In Situ Microgravimetric Study of Ion Exchanges in the Ternary Cu-In-S System Prepared by Atomic Layer Deposition

Reaction mechanisms during the growth of multinary compounds by atomic layer deposition can be complex, especially for sulfide materials. For instance, the deposition of copper indium disulfide (CuInS2) shows a non-direct correlation between the cycle ratio, the growth per cycle of each binary growth cycles, i.e., CuxS and In2S3, and the film composition. This evidences side reactions that compete with the direct Atomic Layer Deposition (ALD) growth reactions and makes the deposition of large films very challenging. To develop a robust upscalable recipe, it is essential to understand the chemical surface reactions. In this study, reaction mechanisms in the Cu-In-S ternary system were investigated in-situ by using a quartz crystal microbalance system to monitor mass variations. Pure binary indium sulfide (In2S3) and copper sulfide (CuxS) thin film depositions on Al2O3 substrate were first studied. Then, precursors were transported to react on CuxS and In2S3 substrates. In this paper, gas-phase ion exchanges are discussed based on the recorded mass variations. A cation exchange between the copper precursor and the In2S3 is highlighted, and a solution to reduce it by controlling the thickness deposited for each stack of binary materials during the CuInS2 deposition is finally proposed.


Introduction
Atomic Layer Deposition (ALD) is a technique based on self-limited surface chemical reactions, where the precursors are injected successively in a reactor under vacuum and heated at a relatively low temperature [1][2][3][4]. By exploiting its unique properties, conformal, pinhole-free, and precisely controlled (thickness, composition and interfaces) ultrathin binary, but also multinary films, can be synthesized. In the case of multinary compounds (ternary, quaternary materials, etc.), the most common ALD approach is the so-called "supercycle synthesis strategy" that consists of alternating binary ALD cycles [5][6][7]. Among its parameters, the cycle sequence (i.e., the order of the precursor doses) and the cycle ratio (i.e., the number of cycles of one binary process vs. the total number of cycles) are essential to obtain a homogeneous film. Added to these, nucleation delays, exchange mechanisms, growth rate, and temperature window differences complicate the deposition of such multinary compounds and make the control of the process parameters crucial [5][6][7]. The sulfide class contains a wide range of materials in which syntheses by ALD have been relatively less explored compared to others, such as oxide and elementary metals. Their interests are particularly driven by the energy-related fields [8]. In particular, Copper Indium Disulfide (CuInS 2 , CIS) is a suitable material to be used as a thin absorber in photovoltaic devices (direct bandgap ≈ 1.5 eV, absorption coefficient > 10 4 cm −1 for λ = 500 nm) [9,10], and CIS-type solar cells with homo- [11] or hetero- [12,13] junction architecture have been reported. CIS synthesis has been reported by various routes, such as thermal evaporation [11], co-evaporation [14], sputtering [15], electrodeposition [16], spray pyrolysis [17], spray ion layer gas reaction (spray-ILGAR) [18], atmospheric pressure spray chemical vapor deposition (APCVD) [19], and ALD [20,21]. The complexity of the ALD deposition process of CIS, as for sulfide multinary compounds in general, lies in the high propensity of the cation exchange mechanisms, the greater diffusion rates and the unintentional annealing of the labile sulfur [8,[21][22][23]. Besides a deep understanding and control of the chemical process, a rigorous cleaning of the various parts of the reactor must often also be carried out to obtain a homogeneous film [21,24]. To date, two teams have mastered these conditions and reported CIS homogenous deposition by employing quartz and small scale (5 × 5 cm 2 ) reactors [20,21,25].
In the present work, an upscale was attempted to deposit CIS in a larger reaction chamber (15 × 15 cm 2 ) using the supercycle strategy of Schneider, et al. [17]. A non-homogeneous film composed of copper sulfide (Cu x S) upstream, indium sulfide (In 2 S 3 ) downstream, and a Cu x In y S 2 film in the middle of the reaction chamber was obtained. The Cu x In y S 2 film area was inversely proportional to Cu and In concentrations in the flux direction from the inlet to the outlet. Variation of process parameters, such as the cycle ratio, the reactor temperature, or the Cu precursor pulse time, showed a shift in the position of the Cu x In y S 2 film area but failed to widen it. Thus, to tailor a working process on a large scale, understanding of the surface chemical reactions was essential. To investigate and control those phenomena, mechanism studies using a quartz crystal microbalance (QCM) system was performed to monitor thickness and mass variations. First, pure films of In 2 S 3 and Cu x S were grown on Al 2 O 3 . While In 2 S 3 had already been investigated with by in-situ QCM [26][27][28], the growth of ALD-Cu x S was investigated, to the best of our knowledge, for the first time with this set of precursors. Then, ion exchanges in the Cu-In-S ternary system were investigated by monitoring the growth of In 2 S 3 on Cu x S and of Cu x S on In 2 S 3 but also the behavior of the Cu precursor on In 2 S 3 . These have evidenced additional reaction paths that need to be controlled to pursue large scale CIS synthesis.

