Solvent-Free Method for Defect Reduction and Improved Performance of p-i-n Vapor-Deposited Perovskite Solar Cells

As perovskite-based photovoltaics near commercialization, it is imperative to develop industrial-scale defect-passivation techniques. Vapor deposition is a solvent-free fabrication technique that is widely implemented in industry and can be used to fabricate metal-halide perovskite thin films. We demonstrate markably improved growth and optoelectronic properties for vapor-deposited [CH(NH2)2]0.83Cs0.17PbI3 perovskite solar cells by partially substituting PbI2 for PbCl2 as the inorganic precursor. We find the partial substitution of PbI2 for PbCl2 enhances photoluminescence lifetimes from 5.6 ns to over 100 ns, photoluminescence quantum yields by more than an order of magnitude, and charge-carrier mobility from 46 cm2/(V s) to 56 cm2/(V s). This results in improved solar-cell power conversion efficiency, from 16.4% to 19.3% for the devices employing perovskite films deposited with 20% substitution of PbI2 for PbCl2. Our method presents a scalable, dry, and solvent-free route to reducing nonradiative recombination centers and hence improving the performance of vapor-deposited metal-halide perovskite solar cells.


Bottom Contact
Films were deposited on indium-doped tin oxide (ITO) glass (AMG). The substrates were cleaned using industrial detergent Decon90 (1% vol in de-ionised water), de-ionised water, acetone, and isopropyl alcohol. Immediately prior to deposition, the substrates were cleaned under UV-ozone for 15 minutes.

P-Type Contact Layer
Poly[bis(4-phenyl)(2,4,6-trimethylphenyl)amine (PTAA) (Xi'an Polymer Light Technology) was spin-coated in an N2 glovebox. PTAA was dissolved in toluene at a concentration of 1.5 mg/mL; 100 µL of solution was applied statically onto the substrate, which was then spun up to 6000 rpm for 30 s, with an acceleration of 2000 rpm/s. After deposition, the substrate with PTAA was annealed in a N2 glovebox at 100 °C for 10 mins.

Perovskite Absorber Layer
CH(NH2)2 (FAI) (Greatcell), PbI2 (99.998% metal base, Alfa-Aeser), CsI (99.998% metal base, Alfa-Aeser), and PbCl2 (99.998% metal base, Alfa-Aeser) were co-evaporated in a custom-built thermal evaporator chamber to deposit compositions of the form FA1-yCsyPb(I1-xClx)3 by progressively substituting PbI2 with PbCl2. Specifically, starting with the neat-iodide aimed composition of FA0.83Cs0.17PbI3, we substituted 0%, 10%, 20%, 30%, 40% of the molar deposition rate of PbI2 with PbCl2, respectively. The FAI and CsI rates were kept the same for all deposition. During the evaporation, the pressure was typically < 1 × 10 -6 mbar. The source rates were kept constant using gold-plated quartz microbalances (QCM) and PID-loop-control software, with set rates and typical deposition temperatures detailed in Table S1. The sublimation rate of the inorganic precursors was controlled using QCMs adjacent to the crucible, while the FAI sublimation rate was controlled using a QCM close to the substrate, following the method we previously used for MAI [1]. Prior to every deposition, the FAI crucible was topped up to 1.1 g, adding fresh FAI to the remainder of previous deposition. For a typical 500 nm perovskite deposition, we evaporated 0.21 -0.23 g of FAI. Our chamber is set up such that the walls of the chamber are cooled to 17°C, while the substrate was held constant at 20°C.
Unless specified otherwise, all samples were annealed after deposition in an N2 glovebox for 30 minutes at 135 °C.

N-Type Contact Layer
C60 fullerene (99.9%, Acros Organics) was vacuum deposited in the same custom thermal evaporator chamber to 25 nm at 0.06 Å/s. The rate was controlled using a gold-plated quartz microbalance. Following C60 deposition, without breaking vacuum, 2 nm of Bathocuproine (BCP) (99.5%, Sigma-Aldricht) were deposited in the same evaporator chamber at 0.1 Å/s.  A mask was used such that the entire photoactive area of the devices was covered.

