The growth of methylammonium lead iodide perovskites by close space vapor transport

Vapor deposition processes have shown promise for high-quality perovskite solar cells with potential pathways for scale-up to large area manufacturing. Here, we present a sequential close space vapor transport process to deposit CH3NH3PbI3 (MAPI) perovskite thin films by depositing a layer of PbI2 then reacting it with CH3NH3I (MAI) vapor. We find that, at T = 100 °C and pressure = 9 torr, a ∼225 nm-thick PbI2 film requires ≥125 minutes in MAI vapor to form a fully-reacted MAPI film. Raising the temperature to 160 °C increases the rate of reaction, such that MAPI forms within 15 minutes, but with reduced surface coverage. The reaction kinetics can be approximated as roughly first-order with respect to PbI2, though there is evidence for a more complicated functional relation. Perovskite films reacted at 100 °C for 150 minutes were fabricated into solar cells with an SLG/ITO/CdS/MAPI/Spiro-OMeTAD/Au structure, and a device efficiency of 12.1% was achieved. These results validate the close space vapor transport process and serve as an advance toward scaled-up, vapor-phase perovskite manufacturing through continuous vapor transport deposition.


Introduction
As the stability 1,2 and efficiency 3-6 of lead halide perovskite solar cells improve, scaling laboratory performance to industrial production becomes increasingly important. The majority of published research currently focuses on scaling established perovskite spin-coating processes to continuous solution processes such as slot-die coating or blade-coating. [7][8][9][10][11] However, high-efficiency perovskites can be fabricated using a variety of vapor deposition techniques with established pathways to solvent-free, scaled-up manufacturing. [12][13][14][15] One such method is "close space vapor transport" (CSVT), which serves as a laboratory precursor for large-area, commercial production through vapor transport deposition (VTD). 16 In CSVT, the source material of interest is evaporated into a carrier gas, transported to a substrate, and condensed to form a thin lm. The technique provides insight into how different temperature and pressure regimes, in various carrier-gas environments, can be used to control the deposition rate, uniformity, and crystallinity of the deposited lms before designing and implementing scaled-up VTD reactors. CSVT has been used extensively for CdTe solar cell research, where it was regularly used to produce highquality lms that resulted in high-efficiency solar cells. 16 This fostered the deployment of a commercial VTD process for largearea, uniform CdTe deposition by First Solar LLC. 16 Despite this success, an all-vapor CSVT process for perovskite thin lm deposition has not been investigated. Therefore, feasible processing temperatures and pressures have not been dened, and their effect on deposition rate, lm quality, and device performance are unknown. There are only a few reports of methylammonium iodide (CH 3 NH 3 I, MAI) reactions in a CSVT conguration with spin-coated PbI 2 followed by annealing to drive a reaction between PbI 2 and MAI vapor. [17][18][19] However, these studies do not provide sufficient information to develop a VTD reactor for scaled-up perovskite manufacturing because they omit the vapor deposition of the PbI 2 lm.
The application of an all-vapor CSVT process to directly deposit methylammonium lead iodide (CH 3 NH 3 PbI 3 , MAPI) thin lms must differ from CdTe deposition by CSVT, where the absorber sublimes from a single molecular source to form Cd and Te 2 vapors that react at the substrate and form a solid lm, 16 because the vapor pressures of the constituent compounds of MAPI perovskites (PbI 2 and MAI) are too dissimilar to deposit stoichiometric lms from a single source. 20,21 Therefore, a two-stage CSVT process has been developed to form the MAPI absorber layer by depositing a lm of PbI 2 and reacting it in MAI vapor as shown in Fig. 1. Here we report the deposition of PbI 2 lms via CSVT, their reaction with MAI vapor, a reaction kinetic model, an upper-threshold to reaction temperature, and functional devices achieving 12.1% efficiency.

Substrate preparation
Soda-lime glass (SLG) substrates were sequentially cleaned by sonication in LiquiNOX® and CitraNOX® soaps, followed by rinsing in DI water and drying at 180 F for 1 hour. Aer drying, a 300 nm layer of ITO was RF sputtered at room temperature onto the substrates through a shadow mask to dene cell area.

CdS deposition by chemical surface deposition
CdS was chosen as the electron transport layer because its conduction band lies 0.3 eV below the perovskite's conduction band and due to its good hole-blocking properties. 22 CdS was deposited on the ITO-coated SLG substrates using chemical surface deposition (CSD). 23 The SLG/ITO substrates were heated on a hot plate to 55 C and a solution was prepared containing 2.2 mL 1.5 mM CdSO 4 (99.996%, Alfa Aesar), 2.2 mL 1.5 M thiourea (99%, Alfa Aesar), 2.8 mL 30% NH 4 OH (JT Baker) and 15 mL DI water. Aerward, 1.5 mL of the solution was dispensed dropwise onto each of the hot substrates. Aer 5 minutes, the substrates were removed from heat, tipped of uid, rinsed with owing DI water, and dried with argon. The process was repeated to apply a second coat of CdS, yielding a uniform, z50 nm thick layer.

