Vapor deposition of metal halide perovskite thin films: process control strategies to shape layer properties

Vapor-based processes are particularly promising to deposit the perovskite thin-film absorber of solar cells. These deposition methods are up-scalable, involve a controlled solvent-free environment, have the ability to conformally coat rough substrates, involve soft, low-energy deposition conditions, are compatible with shadow masks for patterning, and are already widely deployed at the industrial level. Still, solar cells featuring layers processed with these methods have not yet reached the same performance as their solution-processed counterparts, in part due the complexity of controlling the sublimation of the organic precursors. This review will discuss the different vapor-based deposition processes that have been reported to deposit perovskite thin films and will discuss reaction chamber designs that provide an enhanced control over the deposition process. The second part of the review will then link experimental observations regarding layer properties depending on process conditions to theoretical concepts describing the sublimation and condensation of precursors, and the growth of the perovskite thin film.


Introduction
The development of efficient but cheap renewable electricity sources is essential to sustain the momentum towards a carbon-free society and achieve the objectives of the 2015 World Climate Conference agreement.Photovoltaics (PV) is driving the progression of renewable electricity generation technologies, with more than 623 gigawatts (GW) installed worldwide at the beginning of 2020, about 110 GW more than one year earlier [1] .However, realizing these climate goals will require these numbers to quickly increase: terawatts of renewable electricity generation capacity will soon have to be installed every year [2].Even though the market-dominating crystalline silicon (c-Si) photovoltaic technology can sustain a large shift of energy systems alone, it will remain limited in its efficiency.The key issue is that c-Si solar cells, or any single-junction technology, do not fully exploits the sun's energy.A 3 eV blue photon absorbed by a c-Si solar cell will thermalize to the band edge and contribute to around 0.6 eV in electrical energy, hence limiting with other processes the efficiency of c-Si to a theoretical maximum of 29.5% [3].
As previously demonstrated with III-V materials [4], the most promising option to reduce transparency and thermalization losses and surpass the 30% efficiency mark relies on stacking at least 2 semiconductors with different bandgaps on top of each other in a multi-junction cell [5].III-V multijunction designs can surpass this efficiency value but the scarcity of their constituent materials and their high cost do not allow this technology to achieve a competitive levelized cost of electricity (LCOE), restricting their use to niche markets such as space or military applications.They could possibly be used with concentrated photovoltaics.However, this approach did not succeed to take a significant market share of terrestrial PV.The recent emergence of organic-inorganic metal halide perovskite solar cells may change this situation: this material has the potential to yield cost-competitive highly efficient multijunction solar modules.
Perovskite semiconductors crystallize in the ABX 3 lattice structure, where A is usually an organic cation such as methylammonium and/or formamidinium, sometimes mixed with Cs, B a bivalent metal such as lead and/or tin, and X a halide such as iodide, bromide and/or chloride.This composition can be adapted to produce materials with bandgaps from 1.2 to > 3 eV [6], [7].In addition to this bandgap This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642tunability, perovskite materials can be processed at low temperatures and still exhibit excellent optoelectronic properties [8], have the potential to be produced at low-costs [9], [10], and exhibit a high power conversion efficiency of currently up to 25.5 % for small area (cm 2 range) single-junction level [4].These characteristics make perovskite solar cells the ideal building block of multi-junction solar cells, e.g. with other perovskite cells or on top of a c-Si cell.With c-Si solar cells dominating more than 90% of the market, the latter approach is particularly attractive as it has the potential to upgrade c-Si technologies to efficiencies above 30% [11]- [15] without impacting the c-Si process flow.Perovskite multi-junction solar cells hence appear as a promising option to continue the learning curve of PV efficiency.At constant $/W, stacking solar cells on top of each other to form a higher efficiency device is the most efficient approach to reduce the cost of a PV system.This is due to the large portion taken by area-related expenses in the total cost of a system.
The direct deposition of a perovskite top cell onto a c-Si solar cell to form a 2-terminal monolithic tandem induces certain constraints on the processing methods that can be used for the top cell.All monocrystalline c-Si technologies commercialized today are based on 6" wafers (or even larger).The front-side is textured by KOH etching to reduce reflection losses and promote light trapping of infrared light.The challenge is that this pyramidal texture complicates the coating of the c-Si cell with conventional perovskite deposition methods, which are based on solution processing [16].The perovskite layer is about one order of magnitude thinner than the height of the pyramids, typically a few micrometer, it has to cover.Depositing the perovskite absorber from a solution directly on top of this texture results in a discontinuous film, with shunts forming at the position of the summits of the pyramids [17].To circumvent that issue and still benefit from the global expertise in solution processing, most perovskite/c-Si tandem solar cell reported to date have instead employed a bottom c-Si cell mechanically polished on its front side [10], [18]- [29] or that features pyramids downsized to a height in the hundreds of nanometer range to enable their coverage by a solution-processed perovskite [30]- [32] (Figure 1a-b).In both cases, the top surface of the device is flat, leading to lower photocurrents due to an absence of double bounce effects.This optical loss can be only partially mitigated by additional light management schemes (additional antireflective layers, rear-side texturization, front glass texture), which complicate the final product design and manufacturing [19], [21].Based on these considerations, conformal layers on top of large Si pyramids would be the ideal optical system (Figure 1c).Vapor-based deposition methods hence appear particularly suited to achieve this feat.Conformally covering the pyramids of the c-Si cell with a perovskite front cell should eventually facilitate the achievement of tandem efficiencies beyond 30% thanks to short-circuit current densities surpassing 20 mA/cm 2 [12].
Vapor deposition processes have been employed with success to deposit metal halide perovskite materials.Perovskite single-junctions deposited with these approaches reach efficiencies now surpassing 20% [33] and demonstrate a good uniformity when deposited over larger areas (e.g.~100 cm 2 [34]- [36]).In addition to the conformal nature of the deposited layers [37], another advantage of these methods over solution-based routes include an absence of solvents (which need to be managed at the industrial level).Moreover, sublimating the lead precursors in a dedicated vapor deposition chamber in vacuum greatly reduces contamination risks.Vapor deposition processes also benefit from a strong industrial know-how as physical and chemical vapor deposition techniques are widely applied in industrially mature fields: flat panels, photovoltaic thin films, silicon industry etc. Vapor-based processes hence appear particularly promising to accelerate the transfer of perovskite technologies from the laboratory to industry.
In this review, we will discuss the different vapor deposition processes that have been used to deposit organic-inorganic lead-halide perovskite materials.A classification of the different methods is done considering whether the sublimed vapors are brought to the substrates without or with a carrier gas, a classification that we label here "line-of-sight" and "vapor transport deposition" techniques, respectively.Processes relying on the sequential use of techniques belonging to both categories are defined in this review as hybrid vapor deposition processes.This categorization was preferred to the classical chemical vapor deposition (CVD) versus physical vapor deposition (PVD) one, since processes labelled PVD may also involve a decomposition of precursors and chemical reactions.We would like to mention here that the techniques presented below are often unequally investigated, meaning that This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642device properties (active area, efficiency, stability) scatter significantly but not necessarily as a result of inherent weaknesses of certain methods.
After reviewing the wide variety of methods used to deposit perovskite thin films from vapors, we will aim to link experimental observations regarding layer chemistry and morphology to theoretical concepts describing the evaporation of precursors, their condensation on substrates, and the perovskite growth mechanisms.
2 Vapor deposition of metal halide perovskite solar cells 2.1 "Line-of-sight" vapor deposition processes Methods labelled here "line-of-sight" vapor deposition regroup techniques in which precursors are sublimated directly towards the substrates without the intervention of a carrier gas.These evaporations are usually performed in high vacuum conditions to increase the mean free path of the evaporated species, lower sublimation temperatures, and improve the purity of the as-deposited layers.

