Solution-based synthesis of kesterite thin film semiconductors

Large-scale deployment of photovoltaic modules is required to power our renewable energy future. Kesterite, Cu2ZnSn(S, Se)4, is a p-type semiconductor absorber layer with a tunable bandgap consisting of earth abundant elements, and is seen as a potential ‘drop-in’ replacement to Cu(In,Ga)Se2 in thin film solar cells. Currently, the record light-to-electrical power conversion efficiency (PCE) of kesterite-based devices is 12.6%, for which the absorber layer has been solution-processed. This efficiency must be increased if kesterite technology is to help power the future. Therefore two questions arise: what is the best way to synthesize the film? And how to improve the device efficiency? Here, we focus on the first question from a solution-based synthesis perspective. The main strategy is to mix all the elements together initially and coat them on a surface, followed by annealing in a reactive chalcogen atmosphere to react, grow grains and sinter the film. The main difference between the methods presented here is how easily the solvent, ligands, and anions are removed. Impurities impair the ability to achieve high performance (>∼10% PCE) in kesterite devices. Hydrazine routes offer the least impurities, but have environmental and safety concerns associated with hydrazine. Aprotic and protic based molecular inks are environmentally friendlier and less toxic, but they require the removal of organic and halogen species associated with the solvent and precursors, which is challenging but possible. Nanoparticle routes consisting of kesterite (or binary chalcogenides) particles require the removal of stabilizing ligands from their surfaces. Electrodeposited layers contain few impurities but are sometimes difficult to make compositionally uniform over large areas, and for metal deposited layers, they have to go through several solid-state reaction steps to form kesterite. Hence, each method has distinct advantages and disadvantages. We review the state-of-the art of each and provide perspective on the different strategies.


Introduction
In the context of developing a thin film photovoltaic technology that is truly scalable on the terawatt level using low-cost earth abundant elements, the kesterite semiconductor, Cu 2 ZnSn(S,Se) 4 , or CZTSSe, with a tunable bandgap between 1.0 to 1.5 eV has been a key material of interest. It may be seen as a potential replacement or successor absorber material to Cu(In,Ga)Se 2 (CIGS), since both devices have similar architecture and processing. The key advantage of CZTSSe is the absence of indium, which may have issues with price or price volatility given its widespread use in other emerging technologies. Investigations into the use of CZTSSe for photovoltaics began in 1988 [1], and since then, it has been the subject of over 2900 publications to date. Of these publications, over 65% have investigated liquid processed layers. By this we mean that the kesterite layer was formed by annealing a precursor layer deposited from one or more solutions containing the necessary elements. Many of the advances in our knowledge of the kesterite material have come from solution-processed films, and indeed the world record power conversion efficiency (PCE) cell contains a solution processed kesterite layer [2]. Our purpose in this review is to provide a comprehensive overview of the different possible solution-based syntheses of kesterite layers by discussing the various film formation mechanisms, by giving a brief historical background for each, as well as by comparing and contrasting the latest approaches.
Solution methods to produce semiconductor precursor layers are of interest since they may be deposited uniformly at high speed, with nearly 100% material utilization, using low-capital-cost equipment. The kesterite semiconductor is synthesized by annealing a precursor layer that is formed by drying a wet film deposited from 'inks' or solutions. Although the constituents of the semiconductor, namely Cu, Zn, Sn, S, and or Se may be deposited with high purity, the challenge lies in removing all the other components of the solutions from the precursor films. Most semiconductors are notoriously sensitive to impurities, and therefore the removal of all unwanted species is a very high priority. While many CZTSSe devices achieve >8% PCE with >1 atomic% impurities (mainly carbon, nitrogen, and halogens), the best performing devices tend to have the lowest concentrations of impurities, approaching levels from vacuum-based processing. As a result, methods are sought to fabricate kesterite thin films that only contain their constituent elements with the correct composition and any necessary dopants or defect passivation agents (such as lithium and germanium).
The kesterite absorber layer can be formed from dried chemical precursor layers consisting of molecules or salts, of preformed kesterite nanoparticles, of metallic alloys, or even of binary chalcogenide compounds. This review is split into five parts covering the different possibilities. Parts (2.1) through (2.3) review the molecular and salt-based solution approaches, where different solvents may be used including: (2.1) the carbon-free solvent hydrazine, (2.2) aprotic solvents such as di-methyl-sulfoxide, and (2.3) protic solvents that may yield insitu particle formation due to condensation reactions (the so called sol-gel method). Part (2.4) discusses the main methods to form and stabilize nanoparticles. Part (2.2) assesses the use of electrodeposition to form both metallic and metal chalcogenide precursor layers. The review is completed by (3) a brief overview of the different precursor deposition methods, and (4) associated strengths-weaknesses-opportunities-and threat (SWOT) analysis of the different solution-based synthetic routes.