CSVT reactor
A photograph of the CSVT reactor developed for these experiments is shown in Fig. 2a. It consists of a planar sourcesubstrate geometry with the source positioned beneath the substrate. A motorized arm is used to position the apparatus in a 4.5 cm diameter quartz tube between two 1000 W lamps. These lamps independently heat two graphite susceptors, which are in direct contact with the source material and the substrates, as seen in Fig. 2b. Eurotherm 2404 controllers moderate the temperature of each susceptor using embedded thermocouples so that the heating lamps control the source and substrate temperatures between 25 and 500 C. A roughing pump, an argon carrier gas owing at 10 sccm, two Baratron capacitance manometers, and a throttle valve maintain the working pressure between 0.1 and 100 torr. The CSVT reactor is integrated with a multi-function glove box used to prepare, load, and unload the sources and substrates in a nitrogen atmosphere with O 2 < 0.1 ppm and H 2 O < 5 ppm. This allows the perovskite layer to be processed start-to-nish without exposure to ambient air. Fig. 2b shows an exploded schematic of the source and substrate holder. Insulating Macor® ceramic ttings hold the 10.8 cm long source and substrate susceptors in place and maintain a temperature gradient across a 2.2 mm gap dened by the spacer. This gradient promotes mass transport from the source to the substrates, resulting in deposition and lm growth. The ttings also dene the deposition area to be z20 cm 2 , which accommodates three 2.5 cm Â 2.5 cm substrates per run.
CSVT processing of MAPI is shown schematically in Fig. 1. CdS-coated SLG/ITO substrates are loaded face down into the CSVT chamber over a PbI 2 source, and the system is pumped to a base pressure of 10 À7 torr. The chamber is then backlled with argon gas to a xed working pressure, and the system is allowed to equilibrate for 5 min. The lamps ramp the source and substrate susceptors to their temperature setpoints over 5 min at a xed rate of 47 C min À1 for the source and 38 C min À1 for the substrate. The source and substrate temperatures are then controlled to within AE2 C of their setpoints for a specied duration by independently modulating the top and bottom lamp power.
Following PbI 2 deposition, the PbI 2 -coated substrates and the PbI 2 source are unloaded from the CSVT reactor. Then a second source, lled with MAI powder, is loaded into the CSVT reactor with the PbI 2 -coated substrates. Aer pumping and backlling, the source and the substrate susceptors are heated to their setpoints at a rate of 25 C min À1 . MAI sublimes and saturates the headspace surrounding the PbI 2 lm, driving a solid-vapor reaction to form MAPI. Finished MAPI lms are removed from the CSVT chamber into the glove box, washed with isopropanol (anhydrous 99.5%, Sigma Aldrich) to remove residual MAI, blow-dried with argon, and annealed on a hotplate at 100 C for 5 minutes.

Rear contact preparation
Spiro-OMeTAD (Spiro) (99%, Sigma Aldrich) lms were deposited using an established recipe from literature. 24 The Spiro solution was stirred for 10 minutes and ltered through a 0.2 mm PTFE lter to remove large particles prior to spin-coating. Spiro lms  were dynamically spin-coated by aliquoting 100 mL of solution onto the substrate at 4000 rpm for 10 seconds. The lms were dried and stored in the dark in a dry air desiccator (relative humidity < 7.5%) overnight prior to metallization.
Oxidized SLG/ITO/CdS/MAPI/Spiro samples were loaded into an electron beam evaporator and pumped to a base pressure of 10 À6 torr. Gold lms were deposited through a shadow mask at a rate of 6 A s À1 to a nal thickness of 100 nm. The active area of the solar cells was dened by the overlap area between the gold and the ITO to be 0.24 cm 2 . The edges of the perovskite were removed with a razor blade to expose the ITO contacts, and silver paste was applied using Dotite® D-550 Silver Colloid (2SPI).