Thermal co-evaporation in high-vacuum
Co-evaporation consists in simultaneously sublimating multiple precursors in a chamber maintained in high vacuum (around 1.10 -3 Pa).Loaded in crucibles facing a rotating substrate holder, precursors are brought to their respective sublimation point.A schematic of the technique is presented in Figure 2a.In 1997, Era et al. [38] reported the first co-evaporation of two and three dimensional perovskites, including of cubic methylammonium lead triiodide (CH 3 NH 3 PbI 3 , or MAPI).A few years later, Matsushima et al. successfully fabricated field effect transistors by co-evaporating a tin halide perovskite [39], [40].The first co-evaporation of a perovskite layer included in a solar cell was reported in 2013 by Liu et al. [41].In this pioneering work, solar cells were produced by co-evaporating methylammonium iodide (CH 3 NH 3 I, or MAI) and lead chloride (PbCl 2 ) and cells achieved a power conversion efficiency of 15%.These vapor-deposited layers exhibited an uniform microstructure as well as an homogeneous thickness.These thermal co-evaporation processes did not require any post annealing step [42]- [44], a potentially important aspect regarding a potential industrialization of the process.
One issue became however rapidly evident with thermal co-evaporation: the high vapor pressure of organohalides.Indeed, the evaporation rate of a precursor is usually controlled by a quartz crystal microbalance (QCM).Due to their poor sticking on surfaces, organohalides may re-evaporate from chamber surfaces and from the QCM supposed to monitor their evaporation rate, leading to severe reproducibility issues.In addition, these organic precursors usually evaporate following a multidirectional flow, leading to their deposition on QCMs dedicated to other precursors as depicted in Figure 2a.Furthermore, the calibration of the QCM monitoring the organohalide sublimation rate is made complex by their tendency to grow under Volmer-Weber or Stranski-Krastanov modes [45], leading to layers with inhomogeneous thicknesses when deposited on their own.Furthermore, Borchert et al. [46] demonstrated that QCMs used to monitor evaporation rates are very sensitive to the organohalide impurities typically found in commercially available powders.Originating from the synthesis process of MAI, these impurities usually have a higher sticking coefficient on the gold surface of QCMs compared to pure MAI.
Different approaches have been explored to facilitate process control, notably by controlling the vapor pressure during deposition [47], the composition of the evaporated compounds [48], or the substrate temperature to tune the perovskite growth [45].Researchers have proposed to circumvent cross-talk and cross-contamination issues by either using a Knudsen cell evaporator and a QCM facing the opposite direction of the substrates [49], or by adding a shield between the inorganic and the organohalide sources [45].In addition, cooling the evaporation chamber down to -25°C as demonstrated by Roß et al. prevents the re-evaporation organics from chamber walls and QCMs [50].Other research groups calibrated the organohalide evaporation rate off-line by analyzing the perovskite layers with different MAI:PbI 2 ratios by X-ray diffraction (keeping the evaporation temperature of MAI constant) [51].This approach resulted This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642
in 20%-efficient perovskite solar cells.An interesting outcome of this work is that high efficiency cells were obtained with an absorber featuring relatively small grains.The trend in the field is usually to promote films with large grains spanning across the full perovskite absorber thickness.An alternative control strategy was recently proposed by Lohmann et al. [52]: the PbI 2 deposition rate was kept constant while the MAI sublimation rate was varied.The MAI evaporation rate was then deduced from the difference between the thickness measured at a QCM located close to the substrates and that from a QCM close to the PbI 2 crucible.This method presents the advantage that MAI will stick more easily on the PbI 2 -coated QCM.
Despite the general issue of precisely controlling composition, proof-of-concept solar cells produced by co-evaporation and featuring different cations and halides have been demonstrated [42], [53], [54].This evolution from depositing MAPI to mixed perovskites is especially important for perovskite/silicon tandem solar cells, where a fine control of the perovskite top cell bandgap and thickness is necessary to ensure maximum performance [11], [55].In addition, replacing MA with formamidinium ((HC(NH 2 ) 2 or FA) and Cs yields layers that are thermally more stable [56]- [58].Recently published studies report vapor-deposited devices exhibiting a stability comparable to the state-of-the-art.In Ref. [37], 99% of the initial PCE was retained after 1000 h at maximum power point under a 1-sun illumination conditions in nitrogen.Cells deposited via vapor-based processes were also subjected to thermal stresses, with some devices keeping 80% of their initial PCE after 3600h at 85°C in 10% RH [59].Promising efficiency and stability were also reported with mini-modules [60]: 18 % on an active area of 21 cm 2 with good stability in various conditions (>85% of initial PCE after 8000s at MPP in 70% RH, 80% after 100h at 65°C, 95% after 60 days on a shelf in 35% RH).
Compared to solution-processing, one limitation of thermal evaporation is the need to reach a high vacuum and the high source temperatures that are needed (several hundreds of °C depending on the precursor).Pumping down to such pressure and the delay induced by the need to warm/cool sources add to the process time, generally comprised between 30 minutes and 2 hours.It indicates that dedicated production tools with load-lock and linear sources are required to achieve reasonable throughputs.

Laser-induced co-evaporation
Miyadera et al. introduced in 2016 [61] a co-evaporation process in which the precursors are not sublimed by conventional resistive heating but by a continuous laser exposure, leading to a local heating of the MAI and PbI 2 precursors.A schematic of the setup is presented in Figure 2b.Compared to the usual resistive heating of the entire crucible plus the quantity of precursor necessary for a full evaporation, this method greatly reduces the thermal response of the sublimated material and should provide a more precise control over the evaporation rate.The best performing solar cell processed with this route achieved an efficiency of 15.4%.Still, it should be mentioned that this method has remained largely unexplored, likely due to the complexity and cost of such an evaporation system.

Sequential evaporation routes
As a response to the cross contamination and cross-talk issues occurring during co-evaporation, sequential evaporation processes have been developed.Instead of sublimating the precursors simultaneously, metal halide salts are evaporated first, followed by the organohalide salts.

Sequential thermal evaporation in high vacuum
Sequential evaporation in high vacuum is usually performed using the same chamber [62]- [65], except in the notable work of Hsiao et al. [66].There, two vacuum chambers were used, one for the lead halide precursor, transport layers and electrodes and another one specifically dedicated for the sublimation of organohalides.The evaporation of the latter was monitored by controlling the partial pressure of the methylammonium iodide (MAI) vapor.The authors achieved a complete conversion of the lead iodide into a perovskite phase by fine tuning the partial pressure of MAI.The duration of the process is one weakness of such sequential approaches.In this case, the MAI deposition step took two hours to reach optimal results, in part due to the density and thickness (240 nm)of the PbI 2 layer that needed to be converted.This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642
Chen et al. [64] highlighted the tradeoff that must be reached regarding substrate temperature when depositing MAI.As the penetration of MAI in the lead halide template is diffusion-limited, there is an incentive to increase the substrate temperature to promote conversion.However, doing so also decreases the sticking coefficient of the organohalide vapors on the substrate.In this work, the difficulty of controlling the MAI evaporation rate with a QCM also led the authors to evaporate the organic precursor at a constant temperature.Another way of producing a perovskite absorber with suitable thickness and complete conversion is to perform alternated depositions [67], [68].Successive layers of PbI 2 and MAI are sublimated one by one in the vacuum chamber, with varying thicknesses and number of pairs of layers.Meant to facilitate the conversion, this layer-by-layer technique still requires, in the presented cases, a post-annealing step to complete the conversion (typically from one to two hours).Another alternated deposition scheme was demonstrated by Feng et al. [33], who produced cells with an efficiency of 21.3%, the current record for a perovskite solar cell using vapor deposition.Their vacuum deposition setup features three sources: PbI 2 , FAI and CsI.These sources vaporize one at a time their respective species, following which the samples stay in the vacuum chamber for a post annealing to completely form the perovskite.Each deposition is monitored with a dedicated QCM.With this setup, a 10 cm 2 module with an efficiency of 14.6% PCE was also demonstrated, as well as large area (400 cm 2 ) perovskite films with high crystalline quality.

