Transferrable optimization of spray-coated PbI2 films for perovskite solar cell fabrication

Ultrasonic spray coating is a promising pathway to scaling-up of perovskite solar cell production that can be implemented on any scale – from table-top to mass production. However, unlike spin-coating, spray coating processes are not easily described by a set of machine-independent parameters. In this work, in situ measurement and modeling of wet film thickness and evaporation rate are presented as a machine-independent description of the ultrasonic spray coating process, and applied to fabrication process optimization for high-performing perovskite solar cells. Optimization based on physical wet film parameters instead of machine settings leads to better understanding of the key factors affecting film quality and enables process transfer to another fabrication environment. Spray coated PbI2 film morphology is analyzed under a range of coating conditions and strong correlation is observed between spray coating parameters and PbI2 film uniformity. Premature precipitation and sparse nucleation are suggested as causes of film non-uniformity, and optimal process parameters are identified. Device fabrication based on the optimized process is demonstrated under ambient conditions with a relative humidity of 50%, achieving a power conversion efficiency of 13% in 1 cm2 area devices, with negligible hysteresis.


Introduction
Despite impressive power conversion efficiencies achieved in perovskite-based solar cells, 1,2 it remains uncertain how perovskite solar cell fabrication will be scaled to industrial production. Among various methods, ultrasonic spray coating has been proposed as a path to scaling up the production of perovskite based solar cells. Several studies have demonstrated fabrication of one or more layers of perovskite 3-8 solar cells by ultrasonic spray coating. Although the previous studies reported optimized recipes, indicating machine settings, these studies did not quantitatively determine wet lm characteristics that result in high-performing dry lms. If a recipe is to be transferred to a manufacturing environment on a different scale or using equipment of different manufacturers, the properties of the wet lm that are necessary to achieve high-performing functional lms are more important than specic machine settings, which will vary depending on the machine design. In this work, we demonstrate both an optimized process and a method to determine wet lm thickness and evaporation rate, which are key parameters that determine the nal lm quality.
As a new technology reaches a stage where mass production is considered, development focus shis from champion device performance to reproducibility. Reproducibility between different locations on the substrate must be considered when device size is increased, and reproducibility between different batches must be considered when production volume is scaled up. For devices based on thin lm coating, lm uniformity is strongly connected to process reproducibility. In this study we develop the strategy to achieve uniform spray-coated lms as device area is increased above 1 cm 2 . We demonstrate that the mm-scale uniformity of spray coated PbI 2 lms is determined by spray coating conditions. Furthermore, we address the problem of reproducing the result in a different environment and on equipment where current machine settings may not be easily translated. Through a combination of direct measurement and modeling we accurately determine wet lm thickness and evaporation rate that correspond to optimal coating quality. By real-time monitoring of laser light interference and scattering by a wet lm, we are able to measure in situ the wet lm thickness and evaporation rate. Armed with accurate data for a single solvent, we are able to compute an evaporative mass transfer coefficient that is characteristic of the process chamber, and thereon predict evaporation rates for any solvent, based on data available in the literature.
Evaporation of a thin liquid lm is a complicated process that is strongly affected by the properties of the evaporating material, air ow pattern in the process chamber, and the geometry of the evaporating lm. In this work we demonstrate that with solvent properties available in the literature, air ow pattern can be characterized by in situ measurement of lm evaporation rate, and spray pattern geometry can be accounted for by spatially resolved computation using the COMSOL Multiphysics soware suite.
By quantitatively characterizing the wet lm under various coating conditions we can advance from intuition to quantitative prediction of coated lm quality. The method of in situ wet lm characterization that we present requires minimal modi-cation to machinery and can be easily adapted to another spray coating system to determine evaporative mass transfer coefficient and wet lm thickness as a function of specic machine settings. Once these parameters are known, process recipes become transferrable between the two machines. Even differences in spray patterns due to different sample sizes can be accounted for by local evaporation rate modeling described in this work.
The immediate goal of this work is optimization of the spray coating process to achieve the best uniformity of the PbI 2 precursor for the MAPbI 3 perovskite lm fabrication. In the sequential perovskite formation process, 9,10 the nal perovskite lm thickness is determined by the thickness of the PbI 2 layer, as methyl ammonium iodide (MAI) is supplied from solution as much as is necessary to react with PbI 2 . While the size and quality of perovskite crystal grains is driven by the concentration of the MAI solution 16 (see the ESI †), we show that the large scale (mm scale) uniformity of the perovskite lms is determined by the uniformity of the PbI 2 layer. Therefore, fabrication processes capable of achieving high uniformity of PbI 2 lm are crucial to scale-up of perovskite solar cell production.

