Monitoring the Formation of a CH3NH3PbI3–xClx Perovskite during Thermal Annealing Using X‐Ray Scattering

Grazing incidence wide and small angle X‐ray scattering (GIWAXS and GISAXS) measurements have been used to study the crystallization kinetics of the organolead halide perovskite CH3NH3PbI3–xClx during thermal annealing. In situ GIWAXS measurements recorded during annealing are used to characterize and quantify the transition from a crystalline precursor to the perovskite structure. In situ GISAXS measurements indicate an evolution of crystallite sizes during annealing, with the number of crystallites having sizes between 30 and 400 nm increasing through the annealing process. Using ex situ scanning electron microscopy, this evolution in length scales is confirmed and a concurrent increase in film surface coverage is observed, a parameter crucial for efficient solar cell performance. A series of photovoltaic devices are then fabricated in which perovskite films have been annealed for different times, and variations in device performance are explained on the basis of X‐ray scattering measurements.


Introduction
In recent years organometal halide perovskites have become the hot topic of solar cell research within the academic community.
wide angle X-ray scattering (GIWAXS) and small angle X-ray scattering (GISAXS) measurements are used to track the crystallization of the perovskite fi lm in situ. We note that the fi rst use of GISAXS for the investigation of organometal halide perovskite fi lms was made by Schlipf et al., who undertook a detailed investigation of two-step fabrication of planar CH 3 NH 3 PbI 3− x Cl x fi lms and concluded that the length scales present in the fi nal perovskite fi lms were very similar to those observed in the initial PbI 2 fi lm, suggesting that control of PbI 2 morphology is critical in controlling the morphology of the fi nal perovskite fi lm. [ 27 ] Liao et al. have also used ex situ GISAXS to investigate CH 3 NH 3 PbI 3 deposited via both one-step and twostep methods. [ 22 ] Here, evidence of pores was observed within the perovskite crystal grains, with such pores being subsequently fi lled by the electron or hole transporting material that was deposited onto the perovskite fi lm. The perovskite devices fabricated in this study using a one-step process, however, had a low average PCE of only 2.2% as a result of poor surface coverage. In our work, we use in situ GISAXS measurements to explore perovskite materials that were deposited from a onestep process. The inverted, planar-architecture PSC devices made from our fi lms achieve an average PCE of 12.2%. The in situ GISAXS measurements show an evolution of the size distribution of the perovskite crystallites during annealing of a precursor fi lm, which is confi rmed using ex situ scanning electron microscopy (SEM). We also apply a kinetic model to GIWAXS measurements on representative perovskite fi lms to extract the activation energy for the formation of CH 3 NH 3 PbI 3− x Cl x . Our work advances upon existing X-ray diffraction measurements on CH 3 NH 3 PbI 3− x Cl x and CH 3 NH 3 PbI 3 fi lms [22][23][24][25][26][27] by providing detailed access to the dynamics of structural changes during fi lm processing over three orders of magnitude in length scale.

In Situ GIWAXS
We fi rst consider the evolution in fi lm structure as characterized by in situ GIWAXS measurements performed on CH 3 NH 3 PbI 3− x Cl x fi lms having a thickness of ≈400 nm. An MAI:PbCl 2 precursor ink (3:1 molar ratio) was deposited via spin-coating under ambient conditions (relative humidity, RH ≈ 40%-45%) onto Si substrates with a native SiO 2 layer coated with PEDOT:PSS to replicate the hole transport layer employed in our solar cells. After spin-coating, the precursor fi lms were immediately transferred to a sample chamber for GIWAXS measurement and a single diffraction pattern was recorded at room temperature as a reference ( t = 0 min). Films were subsequently heated in situ and annealed under isothermal conditions for a period of up to 3 h (see the Experimental Section). Figure 1 shows a summary of the in situ GIWAXS data recorded during thermal annealing at 80 °C, a temperature at which resultant perovskite fi lms lead to high performance solar cells (see Figure S8, Supporting Information). In Figure 1 a,b (corresponding to t = 0 and 8 min