Close-space sublimation of organohalides
In high-vacuum line-of-sight evaporation processes, having a certain distance between the sources and the substrates improves the homogeneity of the deposition of precursors, as some have a ballistic behavior during sublimation.This distance, generally a few tens of centimeters, must be adapted to the size of the substrate to coat, as longer distances lead to a loss of material.Things however differ for organohalides as they tend to deposit everywhere in the chamber, even on surfaces not directly in the line-of-sight of the crucible [49].In that regard, close-space sublimation methods employ a chamber with a small gap between substrates and source [35], [69]- [73].Substrates can be exposed to a MAI vapor by enclosing substrates and the MAI powder in a closed and heated vessel [74]- [76], or even by heating substrates directly in contact with the organohalide powder [77].Despite their relative simplicity, these processes have resulted in high efficiencies in the range of 15-20%.Various compositions, from simple MAPI to mixed cations perovskites, were demonstrated.Layers were included in mini-modules, which exhibited an efficiency of 12% for an active area of 41.25 cm 2 .The PCE of these devices remained above 80% after 200h of storage in ambient conditions (RH = 30%, 25°C) and when heated to 60°C (nitrogen atmosphere) [35].
In addition to powders, the organohalide precursor used in close space evaporation can be made by solution processing (e.g.spin or spray coating) or dry alternatives such as pressed powder in the form of pellets [70].Most close space evaporation studies report long vapor exposure and/or post annealing steps, and sometimes a washing of the excess of MAI by isopropyl alcohol before annealing.Still the process flow can be accelerated, as demonstrated by Li et al. [70] who presented a process taking about 10 minutes for the evaporation, washing and annealing.
Overall, dissociating the evaporation of the inorganic and organic perovskite precursors enables an easier process control.One issue however is that the already long processing time tends to be exacerbated by the addition of extra steps.

Single-source vapor deposition processes
Another way of facilitating process control is to place all the precursors in a single source.This group of methods evaporate precursors pre-mixed in various forms, namely thin films, powders or targets.

Flash evaporation routes
Mitzi et al. introduced the concept in 1999 with the so-called thermal ablation technique [78].One of the initial purpose of the method was to uniformly coat irregular surfaces with a perovskite for light emitting applications.To circumvent issues linked to controlling the sublimation rate of each precursor separately, both the organic and inorganic precursors were dissolved in a solvent (anhydrous N,Ndimethylformamide, or DMF), before dispersing the resulting solution on a tantalum sheet (chosen for This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642
its high melting point and chemical resistance).The tantalum sheet was then connected to two electrodes inside a vacuum chamber, before passing a large current through it to rapidly sublime the precursors.The deposited perovskite layers did not require any post-deposition treatment.
The first application of perovskite thermal ablation for photovoltaics comes under the denomination "flash evaporation", which was introduced in 2015 by Longo et al. [79].The precursor employed were MAI and PbI 2 dissolved in DMF deposited by meniscus coating on a tantalum sheet and annealed to form MAPI. The tantalum sheet was then loaded in a high vacuum chamber, which was quickly heated to sublimate the film after reaching a pressure of 10 Pa.The setup is presented in Figure 3a.The deposited thickness was controlled by the amount of material on the tantalum sheet.Solar cells made with this approach reached an efficiency of 12.2%.Xu et al. [80] investigated the perovskite formation and highlighted the importance of performing the flash evaporation in high vacuum to form dense MAPI films.Pressures higher than 10 Pa led to perovskite films with irregular grains and voids.
In 2016, Fan and co-workers [81] reported a single-source evaporation method midway between flash evaporation and classical thermal evaporation.A pure MAPI powder synthesized in-house was loaded in a high vacuum thermal evaporation chamber, before rapidly heating the crucible to evaporate the powder.The evaporation lasted three minutes, longer than flash evaporation but much faster than conventional thermal deposition processes.The thin film deposited on the substrates did not require any post treatment annealing and displayed a good uniformity and phase purity.Solar cells reached an efficiency of 10.9%.In 2018, Lan et al. [82] used the same process to deposit formamidinium lead triiodide perovskites (FAPbI 3 or FAPI).

Ball milling of perovskite precursors for evaporation
Despite reducing the processing time and avoiding control issues, single-source processes still require the production of a mixture of the perovskite precursors prior to the deposition process.In that regard, ball milling appears especially promising as it is a solvent-free, simple, low cost and continuous process to synthesize phase pure perovskite powders.During the rotation of a cylinder shell, the friction between grinding balls causes the loaded precursors to be downsized, mixed, which eventually leads to their chemical transformation.The working principle is schematically illustrated in Figure 3b.A direct synthesis of perovskites precursors by ball milling has been the focus of a number of publications [83]- [85].Moreover, in a recent work by López et al. [85], the long-term stability of a MAPI powder obtained by such a mechano-synthesis approach was inferred to be superior compared to a solution-processed powder.This was explained by the absence of solvents and impurities and the high yield of the synthesis.
Single-source evaporation employing ball-milled precursors was demonstrated in 2018 by El Ajjouri et al. [86], who deposited inorganic perovskite layers.In 2019, Crane et al. [87] presented a proof-ofconcept (FA 0.81 MA 0.14 Cs 0.05 )Pb(Cl 0.02 Br 0.14 I 0.84 ) 3 perovskite thin film evaporated by combining ball milling and single-source flash evaporation.While the material performance in solar cells remain to be demonstrated with this approach, the ability of ball milling to synthesize mixed cation/halide perovskite powders that can then be evaporated at once is particularly interesting given the complexity of evaporating each precursor independently in a controlled way.

Sputtering
The development of simple, low-cost processing techniques of pure perovskite powder could benefit another vapor deposition technique: sputtering.This process, widely used in industry, is known for its fast and (relatively) conformal coatings on large surfaces, as well as for the good film properties achievable.With respect to perovskite research, two interesting proof-of-concept process flows were published in 2018 by Bonomi et al. [88] and in 2019 by Borri et al. [89].In the first reference, MAPI thin films were produced in a single step by radiofrequency magnetron sputtering.The target was made by pressing together MAI and PbI 2 powders.The second study reports the use of ball milling to manufacture an inorganic perovskite CsPbBr 3 from PbBr 2 and CsBr precursors.The resulting perovskite powder was pressed to form a target, which was then sputtered.This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642

Laser-based single-step evaporation processes
Laser-induced evaporation is another method that will certainly benefit from the advent of organicinorganic perovskite targets.The one-step laser evaporation of perovskites, similarly to sputtering or even flash evaporation, is another under-investigated yet promising approach.In 2017, Bansode et al. [90] demonstrated the pulsed laser deposition of MAPI and MAPbBr 3 .The evaporation was performed by shining a laser on a target which was made by pressing MAI (or MABr) and PbI 2 (or PbBr 2 ) powders together and annealing them for one hour at 100°C.Solar cells featuring an absorber processed with this approach achieved an efficiency of 10.9%.In a recent work [91], Kiyek et al. presented the pulse laser deposition of a CsSnI 3 perovskite.The target was made by compressing a pure CsSnI 3 powder obtained by ball milling from CsI and SnI 2 powders.While no solar cell was demonstrated, this work further highlights the potential of ball-milled precursors.Overall, high deposition rates were achieved as the evaporation process lasted between 5 and 10 minutes for a 200 nm-thick perovskite films.
Another single-source laser evaporation process is the so-called resonant infrared, matrix-assisted pulsed laser evaporation (RIR-MAPLE) reported by Dunlap-Shohl et al. [92].A pulsed laser ablates a cryogenically frozen solvent matrix containing the already formed perovskite.The energy of the laser is chosen to resonate with vibrational modes specific to the solvent to trigger the evaporation.With this approach, the risk of damaging the perovskite itself is negligible compared to other techniques directly heating the perovskite through heat conduction or laser absorption.A solar cell processed with this approach exhibited an efficiency of 12%.While promising, the process needs to be optimized further as it lasts several hours and a post-deposition annealing is required.
Laser evaporation was combined with flash evaporation in what is called laser printing.In 2017, Wei et al. [93] presented an evaporation setup composed of a laser illuminating a sheet of carbon nanotubes on which a lead iodide solution was dropped and dried, as presented in Figure 3c.An interesting characteristic of the carbon nanotube sheet is its low heat capacity per unit area (several orders of magnitude lower than metal heaters), resulting in an extremely fast temperature increase when illuminated by the laser.In this work, it took four seconds to print a PbI 2 layer of 300 nm on a 20 x 20 mm 2 surface.The lead iodide film was then converted to the perovskite phase by dipping it in a MAI solution.In 2019, Tai et al. [94] pushed the concept further by demonstrating inorganic perovskite layers deposited solely with the laser printing part of the process.To do so, the carbon nanotube sheet was coated with a CsPbI 2 Br solution.The evaporation process took only a few seconds and the resulting layer was then annealed at 280°C during one minute.Perovskite layers printed with this technique were incorporated in solar cells that reached an efficiency of 12.2%.Despite requiring heavy solvents to dissolve the metal halide, the high speed of this method makes it compatible with roll-to-roll processing.
To conclude on this part, single-source vapor deposition methods include a broad range of promising techniques addressing process control issues faced by standard co-evaporation processes.By evaporating the perovskite in a single-step, the deposition process is simplified and accelerated.One issue is that these methods require a pre-existing perovskite precursor of the targeted composition, an aspect mitigated by the advent of ball milling techniques for perovskite processing.While solar cells featuring an absorber processed with these techniques do not exhibit the same efficiency as solutionprocessed cells, this difference is more likely a result of the smaller research community working on these techniques rather than of an intrinsic limitation.