Device fabrication
FTO glass substrates were cleaned by brushing with SDS solution, and washing in deionized water followed by ultrasonic cleaning in IPA for 15 min. A compact TiO 2 layer was fabricated by spray pyrolysis at 475 C, using a titanium diisopropoxide bis(acetylacetonate) precursor. The approximate TiO 2 compact layer thickness was 40 nm (measured by SEM cross-section imaging). Mesoporous TiO 2 was formed by spin coating (4000 rpm, 30 s) TiO 2 nanoparticle paste (Dyesol T-90) diluted 1 : 3 (by weight) with terpineol, dried (100 C for 3 min) and annealed (470 C for 1 hour). Before coating the PbI 2 , substrate was treated with UV-ozone (15 min). PbI 2 was spray coated using a "broad" spray pattern (see below), and heated to 100 C for 5 min to remove the solvent. PbI 2 ink was prepared by dissolving a specied concentration of PbI 2 in a mixture of DMF : DMSO, ratio 14 : 1 by volume. A perovskite layer was formed by immersing the sample in MAI solution (7 mg mL À1 in dehydrated IPA) for 45 s, drying by a nitrogen ow, and annealing at 100 C for 5 min. A hole transport layer was formed by spin-coating (2000 rpm, 20 s) spiro-MEOTAD solution (72 mg spiro-MEOTAD, 28.8 mL 4-tert-butyl pyridine, 17.5 mL Li-TFSI (520 mg mL À1 in acetonitrile), and 1 mL chlorobenzene). A metallic electrode was formed by thermal evaporation of gold through a shadow mask. All processing (except gold evaporation) was performed in ambient air, with temperature of 23-25 C and relative humidity 45-55%. Heat was applied using a hot plate in all processing steps. All materials were stored in dry nitrogen and solutions were prepared in a dry nitrogen atmosphere. Aer preparation, solutions were removed and exposed to air, and stored in sealed containers.

PbI 2 lm fabrication
Films shown in Fig. 1 were spray coated using a "broad" spray process, see below. PbI 2 ink was prepared by dissolving PbI 2 in a mixture of DMF : DMSO, ratio 14 : 1. The mixture of solvents was chosen to make it possible to dissolve a greater concentration of PbI 2 than possible for pure DMF at room temperature. Wet lm drying rate modeling and measurement was performed for the pure DMF solvent. Accordingly, the evaporation rates specied in Table 1 are for pure DMF. The DMF : DMSO mixture was experimentally measured to have approximately 10% lower evaporation rate than pure DMF, therefore it was considered a good approximation to perform modeling for pure DMF.

Spray coating parameters
Spray coating was performed using a USI Prism 300 ultrasonic spray coating system, equipped with a CAT-35 ILDS dual mode spray head (operated at 35 kHz, see Fig. S3 in the ESI †). Substrate temperature was controlled by a heated stage. For all coating processes, the substrate was allowed to equilibrate with stage temperature for 3 min before coating. The spray head lateral movement rate was 200 mm s À1 , it was operated in "wide" spray mode, and the sample was allowed to dry for 2 min aer coating before the process chamber was opened. Coating was accomplished in a single spray pass. For the "broad" spray process, the spray head height above the substrate was 100 mm and the carrier gas (dry N 2 ) pressure was 40 psi, resulting in a wet lm pattern approximately 100 mm wide across the direction of head motion. For the "narrow" spray process the head height above the substrate was 30 mm and the carrier gas pressure was 20 psi, resulting in a wet lm width of approximately 40 mm.