Hydrazine-Based suspensions and solutions
Ever since its conception, CZTSSe has been recognized as a challenging system for thin-film device fabrication using state-of-the-art vacuum-based methods [3]. Many exciting materials and property similarities with CIGS inspired attempts to apply analogous thin film growth approaches, such as two-step co-evaporation or sequential sputtering followed by annealing [4][5][6]. Composition control and secondary phase suppression during the high temperature deposition and/or crystallization is hindered in these approaches by the volatility of Sn chalcogenide phases [7], in addition to the already familiar S and Se. The process control becomes even more difficult due to the interrelation of the vapor pressure of all these three elements and their phase formation. There was therefore a need for a method that could at the same time provide precise, homogeneous target composition and facile crystallization of the desired phase, minimizing secondary phase formation and providing device-quality interfaces.
Hydrazine solution processing was the first method to simultaneously address these challenges to a level enabling CZTS to approach and later surpass the 10% PCE milestone. Despite the extreme safety precautions required for hydrazine manipulation, including flammability and toxicity, this solvent has unique properties for research on multinary chalcogenide materials. Hydrazine is carbon and oxygen-free and, due to its strong reducing properties, has the capability to form soluble chalcogenide complexes with many metals (therefore already providing desired metal-chalcogen linkages in the precursor, as needed in the final product) [8][9][10]. These precursors cleanly decompose at relatively low temperatures, producing highly pure crystalline layers with precise composition control. The method was first developed for SnS-based thin-film transistors [8]. Later it enabled achieving excellent composition and dopant control in some of the most sophisticated multi-element PV absorbers, such as Cu(In,Ga)(S,Se) 2 , producing homogeneous layers with impressive crystalline quality, which were incorporated into high performance devices [9]. The efficiency was further improved to become the highest among all-inorganic solution-processed absorbers at the time (figure 1(a) [11]). Implementation of hydrazine solution processing for CZTSSe was not straightforward, given the poor solubility of zinc chalcogenides in hydrazine (in contrast to the Cu and Sn chalcogenides). However, reaction of Zn nanoparticles with chalcogen solutions leads to the formation of zinc hydrazinate nanoparticles with different shapes [12]. These nanoparticles were found to be sufficiently dispersible in solutions of Cu and Sn chalcogenides to allow the development of hybrid solution-nanoparticle (or slurry) inks. While not yet harnessing the full reaction potential of molecularly intermixed pure solution precursors, the hybrid inks could already be spin-coated, annealed and processed, as for CIGS. Employed in analogous device structures (Glass/ Mo/CZTSSe/CdS/ITO/Ni-Al, figure 1(b)), they soon yielded a breakthrough in CZTSSe performance to 9.6% efficiency [12]. This value, over 40% higher than any previous result with this material, demonstrated for the first time that an atmospheric-pressure liquid-based method was able to not only reach, but even to surpass established vacuum-based processing. The sudden boost in CZTSSe performance motivated intensified efforts to further improve this technology and dedicated projects were launched at IBM, as well as at multiple other research centers worldwide. Incremental improvements of the hybrid approach, including refinement of composition and annealing conditions to suppress Sn loss, soon increased efficiencies to >10% [13], reaching the highest PCE of 11% in 2012 [14]. Despite this success, the presence of nanoparticles and their segregation could not completely suppress localized compositional fluctuations, which limited further efficiency increase by this method.
Development of pure CZTS solutions in hydrazine relied on finding a soluble zinc precursor. This was achieved by two different routes. An approach developed at IBM used zinc formate, which could be directly dissolved in solutions containing the Cu-Sn-S-Se precursors [15]. The presence of Sn was found to be critical for the formation of soluble species, suggesting that it participated in the formation of a carboxylate complex with Zn. Independent work at UCLA used a completely different approach, which however most likely arrived at a similar soluble complex. Hydrazinocarboxylic acid, obtained by carefully reacting solid CO 2 with hydrazine, was reacted with zinc metal to form a stable complex that could be incorporated in the Sn solutions [16]. The zinc formate method soon displaced the hybrid slurry approach due to its advantages of fine composition tuning with excellent uniformity and became the main vehicle of state-of-the-art CZTS device optimization and fundamental materials understanding. Combining advanced thermal processing with precise composition and optical stack optimization of the complete device [17], a world record efficiency of 12.6% was achieved [2]. This Figure 1. SEM cross sectional images of exemplary hydrazine processed absorbers: CIGS yielding 15% efficiency (a) [11], CZTS from hybrid slurry-based ink with 9.6% efficiency (b) [12], and pure-solution CZTS ink offering 12.6% efficiency (c), with corresponding J-V curve of the latter device (d) [2].
2014 milestone remained unchallenged for many years, prompting questions as to whether CZTSSe performance has reached a plateau and if further improvement would be possible. Nevertheless, by providing state-of-the-art CZTSSe absorbers for fundamental studies, the method continued to provide key insights into the CZTSSe material and device understanding, serving as a benchmark for the improvement of a multitude of other fabrication approaches. Some of these are discussed in the following parts of this review and are already approaching and will likely surpass the long-standing efficiency milestone set by the hydrazine approach.

Aprotic molecular inks
Molecular inks are solutions-i.e. single phase homogeneous liquids, as opposed to heterogeneous mixtures, as is the case for slurries, suspensions, or colloidal dispersions of nanoparticles. A key advantage of molecular inks is that they avoid localized compositional fluctuations that are difficult to overcome with suspensions or colloidal dispersions of nanoparticles. The development of hydrazine-based molecular inks was a key advance that led to the current record CZTS device reported in 2014 (discussed above). However, hydrazine presents many challenges due to safety concerns and environmental regulations given its toxicity. As a result, molecular inks for CZTS have been developed based on other solvents, including: (1) polar aprotic solvents di-methylsulfoxide and di-methyl-formamide (DMSO and DMF) [18][19][20][21][22][23][24][25], (2) polar protic solvents (like water and alcohols) [26,27], and (3) amine-thiol solutions [28][29][30]. Aprotic polar solvents like DMSO and DMF have very high pKa and lack a reactive oxygen site (in contrast to protic polar solvents like water and alcohols). Thus, these aprotic solvents avoid the formation of metal-oxygen-metal bonds or metal-oxygen-carbon bonds, which can be difficult to fully eliminate once formed. We also note that thiols (RSH) are more acidic (or equivalently, more protic) than their corresponding ROH alcohol. Amine-thiol routes show promise, but they must contend with a reactive hydrogen that may yield difficult to remove MSR functional groups that result from condensation reactions with the thiol (RSH). In this section, we focus primarily on the aprotic polar solvent route to molecular inks.
A widely studied molecular ink route to CZTS is based on the relatively benign aprotic polar solvent DMSO and a sulfur containing complexation ligand thiourea (TU), which was first reported in 2011 [18]. In DMSO, the hydrogens on TU are not very acidic (note that the pKa of TU in DMSO is 21) [31]. As a result, TU tends to act as a ligand to solvate metal cations with the accessible soft Lewis base S-donor site coordinated to the metal cation. In fact, many TU-metal complexes for soft metal cations are well-known and studied, including CuCl(TU) 3 [32], Cu 2 Cl 2 (TU) 4 [33], AgCl(TU) 2 [34], ZnCl 2 (TU) 2 [35], and SnCl 2 (TU) [35]. For the CZTS system, single crystals of Cu 2 (TU) 6 Cl 2 (DMSO) 2 ( figure 1(b)) have been isolated from complete inks (figure 2(a), far right) 13 . Further, with DMSO as the primary solvent, hydrated metal salt complexes can be used as precursors since even water in DMSO is quite aprotic with a pKa of 31.4 [36]. In fact, experiments show that DMSO can often be replaced with H 2 O at up to 50 vol% in the parent solvent without triggering hydrolysis and condensation reactions that lead to conventional sol-gel chemistry (see footnote 12). The use of water may be desirable to make the solvent and ink even more 'green'.
The desired oxidation state for cations in CZTS (+1, +2, and +4 for Cu, Zn, and Sn, respectively) may be achieved in the molecular ink by dissolving salts where the metals are already in that oxidation state [21,22,24] or by adding salts with metals in other oxidation states, provided that they have a facile pathway for redox reactions to reach to the desired oxidation state. This latter approach is exemplified by Xin et al [19], in which Cu(II) and Sn(II) precursors are added to the DMSO solution (without TU present) and undergo a redox reaction, 2Cu +2 +Sn +2 →2Cu +1 +Sn +4 , as indicated by the transition from a dark-green tinted solution due to Cu(II) to a colorless solution indicative of Cu(I) (figure 2(a)). The additional chloride that accompanies the SnCl 2 and ZnCl 2 also has a beneficial effect, in allowing the formation of a DMSO-soluble copper chloride anion (Cu 4 Cl 6 ) −2 , as determined by single crystal x-ray structure solution of crystals to be Zn(DMSO) 6