Atomic Layer Deposition
In 2 S 3 and Cu x S thin films were grown in a Beneq TFS-200 ALD reactor (15 × 15 cm 2 ; Beneq Oy, Espoo, Finland), at T dep = 180 • C for In 2 S 3 and T dep = 160 • C for Cu x S, respectively. High purity nitrogen 99.9999% (N 2 , Air Liquide, Paris, France) was used as carrier and purge gas. The carrier gas flow rate was set at 600 sccm. In the reaction chamber, the pressure was kept around 1 mbar. Indium(III) acetylacetonate (In(acac) 3 , 98%, Strem Chemicals, Bischheim, France), copper(II) acetylacetonate (Cu(acac) 2 , 98 + %, Strem Chemicals, Bischheim, France), and H 2 S (99.5%, Air Liquide, Paris, France) were used as In, Cu, and S sources, respectively. All chemicals were used as received, without further purification. A mass of 2 g of In(acac) 3 and Cu(acac) 2 was heated in two different hot solid sources (system Beneq HS300, Beneq Oy, Espoo, Finland) at T In(acac)3 = 150 • C and T Cu(acac)2 = 140 • C. To ensure proper mass transport of the metal precursors, "booster" and "combination" modes were chosen. In "booster" mode, N 2 carrier gas is injected in the precursor source before its pulse. "Combined" mode is composed of five steps ([t 1 /t 2 /t 3 /t 4 /t 5 ]). N 2 is first injected in the source (t 1 , as in "booster" mode), all valves are kept closed (t 2 ), precursor is pulsed (t 3 ), and N 2 is injected during the last part (t 4 ) and after (t 5  thick Al 2 O 3 passivation film was deposited on the reaction walls, prior to each experiment, and on the quartz crystal. It mitigates the cation exchange mechanisms [24,29], so that only the reactions between the quartz surface and the precursor(s) pulsed are revealed. Each experiment is referred in the text by the names given in the Table 1. In-situ QCM measurements were acquired with a Colnatec Eon-LT monitor system (Greenville, United States of America), using a HT quartz crystal covered by an alloy (6 MHz initial oscillation frequency, Neyco, Vanves, France) located downstream on the cover lid of the reactor. Signal was recorded every 0.2 s and the mass resolution set in order to have the lowest thickness step precision of 0.04 Å. Measurements were done after a long enough stabilization time to reach a uniform and constant temperature in the whole reaction chamber (∆T ± 1.5 • C). Mass variations were calculated from the thickness values by considering the average density of the films grown by ALD as determined by X-ray Reflectivity (XRR) (4.75 g/cm 3 for ALD-In 2 S 3 and 5.52 g/cm 3 for ALD-Cu x S). Impedance acoustic values were set at 0.27 (calculated from the shear modulus for a spray pyrolysis-In 2 S 3 film obtained in the literature [30]) and 0.82 (considering the covellite phase), respectively.

Thin Film Characterization
Film thicknesses and densities were determined from samples deposited on glass substrates by X-ray Reflectivity using a PANalytical Empyrean equipment (PANanalytical, Orsay, France), equipped with a PIXcel 3D detector and Cu anode providing Cu-Kα radiations. Fitting of the experimental curves were performed with X'Pert Reflectivity Software (PANanalytical, Orsay, France). The morphology of the In 2 S 3 samples was investigated with a field-emission scanning electron microscope (FESEM, Merlin VP Compact, ZEISS, Marly-Le-Roi, France). The electron beam acceleration voltage was set at 15 kV.

In-Situ Microgravimetric Study of In 2 S 3 Material (Experiment #1)
Growth mechanism of In 2 S 3 film was first studied and compared with the literature [26][27][28][31][32][33]. Figure 1a depicts the mass variation of In 2 S 3 deposited on an Al 2 O 3 substrate, monitored by the QCM during 900 cycles. The growth is non-linear as the estimated Growth Per Cycle (GPC) increases progressively until it reaches a constant value approximately equals to 0.32 ± 0.03 Å/cycle (determined by linear regression). This value is similar to what Genevée [31] (0.26 Å/cycle), Sterner [32] (0.30 Å/cycle), and Bugot [33] (0.24 Å/cycle) found using the same deposition temperature. The 400 first cycles, including the nucleation step, are characterized by a lower GPC, meaning that Al 2 O 3 substrate partly inhibits the growth of In 2 S 3 , as observed by Sarkar et al. [27]. Figure 1b shows the mass evolution during a regular growth cycle of In 2 S 3 , i.e., of In 2 S 3 on itself, averaged over 10 cycles. It is characterized by a mass gain during the In(acac) 3 pulse (+(43 ± 2) Materials 2020, 13, 645 4 of 15 ng.cm −2 ), a progressive mass loss during the purge (−(9 ± 1) ng.cm −2 ), and a mass loss during the H 2 S pulse (−(19 ± 1) ng.cm −2 ). These mechanisms correspond to the ligand exchange reactions, as explained elsewhere [27,28]. This description will serve as a reference for the study of ion exchanges with an ultrathin film of Cu x S. explained elsewhere [27,28]. This description will serve as a reference for the study of ion exchanges with an ultrathin film of CuxS.

In Situ Microgravimetric Study of CuxS Material (Experiment #2)
The growth mechanism of the second binary material in the Cu-In-S ternary system, i.e., CuxS, was then investigated. The mass variation over 200 {CuxS} deposition cycles is displayed in Figure  2a. The curve and thus the growth are non-linear, as the estimated GPC varies between low (0.1 Å/cycle) and relatively high values (0.60 Å/cycle). These variations all along the deposition may be partly explained by an instability of the flux, which has been observed during preliminary optimization efforts on Cu(acac)2 transport. Figure 2b presents the mass evolution during one growth cycle of CuxS on itself, averaged over 10 cycles (190th to 199th cycle). Estimated GPC equals 0.54 ± 0.03 Å/cycle. The mass slowly increases all along the pulse of Cu(acac)2 (+ 33 ng.cm −2 corresponding to + 0.6 Å for 10.5 s). The H2S pulse leads to a mass gain followed by a loss of similar value. Thus, this local variation has only a very small effect on the film mass gain.  Based on the theoretical surface chemical reactions that can be predicted and the mass (or thickness) variations obtained during the experimental reaction, it is possible to estimate which CuxS phase has more chance to be deposited. According to Schneider et al. [34], the CuxS synthesized by