respectively), it can be seen that the GIWAXS pattern is characterized by a large number of rings having a nonuniform intensity distribution in the azimuthal direction, indicative of a polycrystalline sample with some degree of crystallite orientation present. In Figure 1 f, we plot the radial line profi les recorded at different points within the annealing process. As reported previously, we observe an unidentifi ed, ordered, crystalline precursor phase during the early stages of annealing as evidenced by the rings located around q ≈ 0.45 Å −1 and ≈0.8-0.9 Å −1 . [ 24,26 ] In line with previous reports, the ring at q ≈1 Å −1 is attributed to the (110) and (002) planes of the tetragonal CH 3 NH 3 PbI 3− x Cl x perovskite. [ 4,9,25,28 ] It is apparent that this broad feature having maximum intensity at q ≈ 1.00 Å −1 is actually composed of a number of narrower peaks, with peaks evident at both q = 1.00 and 1.05 Å −1 . This observation is consistent with the existence of a series of crystallites within the fi lm having different lattice constants, perhaps due to the coexistence of multiple perovskite phases. [ 29,30 ] In fact, the diffraction rings observed in Figure 1 a-e are made up of a large number of diffraction spots having slightly different q values. We note both that the material commonly denoted as CH 3 NH 3 PbI 3− x Cl x actually appears to be structurally similar to CH 3 NH 3 PbI 3 , and that due to the temperature during the GIWAXS measurements (80 °C) being above the reported transition temperature for these materials from the tetragonal to cubic structure (≈55 °C), the diffraction ring at q ≈ 1 Å −1 may in fact be attributable to the (100) peak of the cubic structure. [30][31][32] Whilst it is tempting to estimate domain size from analysis of peak widths, we believe that this process would likely be in error due to (i) differences in lattice constant of the various domains, and (ii) peak broadening effects due to the integration of the scattering signal over the angular range χ = 0° to χ = 90°.
Due to the relatively high ambient humidity in the experimental environment (RH ≈ 40%-45%), it was found that the yellow precursor fi lm rapidly became grey during spin-coating. We attribute this to the formation of a perovskite upon exposure of the rapidly cooling precursor fi lm to ambient atmosphere, as has been reported previously in the Supporting Information of ref. [ 26 ] . Upon heating the fi lm returned to the bright yellow color whilst, as can be seen in Figure 1 a,b,f, the intensity of the q ≈ 1.00 Å −1 diffraction ring dropped to a minimum at t = 8 min and the intensity of the precursor peaks increased, indicating a return to the precursor phase. As the thermal anneal proceeded, the intensity of the q ≈ 1.00 Å −1 diffraction ring increased, reaching a maximum integrated intensity at t = 117 min. This is accompanied by a simultaneous reduction in the intensity of the peaks attributed to the crystalline precursor phase, indicating a transformation of the precursor into CH 3 NH 3 PbI 3x Cl x . In Figure 1 g it can be clearly seen that the precursor peak at ≈0.45 Å −1 has all but disappeared by ≈115 min, an observation that is also seen in the radial profi le of the diffraction image at 117 min shown in Figure 1 f. At the longest annealing times (164 min) a new peak emerges at q ≈ 0.9 Å −1 , attributable to PbI 2 . [ 24,33,34 ] Due to the overlap of this peak with the precursor peaks however, it is diffi cult to quantify the exact time at which it emerges. We fi nd however that the growth of this feature is accompanied by a relative reduction in intensity of the peak at q ≈ 1 Å −1 , indicating a decomposition of the CH 3 NH 3 PbI 3− x Cl x (this effect is particularly clear when annealing at higher temperatures as shown in Figures S1 and S2 in the Supporting Information). The relatively prompt onset of perovskite decomposition observed here is likely attributable to the high relative humidity (RH ≈ 40%-45%) during the experiment. [ 24 ] Although we cannot rule out possible beam damage accelerating this decomposition, it is notable that similar experiments performed under a dry nitrogen atmosphere do not lead to such rapid degradation, implying that humidity and/or oxygen are likely the principal cause of the observed decomposition under ambient conditions (see Figure S4, Supporting Information).
