Quasi real-time Raman studies on the growth of Cu – In – S thin films

In this work annealing and growth of CuInS 2 thin films is investigated with quasireal-time in situ Raman spectroscopy. During the annealing a shift of the Raman A1 mode towards lower wave numbers with increasing temperature is observed. A linear temperature dependence of the phonon branch of22 cm/100 K is evaluated. The investigation of the growth process ~sulfurization of metallic precursors ! with high surface sensitivity reveals the occurrence of phases which are not detected with bulk sensitive methods. This allows a detailed insight in the formation of the CuInS 2 phases. Independent from stoichiometry and doping of the starting precursors the CuAu ordering of CuInS2 initially forms as the dominating ordering. The transformation of the CuAu ordering into the chalcopyrite one is, in contrast, strongly dependent on the precursor composition and requires high temperatures. ©2004 American Institute of Physics. @DOI: 10.1063/1.1667009 #


I. INTRODUCTION
Chalcopyrite semiconductor materials have been well established as absorber layers in high efficient solar cell devices. 1 Among these, CuInS 2 ͑CIS͒ is promising because of its bandgap being well suited to the solar spectrum.As the understanding of this material is far from being complete structural characterization is needed.Real-time x-ray diffraction ͑XRD͒ experiments have already been performed during the growth of the CuInS 2 films 2 which gave an insight into processes in the bulk.But an additional method clarifying the reactions taking place near the surface is as well desirable.Therefore, studies of the vibrational properties via in situ Raman spectroscopy appear promising.
One advantage of Raman spectroscopy is that the different orderings of CIS, such as the chalcopyrite ͑CH͒ and the CuAu ͑CA͒ ordering are observable. 3Furthermore, Raman spectroscopy is a contactless method which can be used for quality assessment. 4Previously it has been found that the presence of CuAu ordering in the CuInS 2 films gives rise to a deterioration of the solar cell performance. 5The presence of this ordering in the completed CuInS 2 layers, however, is dependent on the starting precursor composition. 6Therefore, it is important to investigate the growth of CIS in detail.
It is known that different Cu/In ratios in chalcopyrite films as well as the presence of doping elements have an influence on the structural and electronic properties. 7,8A dopant of primarily importance for chalcopyrite thin films in general is the element Na. 9 The role of Na, in particular for the growth process, is currently not understood.
In this study, we first performed annealing experiments of completed CuInS 2 layers with different compositions and determined the temperature dependence of the main phonon mode of CIS.In a second step, sulfurization experiments of Cu-In precursors have been investigated by quasireal-time Raman measurements.We show that the CIS formation starts with the CuAu ordering being the dominant one independent from the precursor composition.We are also able to follow the transformation of the CuAu ordering into chalcopyrite ordering which is dependent on the precursor starting composition.

II. EXPERIMENT
For the presented in situ Raman experiments a portable vacuum chamber has been constructed.A base pressure of about 1ϫ10 Ϫ6 mbar is provided.Heating of the substrates up to 500-550 °C results from thermal radiation of a graphite heater, while cooling down takes place without any tem-a͒ Author to whom correspondence should be addressed; electronic mail: rudigier@hmi.deperature regulation.For the temperature measurement a thermocouple is situated between the backside of the substrate and the heater.A Knudsen source provides sulfur vapor from elemental sulfur.The working temperature for the sulfur source is at 190 °C providing a sulfur partial pressure of around 1ϫ10 Ϫ4 mbar.The chamber is equipped with an especially designed optical window for excitation and detection of the Raman signal.A T64000 Jobin Yvon Raman setup has been used with an excitation line of 514.5 nm from an Ar ϩ laser.For this wavelength we calculated a penetration length of 150 nm in CuInS 2 . 10The Raman spectra were taken with a time resolution of 30 or 60 s.Fitting of the Raman lines was performed using single or multiple Lorentzians.
In situ Raman experiments were performed during sulfurization of metal precursors ͑growth experiments͒ and during annealing ͑annealing experiments͒ of presulfurized Cu-In-S layers.For the latter the Cu-rich presulfurized Cu-In-S layers had been previously etched in KCN to result in the stoichiometric composition.No additional sulfur was supplied during the annealing experiments.The formation of the precursors was accomplished by sequential dc magneton sputtering of Mo/Cu/In on soda-lime glass as described elsewhere. 11A NaF layer was introduced between Mo and Cu for some samples by evaporation.Different copper-toindium molar ratios ͑Cu-poor: 0.8, Cu-rich: 1.8͒ have been used, resulting in a final CuInS 2 film thickness of 2 m.For the annealing experiments the following temperature profile has been realized: heating up period of 30 min from room temperature up to 500 °C, followed by a holding period at this temperature of 10 min.For the growth experiments a 20 min ramp up to 550 °C was used, followed as well by a holding period of 10 min at top temperature.