In Situ Microgravimetric Study of Cu x S Material (Experiment #2)
The growth mechanism of the second binary material in the Cu-In-S ternary system, i.e., Cu x S, was then investigated. The mass variation over 200 {Cu x S} deposition cycles is displayed in Figure 2a. The curve and thus the growth are non-linear, as the estimated GPC varies between low (0.1 Å/cycle) and relatively high values (0.60 Å/cycle). These variations all along the deposition may be partly explained by an instability of the flux, which has been observed during preliminary optimization efforts on Cu(acac) 2 transport. Figure 2b presents the mass evolution during one growth cycle of Cu x S on itself, averaged over 10 cycles (190th to 199th cycle). Estimated GPC equals 0.54 ± 0.03 Å/cycle. The mass slowly increases all along the pulse of Cu(acac) 2 (+ 33 ng.cm −2 corresponding to + 0.6 Å for 10.5 s). The H 2 S pulse leads to a mass gain followed by a loss of similar value. Thus, this local variation has only a very small effect on the film mass gain. explained elsewhere [27,28]. This description will serve as a reference for the study of ion exchanges with an ultrathin film of CuxS.

In Situ Microgravimetric Study of CuxS Material (Experiment #2)
The growth mechanism of the second binary material in the Cu-In-S ternary system, i.e., CuxS, was then investigated. The mass variation over 200 {CuxS} deposition cycles is displayed in Figure  2a. The curve and thus the growth are non-linear, as the estimated GPC varies between low (0.1 Å/cycle) and relatively high values (0.60 Å/cycle). These variations all along the deposition may be partly explained by an instability of the flux, which has been observed during preliminary optimization efforts on Cu(acac)2 transport. Figure 2b presents the mass evolution during one growth cycle of CuxS on itself, averaged over 10 cycles (190th to 199th cycle). Estimated GPC equals 0.54 ± 0.03 Å/cycle. The mass slowly increases all along the pulse of Cu(acac)2 (+ 33 ng.cm −2 corresponding to + 0.6 Å for 10.5 s). The H2S pulse leads to a mass gain followed by a loss of similar value. Thus, this local variation has only a very small effect on the film mass gain. Based on the theoretical surface chemical reactions that can be predicted and the mass (or thickness) variations obtained during the experimental reaction, it is possible to estimate which CuxS phase has more chance to be deposited. According to Schneider et al. [34], the CuxS synthesized by Based on the theoretical surface chemical reactions that can be predicted and the mass (or thickness) variations obtained during the experimental reaction, it is possible to estimate which Cu x S phase has more chance to be deposited. According to Schneider et al. [34], the Cu x S synthesized by ALD with the precursors Cu(acac) 2 and H 2 S may present a multiphasic state, that complicates the determination of the reactions enabling its synthesis.
Ideal reaction equations and mechanisms of the stoichiometric phases CuS (covellite) and Cu 2 S (chalcocite or digenite) can be hypothesized. For CuS, the overall reaction would be: which can be decomposed in two half-reactions during each Cu(acac) 2 and H 2 S pulses (with " -" depicting a surface group): The ratio of the reaction (R s ) can then be calculated as follows: For Cu 2 S, even though the ALD synthesis of this material has been reported by several groups [24,[35][36][37], the detailed mechanisms is complex and, as far as we know, has never been investigated by QCM. The overall reaction may be: It is based on the assumption of the reduction of copper, from Cu 2+ to Cu + , leading to a Cu-rich phase (chalcocite or digenite) from a Cu-poor phase (covellite) [34], accompanied with the oxidation and subsequent evaporation of sulfur: Though this equation is slightly endothermic (∆G = 17 kJ/mol at 100 • C for x = 0 [38]), it is driven by the non-reversible evaporation of sulfur. The ratio of the reaction (R s ) can be calculated as follows: with k corresponding to the number of ligands desorbed during the Cu(acac) 2 pulse, k = [0; 4]. As a function of k, R s differs (see Table 2). The reaction mechanisms and the phase formations can be discussed in light of the experimental results. According to the theoretical growth mechanisms estimated for the CuS formation, the desorption of ligands during the H 2 S pulse step should be characterized by a thickness drop (as M(acac) > M(SH)), which is not monitored by the QCM. The small variation during this H 2 S pulse may rather be characteristic of the formation of a Cu-rich phase. Indeed, a ratio of reaction of Cu 2 S close to 1, corresponding to an absence of thickness variation, can be determined for a certain value of k. We can estimate at 3.29 the number of ligands desorbed during the Cu(acac) 2 pulse, while 0.71 ligands would be desorbed during the H 2 S pulse.
The growth mechanisms of Cu x S deposited on Al 2 O 3 monitored by the QCM seem to be characteristic of a Cu-rich phase formation. They will serve as a reference for the microgravimetric study on the Cu-In-S ternary system.