Current-Voltage Characterization
Devices were tested in ambient atmosphere using a Wavelabs Sinus-220 solar simulator and a Keithly 2400 sourcemeter at AM1.5 100 mW/cm 2 sunlight. The intensity of the solar simulator was set to produce the same short-circuit current from a KG3-filtered silicon reference photodiode (Fraunhofer) as its certified value. The active area illuminated for each device was 0.25 or 1 cm 2 , which was defined by black anodized metal masks. The measurement routine for each device was as follows: first, the open-circuit voltage (VOC) was measured for 3 s under illumination; then the current-voltage characteristics under illumination were measured scanning first in the reverse scan direction, from 1.2 -0 V, followed immediately by the forward scan, at a scan rate of 0.013 V/s in each direction; next the steady-state maximum power output under illumination was estimated using a maximum point tracking algorithm (gradient descent) over 30 s; finally the short-circuit current (JSC) was recorded over a period of 3 s. The devices tended to improve slightly over the course of 1 week after deposition. All J-V data shown in this work was measured 1 week after deposition of the top contact. We tested 8 devices for each composition other than the control 0% PbCl2 devices, which had 4 devices in this particular batch. The J-V characteristics shown in Figure 3c of the main text do not show shorted devices, defined as having JSC < 1 mA/cm 2 ; there was one such device from all the measurements we took.

X-ray Diffraction Measurements and Analysis
X-ray diffraction measurements (XRD) were performed on a Panalytical X'pert powder diffractometer equipped with a copper X-ray source (Cu-Kα 1.54 Å source set to 40 kV and 40 mA). The data were corrected for sample displacement by shifting the x-axis of the individual profiles such that the reference ITO peak sits at 2θ = 30.4°.
To obtain the peak position, and peak intensity of the PbI2 (001) peak at 12.7°, and the (100) CsPbCl3 peak at 15.8°, we fit a pseudo-Voigt distribution to the data. The pseudo-Voigt distribution was defined as described by Thompson et al. [2]. Figure 2a in the main text only shows a small section of the XRD trace. Figure S1 shows the full spectra for each of the samples, normalized with respect to their highest peak. As discussed in the main text, all films are made to have a small excess of lead-halide. We find this results in a small excess PbI2 peak for all amounts of Cl substitution, as shown in Figure  S2. We believe the PbI2 excess drops initially because of the overall reduction in PbI2 sublimated, while the subsequent increase is due to the formation and re-evaporation of FACl at the substrate, as discussed in Section 6.3. Figure S2 also shows how the XRD peak intensity of the CsPbCl3 peak increases linearly with added Cl, for 20% PbCl2 substitution onwards. Figure S1: X-ray diffraction patterns of vapor deposited FA1-yCsyPb(I1-xClx)3 full devices after testing, grown with varying amounts of the PbI2 substituted for PbCl2. Each trace has been normalized with respect to its highest peak.
In all cases, the perovskite phase was the majority phase but careful inspection, when plotted on a logscale, revealed the presence of peaks due to impurity phases ( Figure S3). Over the series the δ-FAPbI3, δ-CsPbI3, CsPbCl3, and PbI2 impurity phases were identified. In Figure S3, coloured tick marks have been used to identify the phases used in the fittings. The PbI2 is ever-present as expected from the excess PbI2 deposited (red tick marks). δ-CsPbI3 is present in the 0% and 10% PbCl2 samples (green tick marks). A sharp peak due to δ-(FA)PbI3 is observed in the 30% PbCl2 sample (marked with +). The CsPbCl3 phase is present in samples above 20% PbCl2 substitution (blue tick marks).   Figure S3) to the XRD patterns shown in Figure 2a and S1.

Figure S3:
The XRD data (red circles) fitted with a Pawley fit (black line) and the difference (blue line) plotted on a log scale to show impurity phases. Tick marks below the difference line show the peak positions of phases included in the fits; cubic perovskite (black), PbI2 (red), δ-CsPbI3 (green), CsPbCl3 (blue). The peaks due to ITO are marked with * and were included in the fit. A peak due to a δ-(FA)PbI3 impurity which could not be included in the fit is marked with +. All peaks are accounted for.