Vapor transport deposition
A recurrent feature of "line-of-sight" vapor deposition routes is the high vacuum maintained during the evaporation to ensure a high mean free path and purity of the condensed compounds.However, as discussed above, the high vapor pressure and low sticking coefficient of organohalides makes such processes sometimes difficult to control.
In vapor transport deposition processes, the evaporated vapors of the precursors are transported from the sources to the substrates with a carrier gas.Compared to most "line-of-sight" vapor deposition processes, the operating pressure is orders of magnitude higher with vapor transport deposition (in the hundreds of Pa range), which hence requires a simpler pumping system.Physically separating the evaporation from the deposition also provides some additional process flexibility.In addition to This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642
providing a control over the temperature of the source, substrates and sometimes chamber walls, the composition, temperature and flow rate of the carrier gas can also be tuned.Outside of the fields of perovskites, vapor transport deposition are particularly well-suited to process CdTe thin films [95], [96] or layers for organic electronics [97].
In this section, the focus is on processes using solely vapor transport deposition to grow perovskite thin films.Processes featuring a combination of "line-of-sight" vapor deposition and vapor transport deposition are designated here as "hybrid" processes and will be treated in the next section.

Tubular furnace deposition
Tube furnaces are amongst the simplest chambers used for vapor transport deposition.The first perovskite solar cell entirely processed by vapor transport deposition, in a tube furnace, was presented in 2015 by Takavoli et al. [98].The setup features a two-zones tube furnace with different temperatures, one for the substrates and another one for the precursors (lead halide and organohalide salts).An argon carrier gas transports the vapors of the sublimated precursors from the higher temperature zone to the lower temperature zone where the substrates were positioned.The two precursors are placed in a way so that they reach their respective sublimation temperature.While this approach is simple and inexpensive to setup, one major issue is the complexity of controlling individually the evaporation rate of each material, a complexity that increases drastically with the number of precursors present along the tube furnace length, as is the risk of cross-contamination of precursors.
In an attempt to decorrelate the sublimation of different precursors, sequential depositions of perovskites were also demonstrated [99], [100].However, sequentially sublimating the organic and inorganic precursors in the same tube furnace may led to cross-contamination and long cooling times between process steps.

Upgrades to tubular furnaces
Issues related to cross-contamination and individual control of sources was addressed for tin halide perovskites by Clark et al. [101].Their experimental setup is an evolution of a tubular furnace and is inspired by advances in the field of organic vapor phase deposition [102].The system, detailed in Figure 4a, is made of a three zones tube furnace containing a set of sources and an axisymmetrically positioned substrate holder cooled by water.Each tube source contains a porous frit of a given material attached to a linear actuator with a thermocouple.Sources can be translated along the tube to vary their temperatures thanks to a thermal gradient within the furnace.The carrier gas, nitrogen, is independently fed through each tube and the surrounding volume.The influence of each of the many process parameters on stoichiometry, morphology and electrical properties was thoroughly studied, highlighting the robustness of the process.
Another upgrade to tubular furnace systems was presented by Hoerantner et al. in 2019 [103] and is depicted in Figure 4b.The main tube contains the temperature-controlled substrate holder and is connected on one side to three separate tubes, one containing MAI, one PbI 2 and one used as N 2 inlet for pressure control.N 2 flows through each precursor tube to bring their vapors to the substrates.A silicon carbide foam is used to keep the solid precursors in the capsule and homogenize the flow.The deposition is performed sequentially by forming three thin pairs of PbI 2 /MAI layer stacks on the substrate holder heated at 100°C, a strategy that favors the perovskite formation.The as-deposited layers do not require any post-deposition annealing to complete the conversion.Considering the deposition of all the layers and the time required to alternate between PbI 2 and MAI, the process takes only a few minutes, with deposition rates reaching more than one nanometer per second.The study underlined the need to precisely control several parameters (chamber pressure, temperatures of walls, tubes, substrate holder) to produce a high-quality perovskite.A first solar cell made with this approach reached an efficiency of 6.9%.This modest result was partly attributed to unoptimized charge collection layers and a probably suboptimal perovskite composition.A generalization of this approach is presented in a patent filed by the authors [104]: the process is upscaled and features a vaporizer, a precursor mixing zone and a larger reaction chamber.

Showerhead-based reaction chambers
This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642
The last systems discussed are promising to produce perovskite layers in low vacuum with enhanced process control compared to standard "line-of-sight" vapor deposition routes.However, while upscaling these systems is feasible, the homogeneity of the deposited layer may suffer.This issue can be mitigated when forcing the precursor vapors and carrier gas through a showerhead, as used in industry in chemical vapor deposition (CVD) systems or in some plasma-enhanced CVD processes [105], [106].
A showerhead-based vapor transport deposition system for perovskites was reported in 2019 by Sanders et al. [107].Methylammonium bismuth iodide layers were deposited on substrates facing a showerhead located below.Vapors of MAI and BiI 3 were carried by nitrogen from sources separated by a shield to avoid cross contamination.A schematic of the setup is presented in Figure 4c.Incorporating such thin films in a solar cell led to low efficiencies, in part due to the nature of the absorber but also due to the low penetration of the perovskite in the mesoporous TiO 2 electron selective layer.Still, a stoichiometric bismuth-based perovskite could be deposited and links between precursor evaporation rates, substrate temperature, morphology, optical and crystallographic properties of the resulting layers could be established.Later on, the same group presented a similar setup for MAPI deposition from MAI and PbI 2 sources [108].While the system enables the growth of stoichiometric MAPI, first implementations in solar cells led to low efficiencies due to the porosity of the perovskite layer and a high series resistance.An important requirement of the system is that the walls must be heated above the highest vaporization temperature of the precursors, here that of metal halides.A probable consequence of such high temperatures is the thermal decomposition of the organic vapors, as discussed in the second part of this review.
In conclusion, vapor transport deposition as a standalone way of fabricating perovskite thin films is still an under-investigated field.However, the proof-of-concept systems reported to date appear promising due to the enhanced process control they provide and in some cases, the speed of deposition that can be achieved compared to standard thermal evaporation process.

Hybrid vapor deposition processes
Metal halide and organohalide salts, the main perovskite precursors, exhibit opposed behaviors during vapor deposition.Metal halide salts are well suited for "line-of-sight" vapor deposition processes thanks to high sublimation temperatures and low volatility.On the other hand, vapor transport deposition processes appear more suited to deposit the volatile organohalides.This difference has encouraged the development of hybrid sequential processes, where each step is tailored to the properties of the species to evaporate.

Thermal evaporation followed by vapor transport deposition
In most hybrid processes, the metal halide precursor is first thermally evaporated with a precise control over its thickness (by a QCM).Its conversion to a perovskite phase is then performed by exposing it to organohalide vapors brought by a carrier gas.Leyden et al. [109] presented one of the first hybrid vapor deposition process for perovskite solar cells in 2014.A PbCl 2 template was thermally evaporated in a high vacuum chamber, before depositing MAI by vapor transport deposition in a two-zones tubular furnace.A typical setup is presented in Figure 5a.
The choice of substrate temperature is a subtle tradeoff between different phenomena: increasing the temperature promotes the diffusion of the adsorbed species on the template, but it lowers the sticking coefficient of the MAI vapors and, if too high, it may lead to a degradation of the just-formed perovskite.While in an ideal case a high wall temperature would be beneficial to prevent unwanted condensation, in a tube furnace, the wall temperature determines that of the substrates.This compromise of temperature was thoroughly investigated by Moser et al. [110], who studied the vapor transport deposition of FAI on thermally evaporated cesium lead halide templates.They presented an isochronal map featuring source and substrate temperatures, highlighting the different regimes of vapor transport deposition: overconversion, underconversion, perovskite growth and degradation.In another work [111], Leyden et al. demonstrated that, when depositing FAI on substrates placed at temperatures higher than 145°C, a perovskite phase formed directly without the need for a post annealing and without significant organohalide oversaturation.They also highlighted a limitation of tubular furnaces: the high This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.In 2016, Ng et al. [112] investigated the impact of the post-deposition cooling rate.Solar cells fabricated with an absorber cooled faster displayed pinholes and a lower shunt resistance, leading to poorer performances.The authors also studied the impact of the composition of the carrier gas and found that mixing nitrogen with oxygen led to a passivation effect resulting in a higher power conversion efficiency.Other studies reported the use of air as carrier gas with similar benefits [113], [36].It is also worth mentioning that the carrier gas flow rate is another parameter that must be optimized.