Growth of CuInS 2 Films (Experiments #3-5)
As described earlier, CIS thin film can be deposited by ALD by alternating cycles of In 2 S 3 [26] and Cu x S [34] binary materials (supercycle strategy, [CIS] = n 1 × {Cu x S} + n 2 × {In 2 S 3 }), using β-diketonates precursors and H 2 S, at a temperature as low as 150 • C [16]. CIS stoichiometry was obtained only from some cycle ratios (n 2 :n 1 > 5 at 150 • C), which demonstrated that there is a non-direct correlation between the cycle ratio and the film composition and evidenced side reactions that may compete with ALD traditional surface reactions. Indeed, as for other copper-containing sulfide materials [24,39,40], cation exchanges promoted by a rapid diffusion may occur during the CIS synthesis. This complexity has prevented a straightforward upscale of the CIS deposition in a large reaction chamber. To evidence cation exchanges in the Cu-In-S system and identify new paths to allow the deposition of large CIS films, growth of Cu x S and In 2 S 3 on the other binary sulfide was investigated.
3.3.1. Reaction Mechanisms of In 2 S 3 on a Cu x S Substrate (Experiment #3) Figure 3a shows the mass variation of an In 2 S 3 film grown on a Cu x S substrate. It increases all along the 450 cycles, but the gain progressively slows down, with estimated GPC values from 0.71 Å/cycle for the first cycles to 0.38 Å/cycle for the last cycles (averaged values). It is characteristic of a substrate-enhanced growth that occurs over a large thickness. During the last cycles, after 20 nm is deposited, the GPC (0.38 Å/cycle) gets closer to the regular GPC of In 2 S 3 (0.32 Å/cycle).
Materials 2020, 13, x; doi: FOR PEER REVIEW 6 of 14 diketonates precursors and H2S, at a temperature as low as 150 °C [16]. CIS stoichiometry was obtained only from some cycle ratios (n2:n1 > 5 at 150 °C), which demonstrated that there is a nondirect correlation between the cycle ratio and the film composition and evidenced side reactions that may compete with ALD traditional surface reactions. Indeed, as for other copper-containing sulfide materials [24,39,40], cation exchanges promoted by a rapid diffusion may occur during the CIS synthesis. This complexity has prevented a straightforward upscale of the CIS deposition in a large reaction chamber. To evidence cation exchanges in the Cu-In-S system and identify new paths to allow the deposition of large CIS films, growth of CuxS and In2S3 on the other binary sulfide was investigated.
3.3.1. Reaction Mechanisms of In2S3 on a CuxS Substrate (Experiment #3) Figure 3a shows the mass variation of an In2S3 film grown on a CuxS substrate. It increases all along the 450 cycles, but the gain progressively slows down, with estimated GPC values from 0.71 Å/cycle for the first cycles to 0.38 Å/cycle for the last cycles (averaged values). It is characteristic of a substrate-enhanced growth that occurs over a large thickness. During the last cycles, after 20 nm is deposited, the GPC (0.38 Å/cycle) gets closer to the regular GPC of In2S3 (0.32 Å/cycle).
The first cycles were studied, in particular, as ion exchanges are more likely to occur during them. Thus, Figure 3b depicts the five first cycles of the growth and evidences the mass variation during each pulse and purge. The first In(acac)3 pulse leads to a lower mass gain than the following pulses, meaning that the surface of the substrate is less reactive or has a poorer density of absorption sites than the surface of the growing film. The H2S pulse is notably characterized by the absence or a weak mass variation during these first cycles, while a clear drop can be seen during the H2S pulse when In2S3 grows on Al2O3 (see Figure 1b). Two reasons might explain this behavior. First, the surface reaction of H2S gas may be inhibited by the substrate or the In(acac)x-terminated surface fragments. Second, the conversion from a Cu-rich phase to Cu-poor phase or covellite phase by copper dismutation may occur and some of the released sulfur remain at the surface to react with In(acac)3 (hence reducing the following mass gain during H2S pulse). No significant mass (or thickness) change has been observed during one of the precursor pulse, such as when diethyl zinc reacts on Cu2S [24], or when bis(N,N-disec-butylacetamidinato)dicopper(I) (CuAMD) reacts on Sb2S3 [40]. Therefore, there is no evidence that indium does readily ion exchange with copper in a CuxS film at a deposition temperature of 180 °C.  To better understand the growth mechanism changes all along the deposition, first and last cycles are compared (Figure 4). Mass Gain Per Cycle (MGPC) is evidenced to vary and depends especially on the mass variation during the pulse and purge of H2S. Indeed, desorption during the H2S pulse steadily increases when the film grows, so that the MGPC decreases. As a consequence, the shape of the cycle is modified and finally appears to be similar to the cycle of growth of In2S3 on itself The first cycles were studied, in particular, as ion exchanges are more likely to occur during them. Thus, Figure 3b depicts the five first cycles of the growth and evidences the mass variation during each pulse and purge. The first In(acac) 3 pulse leads to a lower mass gain than the following pulses, meaning that the surface of the substrate is less reactive or has a poorer density of absorption sites than the surface of the growing film. The H 2 S pulse is notably characterized by the absence or a weak mass variation during these first cycles, while a clear drop can be seen during the H 2 S pulse when In 2 S 3 grows on Al 2 O 3 (see Figure 1b). Two reasons might explain this behavior. First, the surface reaction of H 2 S gas may be inhibited by the substrate or the In(acac) x -terminated surface fragments. Second, the conversion from a Cu-rich phase to Cu-poor phase or covellite phase by copper dismutation may occur and some of the released sulfur remain at the surface to react with In(acac) 3 (hence reducing the following mass gain during H 2 S pulse). No significant mass (or thickness) change has been observed during one of the precursor pulse, such as when diethyl zinc reacts on Cu 2 S [24], or when bis(N,N-disec-butylacetamidinato)dicopper(I) (CuAMD) reacts on Sb 2 S 3 [40]. Therefore, there is no evidence that indium does readily ion exchange with copper in a Cu x S film at a deposition temperature of 180 • C.
To better understand the growth mechanism changes all along the deposition, first and last cycles are compared (Figure 4). Mass Gain Per Cycle (MGPC) is evidenced to vary and depends especially on the mass variation during the pulse and purge of H 2 S. Indeed, desorption during the H 2 S pulse steadily increases when the film grows, so that the MGPC decreases. As a consequence, the shape of the cycle is modified and finally appears to be similar to the cycle of growth of In 2 S 3 on itself (see Figure 1b), which confirms that approximately 20 nm are necessary for the film not to be influenced by the substrate. (see Figure 1b), which confirms that approximately 20 nm are necessary for the film not to be influenced by the substrate.