XRD Stability Measurements
We investigated the phase stability of the perovskite compositions outlined in this study by measuring the XRD of bare films grown on ITO/PTAA over the course of 30 days, with the films stored in ambient air in the dark (40-50% relative humidity). Figure S4 and S5 show this data for films before and after annealing, respectively. Note that the unannealed 10% PbCl2 substitution film proved particularly unstable, and turned completely yellow after 1 week, so this data point is shown instead. Figure S4: X-ray diffraction patterns of vapor deposited FA1-yCsyPb(I1-xClx)3 thin films grown with varying amounts of PbI2 substituted for PbCl2, on ITO/PTAA. The films were not annealed before or after measurement. (a) shows the films on the day of deposition, (b) shows the films after 30 days stored in the dark, in ambient air (40-50% RH), apart from the 10% PbCl2 film, which is shown already after 1 week.

X-ray Photoemission Spectroscopy
X-ray photoemission spectroscopy measurements were carried out using a Thermo Scientific Kα X-Ray Photoelectron Spectrometer with a monochromatic Al Kα X-Ray source at a take-off angle of 90 degrees. The analysis area was 200 µm x 200 µm and the core level spectra were recorded using a pass energy of 20 eV (an approximate resolution of 0.4 eV). Fitting procedures to extract peak positions from the XPS data were carried out using Casa XPS software suite. A Shirley background was used, and the spectra were fit with a mixture of Gaussian/Lorentzian (Lorentzian = 30%) line shapes. Figure 2c of the main text shows the high-resolution scan of Cl 2p region. Figure S6a shows the XPS survey scans, and Figure S6b shows the corresponding calculated relative atomic composition. The numerical values corresponding to Figure S6b are shown in Table S3.  Figure S7 and S8 show typical top down and cross-section scanning electron microscopy (SEM) images for the series of films investigated in this study. The SEM images were taken on an FEI Quanta 600 FEG, at 5 and 10 keV acceleration voltage for top-down and cross-section images respectively, with current defined by spot size 2.5. Prior to measurement, the SEM chamber was pumped down to high vacuum (< 2x10 -4 mbar).

PbCl2 Substitution
Relative    Figure S6b. The value in parentheses represents the standard deviation from the fit to the XPS spectra used to obtain the relative atomic composition.

Energy Dispersive X-ray Spectroscopy
We performed SEM energy dispersive x-ray spectroscopy (EDX) on the same system described above, using an Oxford Instruments detector. For each sample, we took 9 scans over areas of order 0.01 mm 2 , over a 90 s duration, using a 10 keV electron beam. To correct for differential Cl and I detection, we also prepared 500 nm thick samples of just PbI2 and PbCl2. We measured the EDX spectra for samples from the set of devices presented in the main text (thin films capped with C60/BCP), as well as two thin films grown with 20% PbCl2, before and after annealing. Table S4 shows the average relative atomic composition detected for the reference PbI2 and PbCl2 samples, and the calculated correction factor for I and Cl. Table S5 shows the corrected relative atomic compositions for the 20% PbCl2 thin films and the series of devices, normalized to the Pb detected.

Absorption Coefficient and Bandgap
Transmission and reflection measurements were taken using a modified Bruker Vertex 80v Fourier Transform Interferometer (FTIR), using a near-infrared source, CaF2 beam splitter, and Si diode detector. The absorption coefficient was obtained using the formula = 10 ( 1− ) / , where R is the amount of light reflected off the sample, T is the amount of light transmitted through the sample, and t is the film thickness. Figure S9 shows an example of the transmission and reflection data for the 20% PbCl2 sample, showing the oscillations in R and T indicative of a smooth film. Figure S10 shows the absorption coefficient for all PbCl2 compositions investigated in this work. The electronic bandgap was extracted from the absorption edge using Elliott fits, described in the next section.  Table S4: Relative atomic composition measured using energy-dispersive X-ray spectroscopy (EDX) for the 500 nm PbI2 and PbCl2 reference sample, and the corresponding correction factor Table S5: Relative atomic composition measured using energy-dispersive X-ray spectroscopy (EDX) for 500 nm films of FA1-yCsyPb(I1-xClx)3 grown with varying amounts of PbI2 substituted for PbCl2. The atomic composition values have been corrected using the factors derived in Table S4, and normalized to the measured Pb content. The table entries labeled as device were all annealed, and capped by C60/BCP.

