.1021/acsnano.6b05825 ACS Nano 2016, 10, 10258−10266

Article

ACS Nano

would be reduced. Clearly, further work is needed to optimize the contacts and push devices closer to their theoretical limits. Then, the impact of material heterogeneity will become more severe, and it will be necessary to develop effective passivation techniques that do not also impact carrier extraction.

close to the maximum for the combination of selective contacts, which is likely the case here. Finally, to further test this hypothesis that surface layers such as lead iodide heterogeneity may be responsible for causing (slightly) hindered charge extraction at well-passivated bright regions, resulting in the high PL−low current correlation, we fabricated stoichiometric (1:1 MAI:PbI2) and lead iodidedeficient perovskite devices. These had lower PCE than those with the slight excess, as shown in Figure S11 and consistent with the literature. We then measured PL and LBIC as previously to establish if the same correlation between local PL and local photocurrent was observed. Figure 6 shows PL (Figure 6a) and LBIC (Figure 6b) maps from the same region of a device fabricated with stoichiometric precursors. We plot the PL intensity against binned current in Figure 6c. As we show in Figure 6a and b, we do still observe significant PL heterogeneity and current heterogeneity. However, when we compare the binned current to the PL intensity for these films, there is no longer an obvious trend. These observations were repeatable across multiple samples (see Figure S12). Devices fabricated with a deficit of lead iodide, with much lower PCE, showed a similar lack of correlation (Figure S13). These results seem to support the hypothesis that excess PbI2 could be the component responsible for introducing better extraction in some regions than others, either via forming an insulating extraction barrier at the less efficient regions or by enhancing extraction in the more efficient ones. Overall the lead iodide excess appears beneficial, compared to currently available alternatives.38−40 However, we propose that the ideal solution may be that a new strategy for passivation is needed, one that passivates effectively without forming insulating barriers in some regions or by enhancing all regions uniformly. It is furthermore apparent that more detailed studies on the precise impact of lead iodide excess and accurate determination of its energetics would improve our understanding of interfacial effects in this system. In the best devices, the fractional impact on current is much less than the fractional variation in PL. This implies that something else is limiting the performance rather than nonradiative losses in the active layer. To understand the limiting factor, returning to Figure 4, we notice that the magnitude of PL decreases substantially upon addition of each selective contact. Ideally, this decrease or “quench” would not happen; it implies that many carriers are effectively lost to nonradiative recombination in the contact and can no longer return to the perovskite. We also find that when measuring PL at short circuit compared to open circuit (see Figure S14), the PL does not decrease much. At short circuit, carrier extraction can compete with field gradient; at open circuit carriers should return to the perovskite and emit if they are not lost in the contact. Thus, we would expect a much higher PL at open circuit than short circuit with good contact materials. The small difference confirms that many carriers are lost via nonradiative recombination in the contacts. This is surprising given that the contacts used are state-of-the-art and voltage losses are relatively low. However, it is worth noting that mitigation of nonradiative recombination at the contacts is often a final engineering effort for mature PV technologies and was responsible for pushing silicon solar cells to their world-record efficiencies.41 Large differences between open- and short-circuit PL are also found in other more optimized thin-film solar cells.42 With perfect “nonquenching” contacts, a higher opencircuit voltage would be attained as nonradiative recombination