PLEASE CITE THIS ARTICLE AS
Mixed cation/halide compositions can also be deposited with hybrid processes [36], an important aspect in view of a multijunction integration.Bandgap tuning can be achieved by co-evaporating different metal halide compounds (PbI 2 and CsBr for example), before depositing the organohalide (MAI, FAI) via vapor transport deposition [114].
The up-scalability of the technique was also investigated in different studies [111], [36], with recent results demonstrating modules reaching dimensions of up to 100 cm 2 and an efficiency on the order of 10%.In the same study, a mini-module with an active area of 22.4 cm 2 and an initial efficiency of 10% was reported to degrade linearly at a rate of -0.04abs%/hour after an initial burn-in loss (test conditions of 25°C under one sun illumination at MPP during a 250 hours experiment).Regarding up-scalability, Sahli et al. [115] recently presented a vapor transport deposition system for organohalide deposition.Following the thermal evaporation of a metal halide template, the substrates were placed in a reaction chamber.MAI sublimation products were transported from a distant evaporator by a carrier gas and brought to the substrates through a showerhead.The walls, evaporator and substrate temperatures were independently controlled, which enabled the authors to produce conformal and homogeneous perovskite layers on a textured 6 in.silicon wafer (239 cm 2 ).

Reactive polyiodide melts
An inherent weakness of vapor transport deposition in tubular furnace is the appearance of a gradient in precursor concentration along the flow path.Experimental conditions and substrate positioning are known to affect this gradient [111], [113].Uniformity can be enhanced by increasing the tube furnace temperature to reduce the deposition of the precursors on the tube walls [111].However, the temperature of the walls directly determines that of the substrates: the two cannot be tuned independently, hence limiting the processing window.
The use of reactive polyiodide (MAI 3 , FAI 3 ) melts relaxes some of these constraints.The different steps of the process are presented in Figure 5b.This hybrid deposition method was demonstrated in 2019 by Turkevych and co-workers [116] and involves the thermal evaporation of a metallic lead layer, before thermally evaporating MAI and exposing the layer stack to a vapor of I 2 .This exposure triggers the formation of a liquid polyiodide melt, MAI 3 , which reacts with the lead layer to form the perovskite through a fast redox reaction.The reaction driving force and large volume expansion ensure the absence of intermediate phases and/or pinholes.Uniform perovskite layers could be deposited onto substrate areas up to 600 cm 2 .The method is flexible as MAI can be replaced with CsI/MAI/FAI mixtures.Best cells on the mm 2 scale achieved conversion efficiencies of about 16-17% depending on composition and contact architecture.

Atomic layer deposition of the lead template
While thermal evaporation is well suited to deposit the metal halide template, alternatives have also been explored such as atomic layer deposition.The substrates are sequentially exposed to different gas precursors that react in a self-limiting way.This results in a conformal and precisely controlled deposition.The deposition occurs at low temperature and the working pressure is orders of magnitude higher than in the case of thermal evaporation.In a first attempt, Sutherland et al. [117] proposed in 2015 a three steps method to grow a perovskite layer.An atomic layer deposited PbS template was exposed to I 2 to trigger its conversion to PbI 2 .This lead halide template was then converted to a perovskite by dipping it in a MAI solution.In 2019, Popov et al. [118] reported another ALD process to deposit directly PbI 2 , yet without converting it to a perovskite.This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642

Sputtering of lead-based templates
Radio frequency magnetron sputtering is another interesting way of depositing metal halide template.As presented in a previous section, sputtering is a conformal, low temperature and fast process that is already widespread in industry.Nonetheless, sputtering is a high kinetic energy deposition, capable of damaging soft layers such as the organic charge transport layers commonly used in perovskite solar cells [119].Similarly to the approach of Sutherland et al. [117] with atomic layer deposition, da Silva Filho et al. [120] presented in 2018 a three step process that involves sputtering PbS, its conversion to PbI 2 by an exposure to I 2 , before dipping this template into a MAI solution to trigger the perovskite formation.
Hwang et al. [121] proposed in 2019 an evolution of this method by sputtering a PbO layer and then directly converting it to a perovskite through a MAI vapor exposure in a two-zones tube furnace.In the furnace, the perovskite formation was found to occur in two steps: MAI decomposed notably into HI and CH 3 I, reacting with PbO to form PbI 2 , and finally to MAPI.A 10.2%-efficient cell was demonstrated, and the method produced a conformal perovskite coating on textured silicon over 100 cm 2 .The vapor exposure of the Pb-based sputtered template resulted in oversaturation, requiring an annealing step to remove the (MA) 4 (PbI 6 )•2H 2 O dihydrate that formed on the surface.Most importantly, a complete conversion of the PbO layer was also difficult to achieve.To mitigate the issue, a gas treatment was performed by exposing the layer to methylamine (CH 3 NH 2 or MA), leading to a recrystallization of the perovskite.This effect was first demonstrated by Zhou et al. in 2015 [122]: in a MA vapor, MAPI single crystals were shown to form a metastable MAPbI 3 •x CH 3 NH 2 liquid phase.Removing the MA partial pressure led to a recrystallization of MAPI into a single crystal without any intermediate phase.Figure 5c features optical microscopy images demonstrating each step of this recrystallization.

Vapor treatments of perovskite thin films
Methylamine has been used as a post-deposition treatment to passivate defects and modify the layer microstructure but also to produce perovskite absorbers from lead halide templates [123]- [129].Notably, the reaction between methylamine and presumably HPbI 3 (Cl), or more likely DMAPbI 3 (Cl) based on recent evidence [130], enabled the crystallization of thick perovskite absorbers (1.1 m).When included in small devices of 0.1 cm 2 and 5 x 5 cm 2 mini-modules (active area of 12 cm 2 [131]), solar cells reached efficiencies of 19.1% and 15.3%, respectively.The ability of this approach to crystallize thick absorbers with a high optoelectronic quality is particularly attractive for monolithic tandems.Indeed, it should widen the range of bandgap/thickness combinations that can be used to achieve currentmatching conditions, with each subcell delivering the highest voltage possible [55].
In 2016, Raga et al. [132] used a MA vapor treatment to directly convert PbI 2 into MAPI in ambient air.The presence of water vapor generates a significant amount of lead oxide byproducts, and the authors obtained films of higher quality by exposing simultaneously the lead halide films to MA and hydroiodic acid (HI) vapors.As presented in Figure 5d, HI vapors are used to convert the lead oxide byproducts back to lead iodide.Best solar cells using a MAPI absorber processed with this gas treatment method displayed an efficiency of 15.3%.In 2020, Mortan et al. [133] presented a setup based on the same working principle: a tubular furnace fed with gaseous HI and MA to convert lead halide templates.In their case, they obtained better results by alternatively exposing the lead iodide template to each vapor.They emphasized the importance of using a drying agent between the tube furnace and the homemade HI gas source.On average, cells made with a drying step exhibited a 2% absolute higher power conversion efficiency (12.9%).
Another type of vapor treatments involves the use of formamidine (CH(NH 2 ) 2 or FA), in a vapor phase, to displace to the MA cations of MAPbI 3 .Resulting FA x MA 1-x PbI 3 cells produced with this method reached an efficiency of 18% [134].Advantageously, the method was found to maintain the smooth microstructure of the initial MAPbI 3 layer.Moreover, ion exchange via a vapor phase was also shown to modify the halide anion of MAPI in 2015 by Li et al. [135].Using MABr and MAI vapors, they demonstrated a reversible transformation between MAPbI 3 and MAPbBr 3 .
A vapor treatment was recently used to stabilize at room temperature FAPI perovskite in the photoactive α phase.Pristine FAPI may crystallize in a photo-inactive δ phase below 150°C.So far, this phase This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642
change has often been circumvented by alloying the perovskite with MA, Br or Cs ions.Lu et al. [136] proposed an organic thiocyanate (SCN) vapor exposure at moderate temperature (100°C) to convert FAPI films from the δ to α phase (the treatment works with both MASCN and FASCN).The short exposure (one minute) produced a stable photoactive layers and high-efficiency devices (>23%).
The possibilities offered by vapor treatments are numerous and particularly interesting for device processing.Ion exchange, bandgap tuning or stabilization via such versatile processes may be combined advantageously with the vapor-based deposition techniques mentioned above.