Elliott Fits
The onset in the absorption coefficient spectrum is fitted using Elliott's model that accurately describes the absorption of a semiconductor near its band edge.
The theory expresses the energy-dependent absorption as: The bound exciton contribution has the form: Where 0 is a constant of proportionality that incorporates the electric dipole transition matrix element between the valence and conduction band. is formed of the weighted sum of contributions from the exciton states with positive integer quantum number n and energies − 2 , where is the band gap energy and is the exciton binding energy.
Here, the contribution from the electron-hole continuum states ( ) has the form:  (2 ) 2 ( 2 ℏ 2 ) 3/2 × 2, where μ is the reduced effective mass of the electron-hole system. The term in square brackets is the Coloumbic enhancement factor, which increases the absorption from the continuum states above the square-root form of the JDoS for a direct-gap semiconductor as a result of the Coloumbic attraction between the unbound electrons and holes.
The linear combination of the contributions from the excitonic and continuum states, α(E), is then convolved with a broadening function comprising of a convolution of a normal distribution -representing the homogeneous broadening caused by electron-phonon coupling, and an asymmetric log-normal distribution -accounting for the inhomogeneous broadening caused by disorder and local fluctuations of the stoichiometry of the material. The Broadening is mathematically denoted by ( ) = ℵ(0, 2 ) ⊛ ln(ℵ( , 2 )).
For our series, the log-normal broadening parameters were globally fitted for all the films with different PbCl2 concentrations, while the band gap, binding energy and electron phonon coupling were freely fitted afterwards. Table S6 shows the fit parameters obtained for the Elliott fits to our series.

External Quantum Efficiency and Urbach Energy
The External Quantum Efficiency (EQE) was measured using a Fourier transform photocurrent spectrometer in the FTIR system described in section 5.1. The photocurrent was measured using a near-infrared source and a CaF2 beamsplitter, using a Newport-calibrated reference silicon solar cell of known EQE value for calibration. Figure S11 shows  Urbach energy (EU) measurements were taken on the same setup as the EQE, using a 700 nm low-pass filter, such that the absorption band edge could be measured at a much higher sensitivity. The single exponential = ( − ) was fitted to the data in the region near the band edge to obtain EU. Three measurements were taken per sample to get an average, with the error reflecting the random error in the mean. These values are shown in Table S7.

Photoluminescence Spectra
We performed photoluminescence (PL) measurements on thin films grown on z-cut quartz. These measurements were taken by illuminating the samples with a 470 nm picosecond pulsed diode laser (PicoQuant, LDH-D-C-470M) operated at 1 MHz, at a power of 30 µW. The PL emitted by the films was coupled into a grating spectrometer (Princeton Instruments, SP-2558), and measured using and iCCD (PI-MAX4, Princeton Instruments).
To obtain the peak wavelength position, a pseudo-Voigt distribution was fit to the data, following the same formula as used for the XRD fits. To account for slight asymmetry in the peaks, the pseudo-Voigt distribution was fit to the left edge of the data to get the most accurate peak position. Table S9 shows the extracted PL peak position, as used in Figure 2e.

Time-Resolved Photoluminescence
We performed time correlated single photon (TCSPC) experiments with a photon-counting detector (PDM series from MPD) and a PicoHarp300 TCSPC event timer. The samples were photoexcited by the same 470 nm picosecond pulsed diode laser (PicoQuant, LDH-D-C-470M) operated at repetition rate 1 MHz used for the steady-state PL. The emission was collected using a grating spectrometer (Princeton Instruments, SP-2558). For Figure 1b, the excitation fluence was around 20 nJ/cm 2 respectively. The TCSPC traces from Figure 1b were fit with stretched exponentials to obtain a mean lifetime, following the equations in the work of deQuilettes et al. [12]. Figure S12 shows the traces for the whole range of PbCl2 substitution, along with the extracted average lifetimes which were shown in Figure 3a.