Sample characterization
X-ray diffraction (XRD) measurements were performed using a Bruker D8 Discover instrument (Bruker AXS K. K., Tokyo, Japan). External quantum efficiency (EQE) measurement was performed on an ORIEL EQE 200 system, operated in DC mode. Integrated I sc was computed from EQE data by assuming an AM1.5G spectrum, without any correction factors. Sample crosssection images were obtained by FIB milling and SEM imaging using a FEI Helios G3 dual beam system. Optical microscope images were obtained using a Leica DM4000 B microscope. Solar cell performance was characterized (in ambient air, with no encapsulation) using an ORIEL Sol1A solar simulator calibrated to 1 sun AM1.5 intensity using a calibrated silicon detector. I-V characteristics were measured using a Keithley 2420 source-measure unit. I-V sweeps were performed from À0.1 V to 1.1 V, steps of 1.2 mV, and dwell time of 100 ms. The sample was illuminated under open circuit conditions for 30 s before measurement.

Evaporation rate computation
Evaporation rates shown in Fig. 3f were determined as follows. For laser reection measurements, the average liquid evaporation rate was determined from oscillations of reected beam intensity at room temperature. Using this evaporation rate, the wet lm thickness was determined from the linear t to dry time/pump rate dependence. With known wet lm thickness, the average evaporation rate was determined for all temperatures and solvents from the linear t to the corresponding dry time/pump rate dependence.
Evaporation rate by mass change measurement was obtained by multiplying a directly measured rate of wet lm weight change by a factor that adjusts for different sample geometries (computed by modeling the local evaporation rate using COMSOL Multiphysics, see Section 3.5).
Model curves combine forced convection and free convection models, as described in the ESI. † Forced convection mass transfer constant was obtained from the direct measurement of Fig. 1 Transmission optical microscope images of (dry) spray-coated PbI 2 and corresponding MAPbI 3 films at varying substrate temperatures and ink pump rates. Ink pump rate is changed together with the ink concentration, maintaining a constant average PbI 2 amount per area. Columns indicate varying process temperatures; rows indicate varying ink pump rates. The PbI 2 film was reacted with 7 mg mL À1 MAI in an IPA solution for 45 s to form MAPbI 3 . The scale bar is 1 mm. mass change in the chamber for a specic solvent. Free convection mass transfer coefficient was calculated as described in ESI, † and multiplied by a geometry correction factor. It is important to note that the value of only one tting parameter was used to ne tune the model for all solvents and conditions. The geometry correction factor for forced convection was determined completely through modeling of the local evaporation rate using COMSOL Multiphysics, and the ratio between correction factors for narrow and broad spray patterns was also determined from the local evaporation rate model. For IPA, evaporative cooling was taken into account by using reduced liquid surface temperature in the model, for DMF evaporative cooling was deemed negligible (see the ESI †).

Image processing
Optical images in Fig. 1 were processed to remove intensity variations due to radially varying illumination intensity. Processing was accomplished by subtracting a 3rd order polynomial background automatically computed by GWYDDION 2.44 soware.

Challenges in spray coating
Formation of a coated lm from solution occurs in two stages: wet lm (volatile solvent still present and solids potentially mobile) and dry lm (volatile solvents removed and solids no longer mobile). The wet lm stage is critical to the formation of a uniform dry lm. Prominent causes of non-uniformity that occur in the wet lm are lm discontinuity, coffee stain effect, and Marangoni ow. Film discontinuity occurs when ink droplets impinging on the substrate do not form a at wet lm, but instead dry as separate droplets. A at wet lm will not form when the ink wetting of the substrate is poor (i.e., with a high liquid contact angle) or when the ink dries faster than the time it takes for the droplets to merge. The chief method of ensuring a good contact angle is to select an ink solvent with surface energy lower than that of the substrate. Wetting can also be improved by adding surfactants to the ink, or by oxygen plasma treatment of the substrate (to increase the surface energy).
The coffee stain effect is a term that is applied to describe accumulation of material at the edges of wet lms or droplets. It occurs due to preferential evaporation from a pinned edge of a wet lm, 11 and can cause an overwhelming fraction of the solids to be deposited at the edges of wet lms. The coffee stain effect can be counteracted by inducing Marangoni ow which acts in the opposite direction, for example by addition of a cosolvent with a suitable evaporation rate and surface tension. 12 It can also be minimized by creating a wet lm with a uniform thickness which will minimize the effective "edge" area. Increasing the substrate temperature will accelerate evaporation, reducing the time available for ink redistribution, however it is not suitable for suppressing the accumulation of ink solids at lm edges as it accelerates the coffee stain effect. 13 The Marangoni effect is a ow that occurs on the surface of a liquid in the presence of a gradient of surface tension. A ow will occur from the low surface tension region to the high surface tension region. Generally this will occur in cases of a mixture of liquids, and can be used to engineer a ow that counteracts the coffee stain effect, 12 however it can also occur in droplets of single solvent and cause accumulation of material in the center of the droplet. 14 Another source of non-uniformity that is important to consider is sparse nucleation of precipitate crystals. For salts such as PbI 2 deposition will occur preferentially at sites where nucleation has already occurred. Sparse/slow nucleation and a long drying time will result in accumulation of solids in sparse nucleation sites. To achieve the best lm uniformity, nucleation rate/density needs to be maximized, and the time during which ink solids are mobile (drying time) needs to be minimized.