CONCLUSION We have shown a high-resolution direct correlation between PL heterogeneity and device performance in perovskite solar cells. By comparing pcAFM with LBIC measurements, we find that the top contact is crucial in mitigating current losses in a full device. We observe that in the most efficient devices regions of high PL intensity correlate with areas of lower current and voltage, and hence efficiency. We conclude that the reasons for this correlation are due to variations in the perovskite material, possibly how variations in its surface cause variations in electronic coupling to the extraction layers, rather than being intrinsic to the contact materials. We suggest a possible reason for this correlation, but we conclude that the selective contacts are a significant source of nonradiative recombination losses in the device, meaning that while material heterogeneity may ultimately limit device performance, it is not the dominant factor in the current generation of devices. This conclusion motivates further work on developing contacts that do not lead to “quenched” PL at open circuit; once contact materials have been optimized, material heterogeneity will become more important, and removing it will be necessary to attain the best material quality and charge extraction properties across the device to reach the Shockley−Queisser limit. METHODS AND EXPERIMENTAL SECTION Materials. Unless otherwise stated, all materials were purchased from Sigma-Aldrich or Alfa Aesar and used as received. Methylammonium iodide was purchased from Dyesol Ltd. Perovskite Device Fabrication. Devices were fabricated in the architecture ITO/NiO/perovskite/PCBM/BCP/Au. They were fabricated on prepatterned ITO substrates (TFD Inc.). These were sequentially cleaned by sonicating in dilute detergent, acetone, and propan-2-ol and subsequently treated with oxygen plasma for 10 min. A thin layer of NiO was deposited by spin-coating in air at 5000 rpm a precursor consisting of 129 mg of nickel(II) acetylacetonate dissolved in 5 mL of anhydrous ethanol with addition of 50 μL of 38% HCl. The NiO layer was then annealed at 320 °C for 45 min in air. Subsequently the substrates were taken into a nitrogen-filled glovebox. The perovskite precursor solution (for the PbI2 excess devices) consists of 1 M MAI and 1.1 M PbI2 dissolved in a mixed solvent of 35% DMSO and 65% DMF and was filtered before use. This was spin-coated onto the NiO at 7000 rpm and transferred immediately into a chlorobenzene bath. The films were held there for 20 s before removing and blowing dry with a nitrogen gun. Films were then heated at 60 °C for 3 min followed by 100 °C for 7 min. After perovskite deposition and annealing, a solution of PCBM (Solenne BV) in dichlorobenzene at 30 mg/mL was spin-coated at 1000 rpm. When dry, substrates were gently annealed at 70 °C for 10 min. Subsequently a solution of bathocuproine of 0.5 mg/mL in anhydrous propan-2-ol was spin-coated dynamically at 6000 rpm for 15s, and the films then heated at 70 °C for 10 min. To complete the devices, gold electrodes were thermally evaporated onto the devices at 1 Å/s to give a layer of 80−100 nm thickness. We note that when comparing different samples directly (e.g., for Figure 4), we ensured that all perovskite films were prepared at the same time, in the same batch, to avoid any batch-to-batch variation. However, using this perovskite deposition technique, we do also observe minimal batch-to-batch variation in PL and device performance (see Table ST1 for comparison of coefficient of variation across six devices made at different times). 10264

DOI: 10.1021/acsnano.6b05825 ACS Nano 2016, 10, 10258−10266

Article

ACS Nano J−V Characteristics (Macroscopic). The current density−voltage (J−V) curves were measured (2400 Series SourceMeter, Keithley Instruments) in nitrogen under simulated AM 1.5 sunlight at 100 mW cm−2 irradiance generated by a Solar Light Co. 16S-300 solar simulator, with the intensity calibrated with an IR-filtered Si reference cell. The masked area of the solar cell is 0.013 cm−2. The forward J−V scans were measured from forward bias (FB) to short circuit (SC), and the backward scans were from short circuit to forward bias, both at a scan rate of 0.38 V s−1. There was no stabilization, prebiasing, or light soaking prior to measurement. Laser Beam Induced Current/Voltage and PL Measurements. The sample was mounted in a custom-built nitrogen flow cell on a piezostage Nikon 2000 inverted confocal microscope with a Plan Fluor 40×/0.60NA objective. The piezostage was controlled by an Asylum Research MFP-3D AFM controller. A 488 nm laser pulsed at 1 kHz was coupled into the confocal microscope and focused onto the sample. For LBIC measurements, the sample electrodes were connected to a lock-in amplifier (Stanford Research SR830 with a current gain of 106 V/A) to measure the current, and the sample was raster scanned using the piezostage. For measurements at different biases, the laser was not pulsed and a preamplifier (Stanford Research SR540) was used to apply a bias to the sample. The preamplifier output voltage signal was fed into the scanning control electronics to construct current maps. The voltage maps were generated by finding the applied bias at which the total current across the measured area was approximately zero and binning the current appropriately. For PL mapping, the same 488 nm laser pulsed at 1 kHz was used as for the current measurements. Laser intensity on the sample of 2200 mW/ cm2 (22 suns) was employed in most measurements unless stated otherwise by calibration with a silicon photodiode prior to measurement and attenuating the laser with ND filters. After passing through two 550 nm long-pass filters and an 850 nm short-pass filter, the PL was detected by using a Hammamatsu PMT (H10721-20) fiber-coupled with the microscope, its current amplified, and its output connected to the SR830 lock-in. Scanning Kelvin Probe and Conductive Atomic Force Measurements. Surface potential was measured using an MFP-3D Asylum Research AFM employing a Cr/Pt-coated tip (300 kHz, 40 N/ m) in amplitude-modulated Kelvin probe microscopy mode. Conductive AFM measurements were made using the same system but using Au-coated tips (13 kHz, 0.2 N/m) mounted on a 2 nA/V tip holder. The force applied during the contact imaging was kept below 2.5 nN in order to avoid damaging the sample and the tip.