Reaction Mechanisms of CuxS on a In2S3 Substrate (Experiments #4-#6)
The reaction mechanisms occurring at the surface of a In2S3 substrate exposed to Cu(acac)2 and H2S pulses have been studied. Figure 5 shows the mass variations observed on In2S3 substrates of different thicknesses during the first cycles of CuxS deposition. The curves are not superposed, which indicates that the exchange mechanisms differ as a function of the In2S3 thickness. For the thinner In2S3 substrates, the curves have non uniform variations and the mass decreases to a value lower than the initial one after several CuxS cycles, then the film grows faster. Indeed, on 23 and 50 Å-thick In2S3 substrates, while most of Cu(acac)2 pulses show a mass gain, some depict large mass losses, respectively, the 4th and from the 10th to the 12th pulses. When the In2S3 substrate is thicker (250 Å), the curves have uniform variations. Figure 5. Mass variations measured for three films of In2S3 exposed to successive pulses of Cu(acac)2 and H2S. The apparent short-time mass changes observed during the H2S pulses (estimated thickness of In2S3 = 50 and 250 Å) are probably induced by temperature variation [41] and are considered as artefacts.
Desorption phenomena have been described for multinary sulfide materials and are often linked to high cation mobility [8,42], coupled with fast diffusion of the cations as in ZnInxSy [28] or Cu(Zn,Sn)S2 (CZTS) [39]. Therefore, the large mass losses may be due to a gas-phase ion exchange between the Cu and In cations, from the precursor and the film, respectively. Heavy desorbed molecules (In(acac)3) would be replaced by lighter adsorbed molecules (Cu(acac)2), through an exchange mechanism that has to be determined. According to the shape of the curves in Figure 5, this process seems to slow down when the In2S3 film becomes thicker. As the mass loss does not occur

Reaction Mechanisms of Cu x S on a In 2 S 3 Substrate (Experiments #4-#6)
The reaction mechanisms occurring at the surface of a In 2 S 3 substrate exposed to Cu(acac) 2 and H 2 S pulses have been studied. Figure 5 shows the mass variations observed on In 2 S 3 substrates of different thicknesses during the first cycles of Cu x S deposition. The curves are not superposed, which indicates that the exchange mechanisms differ as a function of the In 2 S 3 thickness. For the thinner In 2 S 3 substrates, the curves have non uniform variations and the mass decreases to a value lower than the initial one after several Cu x S cycles, then the film grows faster. Indeed, on 23 and 50 Å-thick In 2 S 3 substrates, while most of Cu(acac) 2 pulses show a mass gain, some depict large mass losses, respectively, the 4th and from the 10th to the 12th pulses. When the In 2 S 3 substrate is thicker (250 Å), the curves have uniform variations. (see Figure 1b), which confirms that approximately 20 nm are necessary for the film not to be influenced by the substrate.

Reaction Mechanisms of CuxS on a In2S3 Substrate (Experiments #4-#6)
The reaction mechanisms occurring at the surface of a In2S3 substrate exposed to Cu(acac)2 and H2S pulses have been studied. Figure 5 shows the mass variations observed on In2S3 substrates of different thicknesses during the first cycles of CuxS deposition. The curves are not superposed, which indicates that the exchange mechanisms differ as a function of the In2S3 thickness. For the thinner In2S3 substrates, the curves have non uniform variations and the mass decreases to a value lower than the initial one after several CuxS cycles, then the film grows faster. Indeed, on 23 and 50 Å-thick In2S3 substrates, while most of Cu(acac)2 pulses show a mass gain, some depict large mass losses, respectively, the 4th and from the 10th to the 12th pulses. When the In2S3 substrate is thicker (250 Å), the curves have uniform variations. Figure 5. Mass variations measured for three films of In2S3 exposed to successive pulses of Cu(acac)2 and H2S. The apparent short-time mass changes observed during the H2S pulses (estimated thickness of In2S3 = 50 and 250 Å) are probably induced by temperature variation [41] and are considered as artefacts.
Desorption phenomena have been described for multinary sulfide materials and are often linked to high cation mobility [8,42], coupled with fast diffusion of the cations as in ZnInxSy [28] or Cu(Zn,Sn)S2 (CZTS) [39]. Therefore, the large mass losses may be due to a gas-phase ion exchange between the Cu and In cations, from the precursor and the film, respectively. Heavy desorbed molecules (In(acac)3) would be replaced by lighter adsorbed molecules (Cu(acac)2), through an Figure 5. Mass variations measured for three films of In 2 S 3 exposed to successive pulses of Cu(acac) 2 and H 2 S. The apparent short-time mass changes observed during the H 2 S pulses (estimated thickness of In 2 S 3 = 50 and 250 Å) are probably induced by temperature variation [41] and are considered as artefacts. Desorption phenomena have been described for multinary sulfide materials and are often linked to high cation mobility [8,42], coupled with fast diffusion of the cations as in ZnIn x S y [28] or Cu(Zn,Sn)S 2 (CZTS) [39]. Therefore, the large mass losses may be due to a gas-phase ion exchange between the Cu and In cations, from the precursor and the film, respectively. Heavy desorbed molecules (In(acac) 3 ) would be replaced by lighter adsorbed molecules (Cu(acac) 2 ), through an exchange mechanism that has to be determined. According to the shape of the curves in Figure 5, this process seems to slow down when the In 2 S 3 film becomes thicker. As the mass loss does not occur during the first pulse, the reaction is not instantaneous but requires several cycles to be activated. In addition, unlike when Cu x S is grown on Al 2 O 3 (Figure 2b), a mass drop occurs during most of the H 2 S pulses when Cu x S is grown on In 2 S 3 : first cycles for 23 or 50 Å-thick In 2 S 3 and all along the cycles shown in Figure 5. After the mass loss, the shape of the cycle is modified because the mass variation during the H 2 S pulse vanishes. This mass drop occurring during the H 2 S pulse may be a consequence of the exchange mechanism as In(acac) 3 molecules might be removed from the substrate as described by the reaction drawn in Figure 6. This hypothesis can be verified by considering the sign of the mass variations with the theoretical ones of Figure 6 during the first pulse, the reaction is not instantaneous but requires several cycles to be activated. In addition, unlike when CuxS is grown on Al2O3 (Figure 2b), a mass drop occurs during most of the H2S pulses when CuxS is grown on In2S3: first cycles for 23 or 50 Å-thick In2S3 and all along the cycles shown in Figure 5. After the mass loss, the shape of the cycle is modified because the mass variation during the H2S pulse vanishes. This mass drop occurring during the H2S pulse may be a consequence of the exchange mechanism as In(acac)3 molecules might be removed from the substrate as described by the reaction drawn in Figure 6. This hypothesis can be verified by considering the sign of the mass variations with the theoretical ones of Figure 6  . Exchange mechanism that would explain the mass loss during a H2S pulse.