+2
(Cu 4 Cl 6 ) −2 . The crystals were grown by evaporating DMSO from the ink after ZnCl 2 addition. Upon addition of TU to the ink, Cu(I) forms a complex with the TU yielding a stable molecular ink. Single crystal x-ray structure solution of the first crystals that formed upon DMSO evaporation from the complete ink, are determined to be Cu 2 (TU) 6 Cl 2 (DMSO) 2 , as shown in figure 1(b). In contrast to the soft Cu(I) cation, the harder Zn(II) and Sn(IV) preferentially complex with the harder O-donor Lewis base site on DMSO and the O-donor and N-donor Lewis base sites on DMF. Later, Clark et al [37] showed that chloride transfer reactions are extremely common and useful for forming kesterite and chalcogen semiconductor films. Further, by expanding the aprotic solvent suite to include DMF, Clark showed that an even wider variety of metal cations (useful for kesterites more broadly) can be stabilized. Some metal species including Ag(I), Ge(II), and Ge(IV) (which are studied in an attempt to reduce cation site disorder in the kesterite lattice) [38][39][40][41][42][43] have low solubility or low stability in DMSO-TU inks, and can be stabilized by molecular inks based on DMF and TU. Overall, this research illustrates that DMSO-TU and DMF-TU solutions are very flexible chemical routes to solvate a wide spectrum of metal chloride salts due to a variety of soft and hard Lewis base sites available to accommodate both soft cations like Cu(I), Ag(I), and Sn(II) and harder cations like Cu(II), Zn(II), Sn(IV), and Ge(IV).
One of the key challenges to making high quality kesterite semiconductors from any molecular ink is the removal of impurities. For the DMF-TU and DMSO-TU inks, these impurities are primarily carbon, nitrogen and chloride compounds. While there are several pathways that lead to dense large-grain crystalline films, there are many reports of CZTS films from these molecular ink processes that exhibit multilayer morphologies [44][45][46]. These morphologies often show a fine-grained impurity-rich layer trapped in between dense, largercrystalline grains and the substrate. Using composition-calibrated glow-discharge optical emission spectroscopy (cc-GDOES) to reveal the depth-dependent concentration of impurities (see figure 3), it was shown that the process parameters of the annealing and selenization steps are critical to avoid the trapping of impurities. Intermediate annealing temperatures led to the formation of amorphous hydrogenated carbon nitride (a-C:H: N) from the polymerization of TU decomposition products [47]. These impurity-rich layers could be avoided in two ways: (1) by using low annealing temperatures that prevent extended a-C:H:N networks from forming and allow the C and N compounds to be volatilized during subsequent selenization, and (2) by using high annealing temperatures where the a-C:H:N thermally decomposes during annealing (before selenization) [47]. Other parameters, such as film thickness and the alkali environment during selenization also have a major impact on the extent of impurity removal.
The DMSO-TU and DMF-TU chemistry has led to some of the highest quality films and highest PCEs devices for CZTSSe, fabricated by spin coating followed by annealing and selenization (figure 4(a)). In 2015, the DMSO-TU chemistry was used to incorporate lithium doping into CZTS inks, which led to dense impurity-free absorbers (figure 4(b)) and yielded 11.8% active area PCE devices [20]. The DMSO-TU route was also studied by  [47]. (b) Composition-calibrated glowdischarge optical emission spectroscopy (cc-GDOES) depth profiles showing the sensitivity of impurity removal to the annealing profile. Specifically, it shows that 3 min anneal at 500°C after the deposition of each layer is sufficient to remove carbon, nitrogen, and chlorine down to close to the detection limit after selenization. groups at ZSW [44] and Empa [48], each reporting devices with >10% PCEs. More recently in 2018, Haas et al [49] presented a large study of over 700 CZTS devices from DMSO-TU inks, examining the effects of Li, Na, Cs, and Rb doping and the cross-correlation with the metal ratios. The same study also found that Li yielded the best devices and reported a new record of 12.3% active area PCE for hydrazine-free CZTSSe figure 4(c) [49]. Also, the DMF-TU system has been used to spray-coat germanium-alloyed CZGTSSe devices [22] with 11.0% PCE and a record high open circuit voltage (V oc ) of 583 mV for the 1.15 eV bandgap, which is 63% of the Shockley-Queisser (SQ) limit V oc . This is the highest V oc / V oc SQ reported for any kesterite-based device with any bandgap, including the record hydrazine-based devices [2] that only reach 58% of their SQ V oc limit. These percentages are calculated by dividing the measured V oc by the calculated SQ V oc for the bandgap, which is a much more transparent way to quantify the voltage performance for a variable bandgap material, like CZTS. This is in contrast to simply using the measured V oc or even the concept of a voltage deficit, since both values depend on bandgap. Using either V oc or V oc deficit can often be misleading about the voltage performance [22]. Our previously reported detailed balance calculations show that the SQ V oc in units of volts is linearly related to bandgap in units of electron-volts with V oc SQ =0.932×Eg − 0.167, which can then be used to calculate the percentage, 100×V oc /V oc SQ [22].