Device measurement
Current-voltage (J-V) measurements were carried out using a setup maintained in the N 2 -lled glovebox. The intensity of a xenon arc lamp (Newport Oriel 67005 Housing; Newport Oriel 69907 Power Supply) was calibrated using a Si solar cell with a Schott KG-5 IR lter attached to the front. This limits the spectral sensitivity of the Si calibration cell to 800 nm and matches the spectral sensitivity of the calibration cell to test devices. The calibration value is based on quantum efficiency (QE) measurements and J-V measurements on a class A Oriel solar simulator.
J-V measurements were carried out in a 4-point probe conguration with scans between À0.4 and 1.2 V at a sweep rate of 200 V s À1 with 48 data points. Cells were scanned from reverse to forward bias then back. Aerward, cells were prebiased at 1.2 V for 60 seconds prior to measurement from 1.2 V forward bias to À0.4 V reverse bias. The data exhibited minor oscillations due to capacitive coupling between the device and the source measure unit (SMU) or temporal variations in the light intensity, and, therefore, the data was smoothed using a 5-point adjacent average for clarity.

Materials characterization
Three scanning electron microscopes (SEM) were used to collect micrographs of various samples. An AMRAY 1810T Digital SEM was used at an accelerating voltage of 20 kV to image MAPI samples reacted at 160 C and perform energy dispersive X-ray spectroscopy (EDS). A Zeiss Auriga 60 High Resolution Field Emission SEM and a JSM-7400 High Resolution SEM were used at 3 kV to image plan-view and cross-section imaging, respectively.
X-ray uorescence (XRF) measurements were obtained using an Oxford Instruments X-Strata 980 Coating Thickness Analyser to measure the thickness of the PbI 2 lms. Samples were measured over a 0.75 00 Â 0.875 00 area of the lm using a 7 Â 8 grid for a total of 56 equally-spaced data points that were used to determine the average thickness of each lm.
The CSVT lms were analyzed by X-ray diffraction (XRD) using a Philips/Norelco powder X-ray diffractometer using a CuKa X-ray source operating at 35 kV and 20 mA in Bragg-Brentano parafocusing geometry. The scan settings were 0.05 2q per step with a 4 second dwell time, over the 2q range 10-30 to cover the principal peaks of PbI 2 and MAPI. The XRD data was smoothed and processed with the Rachinger correction to remove contributions from CuKa 2 . 25 All peaks were indexed and assigned to phases based on d-spacing conformity with references for ITO (cubic In 2 O 3 , ICDD 00-006-0416), PbI 2 (hexagonal, ICDD 01-080-1000), and MAPI (tetragonal, pattern generated using PowderCell 2.0 for space group I4cm).
The weight percent of PbI 2 (w PbI 2 ) in each sample was estimated from the intensity ratio (R) of the PbI 2 (001) and MAPI (110) reections, corrected for background, using standard quantitative powder diffraction analysis as shown in eqn (1). 26 The detection limit of z0.1% is governed by the signal-to-noise of the data, which is governed by the scattering power of each phase, the instrumental noise, and the scan conditions.

Deposition of PbI 2 thin lms
Previous research indicates that for single-junction, perovskite solar cells, the optimal thickness of the MAPI absorber layer is This journal is © The Royal Society of Chemistry 2020 RSC Adv., 2020, 10, 16125-16131 | 16127 z400 nm, 27 which requires the PbI 2 precursor lm to be z200 nm thick. 28 PbI 2 was deposited using a source temperature (T source ) ¼ 260 C, and a substrate temperature (T sub ) ¼ 215 C, according to a Clausius-Clapeyron t of its vapor pressure ðP vap PbI2 Þ, 20 to produce a vapor pressure difference of two orders of magnitude between the source (P vap source ) ¼ 2 Â 10 À6 AE 8 Â 10 À7 torr and the substrate (P vap sub ) ¼ 6 Â 10 À8 AE 3 Â 10 À8 torr. Depositions were carried out at a pressure (P) ¼ 1 torr and the deposition time (t) was varied to achieve a desired lm thickness. A 20 min deposition (t ¼ 20 min) produced lms with an average thickness of 225 AE 7 nm, measured by XRF.
The XRD pattern of a PbI 2 lm on an SLG/ITO/CdS substrate shown in Fig. 3a is fully indexed using hexagonal PbI 2 and cubic In 2 O 3 standards. Notably, CdS diffraction peaks are not observed due its low thickness (z50 nm) and poor crystallinity.
The PbI 2 XRD pattern shows the strongest diffraction peak at 2q ¼ 12.60 , which corresponds to the (001) basal plane of hexagonal PbI 2 . In a random pattern of PbI 2 powder, the brightest peak occurs at 25.90 and corresponds to the (011) plane. This suggests that the PbI 2 lms deposited on CdS orient around the basal plane during the deposition process. Fig. 3b shows a plan-view SEM micrograph of the same PbI 2 lm. The morphology consists of hexagonal platelets that are consistent with the quasi-two-dimensional, 2H polytype of the PbI 2 hexagonal crystal structure. 29 While the lm appears to have a rough texture, the PbI 2 platelets seem to fully coat the CdS substrate with no visible pinholes.