Reaction Mechanisms of Cu(acac)2 on a In2S3 Substrate (Experiments #7 and #8)
The aim of these experiments was to determine if the Cu(acac)2 alone can activate the cationic exchange mechanism with the In atoms in the substrate. Compared to the {CuxS} program, the H2S pulse is replaced by a purge, leading to the program [Cu(acac)2]/N2 = [0.5/0.2/0.5/10/0.5]/10.5 s. Figure 7a presents the complex variations of the mass during fifty successive pulses of Cu(acac)2 on a 22 Åthick In2S3 film, with a zoom on the first 18 cycles in Figure 7b. A mass increase occurs during the first Cu(acac)2 pulse (+21.2 ng.cm −2 ), that may be ascribed to a surface adsorption of Cu(acac)x groups on In2S3 substrate (phase I, described by the reaction in Figure 8a). During the three following cycles (phase II), the mass progressively decreases, especially during the purge time. Then (phase III, cycle 4 and 5), a mass loss is clearly visible during the pulse of Cu(acac)2 as it suddenly drops (down to −24 ng.cm −2 ). Finally, (phase IV), a mass gain is observed while Cu(acac)2 is pulsed, and this gain is progressively reduced during the following pulses, whereas a mass loss during the purges appears.

Reaction Mechanisms of Cu(acac) 2 on a In 2 S 3 Substrate (Experiments #7 and #8)
The aim of these experiments was to determine if the Cu(acac) 2 alone can activate the cationic exchange mechanism with the In atoms in the substrate. Compared to the {Cu x S} program, the H 2 S pulse is replaced by a purge, leading to the program [Cu(acac) 2 ]/N 2 = [0.5/0.2/0.5/10/0.5]/10.5 s. Figure 7a presents the complex variations of the mass during fifty successive pulses of Cu(acac) 2 on a 22 Å-thick In 2 S 3 film, with a zoom on the first 18 cycles in Figure 7b. A mass increase occurs during the first Cu(acac) 2 pulse (+21.2 ng.cm −2 ), that may be ascribed to a surface adsorption of Cu(acac) x groups on In 2 S 3 substrate (phase I, described by the reaction in Figure 8a). During the three following cycles (phase II), the mass progressively decreases, especially during the purge time. Then (phase III, cycle 4 and 5), a mass loss is clearly visible during the pulse of Cu(acac) 2 as it suddenly drops (down to −24 ng.cm −2 ). Finally, (phase IV), a mass gain is observed while Cu(acac) 2 is pulsed, and this gain is progressively reduced during the following pulses, whereas a mass loss during the purges appears. Cu(acac)2 pulse (+21.2 ng.cm −2 ), that may be ascribed to a surface adsorption of Cu(acac)x groups on In2S3 substrate (phase I, described by the reaction in Figure 8a). During the three following cycles (phase II), the mass progressively decreases, especially during the purge time. Then (phase III, cycle 4 and 5), a mass loss is clearly visible during the pulse of Cu(acac)2 as it suddenly drops (down to −24 ng.cm −2 ). Finally, (phase IV), a mass gain is observed while Cu(acac)2 is pulsed, and this gain is progressively reduced during the following pulses, whereas a mass loss during the purges appears. As described earlier, a desorption of In(acac)3 via an exchange mechanism is highly probable during phase III, as presented in Figure 8c ). This would be characterized by a mass loss equals to 125 ng.cm −2 while the experimental mass loss is 24 ng.cm −2 (end of phase III). It corresponds to only 19% of the mass loss estimated if all the In from the substrate was etched. Thus, it can be estimated than only 4.2 Å over the 22 Å (19%) are impacted by the cation exchange mechanism during the six first cycles of the experiment. In general, the surface saturation is improved by successive pulses of a precursor. For instance, Muneshwar and Cadien [43] have showed that in the case of ZrN and HfO2 depositions, using precursors with a high steric hindrance (Zr(NMe2)4 and Hf(NMe2)4), GPC was increased, respectively, ~46% and ~49% by reproducing at least 3 and 6 successive pulses. However, even when considering a similar steric hindrance for Cu(acac)2, it is surprising in our case to observe such a large mass increase (≈243%) over as much as 43 successive pulses. Hence, this confirms that additional phenomena, such as Cu diffusion, In substitution, and In desorption, occur until the 50th Cu(acac)2 pulse. The desorption of In, in particular, can explain the mass loss during the purges and liberates S atoms, newly available to bond with the adsorbed copper molecules. As described earlier, a desorption of In(acac) 3 via an exchange mechanism is highly probable during phase III, as presented in Figure 8c, as it leads to a mass loss (3 × M(Cu(acac) 2 ) − 2 × M(In(acac) 3 ) < 0). Diffusion of Cu atoms along with a desorption of a small amount of In(acac) 3 may explain the behavior during phase II (Figure 8b).
It is possible to determine if the In 2 S 3 substrate is totally etched. The initial mass of the In 2 S 3 substrate being 1040 ng.cm −2 for 22 Å corresponds to a mass of 733 ng.cm −2 of In (m(In) = m(In 2 S 3 ) × 2M(In)/M(In 2 S 3 )), assuming no contaminants in the structure lattice. According to the Figure 8c process, as three Cu atoms replace two In atoms, a Cu mass equals to 647 ng.cm −2 would be necessary to replace all the In in the 22 Å-thick substrate (m(Cu) = [1.5 × m(In) × M(Cu)]/M(In)). This would be characterized by a mass loss equals to 125 ng.cm −2 while the experimental mass loss is 24 ng.cm −2 (end of phase III). It corresponds to only 19% of the mass loss estimated if all the In from the substrate was etched. Thus, it can be estimated than only 4.2 Å over the 22 Å (19%) are impacted by the cation exchange mechanism during the six first cycles of the experiment. In general, the surface saturation is improved by successive pulses of a precursor. For instance, Muneshwar and Cadien [43] have showed that in the case of ZrN and HfO 2 depositions, using precursors with a high steric hindrance (Zr(NMe 2 ) 4 and Hf(NMe 2 ) 4 ), GPC was increased, respectively,~46% and~49% by reproducing at least 3 and 6 successive pulses. However, even when considering a similar steric hindrance for Cu(acac) 2 , it is surprising in our case to observe such a large mass increase (≈243%) over as much as 43 successive pulses. Hence, this confirms that additional phenomena, such as Cu diffusion, In substitution, and In desorption, occur until the 50th Cu(acac) 2 pulse. The desorption of In, in particular, can explain the mass loss during the purges and liberates S atoms, newly available to bond with the adsorbed copper molecules.