Protic solvents and the sol-gel method
The sol-gel method is a solution-based synthesis technique where one first forms a colloidal suspension of solid particles (a 'sol'), which then undergoes gelation (gel). The resulting gel is a continuous network of connected solid phase particles in a secondary continuous phase, usually a liquid. The predominate reaction mechanisms are hydrolysis of the metals followed by poly-condensation reactions [50,51]. The 'sol-gel' method provides a facile and promising route towards low-cost mass production, which allows for precise control over the microstructure and the film composition.
The sol-gel technique involves three steps for the preparation of thin films. In the first step, the metals are dissolved as ions into solution using precursors such as copper (II) acetate monohydrate, zinc (II) acetate dihydrate and tin (II) chloride dihydrate [50,52]. The solvent is typically 2-methoxyethanol [52] and a stabilizer such as monoethanolamine can be added [50]. Additionally, a sulfur source in the form of thiourea, may be added to the solution [52]. As also described in the aprotic solvent part (2.2) above, thiourea may complex with the metal ions in solution. The chosen salts must completely dissolve and be stable for hours. Any precipitation will lead to a change of bulk solution composition, changing the final kesterite layer composition. If the precipitates get into the final layer, they will form secondary phases within the absorber layer. The second step consists of kesterite precursor thin film formation, which involves (i) spin coating the solution near to room temperature onto a molybdenum-coated glass substrate (ii) a baking process in air using a hot plate to form the metal oxides and remove the residual impurities, and (iii) repeating the coating/baking cycle several times to reach the appropriate thickness of the absorber. The coating solution must not react with the molybdenum, and the film has to be smooth and pinhole-free, since otherwise the solar cell efficiency will be likely deteriorated by shunting. In the final third step, a high temperature thermal treatment in a reducing and/or an aggressively chalcogenizing environment is required to produce the desired CZTSSe phase, remove oxygen and other  [20] (c) highest performance cell from DMSO-TU molecular ink at 11.5% total area PCE with 12.3% active area PCE, adapted from Haas et al [49] (d) J-V of a germanium alloyed CZTGSSe device from the DMF-TU molecular ink process which shows the highest reported V oc /V oc, SQ for any kesterite device [22] of 63%.
impurities, and create crystalline material. During this stage, gas-solid reactions take place at the surface, as well as solid-state reactions in the bulk between the elements, which lead to binary, ternary and finally quaternary chalcogenides. The annealing temperature and atmosphere play a key role in determining the composition and crystallite grain size.
Recently, Sun et al demonstrated a CZTS based solar cell with 8.8% PCE using the solvent, stabilizer and precursors described above [53]. They investigated the solar cell efficiency dependence on the initial metal solution concentrations by varying the Cu/Sn ratio in the following way: 1.69, 1.78, 1.84 and 1.93 labeling the samples A, B, C and D, respectively, while keeping the Zn/Sn=1.25, S/Metal=2 in the precursor constant [53]. The resulting kesterite absorber layer films are illustrated in figure 5. The best cell was achieved with film C, yielding a high V oc of 746 mV, whilst the solar cells A and B were shunted [53]. In a previous work [54], Cu 2 ZnSnS 4 kesterite-based thin films were successfully prepared using copper acetate (1.6 M), zinc acetate (1 M), tin chloride (0.8 M), and thiourea (6.4 M) using 2-methoxyethanol as a solvent, where the authors studied the effect of the film thickness on the kesterite structural and optical properties. Based on the XRD and Raman characterization, it was found that the crystalline quality was improved with increasing film thickness. A bandgap of 1.43 eV was observed for the thickest film with the highest crystallinity. They also noticed that, by increasing the total annealing pressure, a secondary phase SnS 2 can be formed.

Nanoparticle-Inks
Reviews can be found on fabricating nanoparticles (NPs) of various types (e.g. metallic, chalcogenides, oxides) for application in different locations of thin-film photovoltaic devices [55], on the approach of using sintered NPs for CIGS or CZTS [56], and on the evolution of NP-based CZTS device efficiency [57]. Here, we provide an update and comprehensive survey on the CZTS nanoparticle processing.
Nanoparticle-based inks provide the option to distinguish between the solvent(s) used for precursor dissolution and NP synthesis, from the solvent(s) used for NP deposition. This may allow for better processing control (e.g. lower residual contaminants or viscosity adjustment), such as NPs rinsing, protection with shorter ligands, mixing NPs with varying composition and/or size distributions, prior re-formulation in a new, e.g. less toxic solvent. As illustrated in figure 6, precursor film formation from nanoparticles generally proceeds in three steps: (A) colloidal synthesis of solid nanoparticles, (B) re-formulation into an ink, (C) coating into a porous film. Even though a high temperature crystallization step under chalcogen atmosphere (T>400°C, under S and/or Se) is generally required after deposition, the final metal chalcogenide composition can be fixed from the start. CZTSSe device efficiencies in the 9%-11% efficiency range have been demonstrated with the desired Cupoor, Zn-rich metallic ratios, thanks to fine compositional tuning from the initial precursor solutions [58][59][60]. Below, the three steps are discussed in more detail.