Reaction of PbI 2 in MAI vapor
MAI reactions of PbI 2 lms were carried out isothermally at T source ¼ T sub ¼ 100 C. The amount of MAI participating in the reaction is estimated from a Clausius-Clapeyron t of the vapor pressure, 21 which gives P vap MAI ¼ 50 AE 20 mTorr at T ¼ 100 C. The system pressure was maintained at a constant P ¼ 9 torr using an argon ambient, which diluted the concentration of MAI vapor.
A series of reactions were carried out from 25-150 min in 25 min increments. Fig. 4a shows the XRD patterns of the reacted lms at each point in time. The conversion of PbI 2 to MAPI is conrmed by the disappearance of the PbI 2 (001) diffraction peak at 12.60 and the appearance of the MAPI (110) and (220) peaks at 14.05 and 28.40 respectively. The PbI 2 (001) intensity decreases and the MAPI (110) signal increases with increasing reaction time. Aer a 125 min reaction, the PbI 2 signal is indistinguishable from the background. The morphology of the 150 min-reacted lm from Fig. 4a is shown in Fig. 4b. The lm has a dense morphology that appears to be continuous across the CdS substrate without visible pinholes.
Quantitative X-ray diffraction intensity analysis 26 was performed on the XRD patterns shown in Fig. 4a. A comparison of the PbI 2 (001) and MAPI (110) peak intensities yielded the weight percent of PbI 2 as a function of the reaction time, which was converted to moles (N PbI 2 ) and plotted in Fig. 5. As t increases, N PbI 2 decreases as it is converted to MAPI. This shows >99% conversion from PbI 2 to MAPI aer t ¼ 125 min at T ¼ 100 C and P ¼ 9 torr.
A linear t of the data in Fig. 5 is used to approximate the rst-order kinetic expression in eqn (2), where k is the rate constant and t is reaction time.
The associated R 2 -value of 0.958, and an adjusted-R 2 -value of 0.950, indicate that the rst-order kinetic model is in reasonable agreement with the data; however, there is some visual evidence of curvature in the data, which may indicate a slightly more complicated functional relation. Consequently, we estimate the reaction rate constant, k, from the slope of the line as k ¼ 0.056 AE 0.005 min À1 . While this rate parameter is low, leading to reactions requiring 150 min to reach >99% completion, it is likely to be a function of the reaction temperature and the concentration of MAI vapor, which are not explicitly included in the rst-order model of eqn (2). This suggests that the reaction temperature and pressure can be tuned for faster reactions.
To investigate the effect of increased temperature and MAI concentration, another set of MAI reactions were carried out at T ¼ 160 C and P ¼ 9 torr, corresponding to an MAI vapor pressure of P vap MAI ¼ 6 AE 1 torr 21 -approximately two orders of magnitude higher than that of the 100 C reactions. Fig. 6a shows the XRD patterns of lms aer 10 and 15 min reactions. There is a distinct PbI 2 (001) peak at 12.60 for the 10 min reaction, whereas the PbI 2 peak is indistinguishable from the background aer 15 min. According to our quantitative analysis, this indicates >99% conversion from PbI 2 to MAPI. This reaction at T ¼ 160 C is an order of magnitude faster than the reaction at T ¼ 100 C-indicating the rate of reaction is, in fact, a function of temperature. It is worth noting that the MAPI (112), (211), and (202) peaks are present in both the 10 and 15 min reactions. Fig. 6b shows that the resulting MAPI lms have a discontinuous morphology of large agglomerates on the substrate surface. EDS was used to characterize the lms and conrmed that the light gray agglomerates were MAPI and the dark background was the SLG/ITO/CdS substrate. This agglomeration may be caused by de-wetting from the substrate surface during the reaction process-possibly due to high interfacial surface energy between the substrate and the perovskite. Alternatively, it may be due to Ostwald ripening where small, thermodynamically unstable particles are incorporated into larger, more stables particles to minimize interfacial surface energy. 30 Both phenomena typically occur at elevated temperatures due to increased adatom mobility, 31 and suggest that the 160 C reaction temperature is too high for MAPI processing on CdS at P ¼ 9 torr.