On the understanding of perovskite vapor deposition processes
The second part of this review discusses the underlying mechanisms of perovskite deposition from vapors.The growth of thin films from vapors is a succession of steps, from the sublimation of precursors, the adsorption of precursor adatoms/molecules, their surface migration, the formation of the first stable nuclei, their (sometimes preferential) growth, their impingement and possibly their coarsening at the expense of smaller grains.Each of these steps may limit the growth process, an aspect that the final layer morphology may reflect.In this section, we will review simple theoretical concepts describing the main steps of the growth process and attempt to tie these concepts to experimental observations.Readers interested in the mathematical formalism of heterogeneous nucleation, atomistic considerations or simulations are referred to dedicated work [137]- [143].

Layer nucleation and growth: some theoretical considerations
When particles from sublimated vapors are brought in the vicinity of substrates, either by their own ballistic motion or by a carrier gas, they may eventually be adsorbed on the surface.The adsorbed molecules or atoms (adatoms) diffuse on the surface until meeting other adatoms/molecules or reevaporate from the substrate.The adatom density, (, ), can be described using the diffusion equation based on Fick's second law [144]: Where   is the surface diffusion coefficient of adatoms,   the residence time of an adatom before evaporating from the surface, and  the incoming flux of atoms on the substrate from the vapor phase.This diffusion equation is solved by assuming that the diffusion rate is much slower than the rate of attachment and detachment of adatoms, leading to an equilibrium value of adatom density  0 at a given location.When surface diffusion is assumed to be much faster than the motion of already bonded adatom clusters, (,)  = 0.In the case of an isolated cluster at  = 0, the following boundary conditions are used : (0) =  0 , (∞) =   .The following solution is then obtained: Where   = √    is the surface diffusion length of adatoms.This means that only adatoms condensing on the surface within   from an isolated cluster will contribute to the growth of the cluster, while others will re-evaporate.
The complex influence of the substrate temperature on the deposition process is already highlighted by the equation presented above, which only describes the adatom density at the first steps of the process.Indeed, the arriving flux of atoms on the substrate  is proportional to  √ , with  the partial pressure,  the mass of the adatoms/molecules and  the absolute temperature in the vicinity of the substrate.In addition, both the diffusion coefficient   and adatom residence time   follow an Arrhenius dependence to temperature, meaning that both can be expressed as follows :  =  −     , with  the pre-exponential factor,   the activation energy and   the Boltzmann constant.At the experimental level, the influence of the substrate temperature on the process is made even more complex due to the need to consider the This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.These adatoms condensed on the surface may then coalesce into stable nuclei.The nucleation step is most commonly described by the classical nucleation theory (CNT).This theory is based on the socalled capillarity approximation describing the embryonic clusters as spherical drops of an incompressible fluid.The free energy of formation is split into volume and surface components, the latter taken equal to the macroscopic surface energy.The reversible work, or free energy, necessary to obtain a cluster of  adatoms is described by the following expression:

3
With   the saturation vapor pressure,  the surface energy per unit area and  the volume per atom in the cluster.The influences of the bulk and surface contributions are illustrated in Figure 6a, where one can notice that the positive surface component induce an energy barrier impeding nucleation.Finding the solution to ∆()  = 0 and injecting the resulting critical cluster size  * (the size below which the cluster is unstable) in the equation gives the corresponding energy barrier ∆( * ) for the nucleation process: For values above unity, increasing the vapor supersaturation    tends to decrease the energy barrier and the critical cluster size.The influence of temperature in the vicinity of the substrate has various effects on the critical cluster size and barrier height: supersaturation and surface energy decrease with an increasing temperature, while the critical cluster size increases (Figure 6a).Adding the influence of kinetics, the nucleation rate R can be expressed by: Where  is a factor proportional to temperature representing the departure of the cluster distribution from equilibrium and the collision rate between clusters and atoms.These competing influences of temperature lead the nucleation rate to behave as a nonlinear function with a global maximum.
The CNT has been successfully applied to understand and predict simple homogeneous and heterogeneous nucleation cases since decades.However, the capillary approximation is invalid at low cluster sizes.To describe the nucleation of small clusters, extensions to CNT have been developed using statistical mechanics.Instead of relying on macroscopic quantities such as surface energy, atomistic models introduce the energy of adsorption between a single atom and the substrate or the binding energy between two adatoms, for example.However, the difficulty of obtaining precise values of these microscopic parameters coupled to the exponential nature of these atomistic functions hinder the prediction capacity of atomistic models.This means that the extended nucleation model is often used for theoretical interpretation of experimental results while the CNT is favored to evaluate qualitatively the impact of process parameters on nucleation rates, as discussed in the next section.
The overall layer morphology can then follow one of the three thin-film growth modes: layer by layer, layer plus island or island (Frank-Van der Merwe, Stranski-Krastanov and Volmer-Weber modes, respectively) depending on the surface energy difference and lattice misfit with the substrate (Figure 6b).Organohalides typically grow according to Stranski-Krastanov or Volmer-Weber modes when deposited on their own [45], while co-evaporation processes yield perovskite layers with morphologies closer to the Frank-Van der Merwe mode with a smooth topography.A more detailed description of the thin films grown from vapors can be achieved with structure-zone models, which qualitatively predict the grain structure of films as function of a reduced set of parameters: the substrate temperature, the melting point of the material and the deposition pressure [145], [146].These structure-zone models This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.
PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642reflect how each of the three main steps of the growth of a thin film are affected by these process parameters: the transport of species to the substrate in the vapor phase, their adsorption and surface diffusion, and finally their reaction or uptake with/by the thin film.The final layer morphology may also be dominated by the preferential growth of grains with a specific crystallographic orientation (competitive growth models) [147], [148].Finally, depending on the substrate temperature or whether a post-deposition annealing step is applied, a coarsening of the grain structure may occur as the system minimizes its surface energy, as highlighted in Figure 6c [149].
Regarding specifically perovskite materials, computational fluid dynamics studies and approaches using heat and mass transfer equations have been used to investigate their vapor deposition [101], [103], [150]- [153].Although some of the hypotheses adopted do not fit with experimental results (e.g. the laminarity of the flow or the implicit absence of thermal degradation of organohalides into byproducts), these approaches may help in the design of evaporation setups and in the search for process parameters.
The main particularity of perovskite vapor deposition lies in how the layers grow: metal halides follow a classic ballistic motion common in physical vapor depositions, while some of the organohalide thermal byproducts are adsorbed by the deposited metal halide clusters.Once adsorbed, the organohalide molecules diffuse through the metal halide to start the perovskite formation [52], [65], [153].This twostep growth model has been adopted by Kim et al. [153] in their development of a kinetic model of the film growth during MAPI co-evaporation.Based on a sticking probability following an Arrhenius law and Fick's first law of diffusion, their model gave coherent values when fitting experimental data of adsorption.
The theoretical considerations presented in this section highlight how common process characteristics (temperature and surface energy of the substrate, partial pressure, etc.) alter the different steps of a thin film growth.A peculiarity of perovskites vapor growth is the interdiffusion and subsequent reaction of organohalides vapors and metal halides.The next section will treat how this chemical reaction is influenced by the evaporation setup and the process parameters.

Linking theory to experiments
The structure of the perovskite layer is often determined by one step of the growth process, either the interdiffusion and chemical reaction between precursors to form the perovskite, the nucleation of the first perovskite crystalline seeds, their growth until impingement, or their coarsening during the layer growth or a subsequent annealing step.In the next paragraphs, we will discuss how process parameters impact each of these steps of the growth of the perovskite thin films and in turn the layer structure.