and the Alvin L. and Verla R. Kwiram Endowment from the Department of Chemistry at the University of Washington.

REFERENCES (1) Zhang, W.; Eperon, G. E.; Snaith, H. J. Metal Halide Perovskites for Energy Applications. Nat. Energy 2016, 1, 16048. (2) Stranks, S. D.; Snaith, H. J. Metal-Halide Perovskites for Photovoltaic and Light-Emitting Devices. Nat. Nanotechnol. 2015, 10, 391−402. (3) Steirer, K. X.; Schulz, P.; Teeter, G.; Stevanovic, V.; Yang, M.; Zhu, K.; Berry, J. J. Defect Tolerance in Methylammonium Lead Triiodide Perovskite. ACS Energy Lett. 2016, 1, 360−366. (4) NREL. Best Research-Cell Ef f iciencies, Http://www.nrel.gov/ ncpv/images/efficiency_chart.jpg; 2016. Accessed Oct 11, 2016. (5) Sha, W. E. I.; Ren, X.; Chen, L.; Choy, W. C. H. The Efficiency Limit of CH3NH3PbI3 Perovskite Solar Cells. Appl. Phys. Lett. 2015, 106, 221104. (6) Ziffer, M. E.; Mohammed, J. C.; Ginger, D. S. Electroabsorption Spectroscopy Measurements of the Exciton Binding Energy, ElectronHole Reduced Effective Mass, and Band Gap in the Perovskite CH 3 NH 3 PbI 3. ACS Photonics 2016, 3, 1060. (7) Green, M. A.; Emery, K.; Hishikawa, Y.; Warta, W.; Dunlop, E. D. Solar Cell Efficiency Tables (Version 45). Prog. Photovoltaics 2015, 23, 1−9. (8) Shockley, W.; Queisser, H. J. Detailed Balance Limit of Efficiency of P-N Junction Solar Cells. J. Appl. Phys. 1961, 32, 510. (9) de Quilettes, D. W.; Vorpahl, S. M.; Stranks, S. D.; Nagaoka, H.; Eperon, G. E.; Ziffer, M. E.; Snaith, H. J.; Ginger, D. S. Impact of Microstructure on Local Carrier Lifetime in Perovskite Solar Cells. Science (Washington, DC, U. S.) 2015, 348, 683−686. (10) Bischak, C. G.; Sanehira, E. M.; Precht, J. T.; Luther, J. M.; Ginsberg, N. S. Heterogeneous Charge Carrier Dynamics in Organic− Inorganic Hybrid Materials: Nanoscale Lateral and Depth-Dependent Variation of Recombination Rates in Methylammonium Lead Halide Perovskite Thin Films. Nano Lett. 2015, 15, 4799−4807. (11) Draguta, S.; Thakur, S.; Morozov, Y. V.; Wang, Y.; Manser, J. S.; Kamat, P. V.; Kuno, M. Spatially Non-Uniform Trap State Densities in Solution-Processed Hybrid Perovskite Thin Films. J. Phys. Chem. Lett. 2016, 7, 715−721. (12) El-Hajje, G.; Momblona, C.; Gil-Escrig, L.; Á vila, J.; Guillemot, T.; Guillemoles, J.-F.; Sessolo, M.; Bolink, H. J.; Lombez, L. Quantification of Spatial Inhomogeneity in Perovskite Solar Cells by Hyperspectral Luminescence Imaging. Energy Environ. Sci. 2016, 9, 2286−2294. (13) Leblebici, S. Y.; Leppert, L.; Li, Y.; Reyes-Lillo, S. E.; Wickenburg, S.; Wong, E.; Lee, J.; Melli, M.; Ziegler, D.; Angell, D. K.; Ogletree, D. F.; Ashby, P. D.; Toma, F. M.; Neaton, J. B.; Sharp, I. D.; Weber-Bargioni, A. Facet-Dependent Photovoltaic Efficiency Variations in Single Grains of Hybrid Halide Perovskite. Nat. Energy 2016, 1, 16093. (14) Kutes, Y.; Zhou, Y.; Bosse, J. L.; Steffes, J.; Padture, N. P.; Huey, B. D. Mapping the Photoresponse of CH3NH3PbI3 Hybrid Perovskite Thin Films at the Nanoscale. Nano Lett. 2016, 16, 1−28. (15) Eperon, G. E.; Ginger, D. S. Perovskite Solar Cells: Different Facets of Performance. Nat. Energy 2016, 1, 16109. (16) Ostrowski, D. P.; Glaz, M. S.; Goodfellow, B. W.; Akhavan, V. a; Panthani, M. G.; Korgel, B. a; Vanden Bout, D. a. Mapping Spatial Heterogeneity in Cu(In(1-x)Ga(x))Se2 Nanocrystal-Based Photovoltaics with Scanning Photocurrent and Fluorescence Microscopy. Small 2010, 6, 2832−2836. (17) Rösch, R.; Tanenbaum, D. M.; Jørgensen, M.; Seeland, M.; Bärenklau, M.; Hermenau, M.; Voroshazi, E.; Lloyd, M. T.; Galagan, Y.; Zimmermann, B.; Würfel, U.; Hösel, M.; Dam, H. F.; Gevorgyan, S. a.; Kudret, S.; Maes, W.; Lutsen, L.; Vanderzande, D.; Andriessen, R.; Teran-Escobar, G.; et al. Investigation of the Degradation Mechanisms of a Variety of Organic Photovoltaic Devices by Combination of Imaging Techniquesthe ISOS-3 Inter-Laboratory Collaboration. Energy Environ. Sci. 2012, 5, 6521.