Particle synthesis
At least two different reactant solutions are needed for NP formation. Depending on their mixing conditions (nature of solvents, precursor concentrations, additives, time, and of prime importance the temperature), the reaction can result either in well-crystallized or amorphous-like NPs.
The first CZTS nanoparticles were reported in 2009 by various groups based on a high temperature nonaqueous co-precipitation method [64][65][66], similar to CIGS NPs reported in 2008 [67]. This non-aqueous hot injection (HI) method is the most abundant in literature. Previous reviews and perspectives on kesterite NPs and associated CZTSSe devices can be found in Bucherl et al [56] from 2013, Ghorpade et al [68] from 2014, and Hages and Agrawal [61] from 2015. In a typical reaction, for the synthesis of CZTS nanocrystals as reported by Guo et al [64], oleylamine (OLA) serves as surfactant and solvent to dissolve cation precursors at temperatures of around 225°C. The nanoparticle nucleation is initiated by injecting a 'S dissolved in OLA' solution into a preheated cation solution, and these nuclei are allowed to grow by maintaining the elevated reaction temperature for a certain period until the desired particle size is achieved (typically 0.5-1 h). These OLA-capped nanoparticles after proper washing are then dispersed in the desired solvent to formulate a nanocrystal ink for coating films. Riha et al have reported a slight variation to this route, which involves use of tri-octyl-phosphine oxide (TOPO) as a solvent along with OLA, giving better particle monodispersity. Another similar synthesis of CZTS nanoparticles is achieved via the heat-up route, for which cations and S are mixed together in OLA and heated to a reaction temperature of around 280°C, to yield monodisperse CZTS nanoparticles and to avoid the need for HI. Along with pure CZTS nanoparticle synthesis, the HI process has also been successful in synthesizing cation-alloyed CZTS nanoparticles, including Ge-doped CZTS [69,70] and Ag-doped CZTS [71], by using corresponding metal precursors for the reaction, and has shown improved photovoltaic performance compared to pure CZTS nanoparticle devices [59,71]. The CZTSe layers in high efficiency devices generally exhibit the kesterite crystal structure, but NP synthesis of CZTS materials can also yield a meta-stable wurtzite phase, e.g. through the choice of solvent, such as dodecanethiol [72,73], which can then be converted into the kesterite phase by performing an annealing step under Se environment [74].
The Metathesis method (also called instantaneous precipitation) is another route to synthesize nanoparticles and is performed at room temperature or even within an ice bath. Applied initially for instance to MoS 2 [75], CdTe and Cu(In,Ga)Se 2 [76], this approach often results in NPs with comparatively poor crystallinity. A similar room temperature process for Cu 2 ZnSnS 4 NP metathesis was later reported by IMRA Europe in 2014 [62,63], based on comparatively benign solvents (water and acetonitrile). This approach is also applicable to a range of sulfide compounds, such as Sb 2 S 3 or SnS [62]. In a typical CZTS metathesis, a de-oxygenated aqueous solution of NaHS is poured into an acetonitrile (CH 3 CN) solution of the three metal chlorides SnCl 4 , ZnCl 2 and CuCl. Performed under ambient conditions (preferentially under nitrogen), the mixing instantaneously results in an aqueous colloid comprised of amorphous-like nanoparticles, with a size<10 nm ( figure 6, bottom). In these round nanoparticles, TEM EDS indicated that the four elements Cu, Zn, Sn, S are intimately mixed at tunable compositions, together with traces of Na; the latter being (intentionally not completely) removed afterwards by water rinsing steps prior formulation.
Although, use of the nanoparticle HI approach suggests nanoscale film uniformity, a STEM-EDS study by Yang et al shows inter-particle (within a single particle) and intra-particle (between two particles) compositional non-uniformities for CZTS nanoparticles synthesized from various recipes involving the HI approach [77]. This compositional non-uniformity, which also extends to non-uniform CZTS nanoparticle size distribution, can be reduced to some extent by increasing the reaction temperature and time [70,78]. Films casted using such nonuniform particles exhibit composition variation across the film but reactive annealing performed in a selenium environment at temperatures above 400°C results in redistribution of the elements and ultimately relatively uniform CZTSSe film composition [77].

Formulation
As illustrated above, nanoparticles for fabricating CZTSSe active layers are generally synthesized as pure sulfide NPs. Indeed, to our knowledge, >9% PCE devices have been demonstrated only with films based on puresulfide NPs and with the S/Se ratio adjustment being performed during the high temperature selenization of the porous films from deposited/dried ink. Nanoparticles synthesized using the HI approach generally yield OLA capping on the particle surfaces, which requires the use of non-polar or low polarity solvents like hexane, toluene and hexanethiol, in order to disperse the particles for ink formulation [79]. However, an advantageous characteristic of the sulfur-terminated CZTS surface is its good wettability with water (figure 7) [80]. At the ink formulation stage, pure-sulfide NPs that were fabricated from metathesis can thus be dispersed as stable inks into water, and more generally into polar solvents (figure 8). This flexibility offers a broad range of solvents for ink formulation, including environment-friendly solvents [62].

Deposition methods
Porous film deposition from NP-based inks simply involves the controlled evaporation of the solvent during deposition (either by spin coating, doctor blading, spraying or cousin coating methods -see the overview of coating methods in section 3), nominally without the concern of removing by-products seen in reactive deposition methods, like spray pyrolysis [81]. For photovoltaic thin film absorber application, a final thickness of 1-2 μm is needed, which in any case implies the need for repeated coating steps (from two to ten steps). Typical film fabrication for OLA-capped CZTS NPs using the doctor blading technique involves dispersion of NPs in hexanethiol solvent with mass concentration between 200 and 300 mg ml −1 , which results in ∼1 μm thick films with two consecutive coatings. Each coating is generally followed by an annealing step at around 300°C to remove the excess solvent from the layer, yielding a porous nanoparticle film [78].
Non-reactive spray deposition of porous films from inks made of inorganic nanoparticles (metathesis) has been reported for e.g. CIGS [76], CdTe [82] or CZTS [63]. In the latter case, the formulation ink (10 mg ml −1 ) was diluted in a mixed 90% water/10% ethanol solvent, a ratio that was adjusted for optimum evaporation rate during spray deposition (at a temperature of ∼300°C). To make the approach successful for photovoltaic applications, the subsequent conversion of such porous films (figure 9) into a dense CZTS film with large grains (on the order of the film thickness) is critical. In contrast with the case of sprayed Cu(In,Ga)Se 2 films, which always resulted in poor morphology and <5% PCE when starting from NPs [83], large CZTS grains could be achieved by annealing under nitrogen, with PCEs of 5%-7% for the corresponding devices. The likely reasons for this achievement are the absence of carbon-containing ligands, an oxidation-reduction reaction between Sn from CZTS NPs and the Mo substrate [63] and the presence of Na traces. A second annealing step under Se atmosphere was then developed, among other refinements, to form CZTSSe films (∼65 at% Se-rich) with 9%-11% record PCEs [58,84].
Typically, a selenization is used to convert the CZTS (amorphous or crystalline NPs) film into a CZTSSe layer containing large grains. This process occurs via liquid-assisted growth, in which the film is identified to undergo a sintering mechanism that involves (figure 9): (1) condensation of Se in the CZTS NP film (2) dissolution of Cu in liquid Se (3) nucleation of Cu 2−x Se grains on the surface (4/5) incorporation of Sn/Zn via transport through liquid phase/solid-state diffusion as shown in figure 9 [85].