III. RESULTS
In Fig. 1, a survey of Raman spectra from samples during different processing steps is presented.In the spectral range of 300 cm Ϫ1 we find two distinct peaks at around 290 and 305 cm Ϫ1 which can be assigned to the A 1 modes of the chalcopyrite and CuAu ordering, respectively. 12,13If this spectral range is fitted by a single line, different peak positions would be obtained depending on the prevailing ordering.Accordingly, an intermediate peak position indicates the two orderings occurring in one sample.Certainly, the shape of the peak is also affected by the presence of these two contributions.In some cases it is even possible to detect both peaks well separated, as shown in spectrum ͑b͒ of Fig. 1.In addition, peaks at 260, 340, and 470 cm Ϫ1 can be discerned for different growth experiments which we assign to the phases, CuS, CuIn 5 S 8 , and CuS, respectively.Some spectra also show a very weak contribution around 140 cm Ϫ1 ͓see Fig. 1͑e͔͒, which can be attributed to vibrational modes from the ␤-In 2 S 3 phase.with more noise.Remarkably, the CIS peak positions differ: starting from sample 1 ͑Cu rich͒ with the lowest Raman shift ͑289 cm Ϫ1 ͒, continuing with the Cu-poor sample 2 with sodium ͑297 cm Ϫ1 ͒, up to sample 3 ͑Cu poor without Na͒ at around 308 cm Ϫ1 .A fit of the data as depicted in Fig. 2 yields for the Cu-rich sample a temperature gradient of We compared the peak position at room temperature with the peak position at top temperature and calculated a relative thermal change in the vibrational frequency of Ϸ6.3 ϫ10 Ϫ5 K Ϫ1 .For the Cu-poor sample with sodium the relative temperature gradient is slightly higher ͑Ϸ1.9 cm Ϫ1 /100 K͒, while the Cu-poor sample without any doping shows a significant higher value of approximately 2.3 cm Ϫ1 /100 K.The variance of the peak positions as well as of the temperature gradients indicate the presence of a second phase besides the CH ordering for the Cu-poor prepared samples.For the Cu-rich experiment at top temperature a small shift towards higher wave numbers has been observed, which remains present in the cooling-down period.This may be due to a small temperature difference between the substrate and the thermocouple.A different explanation could be a real change in the lattice parameter.The temperature gradients and peak positions from the annealing experiments in the following are useful to correct the data obtained from the growth experiments.Not considered here is the increase in intensity of the Raman modes with increasing temperature.This increase is to be expected according to the Bose-Einstein statistic of phonons.Figure 3 shows the CIS peak positions versus the process temperature for the sulfurization of the precursors ͓Cu rich ͑Cu/Inϭ1.8͒͑1͒, Cu poor ͑Cu/Inϭ0.8͒with Na ͑2͒, Cu poor ͑Cu/Inϭ0.8͒without Na ͑3͔͒.The temperature of the initial formation of CIS can be judged from the first appear-ance of the A 1 mode.The course of the growth process is indicated by the arrows.Hereby, the peak positions have been corrected by the linear temperature dependence of the corresponding annealing experiments from Fig. 2. Having substracted the temperature dependence, constant peak positions are to be expected.For the Cu-rich experiment CIS formation starts with high wave numbers ͑around 305 cm Ϫ1 ͒ at approximately 165 °C.Increase of the temperature above Ϸ380 °C induces a decrease of the A 1 mode position ͑down to around 292 cm Ϫ1 ͒.During the holding and cooling down period no further significant change has been observed.For sample 2 ͑Cu poor with Na͒, a similar behavior is found: starting from 200 °C up to 400 °C almost constant values around 310 cm Ϫ1 for the CIS main Raman mode are observed, while for temperatures Ͼ400 °C a decline down to around 295 cm Ϫ1 occurs; during the remaining holding period and cooling down period the A 1 mode does not shift significantly.For the Cu-poor experiment without Na the position of the A 1 mode starting from its first appearance of around 180 °C remains at 308 cm Ϫ1 for both the heating and the cooling periods.The initial formation of CIS is detected at higher temperatures for the two Cu-poor samples, especially the sample containing Na.