PbI 2 spray coating optimization
Measurement of the contact angle of solvents suitable for dissolving PbI 2 (DMF, DMSO, and NMP) shows that the liquid contact angle on a TiO 2 surface is too low to be measurable (near-perfect wetting). Imaging of individual dried droplets of PbI 2 solution (Fig. S1 †) shows that visible rings of PbI 2 form at the droplet edges and no material accumulates in the droplet center. This indicates that the coffee stain effect is strong for PbI 2 dissolved in DMF (or DMSO, NMP), but the Marangoni effect is not observable.
To counteract the coffee stain effect by a co-solvent system 12 a solvent with a lower surface tension and a lower evaporation rate is required. The surface tensions of DMF, DMSO, and NMP are 34.4 dyn cm À1 , 42.9 dyn cm À1 , and 40 dyn cm À1 , respectively. The boiling temperatures of DMF, DMSO, and NMP are 153 C, 189 C, and 202 C respectively. In this case, low evaporation rate (usually corresponding to a solvent with a high boiling temperature) solvents have a higher surface tension as well, so a co-solvent mixture to counteract the coffee stain effect cannot be prepared from these solvents, and other solvents that dissolve PbI 2 are not readily available. Therefore, the remaining pathway to suppressing the coffee stain effect is to minimize the wet lm thickness to achieve a faster drying time without increasing the temperature.
Keeping the above guidelines in mind, it is intuitive that the best dry lm uniformity would be achieved at a low temperature and a low wet lm thickness. Fig. 1 shows transmission light microscope images of spray coated PbI 2 and resulting MAPbI 3 lms fabricated at varying substrate temperatures and wet lm thicknesses. The ink concentration is adjusted to maintain the same amount of PbI 2 per area in all samples. Table 1 gives the spray coating parameters for each sample in Fig. 1, as well as the calculated wet lm thickness and DMF evaporation rate. It is observed that at elevated temperatures (samples B, C, E, F, H and I), thin wet lms result in dry lms with mm-scale nonuniformity, indicating that precipitation occurred before droplets could form a at wet lm. At the process temperature of 30 C, lms appear uniform in optical images. Detailed examination by AFM (Fig. 2a-c) reveals that perovskite lms formed from PbI 2 lms sprayed at high wet lm thicknesses have reduced surface coverage, possibly due to sparse nucleation of PbI 2 during lm drying and resulting in microscopically uneven lms. To quantify this observation, solar cell samples (fabricated as described in Materials and methods) were imaged in AFM aer MAPbI 3 lm formation, before the deposition of the hole transport layer and metal electrode. Fig. 2f shows a clear negative correlation between wet lm thickness of the PbI 2 spray coating process and surface coverage fraction by MAPbI 3 .
Data in Fig. 1 demonstrate that the mm-scale uniformity of coated PbI 2 lms is strongly correlated with spray coating conditions, and that the uniformity pattern of PbI 2 lms is re-ected in the resulting MAPbI 3 lms. Fig. 2f indicates that a larger wet lm thickness results in poorer surface coverage by MAPbI 3 . A low wet lm thickness and a low evaporation rate are optimal for lm uniformity. A high evaporation rate results in the mm-scale non-uniformity (i.e., macroscopic non-uniformity), whereas a high wet lm thickness results in reduced surface coverage on the mm scale (i.e., microscopic non-uniformity).