Optical Pump THz Probe Spectroscopy
Optical-Pump THz-Probe (OPTPS) experiments were performed as follows. A Spectra Physics Mai Tai-Empower-Spitfire Pro Ti:Sapphire regenerative amplifier provides 35 fs pulses centered at 800 nm at a repetition rate of 5 kHz. 400 nm photoexcitation is obtained by frequency doubling the fundamental laser output through a BBO crystal. THz probe pulses are generated by a spintronic emitter composed of 1.8 nm of Co40Fe40B20 sandwiched between 2 nm of tungsten and 2 nm of platinum, all supported by a quartz substrate. Detection of the THz pulses was performed using electro-optic sampling in a ZnTe crystal (1 mm (110)-ZnTe). The samples studied were bare perovskite films deposited on z-cut quartz. The sample, THz emitter and THz detector were held under vacuum (<10 -2 mbar) during the measurements. Figure S13 shows the OPTPS photoconductivity over pump delay for the samples studied here, with increasing content of PbCl2. The effective charge-carrier mobility was extracted from the amplitude of the OPTP signal immediately after photoexcitation (i.e. before charge-carrier recombination occurs). The sheet photoconductivity, ΔS, of a material with a thickness much shorter than the wavelength of the THz radiation can be expressed as ΔS = −ϵ 0 c(n a + n b )(Δ / ) (S1) where na and nb are the THz refractive indices of the materials interfacing the perovskite layer at the front and rear respectively. The quantity ΔT/T is the ratio of the photo-induced change in THz electric field to the transmitted THz electric field in the dark. The initial number of photo-excited charge carriers N is given by with E being incident pump pulse energy, λ the excitation wavelength, φ the ratio of free charges created per photon absorbed, and Rpump and Tpump being the reflected and transmitted fractions of the pump beam intensity. These two equations can be used to extract the charge-carrier mobility µ through where Aeff is the effective area from the overlap of the pump and probe beams and e is the elementary charge. Substituting Equations S1 and S2 into Equation S3 we obtain from which the effective charge-carrier mobility φµ may be determined based on the pump beam parameters and the initial measured ΔT/T of the sample. Here, µ is the charge-carrier mobility, and φ is the charge-to-photon branching ration which is assumed to be unity at room temperature.
The fluence dependence data for each sample was globally fitted to Where n is the carrier density and k1, k2, and k3 are the first, second, and third order recombination rate constants. Each film was modelled as a stack of 100 slices with initial carrier distribution in each slice calculated from the absorption profile of the 400 nm pump light. To account for the diffusion of charge-carriers, the population distribution across the film was allowed to vary as described previously by Crothers et al. [13]. The diffusion coefficient is calculated using the mobilities obtained experimentally. For simplicity, k3 was fixed to 1x10 -30 cm 6 s -1 . The values for k1 were adjusted to the decays at the lowest fluences, and k2 was optimized in the fitting routine. The fits are shown alongside the OPTPS transients in Figure S13. Table S10 shows the values for the effective charge carrier mobilities, as well as the recombination rate constants obtained from the OPTPS measurements.

Photoluminescence Quantum Yield
We performed photoluminescence quantum yield (PLQY) measurements following the procedure developed by de Mello et al. [14]. The samples were placed in an integrated sphere and photoexcited by a 532 nm laser the samples and the signal was collected by a QEPro spectrometer. The illumination intensity was 24 mWcm -2 which corresponds to approximately 0.5 suns equivalent photon flux [15], therefore the PLQY values presented here are underestimates of their value at one-sun equivalent conditions. Table S11 shows the obtained PLQE values plotted in Figure 3b of the main text. 2) React with CsI to form PbI2 and CsCl, according to the following reaction 2 + 2 → 2 + 2 In general we can combine these first two equations by introducing a factor representing the preference for CsCl formation over FACl, with 0 < < 1:

3) Remain unreacted
From our XRD results, we find the phases present in the final film to be PbI2, CsPbCl3, and some FA/Cs I/Cl mixed perovskite with general formula 1− ( 1− ) 3 , where 0 < , < 1. As such, we can rule out large amounts of PbCl2 remaining unreacted. Instead, the possible formation of CsPbCl3 opens a new pathway for any CsCl formed, namely: We can now write down a general equation for the ratio of products based on the factors , , , , which define the relative amount of CsI to FAI, the relative amount of PbI2 to PbCl2, the relative amount of CsCl formed to FACl formed, the relative amount of CsPbCl3 formed for each CsCl. First, we start from the precursors evaporated: Here, we have added the terms in the orange box (which sum to 0) to obtain a multiple = 1+2 of equation (3) with the terms from the grey box. Note that this equation is only valid for sufficient amounts of CsI, i.e. ≥ 2 . This represents the moment when all CsI has reacted to form either CsCl or CsPbCl3. These cases will be discussed later on. From equation (4), we can see that Cl could enter the bulk perovskite through CsCl or FACl. If we assume that FACl is volatile enough to leave the film during deposition or annealing, but CsCl is able to integrate the bulk perovskite, equation (4) can be rearranged into the following products: From this, we can deduce the fraction of Cs to FA and Cl to I in the bulk perovskite, denoted by and respectively in a composition of the form 1− ( 1− ) 3 : These values can then be used to determine the expected bandgap and lattice parameter of the resulting perovskite.
As mentioned earlier, this relationship only holds true while there is enough CsI in the precursors to allow both the desired CsCl and CsPbCl3 fraction to form. Should this no longer be the case, when < 2 , there are two scenarios available: 1) Fix the current fraction of CsCl and CsPbCl3, meaning all additional PbCl2 reacts with FAI only 2) Allow the CsPbCl3 to continue to grow, now removing CsCl from the bulk perovskite, until all Cs is in the CsPbCl3 phase, at which point all additional PbCl2 reacts with FAI only.
Our data shows an initial increase in the bandgap and lattice constant, with the trend reaching a maximum around 20-30% PbCl2 substitution, followed by a reversal in the trend. Because the only process that could reduce the bandgap and increase the lattice size in our system is the removal of Cs and Cl from the bulk perovskite, we expect the second process to be dominant in our system. With these assumptions we can do simple modelling of the expected lattice parameter and bandgap as a function of PbCl2 substitution. Figure S14 shows 4 different scenarios based on the fraction of CsPbCl3 formed. The model calculates the Cs and Cl fraction in the bulk perovskite according to the relationships derived above. The bandgap is then calculated assuming linear shifts from one material to the next. Here, we use values of 1.54 eV for FAPbI3, previously reported for vapor-deposited films using Elliott fits [16], 1.81 eV for CsPbI3, previously reported at room temperature from Elliot fits [17], and 3.0 eV for CsPbCl3 [18]. To summarize, we obtain the following expression for the bandgap as a function of Cs and Cl substitution (denoted by 0 < , < 1 respectively) of FAPbI3. We note that our coefficient of y, 0.27, is very similar to that previously reported by Li et al. of 0.28 [19], even though we derive ours from Elliott fits where theirs comes from Tauc plots. In reality, there may be band-bending occurring, especially for the poorly understood I-Cl substitution, since it is observed for I-Br substitution [20]. Band-bending for I-Cl substitution in the same direction as I-Br substitution would result in an underestimate of I-Cl substitution for low x values, so we expect the numerical values obtained to be approximate. Nevertheless, the trends obtained should still reflect the ones observed in our material. The resulting bandgap was calculated for a variety of α and γ values across a range of PbCl2 substitution. For ease of understanding, the curves were grouped according to the fraction of CsPbCl3 grown per 3 Cl atoms in the starting products. The resulting bulk perovskite can be seen to vary wildly depending on the formation of CsPbCl3, FACl, and the latter's volatility.
Comparing to the trends found for our samples in Figure 2b and 2e, we see that we are likely forming a medium amount of CsPbCl3 per 3 Cl atoms in the starting products, between 0.4 and 0.8 CsPbCl3 per 3 Cl, depending on how much FACl is formed versus CsCl. Importantly, we can see that, in order to obtain the trends we observe, we need some amount of Cl to incorporate into the lattice and increase the bandgap, such that not all Cl we detect in the films can come from CsPbCl3. In the following sub-sections we will explore what we can deduce about the formation of FACl and CsCl. Effectively, the "all FACl" scenario depicted by the blue line represents no Cl substituting into the bulk perovskite, while the "all CsCl" scenario represents all Cl not forming CsPbCl3 substituting into the bulk perovskite.