Figures 4͑a͒ and 4͑b͒ give the areal intensity of several Raman lines as a function of the process time for a Cu-rich and a Cu-poor sulfurization, respectively.Here, the CIS signals are on the one hand fitted using a single Lorentzian and on the other hand split up into CH and CA ordering using two Lorentzians.For reference purposes, the substrate temperature profile is plotted in the header of these figures.In the heating up period of the Cu-rich experiment ͓Fig.4͑a͔͒ the 470 cm Ϫ1 line of CuS is detected at around 150 °C.At a slightly higher temperature ͑165 °C͒ signals belonging to CIS are found, while simultaneously the CuS phase vanishes.Up to 500 °C the CA ordering is highly dominating the CIS signal.Further increase of the temperature is accompanied by a decreasing CA signal until it vanishes as soon as the top temperature is reached.The peak belonging to the CH ordering can be observed for the whole remaining process, with a maximum in intensity at 550 °C.In the cooling down period again peaks of the CuS phase appear ͑at around 180 °C͒.The data in Fig. 4͑b͒ are given for a growth experiment of a Cu-poor precursor with no additional sodium.Here, the first observed phase in the heating up period is CuS ͑around 165-210 °C͒.Initial CIS formation takes place at 180 °C.Between 270 and 350 °C a weak peak at 140 cm Ϫ1 is found, which we assign to ␤-In 2 S 3 ͓cf.sample ͑d͒ in Fig. 1͔.Signals of both the CH and CA ordering can be found down to 200 °C in the cooling period.The CA ordering appears to be dominant.With the vanishing CIS signals the appearance of the In-rich phase CuIn 5 S 8 is indicated by the broad signal at around 340 cm Ϫ1 ͓cf.Fig. 1͑e͔͒.This phase remains as the only detectable one until the end of the process.
In principal the growth process of the Cu-poor sample with sodium is similar to the one of the Cu-poor sample without sodium.The main difference is that at the end of the process ␤-In 2 S 3 instead of the CuIn 5 S 8 phase occurs near the surface.

IV. DISCUSSION
The annealed samples differ in their absolute peak position of the A 1 mode if this mode is fitted by a single line.The value for the Cu-rich sample ͑289 cm Ϫ1 ͒ fits with the theoretical value for the A 1 mode of the CH ordering ͑294 cm Ϫ1 ͒, and even better agreement is found with experimentally determined values ͑290 cm Ϫ1 ͒ from other groups. 13,14For the Cu-poor CuInS 2 layer without Na the experimental value of 308 cm Ϫ1 is in good agreement with the position of the A 1 mode for the CA ordering of 305 cm Ϫ1 .This difference between Cu-rich and Cu-poor films has already been reported for coevaporated films. 15The determined peak position for the Cu-poor sample with Na lies between the values of CH and CA.Obviously, in this case the two orderings of CIS coexist with a slightly higher share of the chalcopyrite ordering.Therefore, we conclude that doping of CuInS 2 with sodium increases the amount of the CH ordering in Cu-poor films.
The annealing experiments show that an increase of temperature is combined with a shift of the peak position towards smaller wave numbers.In general the amplitudes of the lattice vibrations increase with temperature and the influence of the potential modified by the anharmonicity of the lattice on the atoms is becoming stronger.In the Gru ¨neisen approach the relative change in the phonon frequency, ␦(⌬ ˜q,j )/ ˜q,j , is assumed to be proportional to the relative change of the volume (␦(⌬ ˜q,j )/ ˜q,j ϭϪ␥(␦V/V)).This expression, where ␥ describes the Gru ¨neisen parameter, is independent from the absolute phonon frequency and the type of the mode.It follows that the thermal expansion coefficient ␣ can be expressed as with the compressibility, C V the specific heat capacity, and V the volume.With the definitions for and C V and the above expression for ␣, the temperature dependence of the phonon frequency is approximately characterized by Thus, a decrease of frequency with temperature will occur.