By comparing Figure 1 (a) and 1(d) it can be seen that some of the preferential crystallite orientation detected at early times persists throughout the annealing process. To determine the extent of this observation, we plot azimuthal profi les of the X-ray scattering intensity for the feature at q ≈ 1 Å −1 in Figure 2 (i.e., intensity vs azimuth χ , where χ is the azimuth of q z and q xy in scattering vector polar coordinates with χ = 0° at positive q z and χ = 90° at positive q xy ). [ 35 ] Data is presented at various times during the annealing process, with each trace normalized to its maximum intensity (at χ ≈ 6°). We fi nd that scattering intensity is initially greatest close to χ = 0°, indicating preferential orientation of the (110) and (002) perovskite planes in the out-of-plane direction. Note however that a relatively small population of CH 3 NH 3 PbI 3− x Cl x crystallites have in-plane orientation, as evidenced by the intensity of X-ray scatter close to χ = 90°. It can be seen that this preferential orientation persists during annealing, although X-ray scattering intensity increases slightly at intermediate angles of χ , implying that annealing drives a somewhat broader distribution of crystallite orientation relative to the structures formed during the preparation of the precursor fi lm. Preferential out-of-plane orientation of the (110) and (002) diffraction peak has previously been observed in planar CH 3 NH 3 PbI 3 fi lms on compact TiO 2 . [ 36 ] Very recently it has been reported that Images are presented at a) t = 0, b) t = 8 min (corresponding to the lowest integrated intensity of the q ≈ 1 Å −1 perovskite peak), c) t = 45 min (corresponding to 50% of the maximum integrated intensity of this peak), d) t = 117 min (the maximum integrated intensity of the q ≈ 1 Å −1 peak), and e) t = 164 min (the end of the annealing experiment). f) Azimuthally integrated intensity profi les determined at these times; vertical lines indicate peaks associated with perovskite (*), precursor phase ( †), and PbI 2 (•). g) The evolution of the GIWAXS profi le through the annealing process as a contour plot of (normalized) azimuthally integrated intensity.
planar fi lms prepared in a similar manner to that reported here had strong preferential out-of-plane orientation and attained high performance in a PSC. [ 37 ] In contrast, preferential in-plane orientation has been observed for CH 3 NH 3 PbI 3− x Cl x fi lms fabricated on mesoporous Al 2 O 3 scaffolds. [ 38 ] Following the approach of Moore et al., [ 25 ] we determine the activation energy E A for the formation of CH 3 NH 3 PbI 3− x Cl x following the isothermal annealing of an MAI:PbCl 2 precursor fi lm. Here, fi lms were annealed at three different temperatures (80 °C, 90 °C, and 110 °C) with relative changes in integrated intensity of the q ≈ 1 Å −1 peak on the GIWAXS data used to gauge the fraction of material transformed into the perovskite structure. Full details are provided in the Supporting Information. We fi nd E A = (85 ± 9) kJ mol −1 , a value in good agreement with the (86 ± 6) kJ mol −1 reported by Moore et al. for the same precursor system. Note that this activation energy corresponds to the entire transformation from precursor to perovskite and does not decouple activation energies relating to nucleation and growth modes.

In Situ GISAXS and Ex Situ SEM
We now discuss the evolution in fi lm structure during thermal annealing as characterized through in situ GISAXS measurements. These measurements complement the wide angle scattering dataset by characterizing the sample at q values as low as 10 −3 Å −1 which in real space corresponds to length scales as large as 400 nm, a size that is potentially commensurate with that of individual perovskite grains. [39][40][41][42][43] Measurements were performed under nominally identical conditions to the GIWAXS measurements, although the relative humidity was slightly higher (RH ≈ 45%-50% as opposed to 40%-45% during the GIWAXS experiments). Sample preparation followed the same procedures as used in the GIWAXS experiment described above, except that the spin-coated precursor samples were directly placed onto a preheated hotplate rather than being ramped from room temperature to the desired anneal temperature. Full technical details are provided in the Experimental Section.
Typical 2D GISAXS patterns collected at t = 0 and 120 min are displayed in Figure 3 e,f, respectively. Line profi les in q y were taken at the Yoneda Peak for each image. [ 44 ] Owing to a slight shading from the beam-stop, visible on the right-hand side of Figure 3 e,f, all line profi les presented are taken from the left side of the images. Quantitative analysis of GISAXS measurements has been undertaken using the distorted wave born approximation to simulate scattering from periodic features in the fi lm under investigation. Unfortunately, it did not prove possible to achieve a good fi t to the data at low q values due to the irregular shape and large size distribution of grains in the fi lms (as can be seen from the SEM images of nominally identical fi lms presented in Figure 4 ). Nevertheless, whilst a quantitative analysis has not been possible, we show below that a qualitative analysis can still provide useful information on the evolution of the fi lm during the annealing process.