In situ measurement of wet lm thicknesses and evaporation rates
As demonstrated above, the optimal coating uniformity is achieved in a narrow range of wet lm thickness and evaporation rate. It is then necessary to accurately determine wet lm thickness and evaporation rate, so that the optimized recipe can be transferred to other fabrication sites or coating machines. Determining the thickness of the wet lm in a spray-coating system is a non-trivial task. While it may appear that with a known spray prole the wet lm thickness can be calculated from the amount of liquid ink applied per area, in practice that is not the case. The spray prole is generally unknown and difficult to characterize, the solvent will evaporate from the ink during ight from the spray head to the coating surface, a fraction of the droplets may not reach the substrate, and ink may ow laterally before it achieves a stable thickness. These factors combined make it impractical to approximate the wet lm thickness by simple mass conservation, and demand a direct measurement. Fig. 3a shows the principle of operation for a wet lm measurement system that allows in situ measurement of wet lm thickness in a spray coating system. A laser beam is set up to be reected from the substrate that is being spray coated. The substrate must be at least partially reective, a requirement that is readily satised by transparent glass-FTO, or glass-ITO substrates typically used for perovskite solar cell fabrication. When a lm of pure solvent is sprayed onto the substrate, a fraction of the light is scattered or absorbed, reducing the total amount reaching the detector. Additionally, the wet lm creates a second partially reective surface. The laser beams reected from the liquid lm surface and the substrate surface will interfere constructively or destructively depending on the thickness of the wet lm (principle of operation of anti-reective coatings). If the intensity is tracked in real time, the reected beam intensity will oscillate. The reected beam intensity will go through a complete oscillation when the path difference between the beams reected from the liquid and from the substrate will change by one wavelength. Note that scattering and absorption do not mask the interference as long as they do not change on exactly the same time-scale. If the change of lm thickness in one complete oscillation is denoted as Dh then where l is the laser wavelength, b is the angle of incidence (to the normal, see Fig. 3a), and n is the index of refraction of the solvent. The formula above takes into account the change of the wavelength of light inside the liquid lm and refraction of the laser beam (Snell's law). Measuring the time between oscillations will yield the rate of lm thickness change. The total lm drying time can be determined by measuring the time that the reected laser light intensity is reduced by scattering. The wet lm thickness can be computed by multiplying the drying time by the thickness change rate. This method assumes that during the majority of the drying time the evaporation rate of the solvent lm is constant. This is a good assumption as long as the lm thickness does not limit the heat ow available for evaporation. Experimental observation of evenly spaced reected intensity oscillations supports the assumption that the evaporation rate of a at wet lm is constant once the lm is stabilized (see Fig. 3b). Fig. 3c-e show that the drying time depends linearly on the amount of dispensed ink, further supporting the assumption that evaporation rate is independent of lm thickness. It is notable that the best accuracy wet lm drying time and evaporation rate measurement are obtained when the total drying time is longer than 20 s. This time frame ensures that the measurement time is much longer than the spray coating time (about 0.5 s). Fig. 3c-e show that at temperatures above 40 C drying time can become less than 10 s. To accurately estimate wet lm evaporation rate in this regime modeling is employed as described in Sections 3.4 and 3.5. To further validate evaporation rates determined from laser reection and interference, the evaporation rates of DMF and IPA were determined by directly measuring the rates of mass change of a liquid lm (with a known area) in the process chamber. Rates measured by laser reection and direct mass change are shown in Fig. 3f (see Materials and methods for details).
The method based on laser light reection is effective if the reectivity of the surfaces is constant and the wet lm evaporates completely (no precipitation). This means that ink properties have to be inferred from pure solvent properties. This assumption is valid if the solvent viscosity and evaporation rate are not strongly affected by the presence of the solute. For nonpolymer materials such as PbI 2 this is a good assumption as even at a saturated concentration, the solvent viscosity is not strongly affected. The evaporation rate will be slightly affected by the presence of the solute in the ink; if necessary, this can be accounted for by Raoult's law. 15 It can also be noted that in the presence of the precipitate, aer most of the solvent has evaporated, solvent evaporation rate may be substantially reduced if it is absorbed into a porous precipitate and is no longer accurately described by a thin liquid lm on a at surface. Our model addresses only the condition where a continuous liquid surface is present, as in this regime ink solids are highly mobile. We do not address the condition of solvent absorbed in a porous precipitate because in that condition ink solids are no longer mobile over macroscopic (mm-scale) distances.