Sublimation, adsorption, nucleation: first stages of the growth
Before condensing on substrates from a vapor phase, perovskite precursors must be sublimed.Apart from inducing a process control challenge as discussed beforehand, organohalides tend to decompose into different byproducts upon sublimation.As underlined by different reports [154]- [156], the main products of MAI evaporation are methyl iodide (CH 3 I), hydrogen iodide (HI), ammonia (NH 3 ), methylamine (CH 3 NH 2 ) and, in smaller quantities, its dimer form (CH 3 NH 2 ) 2 [154].Experiments and first-principles density functional theory (DFT) calculations [156] suggest that the kinetically favored degradation pathway is the reaction leading to the thermal decomposition of the parent molecule into CH 3 I and NH 3 .The degradation upon sublimation depends on the type of organohalide precursor as demonstrated by Williams et al. [157].The thermal decomposition of MAI at high temperature (>170°C) mainly results in methyl iodide and ammonia, while traces of methylammonium (CH 3 NH 3 , or MA) can be detected at lower temperatures [155], [158].On the other hand, MACl thermally degrades into MA and HCl.The evaporation of the other organohalide precursor, e.g.FAI, also results in the fragmentation of the parent molecule.In this case, the main decomposition byproducts are hydrogen cyanide (CHN), ammonia (NH 3 ), sym-triazine (C 3 H 3 N 3 ) and formamidine (CH(NH 2 ) 2 ) [159].The formation of a perovskite thin film from these different byproducts seems to mostly result from an acid/base reaction between organic cations and hydrogen halides in the presence of a metal halide [132], [154], [156], [159], [160].In Ref. [154], it was also hypothesized that the small amount of MAI dimer detected by mass spectrometry could participate in the perovskite formation process.
This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.

PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642
The decomposition of organic precursors upon sublimation has consequences on the layer nucleation and layer growth process.Indeed, the growth of the perovskite layer can be slowed down if one byproduct does not reach the reaction site, e.g. as a result of a lower residence time or low diffusion length on the substrate surface.In 2015, Liu et al. tracked the thermal evaporation of MAI on an already formed PbI 2 layer and the resulting MAPI formation with X-ray photoelectron spectroscopy (XPS) [161].From the XPS spectra, a carbon containing species was detected from the very beginning of the evaporation of MAI, while nitrogen was only detected after a certain incubation period.This carbon signal likely originated from methyl iodide, a known decomposition byproduct of MAI.This result would imply that methyl iodide is present in the bulk of their perovskite, as also inferred by data retrieved from solution-processed layers [162].Olthof and Meerholz performed in 2017 in situ XPS and ultraviolet photoelectron spectroscopy (UPS) at different steps of the growth of co-evaporated MAPI [163].One of the main finding of their XPS experiments is that the evaporation is characterized by a substrate-dependent induction period.Figure 7 displays the XPS spectra measured after 0.5 and 10 nanometers of deposition on PEDOT:PSS, PEIE, MoO x and indium tin oxide (ITO).While organic substrates such as PEDOT:PSS and PEIE are covered by a perovskite from the first nanometers, early stages of the growth on MoO x and ITO are characterized by lead or nitrogen signals, respectively.The catalytic nature of metal oxides was presented as the main factor explaining the difference with organic substrates.XPS data taken from layers thicker than 10 nm displays a strong MAPI signal on all substrates.Moreover, the authors mentioned that surface reactions occurring during the first few nanometers of the growth could be triggered by surface hydroxyl (-OH) groups.This means that, at least with co-evaporation, decomposition byproducts may preferentially deposit during the very first steps of the growth depending on the nature of the substrate.In [164], Thampy et al. demonstrated that these decomposition byproducts impact the long-term stability of the layer.A solution to mitigate this byproducts incubation layer was presented by Moser et al. with a custom-made vapor transport deposition system [110].They reversed the carrier gas flow (from substrates to source to the pump) during the heating phase of the crucible in tubular furnace to ensure that fragments sublimating at lower temperatures do not reach the substrate.When the evaporator reached its temperature setpoint, they switched the flow direction from source to substrate to trigger the deposition process.
Experiments show that, during both co-evaporation and sequential deposition processes, the final perovskite thickness is proportional to the evaporation rate of the metal halide or the thickness of the metal halide template, respectively [64], [161], [165].During co-evaporation, it appears that organohalide vapors preferentially adsorb on metal halides to form a perovskite [38], [52], [153].In their study on co-evaporation, Roß et al. [50] reported that the adhesion of PbI 2 on substrates is constant for substrate temperatures between -30 and 60°C, while that of MAI decreases by 70% when increasing temperature in this range.
To summarize, apart from the notable exception of the first nanometers of growth on metal oxides as reported by Olthof and Meerholz [163], the general trend is that perovskite growth during coevaporation occurs through a two-stage process, at least on a local scale.This behavior is depicted in Figure 8a, which illustrates how metal halides preferentially stick onto the substrate.Some of the organohalide degradation byproducts then react with these metal halides to form the perovskite phase.Similarly, Figure 8b shows the growth sequence for sequential processes, where a metal halide film is first grown and then exposed to the organohalide vapors to eventually form a perovskite layer (Figure 8c).

Layer growth and resulting thin film structure
Various types of perovskite layer morphologies can be obtained with vapor-based processes: perovskite thin films featuring small grains stacked on top of each other [51], duplex microstructures with small equiaxed grains covered by larger columnar ones [52], narrow columnar grains [50], large grains spanning across the full layer thickness [52], [66], etc. Depending on processing conditions, one step of the growth often dominates the process and dictates the final layer morphology as discussed below.
Substrate temperature is one key parameter that may lead to significant changes in growth regime.Lohmann et al. [52] presented perovskite films exhibiting large grains films without unconverted lead iodide when co-evaporating MAI and PbI 2 on substrates kept at -2°C, a morphology counterintuitive This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.
PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642 with respect to structure-zone models [147].As grain coarsening was likely limited in these low temperature conditions, such a morphology would be indicative of a low density of nuclei on the substrate and an impeded nucleation of new grains (Figure 9a).When increasing the substrate temperature or applying specific temperature profiles, films with equiaxed or duplex grain structures could be observed as the nucleation of new grains, and in some cases their growth, became facilitated.The growth rate was also reduced when increasing the substrate temperature, highlighting the critical impact of the adsorption/desorption of the organohalide fragments (factor   in Equation 3.2).Roß et al. [50] cooled the substrates to even lower temperatures (-25°C), which resulted in amorphous and nonphotoactive co-evaporated films.While the desorption of precursors was made less likely, the interdiffusion of species and their reaction to trigger the formation of the perovskite phase became the limiting factor.After a post-deposition annealing step at a moderate temperature (40°C), XRD diffractograms demonstrated the presence of a mixture of crystalline lead iodide and perovskite phases.
The authors mentioned that a substrate temperature close to room temperature is typically an optimal trade-off between increased adatoms mobility and diffusion length (higher temperatures) and higher organohalide fragment residence time (lower temperatures).
The deposition pressure also influences the growth process.Hsiao et al. noticed that increasing the partial pressure of organohalides during a sequential deposition process (from 1.3 10 -2 to 1.3 10 -1 Pa) multiplied the number of grain nucleation sites in these deposition conditions (with the substrates and walls heated at 75°C), producing perovskite layers with smaller grains.Clark et al. [101] studied the impact of varying substrate temperature (30 -70°C) and pressure (40 -1333 Pa) during the vapor transport deposition of MASnBr x I 3-x .It should be noted that, in this case, a higher pressure corresponds to an increased carrier gas/evaporated precursor ratio and thus a decrease in reactant concentration.Their experiments showed an increase in grain size with increasing pressure and substrate temperature, in line with trends given by structure-zone models (regarding the impact of the substrate temperature).On the other hand, morphological changes differ when performing co-evaporation at lower pressures.In these conditions, increasing the substrate temperature leads to a decrease in grain size.Figure 9b-c shows topview images of co-evaporated FAPI [166] and MAPI [51] perovskite films, respectively.In both cases, the substrates were at room temperature and the total pressure is below 10 -3 Pa.Films grown in these conditions typically display smaller grains than films co-evaporated under high vacuum but at lower substrate temperatures, as shown in Figure 9a and by Roß et al. [50].
The film morphology is also severely impacted by the substrate itself as it directly impacts parameters such as surface energy, surface diffusion length of precursors adatoms and molecules   , and their residence time   .Lohmann et al. [52] showed that distinct morphologies could be produced when coevaporating in the same conditions on various substrates at low substrate temperatures.The large grains presented in Figure 9a were grown on FTO/C60.Layers grown on ITO/C60 and crystalline silicon exhibited smaller columnar grains, while growth on FTO alone results in small and compact grains.Roß and co-workers also observed a difference in layer morphology depending on the contact below the perovskite [50].For sequential processes, the morphology of the lead halide template also influences the perovskite process: the use of a porous template can facilitate the perovskite crystallization process [113], [167], [168].
Another way to influence the growth and the final layer is to change the environment in which the layer is grown or treated after its deposition.Several reports tend to show that the presence of oxygen may passivate perovskite trap states [109], [112], [169].Humid air has a large impact on the perovskite conversion process as it favors interdiffusion and the subsequent crystallization process by making MAI more mobile within the perovskite [65].Still, exposure to moisture must be controlled to avoid any formation of hydrates and/or a degradation of the perovskite [170]- [172].Still, hydrates or excess organohalide can be removed from the layers to some extent by a washing step using solvents [65], [173], [174].In general, the presence of an excess in metal halide in the final layer may accelerate the degradation of the perovskite layer in operational conditions [175].
Among the different post-deposition treatments, annealing in a moderate temperature range (100-150°C) is known to promote a better interdiffusion of the remaining unreacted species.It may also lead to a coarsening of the perovskite grains depending on the annealing temperature [176].However, postannealing treatments must be compatible with the different layers of the solar cell stack to avoid This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.
PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642detrimental effects.As already presented earlier in this review, the post-deposition cooling rate of the substrate also has a significant impact on the final film morphology: Ng et al. [112] reported that films cooled faster displayed pinholes and a lower shunt resistance.
In the end, numerous parameters may be adapted to tune layer properties.This flexibility also means that a similar film structure can be obtained with different processing routes.Figure 9a and d shows topview and cross-section scanning electron microscopy (SEM) images of layers obtained by coevaporation [52] and sequential evaporation [66], respectively.Both images display a similar fully converted perovskite morphology with smooth and micrometer-sized grains.The first example was obtained by co-evaporating precursors at low substrate temperatures (-2 °C) without any post-deposition annealing, while the second one resulted from the sequential deposition of MAI on PbI 2 templates kept at 75 °C.
Overall, the competing influences of the different process parameters offer numerous pathways to tune the microstructure of the perovskite layer and hence its optoelectronic properties.The critical step determining the perovskite layer morphology may quickly change from one set of process parameter to another one.