The evolution of the GISAXS scattering intensity in q y during thermal annealing at 80 °C is shown in Figure 3 a. It can be seen that for annealing times longer than around 60 min, we observe an increase in intensity at lower q values although the shape of the profi le remains quite similar throughout the process. By plotting the data in terms of the change in scattering intensity relative to that recorded at t = 0 min, it is possible to more easily observe changes in the scattering profi le as a function of time during thermal annealing as shown in Figure 3 b. It can be seen that during the fi rst 30 min of annealing, the intensity at all q values is slightly lower than that observed at t = 0, and that no clear features can be identifi ed. At t > 40 min, we observe a gradual increase in intensity at q values below ≈0.02 Å −1 , corresponding to length scales greater than ≈30 nm. This increase in intensity gradually manifests itself as a peak in the profi le centered at ≈0.005 Å −1 , corresponding to a length scale of ≈130 nm. Note that the lower bound of our data is at q = 1.54 × 10 −3 Å −1 , corresponding to a length scale of ≈400 nm. Our measurements thus suggest that as the perovskite is formed from the crystalline precursor phase, there is an increase in the fraction and number of structures within the fi lm having length scales in the range 30-400 nm, and in particular an increase in the number of structures having length scales between 100 and 150 nm.
Plotting the logarithm of scattering intensity against q 2 at low q values (Guinier analysis) allows for further qualitative analysis. [ 45 ] This is shown for q y in Figure 3 c, while Figure 3 d presents a similar plot for q z (taken at q y = 0.003 Å −1 ). As the annealing process proceeds, the slope of the curve close to q y = 0 (Figure 3 c) decreases, attributable to a decrease in the average (lateral) diameter of the scatterers in the fi lm. Similarly, an increase in the slope after the Yoneda peak in q z , as seen in Figure 3 d, indicates an increase in the average height of the grains. Taken together, these observations suggest a change in the aspect ratio of crystal grains during the annealing and transformation process. Finally, the shift of the Yoneda peak toward larger q z values can be attributed to an increase in the average density of the fi lm.
In order to further investigate these changes in length scales, we have used SEM and atomic force microscopy (AFM) to image nominally identical fi lms that have been annealed Adv. Funct. Mater. 2016, 26, 4934-4942 www.afm-journal.de www.MaterialsViews.com at 80 °C for various lengths of time. Here we emphasize that while SEM and AFM measurements probe only the surface of the sample, the GISAXS measurements presented above probe the bulk of the fi lm since all measurements are made above the critical angle for the samples. From the SEM images presented in Figure 4 , it can be seen that as the annealing time is increased from 60 to 120 min, the fi lm evolves from a structure that is dominated by micron-sized grains separated by voids to one composed of many features at smaller length scales (hundreds of nanometers). At annealing times longer than 120 min, the fi lm morphology appears to stabilize and remains comparatively unchanged following further annealing. The decrease in the apparent diameter of the grains and the signifi cant increase in the fraction of grains having length scales of hundreds of nanometers during the period from t = 60 to 120 min is consistent with the GISAXS data described above. Further SEM images demonstrating this evolution in fi lm morphology are shown in Figure S5 (Supporting Information). Atomic force microscopy images of the perovskite surface, as shown in Figures S6 and S7 (Supporting Information), further confi rm such an evolution in lengths scales within the perovskite fi lm during annealing.
The change in length scales around t = 60-120 min is accompanied by an increase in the fi lm surface coverage from ≈80% Adv. Funct. Mater. 2016, 26, 4934-4942 www.afm-journal.de www.MaterialsViews.com Figure 3. a) Evolution of the GISAXS data slice in q y during an 80 °C thermal anneal and b) the same data presented as a relative intensity as compared to the data at 0 min. c,d) The Guinier analysis for data slices in q y and q z is shown respectively in parts (c) and (d). e,f) Detector data for 0 and 120 min; the slight shading on the right side of the feature is due to shading from the beam-stop, data were thus taken from the left side of the feature.
to ≈95%, which is consistent with the shift of the Yoneda peak toward larger q z values. This process is of particular importance due to the necessity of high surface coverage for the fabrication of high performance solar cells. Low surface coverage due to large voids in the perovskite fi lm leads to losses in solar cell efficiency due to both reduced light absorption (reduced short circuit current density, J SC ) and additional leakage paths resulting in a reduction in shunt resistance (reduced open circuit voltage, V OC and fi ll factor, FF). [ 9,46 ] The evolution of the fi lm to the more continuous morphology as observed at longer annealing times appears to be linked to the change in length scales as evidenced by the GISAXS data. In particular, the largest changes in GISAXS data and observed fi lm morphology both occur in the period from t = 60-120 min. We speculate that this evolution is critical in the formation of continuous and uniform fi lms, with such properties being necessary in high performance PSC devices prepared using a one-step process.