Evidence for Cl in the Final Films
To determine how much Cl is present in the final films, we performed XPS and EDX measurements on the series of films going from 0% to 40% PbCl2 substitution. The highresolution spectra of the Cl 2p region measured in XPS are shown in Figure 2c. Figure S15a shows the calculated I:Pb:Cl ratios from the XPS measurements, normalized to the amount of Pb, and Figure S15b graphically shows the calculated Cl/(I+Cl) ratio as a function of PbCl2 substitution. We can see that we are clearly detecting substantial amounts of Cl in the final films, of the order of 1-10%, with the amount of Cl detected increasing as more PbCl2 is added. However, the numerical value of Cl detected is only about half of what was originally evaporated. Lab-based XPS measurements, like the ones carried out in this work, probe depths of up to 10 nm therefore to ensure the detection of Cl was not limited to the surface of the films we also performed EDX measurements on another set of films. Figure S16a shows the calculated Cl/(I+Cl) ratio as a function of PbCl2 measured from films used for devices (ITO/PTAA/Perovskite/C60/BCP), while Figure S16b shows EDX measurements for a bare film of the champion 20% PbCl2 composition before and after annealing. We observe a similar trend to the XPS measurements, with the Cl detected increasing with added PbCl2, but also more Cl in general, up to 20% for the film with 40% PbCl2 substitution. We also find only a small change in Cl content before and after annealing, with the final annealed 20% PbCl2 thin film showing 11.5% Cl, very close to the amount of Cl evaporated, 13.3% ( Figure S16b). Overall, our results show that most of the Cl sublimated ends up in the final film, with only a few percent lost during deposition or annealing.

Evidence for FACl Loss
The relative amount of FACl formed during growth and how much of it is re-sublimated has important consequences on the amount of Cl and I present in the final films. We believe that, if formed, FACl is likely volatile enough to escape the film during deposition or annealing, based on the behavior observed for MACl in MAPbI3 grown from MAI and PbCl2. Specifically, early solution processed perovskite recipes [21], and indeed the first thermal evaporation perovskite PV publication [22] employed PbCl2 as the sole Pb source, and relied on volatilization of excess MACl to result in a post annealed film comprising predominantly MAPbI3, with only a small amount of Cl substitution. Figure S17 shows the neat Cl perovskite phase XRD peak for films before and after annealing, and reveals that, while after annealing the position matches that of CsPbCl3 in all cases, the position shifts to lower angles before annealing for higher PbCl2 substitution. This reduction in XRD peak position implies an increase in lattice size of the neat Cl perovskite phase, consistent with the increased presence of FA in this phase. As such, this shows this is evidence that, for higher amounts of PbCl2, we are indeed forming FACl during deposition, that not all FACl is lost through the high-vacuum during deposition, but that it is all lost through the annealing process. Equation (5) shows that PbI2 formation is associated with loss of FACl. We find that the XRD PbI2 peak intensity drops for low amounts of PbCl2 substitution but increases again from 30% PbCl2 onwards ( Figure S2). This is in good agreement with our evidence from the neat Cl phase XRD peak, specifically that there is little FACl loss for PbCl2 substitution below 30%, but noticeable FACl formation and loss for PbCl2 substitution over 30%. Finally, we expect the thickness of the resulting films to decrease as more FACl is lost, which we can observe in the SEM cross-sections ( Figure S8). It is clear that, while the films from 0% to 30% PbCl2 substitution generally have the expected thickness of 500 nm, the 40% PbCl2 film is noticeably thinner than the others, consistent with increased loss of material.
The most striking observation is that, while the 30% PbCl2 and especially the 40% PbCl2 films show evidence for FACl formation and loss, this is not the case for the 10% and 20% PbCl2 films. From equation (5), we can deduce that this is likely related to preferential formation of CsCl over FACl i.e., larger values of α. Since the amount of Cs in the film is limited, FACl formation will dominate once all CsI has reacted with PbCl2, and this is exactly what we observe.