Evaporative mass transfer coefficient
To complement and extend the measurement of the evaporation rate by laser reection from the substrate, we also demonstrate that the evaporation rate of a single solvent (e.g. DMF) at low temperatures can be used to compute the evaporative mass transfer coefficient of the process chamber and predict the evaporation rates of other solvents and temperatures using solvent properties available in the literature.
The evaporation rate of a single solvent can be computed as follows: where C is the evaporation rate in mol s À1 m À2 , C m is the evaporative mass transfer coefficient, P sat is the (temperature dependent) solvent saturation pressure close to the liquid surface, and P inf is the solvent partial pressure at innity, R is the ideal gas constant, and T is the temperature (in K). Saturation pressure values can be obtained from the literature, whereas C m must be computed taking into account the air ow pattern in the process chamber and evaporating surface geometry. The evaporative mass transfer coefficient without external heating is dominated by the air ow due to process chamber exhaust. Computation of the mass transfer coefficient due to a forced air ow requires detailed knowledge of the air ow rate above the evaporating surface, which is very difficult to determine. Instead, the evaporative mass transfer coefficient at room temperature was directly measured by tracking the weight change of the sample placed on a precision scale in the spray coating chamber. Because the evaporation rate is signicantly elevated at sample edges, the average evaporation rate depends on the evaporating surface geometry. Therefore, to determine the evaporation rate of a spray-coated wet lm, the local evaporation rate was determined by modeling in COMSOL Multiphysics (see details below).
At elevated temperatures buoyant convection dominates the air ow. The mass transfer coefficient for buoyant convection can be computed from the sample geometry, air properties, and heated surface temperature 15 where Sh is the dimensionless Sherwood number, D is the diffusivity of the evaporating species in air, and L is the characteristic length. Since the real chamber geometry is not always well described by idealized assumptions of analytical calculations, the characteristic length in eqn (3) was adjusted to maximize the agreement between the model and data from laser reection. It is important to note that only one value of characteristic length was used for two different spray widths and two solvents (IPA, DMF) of very different evaporation rates and excellent agreement between the data and model was obtained. Therefore, once the model is rened by tting to the data from one solvent and one spray condition, it can be used to predict evaporation rates for other solvents and conditions without the need to introduce more tting parameters. Details of the computation of C m can be found in the ESI. †

Modelling of the local evaporation rate
For a nite-sized evaporating surface, there is a signicant difference between evaporation near and far from the edges. Therefore, if the evaporative mass transfer coefficient is known for a specic evaporating surface shape, it is necessary to introduce a correction factor to determine the evaporation rate for a different evaporating surface geometry. To compute the correction factor, COMSOL Multiphysics 5.2 soware was used to determine the local evaporation rates for an arbitrary-shaped surface. Evaporation was modeled by constraining the concentration of evaporating species to match the saturation pressure across the evaporating surface and computing the ux of evaporating species, with appropriate temperature and air ow distributions. Air ow distribution in a realistic geometry is very difficult to compute, therefore it is approximated by a laminar ow set to achieve an experimentally measured average evaporation rate. The experimental value of the average evaporation rate was determined by tracking the change of weight of a square substrate completely covered with DMF, placed on a precision scale in the spray process chamber. Fig. 4 shows an example computation of a difference between evaporation rates at the sample edges and center. The average evaporation rate for a 15 Â 15 mm area at the center of a 50 Â 50 mm surface of evaporating liquid surface is 0.77 of the rate averaged over the entire liquid surface (including the edges). Local evaporation rate modeling can also be used to compute the evaporation rate adjustment when spray patterns are changed, as would be expected to happen when scaling-up to higher volume production. Fig. 3f shows evaporation rates computed for a broad DMF spray pattern (100 mm) and a narrow spray pattern (40 mm).
To summarize, results of three complementary approaches to determining the evaporation rate of a lm are shown in Fig. 3f. The lm evaporation rate and lm thickness can be completely determined immediately following spray coating by interference and scattering of a reected laser beam. Once the laser reection measurement is accomplished for a single solvent and several temperatures, the evaporative mass transfer coefficient of the process chamber can be computed and applied to predict evaporation rates of different solvents and at temperatures where the lm evaporation rate is too fast for accurate measurement. The laser reection measurement and modeling are veried by a direct measurement of mass change of a wet lm of a known area. The set of process parameters to achieve optimal PbI 2 lm uniformity are: temperature ¼ 30 C and ink pump rate ¼ 3.5 mL min À1 , corresponding to wet lm thickness 2.1 mm and DMF evaporation rate of 0.77 mmol s À1 m À2 .