To investigate the impact of the thickness on the growth, the same experiment has been conducted on a 250 Å-thick In 2 S 3 substrate (experiment #8), and both are plotted in Figure 9. It shows that no mass loss is observed in the case of 250-Å thick substrate. This evidences that the growth and exchange mechanisms differ as a function of the In 2 S 3 thickness, as represented Figure 10. This at first surprising observation can be explained by the impact of the In 2 S 3 film morphology. Indeed, the morphology of the surface (Figure 11) and the crystalline character (as determined by XRD under grazing incidence, not shown here) of the film has been demonstrated to be dependent of the thickness. For the 15 nm-thick film, a Volmer-Weber growth mechanism can be considered to explain the island covered looking surface. For the thicker film, grains seem to cover the entire surface. Hence, it is possible that the cation exchange mechanism between Cu and In occurs preferentially on In 2 S 3 islands that may have more reactive surface sites energetically favorable ( Figures 5 and 9). surprising observation can be explained by the impact of the In2S3 film morphology. Indeed, the morphology of the surface ( Figure 11) and the crystalline character (as determined by XRD under grazing incidence, not shown here) of the film has been demonstrated to be dependent of the thickness. For the 15 nm-thick film, a Volmer-Weber growth mechanism can be considered to explain the island covered looking surface. For the thicker film, grains seem to cover the entire surface. Hence, it is possible that the cation exchange mechanism between Cu and In occurs preferentially on In2S3 islands that may have more reactive surface sites energetically favorable (Figures 5 and 9).   surprising observation can be explained by the impact of the In2S3 film morphology. Indeed, the morphology of the surface ( Figure 11) and the crystalline character (as determined by XRD under grazing incidence, not shown here) of the film has been demonstrated to be dependent of the thickness. For the 15 nm-thick film, a Volmer-Weber growth mechanism can be considered to explain the island covered looking surface. For the thicker film, grains seem to cover the entire surface. Hence, it is possible that the cation exchange mechanism between Cu and In occurs preferentially on In2S3 islands that may have more reactive surface sites energetically favorable (Figures 5 and 9).     Finally, the reaction of Cu(acac) 2 alone or Cu(acac) 2 /H 2 S on a 23-Å thick In 2 S 3 substrate are compared in Figure 12. The slopes of the mass loss during the Cu(acac) 2 pulse and of the large mass gain that follows are similar whether H 2 S is used or not (see dashed lines on Figure 12). It suggests similar desorption and adsorption rates. The minimal mass values reached are also similar (−24 ng.cm −2 for Cu(acac) 2 , −18 ng.cm −2 for Cu(acac) 2 /H 2 S). In the latter case, about 14% of the In contained in the film is impacted by the exchange mechanism (3.2 Å over the 23 Å, for an initial mass = 1107 ng.cm −2 ) versus 19% for Cu(acac) 2 alone. This difference can be explained by an enrichment of the film in sulfur, while H 2 S is pulsed, characterized by a mass loss and could be written as follows, with || representing a surface group: The desorption of acetylacetone during the first cycles would deprive the indium in the film of acac ligands and thus reduce the desorption of In(acac) 3 during the fourth cycle. The H 2 S pulse seems to accelerate the cation exchange mechanism. Indeed, the mass loss phenomenon that occurs during a pulse of Cu(acac) 2 happens two cycles earlier when H 2 S is used. Finally, the reaction of Cu(acac)2 alone or Cu(acac)2/H2S on a 23-Å thick In2S3 substrate are compared in Figure 12. The slopes of the mass loss during the Cu(acac)2 pulse and of the large mass gain that follows are similar whether H2S is used or not (see dashed lines on Figure 12). It suggests similar desorption and adsorption rates. The minimal mass values reached are also similar (−24 ng.cm −2 for Cu(acac)2, −18 ng.cm −2 for Cu(acac)2/H2S). In the latter case, about 14% of the In contained in the film is impacted by the exchange mechanism (3.2 Å over the 23 Å, for an initial mass = 1107 ng.cm −2 ) versus 19% for Cu(acac)2 alone. This difference can be explained by an enrichment of the film in sulfur, while H2S is pulsed, characterized by a mass loss and could be written as follows, with || representing a surface group: || − Cu(acac) + H S (g) → || − Cu(acac) (SH) + Hacac (g).
The desorption of acetylacetone during the first cycles would deprive the indium in the film of acac ligands and thus reduce the desorption of In(acac)3 during the fourth cycle. The H2S pulse seems to accelerate the cation exchange mechanism. Indeed, the mass loss phenomenon that occurs during a pulse of Cu(acac)2 happens two cycles earlier when H2S is used. The nucleation behavior and especially the consequence of the first Cu(acac)2 exposure seems also to differ. To estimate the coverage of Cu(acac)2 species, the number of In2S3 surface sites can be approximated by using the In2S3 average density (4.75 g/cm 3 as determined by XRR, equivalent to ρ = 8.8 × 10 21 "In2S3 units"/cm 3 ), and assuming a square lattice, leading to ρ 2/3 = 4.3 × 10 14 "In2S3 units"/cm 2 .
The coverage of Cu(acac)2 can then be approximated based on the mass gain during the first exposure. In the case of the experiment using only Cu(acac)2 precursor, 21.2 ng/cm² is deposited that corresponds to 4.88 × 10 13 "Cu(acac)2 molecules"/cm². The normalized coverage of Cu(acac)2 species relative to "In2S3 units" on the surface is [4.88 × 10 13 "Cu(acac)2 molecules"/cm²]/[4.3 × 10 14 "In2S3 The nucleation behavior and especially the consequence of the first Cu(acac) 2 exposure seems also to differ. To estimate the coverage of Cu(acac) 2 species, the number of In 2 S 3 surface sites can be approximated by using the In 2 S 3 average density (4.75 g/cm 3 as determined by XRR, equivalent to ρ = 8.8 × 10 21 "In 2 S 3 units"/cm 3 ), and assuming a square lattice, leading to ρ 2/3 = 4.3 × 10 14 "In 2 S 3 units"/cm 2 .
The coverage of Cu(acac) 2 can then be approximated based on the mass gain during the first exposure. In the case of the experiment using only Cu(acac) 2 precursor, 21.2 ng/cm 2 is deposited that corresponds to 4.88 × 10 13 "Cu(acac) 2 molecules"/cm 2 . The normalized coverage of Cu(acac) 2 species relative to "In 2 S 3 units" on the surface is [4.88 × 10 13 "Cu(acac) 2 molecules"/cm 2 ]/[4.3 × 10 14 "In 2 S 3 units"/cm 2 ] = 0.11 "Cu(acac) 2 molecules"/"In 2 S 3 units", i.e., a coverage of 11%. In the case of alternated Cu(acac) 2 and H 2 S precursor, a mass gain of 12.9 g/cm 2 is obtained, that yields a coverage of 7%, based on similar assumptions. These coverage values are reasonable, as they are significantly smaller than the maximum coverage calculated to be 180% ([7.76 × 10 14 "Cu(acac) 2 molecules"/cm 2 ]/[4.3 × 10 14 "In 2 S 3 units"/cm 2 ]) according to the steric hindrance of the Cu(acac) 2 molecules (see Appendix A). As the same Cu(acac) 2 pulse time is used, the coverage is indeed similar, and the observed small variations may be due to slightly different In 2 S 3 initial surface state.