Electrochemical coating
Among non-vacuum technologies for metal film formation, electrodeposition stands out as a mature technology and one of the most effective, as well as low-cost, in industrial sectors worldwide. For the semiconductor sector, it has been used successfully by BP for the commercial deposition of CdTe thin film PV modules and in other PV technologies [86]. Since the pioneering work published by Scragg et al [87], more than 150 papers have been dedicated to the electrochemical preparation of kesterites, including several review papers [88,89].
As a shared feature with other coating methods, the classical 2-stage approach employed for the synthesis of kesterite films remains for electrochemical routes, which involves an electrodeposited precursor film followed by chalcogenization. It is in the precursor deposition stage where different electrodeposition routes have been targeted. There are different ways of classifying the electrochemical approaches used to date. One way is to Figure 8. Solvent selection criteria for stable CZTS colloids made of CZTS NPs (metathesis method) re-dispersed in various solvents, at 5 mg ml −1 concentration. Green dots correspond to colloidal dispersions that have not sedimented after 24 h, at ambient temperature. Each tested solvent is characterized here by two Hansen Solubility Parameters: its hydrogen bond δ H and polarity δ P parameters (unpublished data) 14 . Figure 9. Diagram of the proposed liquid-assisted growth mechanism for large-grain CZTSe absorbers from CZTS nanoparticle films [85].
differentiate them according to the number of 'electrochemical' steps (or number of electrodepositions) required for the deposition of the precursor. According to this classification scheme, two approaches can be differentiated. The first approach contains those methods where multiple layers are electrodeposited to form the precursor. This group comprises the stacked elemental layer (SEL) method and includes the sequential electrodeposition of elemental metallic layers [87,90,91], sequential electrodeposition of stacked alloys in the form of CuZn/CuSn [92], and a combination of these, involving the successive electrodeposition of elemental and binary metallic layers [93,94]. The second approach involves those methods where one single electrochemical step is used to form the precursor. This group includes the co-electrodeposition method involving the deposition of a CuSnZn compound; and the so-called 'one pot' electrodeposition where the chalcogen source (selenium or sulfur) is also incorporated in the electrolyte, with the aim of directly obtaining CZTS or CZTSe. A comprehensive review of most of these electrochemical approaches was previously published by Colombara and coworkers [89]. In light of this, this section will focus on more recently published work so as not to duplicate past review efforts. Recent progress in terms of material quality, new insights and solar cell performance will be highlighted for each methodology.

Multiple layer deposition
The sequential electrodeposition of metallic layers has historically been the approach that leads to the highest PCEs [91,95]. An 8.2% record PCE for pure selenide kesterites has been achieved so far. First, Vauche et al [96] have reported this result for large area electrodeposited Cu-Sn-Zn stacked precursors, followed by selenization. A remarkable improvement in device performance was attainable after the incorporation of etching procedures (for removal of zinc and tin selenide secondary phases) and buffer layer optimization (figures 10(a) and (b)). Soon after, Yao et al [97] achieved the same efficiency by optimization of the selenization process and without etching steps. A three-stage annealing at low Se-pressure prevents tin loss and the formation of a thick MoSe 2 layer, thereby enhancing the device efficiency. In terms of sulfurized SEL precursors (CZTS), a record PCE of 8.1% was reported by Jiang et al [98,99]. The authors point out the significance of a long preheating treatment (≈200 min) of the electrodeposited Cu/Sn/Zn stacks, in order to achieve the highest quality of CZTS films and PV devices (figures 1(c) and (d)).
More recent contributions for the SEL approach have been focused on replacing water electrolytes by nonaqueous solvents. For instance, ionic liquids usually offer higher temperature operation and expand the electrochemical potential window making possible the use of higher current densities during electrodeposition because no competing electrochemical hydrogen evolution (reduction of protons) occurs [100]. Recently, Steichen and coworkers have reported a high speed electrochemical SEL procedure achieving average metallic deposition rate of 100 nm s −1 through the use of liquid metal salts [101]. This SEL approach combined with a rapid thermal process (RTP) is capable of producing a promising 5.5% CZTSe PV device.
The synthesis of Ge-alloyed kesterite films has also been explored for higher bandgap using multilayer electrodeposition [102]. In this work brass/tin/germanium stacks were prepared, using pyrophosphate electrolytes for brass and tin electrodeposition and an organic electrolyte for Ge electrodeposition. Despite only minor improvement in terms of PCE, this first attempt expands the capacity of electrochemical methods to obtain kesterites of varied compositions.

Compound electrodeposition
In the last years, most of the published works related to co-electrodeposition have addressed the replacement of citrate salts [103][104][105] as a complexing agent by pyrophosphates [106][107][108]. Other work has studied the effect of cell configuration on the compositional homogeneity of electrodeposited Cu-Zn-Sn alloys [109]. Also, sulfurized and selenized compound electrodeposited precursors were successfully tested in other types of PV devices, like DSSCs [110] or photocathodes for water splitting [111]. In terms of PCE, the record for the coelectrodeposition approach is still 8.0%, as reported by Jeon and coworkers [112]. More recent contributions have explored the combination of this approach with a RTP annealing, which is more attractive from an industrial perspective, and reported a 5.2% PCE [113].
The use of a quaternary-salt containing electrolyte has been the approach that has progressed the most in terms of increasing device efficiency. From a 5.5% PCE reported by Ge et al [114] in 2014, a new record 8.7% PCE CZTS solar cell has been recently reported by Zhang and coworkers [115]. Similarly, Ge et al have improved their previous solar cells by reporting a 7.1% CZTS PCE. In this work, after a systematic investigation, refined plating parameters and an optimized electrolyte led to the obtaining of electrodeposits with outstanding layer uniformity in terms of film appearances and compositions, of comparable quality to those obtained by vacuumbased methods ( figure 11).