Device fabrication
In this study we focus on the optimization of process parameters to achieve optimal uniformity of spray-coated PbI 2 lms and characterization of the process in terms of machine-independent parameters. As the ultimate application of this work is the fabrication of perovskite solar cells, it is necessary to demonstrate that the analyzed lms indeed result in high performance solar cells.
To demonstrate complete devices incorporating a spray coated PbI 2 layer, we chose a commonly used perovskite solar cell architecture that is known to be compatible with ambient processing. 10 Fig. 5a shows the layer structure of the cell and a scanning electron microscopy (SEM) cross-section image of the optimized design. The spray coating process for PbI 2 lms was selected to match the lm with the best uniformity (sample D, temperature ¼ 30 C and ink pump rate ¼ 3.5 mL min À1 ). Device performance data for non-optimal wet lm thickness can be found in ESI Fig. S8. † As expected, the perovskite solar cells fabricated using the optimal parameters show not only the highest PCE, V oc , J sc , and FF, but also the lowest standard deviations, strongly suggesting the signicantly improved device reproducibility under the optimal coating conditions. Fig. 5c shows the XRD diffractogram of a typical device. Peaks at 2Q ¼ 14.1 corresponding to MAPbI 3 and 2Q ¼ 12.7 corresponding to unreacted PbI 2 are observed. The external quantum efficiency (EQE) spectrum shown in Fig. 5d shows slightly greater EQE at shorter wavelengths, also suggesting the presence of unreacted PbI 2 . In several recently published studies it has been suggested that residual PbI 2 may be difficult to eliminate in a 2-step process, and may not necessarily lead to substantial performance loss. Ko et al. 10 implemented a 2-step perovskite formation process under ambient conditions and found that optimal performance was obtained with PbI 2 peak in the XRD pattern around 50% as high as MAPbI 3 peak. Kim et al. 17 discussed in detail the formation process of the MAPbI 3 in a 2-step process, and concluded that excess PbI 2 cannot be eliminated by simple optimization of MAI solution concentration and immersion time, without compromising the performance. On the other hand, it is expected that further enhancement of MAPbI 3 crystallinity can lead to increased performance. Table 2 shows performance statistics for devices incorporating a spray-coated PbI 2 layer. Maximum power conversion efficiency (PCE) of 13.0% is achieved with the average PCE of 10.2%. Device performance is on par with the best performance achieved for spray coated devices with an active area of 1 cm 2 . 8 We also observed that these devices could be operated at the maximum power point for over a 100 hours preserving 60% of initial PCE. See Fig. S7 in the ESI † for steady state measurement of device performance. Thus, we can conrm that the PbI 2 lms for which uniformity was optimized are suitable for highperformance perovskite solar cell fabrication.

Conclusions
We have demonstrated an optimized set of parameters for achieving the best uniformity of ultrasonic spray-coated PbI 2 lms, at a thickness suitable for high performance perovskite solar cell fabrication. We observe a strong correlation between wet lm thickness, evaporation rate and uniformity of dry lms, and identify wet lm thickness and evaporation rate as key parameters to dry lm quality. To enable expedient transfer of a recipe from the specic equipment in the laboratory to the manufacturing environment, we demonstrate a simple way to in situ measure wet lm thickness and evaporation rate. Furthermore, we demonstrate that evaporation rate measurement of a single wet lm can be used to determine the evaporative mass  transfer coefficient of the process chamber. Knowledge of evaporative mass transfer coefficient allows us to compute evaporation rates for inks based on any solvent. We also demonstrate modelling of the local evaporation rate which makes it possible to compute necessary adjustment to process parameters when changing the sample geometry. Using optimal process conditions, we fabricated perovskite solar cells, device performance of which is on par with the best performance achieved for spray coated devices with an active area of 1 cm 2 .