Importance of CsPbCl3 Formation
In general, we found the films grown with small amounts of Cl, before the formation of CsPbCl3 to be less crystallographically stable than films with CsPbCl3. Notably, the 10% PbCl2 film, which does not show a CsPbCl3 peak in XRD, is very unstable before annealing, in contrast to the other films ( Figure S4). We were able to replicate this instability in films where the PbCl2 was introduced separately from the bulk perovskite deposition, as a thin interlayer between the ITO/PTAA and the perovskite. Specifically, we deposited 3 nm and 9 nm of PbCl2 followed by 500 nm of FA0.83Cs0.17PbI3, grown without PbI2 excess to account for the additional PbCl2. We found that, for 3 nm of PbCl2, the films are very unstable, quickly turning into δphase CsPbI3 ( Figure S18b), in contrast with the films grown on 9 nm of PbCl2, which showed improved photovoltaic performance ( Figure S19), accompanied by the appearance of a CsPbCl3 peak ( Figure S18a). As such, we speculate that the formation of at least some CsPbCl3 is necessary to stabilize a mixed I-Cl perovskite phase.
Figure S18: X-ray diffraction (XRD) patterns of vapor-deposited FA1-yCsyPb(I1-xClx)3 full devices after testing, grown either on top of a PbCl2 interlayer or with 20% PbI2 substituted for PbCl2, as denoted by the legend. The y-axis shows the square root of the measured intensity to be able to show both the large perovskite peak and the much smaller CsPbCl3 or δ-phase CsPbI3 peak. (a) shows the XRD pattern around the bulk perovskite (100) peak, with the inset showing the same pattern zoomed in on the CsPbCl3 (100) peak. (b) shows the XRD pattern around the (200) bulk perovskite peak Figure S19: J-V characteristics of the reverse scan for co-evaporated FA0.83Cs0.17PbI3 devices, grown with either a 0, 3, 9 nm PbCl2 interlayer before the perovskite, or with 20% of the PbI2 substituted for PbCl2. Table S12 shows the numerical values of the J-V characteristics of the reverse scans for the FA0.83Cs0.17Pb(I1-xClx)3 devices shown in the main text. Figure S20 shows the reverse scan for the champion device of each Cl composition investigated in the main text. We measured both 0.25 cm 2 and 1 cm 2 devices, and found very similar performance in general. Figure S21 shows a comparison of the 1 cm 2 device and the average J-V of the 0.25 cm 2 device for the 20% PbCl2 composition. Next, as mentioned in the main text, some of the devices made with 30% PbCl2 ended up phase unstable under sunlight illumination, turning yellow-colored, resulting in significantly reduced performance. Figure S22 shows the PCE of these discolored devices as compared to the normal ones shown in the main text, as well as an image of a film showing the very noticeable phase-transition of these pixels. Finally, Figure S23 shows an example of the batch-to-batch reproducibility       Figures S4 and S5 show the XRD patterns of our films before and after annealing at 135 °C for 30 mins, and show that for all compositions except the 10% PbCl2 film, the MHP is already in the desired phase before annealing, and stable in ambient air. In the case of the 10% PbCl2 film, it is only after annealing that the film is in the desired phase, and stable in air. The annealing process mainly increases the height of the XRD perovskite peaks, as shown in Figure S24a. We observe a similar behavior for the optical absorption coefficient, which already shows the desired absorption onset prior to annealing, with the latter slightly sharpening the onset ( Figure S24b). Moreover, we found that annealing is required to ensure stability of the films under AM1.5 and applied bias, as solar cells made from unannealed films turned into the yellow δ-phase under testing conditions. Finally, as discussed in Section 6.3 of the SI, the annealing also shifts the pure Cl peak phase to 15.8% for compositions with PbCl2 ≥ 30%, indicating loss of FACl. Figure S24: (a) X-ray diffraction (XRD) pattern of vapor-deposited FA 1-y Cs y Pb(I 1-x Cl x ) 3 thin films grown with 20% of the PbI 2 substituted for PbCl 2, on ITO/PTAA, as deposited (red) or after 30 minutes annealing at 135 °C. (b) Absorption coefficient of similar vapor-deposited films on z-cut quartz, as deposited (red) or after 30 minutes annealing at 135 °C.