The Role of Annealing Treatment on Device Effi ciency
In order to correlate solar cell performance with the evolution of perovskite fi lms during the annealing process, fi lms that were nominally identical to those used for the X-ray scattering measurements were fabricated into complete perovskite solar cells having the structure ITO/PEDOT:PSS/ CH 3 NH 3 PbI 3− x Cl x /PC 70 BM/Ca/Al. Note we can confi dently relate the structures studied using GIWAXS and GISAXS to the device studies presented here, as in all cases the perovskite layer was prepared on a PEDOT:PSS coated substrate. In particular, fi lms prepared for GIWAXS measurements were fabricated on Si/SiO 2 /PEDOT:PSS substrates, while fi lms studied using GISAXS measurements and fabricated into devices were fabricated on glass/ITO/PEDOT:PSS substrates. The relative humidity during fi lm preparation and measurements using GIWAXS, GISAXS, and device fabrication was 40%-45%, 45%-50%, and ≈45% respectively. Perovskite fi lms were annealed at 80 °C for varying lengths of time from 80 to 180 min. Device metrics for the resulting solar cells are presented in Table 1 , with current-density against voltage plots for typical devices prepared using each annealing condition shown in Figure 5 a. For completeness, we plot histograms showing the spread of device performance for each annealing condition in Figure S8 (Supporting Information). It was found that fi lms annealed for times less than 80 min could be seen by eye to not have fully converted to a perovskite and are thus expected to exhibit poor device performance as well as limited device lifetime. As the annealing time increased from 80 to 100 and then 120 min, the average device PCE increased from 7.8% to 10.7% and fi nally to 12.2%. This improved effi ciency arises due to an increase in J SC from 15.8 to 17.2 and fi nally to 18.4 mA cm −2 , as well as an increase in FF from 50.6 to 64.0 and then 68.8%. At annealing times longer than 120 min, PSC performance begins to reduce with all device metrics undergoing a reduction. Our high effi ciency devices exhibit minimal hysteresis in their current density against voltage behavior, as can be seen in Figure S9 (Supporting Information). The observed trends in device performance can be understood with reference to the GISAXS, GIWAXS, and SEM data presented above. Initial increases in performance can be attributed to the combined effect of two processes. First, the increased annealing time leads to an increase in surface coverage, resulting in an enhanced J SC and FF. Concurrently we observe a more complete conversion of precursor materials to the perovskite, and indeed the maximum effi ciency of devices corresponds closely to the annealing time (117 min) at which the integrated intensity of Adv. Funct. Mater. 2016, 26, 4934-4942 www.afm-journal.de www.MaterialsViews.com the q ≈ 1 Å −1 GIWAXS diffraction peak takes its maximum value. This correspondence is confi rmed in Figure 5 b, where we present PCE and the transformed fraction of material [ x ( t )], determined by tracking the integrated intensity of the q ≈ 1 Å −1 diffraction peak as a function of annealing time (see the Supporting Information). Finally, at long annealing times the perovskite begins to undergo a decomposition into various products including lead iodide, as observed via the in situ GIWAXS data, leading to a reduction in device performance, as might be expected.

Conclusion
We have applied a variety of in situ and ex situ techniques to observe the formation of CH 3 NH 3 PbI 3− x Cl x from a precursor fi lm during a thermal annealing step. The activation energy of this transformation has been calculated to be 85 kJ mol −1 , a value in line with a previous measurement. Preferential orientation for precursor and perovskite crystallites in the out-of-plane direction is observed throughout the annealing process. In situ GISAXS measurements show that during a thermal anneal at 80 °C, there is an evolution in length scales within the fi lm, with a large increase in features having sizes in the range 30-400 nm. This evolution in length scales is correlated with an increase in fi lm surface coverage, which is known to lead to improved device effi ciency. Finally, devices with an average effi ciency of 12.2% have been fabricated for nominally identical fi lms having high surface coverage due to an optimization of the annealing time.
Preparation of Substrates and Films : Substrates were cleaned by sequential sonication in warm Helmanex, DI water, and isopropyl alcohol (IPA) before being rinsed in deionised (DI) water and dried with compressed nitrogen. A PEDOT:PSS (poly (3,4-ethylenedioxythiophene) polystyrene sulfonate) (Ossila Ltd.) layer was then spin-coated at 5000 rpm and annealed at 120 °C for 30 min before use, resulting in a layer with thickness ≈30 nm as measured by surface profi lometry. The perovskite solution was deposited via spin-coating at 4000 rpm in a onestep process. Here, the precursor solution was held on a 90 °C hotplate for ≥10 min before deposition, with substrates placed on a 90 °C hotplate for ≈5 min before deposition and then transferred immediately to the spin-coater for fi lm casting.