Overview of coating methods
CZTSSe absorbers have been deposited from liquid-based precursors using a large range of deposition techniques. The quality of the final material is affected by all process parameters, such as compositional control, extrinsic doping or annealing parameters, in a rather complex way. Nonetheless, the deposition method itself is highly important, since compact and uniform layers (both in terms of thickness and stoichiometry) are required to obtain high performance solar cells in the subsequent steps. For a uniform and crack-free coating, the deposition method should be carefully combined with an ink that exhibits the appropriate rheological properties, such as the viscosity, density and surface tension. The compactness of the film must be maintained after any intermediate drying step, which is performed to evaporate the solvent and remove impurities. Drying of the film often results in the formation of cracks, which can act as shunting paths in the solar cell and reduce its performance. To eliminate this effect, the absorber is typically deposited through a series of deposition/drying cycles, such that the deposited film does not exceed a 'critical' thickness, related to the tensile strength and the Young's modulus of the film [117].
Typical deposition methods for CZTSSe thin films involve two separate steps-i.e., the material coating followed by a post-deposition step, in the form of a thermal or chemical treatment. The processes where the high quality material is obtained in a separate step from the coating, have also been referred to as sequential or 'two step' approaches [118]. The benefit of sequential approaches is that the deposition step can be performed using a low capital cost method and/or at rather low temperatures. The ideal deposition method would have a low thermal budget, high deposition rates and low operating costs to be industrially relevant. This section will give an overview of the most popular coating methods for research-scale CZTSSe precursor layer deposition, and the highest PCE values for the given methods are compared in table 1.

Spin coating
Spin coating has been extensively explored in kesterite research, as it is well suited for processing of small substrates [2]. It typically involves the application of a liquid to a substrate, followed by the acceleration of the substrate up to a certain angular velocity rotational speed [120]. This method is typically combined with solutions of fairly low viscosity (< 0.01 Pa s) and it requires multiple depositions to build-up the necessary film thickness [117,120]. The obtained film thickness depends on the angular velocity during deposition and the physical properties of the ink [120]. For obtaining a final film thickness on the order of 1-2 μm, a minimum of five successive layers are typically needed [12]. Despite the process simplicity, the low material utilization and the restriction to small substrates makes this method industrially less attractive [120]. Table 1 summarizes the record efficiencies obtained in CZTSSe to date with each coating method. Among the solution-based methods, the highest reposted efficiency of 12.6% was obtained using spin coating, and with hydrazine as the solvent [2] (see part (2.1) of the thin film synthesis section).

Spray coating
Spray coating is a method that is also compatible with low viscosity inks and it is suitable for large area deposition. This method involves forcing the ink through a nozzle to form a fine aerosol [120]. For the ink transport to the substrate, a carrier gas or electrostatic charging (electrostatic spraying) may be applied, and the spray jet is scanned continuously to cover the entire substrate area [81]. Although water would be the most benign and safest choice as the solvent, it may introduce substantial oxygen content in the final layer. Water also tends to be unsuitable for spray deposition due to its high surface tension, causing large droplet size and film disjoining upon drying. As a result, alcohol-based protic solvents have been the most commonly explored for spray coating, which allow for uniform films and fast solvent evaporation. However, the highest efficiency CZTSSe devices made by spray-coating employed either DMSO or DMF as a solvent, both of which have a surface tension substantially less than water, but higher than ethanol. The large number of drops in a spray deposition process, their varying velocity and droplet size with time due to evaporation and coalescence, and the temperature gradient between neighboring droplets, are typically so complex that computational resources fail to predict the dynamical and thermal interactions involved. Therefore, the optimum spray process parameters for obtaining device-grade kesterite films tend to be experimentally defined.
Spray coating can theoretically be used as a single coating method, as the solvent evaporates upon contact or in proximity to the substrate surface. In the CZTS system, however, it is only exploited as a two-step approach. The synthesis of kesterites-based absorber layers in an ambient environment using this technique was first reported by Nakayama in 1997 [121]. Following this work, different reports on the spray pyrolysis technique have been published, where the precursor type, substrate temperature, post thermal treatments and solution pH have been investigated [122][123][124][125]. The substrate temperature in particular, is a key process parameter, as it can influence both the morphology and the stoichiometry of the final film [126]. A drawback of this method is the relatively small grain size and high surface roughness, in comparison to other techniques [127]. The absorber roughness can directly influence the absorber/CdS interface, for solar cells prepared in the substrate configuration. This effect can be less detrimental for superstrate solar cells, where the kesterite is deposited on the CdS buffer layer. The relationship between secondary phase formation and chemical composition was also investigated [128]. It has been shown that non-optimum Cu/(Zn+Sn) and Zn/Sn compositional ratios can result in the formation of conductive secondary phases related to copper and sulfur compounds [128]. Spray processing generally leads to rougher films than other deposition methods, but offers several advantages over the more common spin-and dip-coating procedures, including reduced material loss during fabrication, higher sample throughput, and deposition over a larger area [81].
The highest efficiency CZTSSe PV device from a spray-coated molecular ink is 11.0% PCE and is based on DMF as a solvent [22]. Previous reports by the same group using DMSO achieved 9.5% PCE [129]. With regards to nanoparticle-based precursors, Larramona et al have reported an efficiency of 10.8% by (non-pyrolytic) spraying of a colloidal dispersion of CZTS nanoparticles in a solvent comprised of 90% water and 10% ethanol [60]. It is likely that these efficiencies (for both molecular inks and NP inks) will increase to similar levels as spin coated devices. As a case in point, a recent private communication reports that a 11.5% PCE device was achieved by the same team at IMRA for a 0.25 cm 2 total area small cell (so far unpublished data). This process was scaled to 1 cm 2 for a certified large area CZTSSe record of 9.8% PCE [84], and has now been pushed to 50 cm 2 [130]. The latter method has a high commercialization perspective due to the low toxicity of the precursors used.