ASSOCIATED CONTENT S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acsnano.6b05825. Supporting LBIC and PL data, repeat PL/LBIC correlations, current−voltage characteristics, external quantum efficiency, correlation coefficients, and other supplementary figures and information (PDF)

AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected]. Notes

The authors declare no competing financial interest.

ACKNOWLEDGMENTS We thank Dr. Zonglong Zhu for valuable input on selective contact deposition techniques and Dane deQuilettes for valuable discussions. Funding was provided primarily by the Department of Energy BES DE-SC0013957, with fellowship and infrastructure support from the University of Washington Clean Energy Institute, the Washington Research Foundation, 10265

DOI: 10.1021/acsnano.6b05825 ACS Nano 2016, 10, 10258−10266

Article

ACS Nano (18) Rivière, G. A.; Simon, J. J.; Escoubas, L.; Vervisch, W.; Pasquinelli, M. Photo-Electrical Characterizations of Plastic Solar Modules. Sol. Energy Mater. Sol. Cells 2012, 102, 19−25. (19) Mastroianni, S.; Heinz, F. D.; Im, J.-H.; Veurman, W.; Padilla, M.; Schubert, M. C.; Würfel, U.; Grätzel, M.; Park, N.-G.; Hinsch, A. Analysing the Effect of Crystal Size and Structure in Highly Efficient CH 3 NH 3 PbI 3 Perovskite Solar Cells by Spatially Resolved Photoand Electroluminescence Imaging. Nanoscale 2015, 7, 19653−19662. (20) Song, Z.; Abate, A.; Watthage, S. C.; Liyanage, G. K.; Phillips, A. B.; Steiner, U.; Graetzel, M.; Heben, M. J. Perovskite Solar Cell Stability in Humid Air: Partially Reversible Phase Transitions in the PbI 2 -CH 3 NH 3 I-H 2 O System. Adv. Energy Mater. 2016, 1600846. (21) Bauer, G. H.; Gütay, L. Lateral Features of Cu(In0.7Ga0.3)Se2Heterodiodes in the Mm-Scale by Confocal Luminescence and Focused Light Beam Induced Currents. Thin Solid Films 2007, 515, 6127−6131. (22) Coffey, D. C.; Reid, O. G.; Rodovsky, D. B.; Bartholomew, G. P.; Ginger, D. S. Mapping Local Photocurrents in Polymer/fullerene Solar Cells with Photoconductive Atomic Force Microscopy. Nano Lett. 2007, 7, 738−744. (23) Moerman, D.; Sebaihi, N.; Kaviyil, S. E.; Leclère, P.; Lazzaroni, R.; Douhéret, O. Towards a Unified Description of the Charge Transport Mechanisms in Conductive Atomic Force Microscopy Studies of Semiconducting Polymers. Nanoscale 2014, 6, 10596. (24) Giordano, F.; Abate, A.; Pablo, J.; Baena, C.; Saliba, M.; Matsui, T.; Im, S. H.; Zakeeruddin, S. M.; Nazeeruddin, M. K.; Hagfeldt, A.; Graetzel, M. Enhanced Electronic Properties in Mesoporous TiO2 via Lithium Doping for High-Efficiency Perovskite Solar Cells. Nat. Commun. 