GIWAXS : Experiments were performed at the XMaS beamline at the European Synchrotron Radiation Facility (ESRF), Grenoble, France. Beam energy was 10 keV and the incident angle was 0.24°. Relative humidity was 40%-45% during the experiments. A custom made annealing chamber was mounted on a six circle Huber diffractometer. A 2D MAR detector was mounted on the diffractometer arm with calibration performed by scanning the detector vertically and horizontally and tracking the direct beam position. [ 47 ] Perovskite fi lms were spin-cast onto Si/SiO 2 /PEDOT:PSS substrates and immediately transferred to the experimental chamber. There was approximately a 2 min delay between spin-coating and the acquisition of the fi rst image on the cold hotplate ( t = 0). After acquisition of the initial image the hotplate temperature was subsequently raised to the desired value. The hotplate experienced an overshoot when heating (by ≈10 °C), with its measured temperature exceeding the target temperature within 1 min in the 80 °C and 90 °C cases and within 2 min in the 110 °C case. In all cases the temperature dropped to its intended temperature within 10 min of the acquisition of the fi rst image. Further details on how this has been accounted for in the determination of activation energy are given in the Supporting Information. Image processing and analysis was performed using the GI-XRD-GUI program. [ 47 ] GISAXS : Experiments were performed at the I22 beamline at the Diamond Light Source, Didcot, England. Here, the beam energy was set at 12.4 keV and was incident on the sample at an angle of 0.2°. Scattered X-rays were then detected using a Pilatus P3-2M detector. For measurement, perovskite fi lms were spin-cast onto glass/ITO/ PEDOT:PSS substrates before being transferred to the hotplate which was already held at the desired temperature. There was approximately a 90 s delay between spin-coating and the acquisition of the fi rst image ( t = 0). The relative humidity in the lab was measured at 45%-50%. In order to minimize exposure of the sample to X-rays, a fast shutter was used whereby the sample was only exposed to the beam during measurements (corresponding to a 5 s exposure once every 5 min).  www.afm-journal.de www.MaterialsViews.com Calibration was performed using a collagen sample. Image processing and analysis were performed using the DAWN software package. [ 48 ] Data slices in q y were taken at the Yoneda Peak for each image, and were an average of nine pixels in q z in order to improve measurement statistics.
Film Characterization : SEM images were taken using an FEI Inspect F scanning electron microscope. Here an incident beam energy of 5 kV was used in combination with a secondary electron detector. AFM measurements were performed using a Veeco Dimension 3100 in tapping mode with Budget Sensors Tap300-G tips. Gwyddion software ( http://gwyddion.net/ ) was used for image analysis.
Fabrication of Solar Cells : Prepatterned glass-ITO substrates were purchased from Ossila Ltd. Substrate cleaning and deposition of the PEDOT:PSS hole transport layer and perovskite active layer were as described above, with perovskite fi lms formed having a thickness of ≈380 nm. Following the thermal annealing step, the perovskite fi lms were moved into a nitrogen glovebox where PC 70 BM (99%, Ossila Ltd) was deposited by spin-coating of a 50 mg mL −1 solution in chlorobenzene (heated for 12 h at 70 °C and fi ltered using a 0.45 µm PTFE fi lter) at 1000 rpm, creating a layer having a thickness of ≈120 nm. Devices were then completed with 5 nm Ca and 100 nm Al cathode deposited by evaporation, before being encapsulated using UV epoxy and a glass slide. Three separate substrates were fabricated for each condition with each substrate comprising of four devices, giving a total of 12 devices per condition.
Device Characterization : Current-voltage characteristics were measured in ambient conditions using a Newport 92251A-1000 solar simulator (AM1.5 spectrum at an intensity of 100 mW cm −2 ). JV scans between 1 and −1 V were measured at a scan rate of 0.5 Vs −1 , with no light soaking or prebiasing used before measuring current-voltage characteristics. An aperture mask of 0.025 cm 2 was used to defi ne the device area. The effi ciency of the best 50% of devices was reported, thus excluding pixels containing a signifi cant defect whilst avoiding the use of a statistical method that is open to user bias.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.