Printing
Printing techniques, such as screen, pad and gravure printing describe a group of techniques where a layer of ink is transferred from a stamp to a substrate by a reversing action [120]. These techniques can have a very high throughput and offer the possibility of patterning. According to this definition, doctor blade and inkjet printing may rather be regarded as coating techniques, since they involve pouring or casting the ink, rather than stamping. However, they are included in this section as they share important features with more traditional printing methods; namely the compatibility with complex patterns, for the case of inkjet printing, and the necessity for high viscosity inks, for the case of doctor blading [131]. Inkjet printing is a highly versatile coating technique, as it allows deposition of both dissolved and dispersed inks, on any substrate. The working principle consists of the droplet formation and ejection, the spreading and coalescence of the droplets on a surface, and finally the solvent evaporation [131]. Inkjet printing offers the advantage of extremely high lateral resolution and roll-to-roll (R2R) compatibility. Its main disadvantages are its slow drying (compared to techniques like spin coating) and the requirement for very specific ink propertiesnamely viscosity and surface tension. Doctor blading involves a blade that moves over the substrate surface at a certain distance and at a constant velocity, spreading the ink. It offers high material utilization and it is easily transferrable to R2R, in the form of 'knifeover-edge' coating. Its main disadvantage is the necessity for high viscosity inks (> 100mPa s), which generally requires additives [131].
Despite the versatility of these approaches, printing has not yet been substantially explored on the research scale for the CZTSSe system. In comparison to other coating methods, printing generally requires higher viscosity inks, in which case the viscosity adjustment is done using organic additives. The presence of the additives may leave carbon impurities in the film, thereby necessitating an additional high temperature annealing step for their removal. Especially when combined with the use of suspensions rather than solutions, long chain molecules are also used for the stabilization of the particles, which is an additional source of carbon in the film. This, together with the low yield for fabricating nanoparticles [132] can be regarded as the limitations for printing dispersed inks. The highest PCE achieved for CZTSSe films prepared using nanoparticle precursors is 9.3%, obtained using a doctor blading technique [85] (see the synthesis section nanoparticle part 2.4).

Electrodeposition (ED)
ED methods involve electron transfer reactions between electrodes and dissolved reactant ions, the latter usually being in an aqueous solution phase. To achieve uniform layers the reactant ions must arrive and reduce everywhere on the electrodes surface at the same rate, failure to do so leads to compositional and thickness nonuniformities likely resulting in secondary phase formation. The biggest advantage of these approaches is the lowcost, high uniformity and high material utilization efficiency exceeding 90% [133]. ED methods have already been successfully implemented in industry, which confirms their large area compatibility is possible [133]. In addition to the deposition of SELs, a 'one-pot' ED approach can be used, where all the elements (including S) are deposited in a single step. Still, a post-deposition annealing step must be performed to obtain highly crystalline and single phase material quality. The highest efficiency to date from ED was obtained using the coelectrodeposition approach, followed by RTP sulfurization [115]. This improved efficiency of 8.7% was achieved by increasing the background pressure during sulfurization. The increased partial pressure of sulfur during annealing has been shown to improve the crystallinity of the absorber and reduce the thickness of fine-grain layer [115].

SWOT analysis
Having reviewed the state-of-the-art solution-based strategies and the principal coating methods, it is worth comparing their relative strengths and weaknesses, as well as the possible opportunities and threats to future Table 2. Comparison of the strengths, weaknesses, opportunities and threats for each of the different solution-based methods. It offers guidance for future research in that it highlights the relative promise and challenges that must be overcome for increased efficiency or for widespread implementation.

Strength
Weakness -Only applicable to conductive substrates -Roll-to-roll capabilities -Contaminants from bath or electrodes incorporated during the plating process -Low temperature operation -Some approaches yield extremely uneven coating even in small areas -For larger areas develop new (ED) cell geometry to improve current distribution and coating uniformity -Uncontrollable precursor composition leading to detrimental secondary phase formation -Stable benign precursor solutions which are recyclable -Stacked elemental layer precursor can be used as a replacement to vacuum sputtered precursors research in these directions. A SWOT analysis is presented in table 2. The aim of this analysis is to highlight and offer guidance about the relative challenges faced by each route, rather than identifying a more suitable way over another. In general, except for perhaps the hydrazine-based approach, the detection/measurement and removal of impurities is still a topic which requires further research effort for all solution-based routes. This analysis is explicitly based on solution methods, thus a direct comparison with vacuum or single crystal methods is avoided. However, even if not included in the table, an inherent strength of solution methods over vacuum methods is the high material utilization of Cu, Zn, and Sn. In comparison to vacuum-based methods, solution processes exhibit two main challenges. The first is to achieve reduced or impurity-free absorber layers, since impurities seem to be one of the factors which limit PCEs. The second relates to upscaling to industrially relevant sizes in an economically viable manner, which is proven for vacuum-based methods, but only somewhat for solution-based methods. Of course, the main challenge for all scientists working on the kesterite material is to further improve the PCE towards 20%.

Conclusion
For any photovoltaic absorber technology to succeed in the global marketplace, it must offer high performance, low-cost processing, and facile integration within existing manufacturing infrastructure. Kesterite-related technologies, most notably CZTSSe, have recently received a great deal of interest. The main reason for this is the performance increase achieved in laboratory-scale devices, as well as the high potential of this material system to serve as a 'drop-in' replacement for already commercialized CIGS absorbers. Key to this argument is the abundance/ low-cost of Zn, Sn raw materials (for substitution of rare and expensive In, Ga elements), and availability of cost-competitive approaches for fabricating the CZTSSe absorber layers. In this review, we have explored only an aspect of CZTSSe film processing that depends on liquid-based deposition. Such approaches not only provide a prospective pathway for low capital cost fabrication, but also the ability for depositing films of varying composition/morphology for defect and other chemical/physical studies, which are critical for overcoming current performance bottlenecks. While hydrazine-based approaches provided an early lead in terms of PV performance, at least in part because of the near-ideal solvent properties of hydrazine for chalcogenide-based materials, other less-toxic alternatives (based on protic, aprotic or nanoparticle-inks) have reached near parity. Overall, the leadership shown by liquid-based processing in terms of performance for the kesterite absorbers shows that such processing approaches provide a viable pathway for achieving state-of-theart PCE values relative to vacuum-based approaches, which more typically are associated with high performance. However, in order for solution-based methods for CZTS to have further impact on the technology, an improvement in PCE towards >18% is required. Assuming this point is addressed, the combination of high performance and lower-cost liquid-based processing, coupled with readily-available metal component elements, should lead to a promising future for these technologies. ORCID iDs Z Sekkat https:/ /orcid.org/0000-0002-2808-829X S D Deshmukh https://orcid.org/0000-0001-7101-7393 R Agrawal https:/ /orcid.org/0000-0002-6746-9829 P Arnou https:/ /orcid.org/0000-0001-7992-0632 D B Mitzi https:/ /orcid.org/0000-0001-5189-4612 P J Dale https://orcid.org/0000-0003-4821-8669