2016, 7, 1−6. (25) Bi, D.; Tress, W.; Dar, M. I.; Gao, P.; Luo, J.; Renevier, C.; Schenk, K.; Abate, A.; Giordano, F.; Correa Baena, J.-P.; Decoppet, J.D.; Zakeeruddin, S. M.; Nazeeruddin, M. K.; Gra tzel, M.; Hagfeldt, A. Efficient Luminescent Solar Cells Based on Tailored Mixed-Cation Perovskites. Sci. Adv. 2016, 2, e1501170−e1501170. (26) Zhu, Z.; Bai, Y.; Liu, X.; Chueh, C.-C.; Yang, S.; Jen, A. K.-Y. Enhanced Efficiency and Stability of Inverted Perovskite Solar Cells Using Highly Crystalline SnO 2 Nanocrystals as the Robust ElectronTransporting Layer. Adv. Mater. 2016, 28, 6478−6484. (27) Jeon, Y.-J.; Lee, S.; Kang, R.; Kim, J.-E.; Yeo, J.-S.; Lee, S.-H.; Kim, S.-S.; Yun, J.-M.; Kim, D.-Y. Planar Heterojunction Perovskite Solar Cells with Superior Reproducibility. Sci. Rep. 2014, 4, 6953. (28) Eperon, G. E.; Habisreutinger, S. N.; Leijtens, T.; Bruijnaers, B. J.; van Franeker, J. J.; DeQuilettes, D. W.; Pathak, S.; Sutton, R. J.; Grancini, G.; Ginger, D. S.; Janssen, R. A. J.; Petrozza, A.; Snaith, H. J. The Importance of Moisture in Hybrid Lead Halide Perovskite Thin Film Fabrication. ACS Nano 2015, 9, 9380−9393. (29) Pearson, A. J.; Eperon, G. E.; Hopkinson, P. E.; Habisreutinger, S. N.; Wang, J. T.-W.; Snaith, H. J.; Greenham, N. C. Oxygen Degradation in Mesoporous Al 2 O 3 /CH 3 NH 3 PbI 3- X Cl X Perovskite Solar Cells: Kinetics and Mechanisms. Adv. Energy Mater. 2016, 6, 1600014. (30) Matsumoto, F.; Vorpahl, S. M.; Banks, J. Q.; Sengupta, E.; Ginger, D. S. Photodecomposition and Morphology Evolution of Organometal Halide Perovskite Solar Cells. J. Phys. Chem. C 2015, 119, 20810−20816. (31) Ball, J. M.; Stranks, S. D.; Hörantner, M. T.; Hüttner, S.; Zhang, W.; Crossland, E. J. W.; Ramirez, I.; Riede, M.; Johnston, M. B.; Friend, R. H.; Snaith, H. J.; Hoerantner, M.; Hüttner, S.; Zhang, W.; Crossland, E. J. W.; Ramirez, I.; Riede, M.; Johnston, M. B.; Friend, R. H.; Snaith, H. J.; et al. Optical Properties and Limiting Photocurrent of Thin-Film Perovskite Solar Cells. Energy Environ. Sci. 2015, 8, 602− 609. (32) Rau, U. Reciprocity Relation between Photovoltaic Quantum Efficiency and Electroluminescent Emission of Solar Cells. Phys. Rev. B: Condens. Matter Mater. Phys. 2007, 76, 85303. (33) Miller, O. D.; Yablonovitch, E.; Kurtz, S. R. Strong Internal and External Luminescence as Solar Cells Approach the Shockley-Queisser Limit. IEEE J. Photovoltaics 2012, 2, 303−311.