Characterization of solar cells
Fully-reacted MAPI lms from 150 min reactions at 100 C were integrated into functional solar cells with an SLG/ITO/CdS/ MAPI/Spiro/Au structure as shown in the cross-sectional SEM image of Fig. 7. The MAPI and Spiro layers appear to form pinhole-free lms with only small thickness variations due to the intrinsic roughness of the MAPI layer. The MAPI grains appear to be dense and columnar with a grain width of z150-200 nm. Furthermore, the reaction appears to have doubled the thickness of the original PbI 2 lm from z225 nm of PbI 2 to z450 nm of MAPI.
A total of 50 solar cells were fabricated with identical PbI 2 deposition, MAI reaction, and device processing conditions. A champion cell performance of 12.1% efficiency was achieved on a reverse sweep with open-circuit voltage (V OC ) ¼ 980 mV, shortcircuit current (J SC ) ¼ 21.9 mA cm À2 , and ll factor (FF) ¼ 56.6% Fig. 5 The natural log of N PbI 2 normalized to N PbI 2 ,o versus reaction time. The error bars show 95% confidence intervals with increasing uncertainty with increasing reaction time, due to a decreasing signalto-noise ratio in the XRD measurement. A linear fit of the data is shown along with its slope, R 2 , and adjusted R 2 . as shown in Fig. 8. However, this device exhibited hysteresis as evidenced by a lower efficiency of 8.4% during the forward sweep. This lower efficiency is due primarily to a loss in FF with V OC ¼ 950 mV, J SC ¼ 21.1 mA cm À2 , and FF ¼ 41.9%. The higher performance in the reverse sweep is consistent with previous reports, which have attributed hysteretic effects to ion migration within the MAPI layer, where a positive voltage bias leads to better band alignment and improved carrier collection. 32 The compiled cell parameters for all 50 devices are shown in Table 1. On average these cells exhibit a reasonable V OC and J SC of 968 mV and 18.3 mA cm À2 that can be incrementally improved upon to achieve high efficiency solar cells. However, there are substantial losses in FF that limit the efficiency. These losses can be attributed primarily to the CdS electron transport layer because, in addition to parasitic absorption and a small conduction band spike, 33 it has been reported that CdS can interact with excess MAI to form an interfacial Cd perovskite (MA 2 CdI 4 ). 34 This would create a blocking barrier and reduce both the J SC and the FF. 34 Therefore, it is likely that replacing CdS with a more appropriate electron transport layer, such as SnO 2 or TiO 2 , could be used to improve the FF and the overall cell efficiency.

Discussion
This all-vapor CSVT process serves as an advance toward developing a commercial vapor deposition method for the production of perovskite thin lms. However, there are several concepts that must be better understood before designing and implementing scaled VTD reactors. First, the perovskite eld has moved away from the CH 3 NH 3 PbI 3 composition toward alloys that include formamidinium, Cs, and Br because they have proven to be more stable and more efficient. 3 It is apparent that a commercial perovskite production process will require the exibility and methods to incorporate alloys into the material. Second, a quantitative mass transport model of the vapor deposition process is necessary to develop a scaled-up process. The composition of the alloyed perovskite lm is essential to its performance and stability; therefore, a model is needed to guide experimentation toward a targeted composition. Finally, VTD of perovskites could be carried out either via a sequential process, as illustrated through CSVT, or via a simultaneous process. While the sequential process is limited by the rate of PbI 2 conversion, a sequential process could circumvent this issue by co-depositing the perovskite's constituent materials through a low-vacuum and high-throughput process. Any effort to address one of these outstanding issues would be of great importance in developing perovskite vapor processing at the commercial scale.

Conclusions
This work establishes a sequential, all-vapor CSVT process to fabricate single-phase CH 3 NH 3 PbI 3 perovskite thin lms. Vapor-processed PbI 2 lms deposited on a CdS substrate possess a hexagonal, platelet-like morphology oriented about the (001) basal plane. Reacting these lms at 100 C and 9 torr produces continuous MAPI lms, but this reaction temperature yields a slow reaction that requires at least 125 min to reach complete conversion. The change in PbI 2 phase content versus time was used to quantify the rate of reaction, however further study is needed to develop a more accurate kinetic mechanism. Reacting the PbI 2 lms at 160 C and 9 torr yields complete conversion to MAPI in 15 min, but this elevated reaction temperature promotes de-wetting from the substrate which inhibits the formation of a continuous MAPI lm. Devices fabricated from MAPI processed at T ¼ 100 C in the conguration SLG/ITO/CdS/MAPI/Spiro/Au achieved a champion cell efficiency of 12.1% with V OC ¼ 980 mV, J SC ¼ 21.9 mA cm À2 , and FF ¼ 56.6%. These results validate CSVT as a viable processing technique for perovskite solar applications and establish   a foundation for further research into scaling perovskite production through VTD.

Conflicts of interest
There are no conicts to declare.