Conclusion
Vapor-deposited perovskite layers have shown good performance when included in single-junction solar cells, also when up-scaled to larger area active areas a few tens of cm 2 .While solution-processing perovskite solar cells have so far exhibited a higher efficiency at the single-junction level, this difference results mostly from the fact that a smaller community is exploring vapor-phase routes rather than from any fundamental aspect.Still, the volatile behavior of organohalides and their tendency to decompose into several compounds upon sublimation complicates the deposition of high-quality layers by evaporation processes.As discussed in the first part of this review, innovative deposition chambers have been developed to improve the control over the deposition process and to account for the peculiarities of organohalides.Notable examples include cooled shields in multi-source thermal evaporators, vapor transport systems decorrelating evaporation and deposition conditions, showerheads to improve deposition homogeneity, tube furnaces with variable flow directions to coat substrates when reaching steady state sublimation conditions, etc.
In the second part of this review, we discussed the growth mechanisms of perovskite thin films grown from vapors and detailed the critical steps of the process which may impact the final layer structure.
Experimental observations were related to simple experimental concepts in an attempt to understand the impact on the layer properties of the process parameters accessible to the operator.Table 1 summarizes the impact of key process parameters on the deposition process.Overall, one should keep in mind that small changes to a specific process condition can lead to significant differences in the layer properties, hence complicating the identification of clear process-property links.Still, the impact of certain process parameters is rather straightforward to establish, e.g., the usage of precursors increases with chamber wall temperature, or the evaporation rate increases with source temperature, two temperature settings that nonetheless should be selected to avoid/minimize the degradation of the organic precursors.Low deposition pressures increase the mean free path of the precursors, impacting how precursors reach the surface and possibly the microstructure.The impact of other process parameters is more difficult to predict.For example, the substrate temperature will impact the adsorption of the precursor species, their surface diffusion, interdiffusion and reaction to nucleate the first perovskite domains, grain coarsening, and a trade-off condition between these processes will need to be selected.In the case of VTD, the carrier gas flow rate and its temperature may also impact the final layer properties.In the end, different processing routes may lead to similar layer properties if precisely adjusted, e.g.large columnar grains of high optoelectronic quality spanning across the full layer thickness.
This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.
PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642 8  Figure 3: a) Flash evaporation setup.Reproduced from Ref. [79] with permission from the Royal Society of Chemistry, b) Working principle of ball milling [83], c) Flash laser printing [93].Reproduced from Ref. [111] with permission from the Royal Society of Chemistry.b) Reactive polyiodide melts process [116].c) Recrystallisation of MAPI upon MA exposure [122].d) Subsequent reactions of lead halide upon exposure to MA and HI vapors in ambient air [132].
Figure 6: a) Free energy of formation as a function of the cluster size.Effect of an increasing temperature on the free energy of formation, with T 1 <T 2 <T 3 b) Film growth modes c) Nucleation, grain growth, coalescence and coarsening Figure 7: Carbon, nitrogen, iodine and lead XPS spectra of co-evaporated MAPI on 4 different substrates measured after 0.5 and 10 nm of deposition.Solid and dashed vertical lines show expected binding energies characteristic of the perovskite and other molecules, respectively.The table lists possible reaction products responsible for the additional peaks, divided into products (P), educts (E), decomposition products (D), and surface bonds (S).Adapted from [163]. Figure 9: a) Cross-sectional and top-view scanning electron microscopy image of a perovskite film obtained by co-evaporation with substrates kept at -2°C and 10 -3 Pa, adapted from [52], b) Co-evaporated FAPbI 3 film at room temperature and 10 -3 Pa.Reprinted with permission from [166].Copyright 2020 American Chemical Society.c) Co-evaporated MAPbI 3 film at room temperature and 10 -3 Pa, adapted from [51].d) Cross-sectional and top-view scanning electron microscopy image a perovskite thin films fabricated by exposing a 240 nm-thick template of PbI 2 to a MAI vapor at a partial pressure of 10 −2 Pa, adapted from [66].1: Influence of process parameters on a vapor deposition of perovskites.Note that we refer here to the partial pressure of organohalides and not the one of the system (with or without carrier gas).In the latter case, the advantage of a lower system pressure is a higher homogeneity of vapors thanks to a higher diffusivity (cf Chapman-Enskog theory).This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.

Impact on the deposition process
PLEASE CITE THIS ARTICLE AS DOI:10.1063/5.0060642 DOI:10.1063/5.0060642concentration gradient of vapors.Lead halide templates placed along the tube length reached different levels of conversion, as highlighted by XRD diffractograms.
DOI:10.1063/5.0060642temperature-dependent intercalation of organohalides into metal halide clusters, a key step of the perovskite growth process.

Figure 1 :
Figure1: Schematics and scanning electron microscopy (SEM) cross sections of silicon heterojunction/perovskite tandem solar cells featuring : a) a solution-processed perovskite cell on a front side-polished structure[24], b) solution-processed perovskite covering a downsized c-Si texture[31], and c) a perovskite solar cells deposited by a hybrid vapor/solution process to conformally coat the texture of the c-Si cell.TCO stands for transparent conductive oxide.Reprinted with permission from[177].Copyright 2019 American Chemical Society.

Figure 4 :Figure 5 :
Figure 4 : a) Tubular furnace setup with independent tubes for each precursor [101] b) Cross section views of the trident-shaped reactor presented by Hoerantner et al. [103] c) Schematic of the showerhead-based reaction chamber reported in Ref.[107]This is the author's peer reviewed, accepted manuscript.However, the online version of record will be different from this version once it has been copyedited and typeset.

Figure 8 :
Figure 8: Schematic description of the perovskite growth process from vapors.The metal halide preferentially sticks onto the substrate and some of the organohalide byproducts react with it to form a perovskite.Sequences describing a) co-evaporation b) sequential evaporation processes.The resulting perovskite seed is shown in c).