(34) Ross, R. T. Some Thermodynamics of Photochemical Systems. J. Chem. Phys. 1967, 46, 4590. (35) Breitenstein, O.; Höffler, H.; Haunschild, J. Photoluminescence Image Evaluation of Solar Cells Based on Implied Voltage Distribution. Sol. Energy Mater. Sol. Cells 2014, 128, 296−299. (36) Shimokawa, R.; Tajima, M.; Warashina, M.; Kashiwagi, Y.; Kawanami, H. Correspondence among PL Measurement, MBIC Measurement and Defect Delineation in Polycrystalline Cast-Si Solar Cells. Sol. Energy Mater. Sol. Cells 1997, 48, 85−91. (37) Nakanishi, H.; Ito, A.; Takayama, K.; Kawayama, I.; Murakami, H.; Tonouchi, M. Comparison between Laser Terahertz Emission Microscope and Conventional Methods for Analysis of Polycrystalline Silicon Solar Cell. AIP Adv. 2015, 5, 11712910.1063/1.4935913. (38) Jacobsson, T. J.; Correa-Baena, J.-P.; Halvani Anaraki, E.; Philippe, B.; Stranks, S. D.; Bouduban, M. E. F.; Tress, W.; Schenk, K.; Teuscher, J.; Moser, J.-E.; Rensmo, H.; Hagfeldt, A. Unreacted PbI2 as a Double-Edged Sword for Enhancing the Performance of Perovskite Solar Cells. J. Am. Chem. Soc. 2016, 138, 10331. (39) Chen, Q.; Zhou, H.; Song, T.-B.; Luo, S.; Hong, Z.; Duan, H.-S. S.; Dou, L.; Liu, Y.; Yang, Y. Controllable Self-Induced Passivation of Hybrid Lead Iodide Perovskites toward High Performance Solar Cells. Nano Lett. 2014, 14, 4158−4163. (40) Cao, D. H.; Stoumpos, C. C.; Malliakas, C. D.; Katz, M. J.; Farha, O. K.; Hupp, J. T.; Kanatzidis, M. G. Remnant PbI2, an Unforeseen Necessity in High-Efficiency Hybrid Perovskite-Based Solar Cells? APL Mater. 2014, 2, 1−8. (41) Green, M. A. The Path to 25% Silicon Solar Cell Efficiency: History of Silicon Cell Evolution. Prog. Photovoltaics 2009, 17, 183− 189. (42) Shirakata, S.; Nakada, T. Near-Band-Edge Photoluminescnce in Cu(In,Ga)Se2 Solar Cells. Sol. Energy Mater. Sol. Cells 2011, 95, 219− 222. (43) Nelson, J. The Physics of Solar Cells; Imperial College Press, 2004.

10266

DOI: 10.1021/acsnano.6b05825 ACS Nano 2016, 10, 10258−10266