Conclusions
This study is a deep investigation of the reaction mechanism and ion exchanges occurring in the ternary Cu-In-S system prepared by ALD from metal acetylacetonate precursors and H 2 S. This was performed using an in-situ quartz crystal microbalance system. For this, growth of In 2 S 3 and Cu x S binary compounds on a Al 2 O 3 substrate were first studied. It strengthened the postulated mechanism for the ALD-Cu x S formation via reduction of Cu, oxidation and subsequent evaporation of elemental sulfur. Then, mass variations were monitored during the synthesis of In 2 S 3 on a Cu x S substrate, of Cu x S on a In 2 S 3 substrate, and pulsing Cu(acac) 2 on a In 2 S 3 substrate. Mass variations have evidenced gas-phase ion exchanges when Cu(acac) 2 reacts on an In 2 S 3 substrate while In 2 S 3 growth is enhanced by a Cu x S substrate. It was particularly highlighted by a large mass loss that occurs during one or two of the first Cu(acac) 2 pulses. It has also been noted that it may be possible to reduce the cation exchange by controlling precisely the In 2 S 3 thickness. Indeed, from a certain thickness value (and morphology), the reaction of Cu(acac) 2 on a In 2 S 3 substrate is even hindered, as in our case from 250 Å. Indeed, the crystalline character of In 2 S 3 is enhanced by the thickness, and it was shown to be more stable when Cu(acac) 2 reacts on its fully covered surface rather on growth islands. To go further, analytical tools, such as XPS (ex-situ or in-depth) or in-situ XRF, may be employed in order to determine precisely how deep the In 2 S 3 film is impacted by the exchange mechanism (in-depth In and Cu concentration) as a function of the number of cycles of Cu x S. In addition, Quadrupole Mass Spectrometry measurements could help to identify the reaction by-products.
Finally, beyond the fundamental interest in understanding the ever-challenging growth of ALD multinary compounds, it also aims at allowing the upscale of an ALD recipe. Indeed, it explains why the film synthesized during the CIS deposition in the Beneq TFS-200 is not homogeneous and does not contain any indium upstream, next to the gas inlet.
Based on these observations, a nanolaminate strategy that controls finely the In 2 S 3 thickness deposited between each Cu x S cycle, should rather be chosen to deposit homogeneous CIS film on large area. Another solution would be to substitute one of the metal precursor by a precursor without acac ligands as this may lead to different dynamics. Funding: This research was supported by the French Environment and Energy Management Agency (ADEME) via a Ph.D. grant to H. Le Tulzo and EDF and partly carried out in the framework of a project of IPVF, which has been supported by the French Government in the frame of the program of investment for the future (Programme d'Investissement d'Avenir ANR-IEED-002-01).