For most solids such behavior has been experimentally observed.The above equation yields a temperature dependence of the frequency shift for the A 1 mode of CIS of Here, the thermal linear expansion coefficient for CIS ͑CH ordering͒ in the temperature range Tϭ200-600 °C of ␣ ϭ1.0ϫ10 Ϫ5 K Ϫ1 , 16 and the Gru ¨neisen parameter for CuInSe 2 ͑␥ϭ1.3͒ 17have been used.The latter has been chosen because of a lack of data for CIS in the literature.As the thermal expansion coefficients for CuInS 2 and CuInSe 2 are nearly identical, the Gru ¨neisen parameter supposedly has a similar value.The so determined frequency shift is an approximation, because for the calculation no differences between the phonon branches have been considered.Furthermore, only the first term of the Taylor expansion for the peak shift with temperature has been taken into account.Terms with a higher order of the Taylor series describe phonon decay and four-phonon processes.Despite all these simplifications, a fair agreement between the theoretical and experimental temperature coefficient has been found for the CH ordering of the Cu-rich sample.Higher values of ‫⌬(ץ‬ ˜)/‫ץ‬T for the Cu-poor experiments ͑with and without Na͒ are due to the coexisting CA ordering.It is likely that the phonon frequencies of this phase show a different temperature dependence.Therefore, for the Cu-poor samples containing Na the experimental ‫⌬(ץ‬ ˜)/‫ץ‬T can be interpreted as a mixture of the two orderings.Supposedly, for the Cu-poor sample without Na the behavior is dominated by the CA ordering.Because of the relatively small expansion coefficient in the investigated temperature range it is valid to assume the frequency shift to be proportional to the temperature.This is indeed observed in the experiments.
In situ XRD measurements have previously been performed on the sulfurization of Cu-rich and Cu-poor precursor films. 18For both types of precursors the starting metallic phases are CuIn 2 and Cu. 19At around 140 °C CuIn 2 transforms into Cu 11 In 9 under the release of liquid In.From the intermetallic Cu 11 In 9 phase and from elemental sulfur CIS forms at approximately 200 °C in the heating period.In the Cu-rich experiment the In can react completely with the abundant elemental Cu.But due to the excess of Cu in this experiment a CuS phase appears in the cooling down period.
For the Cu-poor experiment an additional In-S compound was found as an intermediate phase.This phase may have formed from the liquid In released by the CuIn 2 →Cu 11 In 9 transformation.The further reaction to CuInS 2 consumes all elemental Cu now leaving no abundant Cu for the In to react.Thus In is sulfurized and is accommodated by the spinel phase CuIn 5 S 8 .The latter is only observed in Cu-poor films.Apparently, this spinel phase segregates during the cooling down period completely covering the two CIS orderings.From the XRD studies we know that CIS exists until the end, i.e., no complete transformation into this In-rich compound took place, and the assumption of CuIn 5 S 8 as a surface coverage seems to be likely. 20n contrast to the Raman data presented here no CuS phase has been observed with XRD during the heating up period.Therefore, it is likely that CuS forms near the surface with a relatively small volume percentage and can only be detected by Raman.From XRD sulfurization experiments of pure Cu layers it is known that the hexagonal CuS transforms into the cubic Cu 2 S. 14 However this phase appears to have a too low Raman scattering cross section as it is not observed in our experiments.Based on the phase diagram 21 the transformation CuS→Cu 2 S should appear at approximately 507 °C.But as indicated by the vanishing of the CuS phase, it is likely that here the transformation occurs at a lower temperature ͑Ϸ200 °C͒.For thin films such a reaction has already been observed to take place at lower temperatures, 22 and to be dependent on the sulfur partial pressure. 23ll growth experiments have in common that the CIS formation starts with the CA ordering being highly dominant, which partially transforms into CH.As the CIS formation proceeds at the very surface, i.e., the growth front is in the direction of the surface normal, 24 it may be argued that initially formed CA ordering at the end of the growth is covered by newly grown CH ordering, i.e., no transformation took place.But from Raman depth dependent experiments of finished layers it is known that for the Cu-rich case no CA ordering is left at all. 13 Therefore, we interpret the data in the sense of a phase transformation.
For the Cu-rich experiment the transformation CA→CH starts at around 380 °C and is completed with reaching the top temperature.The Cu-poor experiment with Na also shows the CA→CH transformation.But as can be seen at the peak position ͑297 cm Ϫ1 ͒ here the transformation takes place only partially.For the remaining process CH and CA ordering coexist.No such transformation ͑CA→CH͒ has been observed for the Cu-poor experiment without any sodium.Here, the CIS signals are clearly dominated by the CA ordering over the whole process.
Two questions arise: First, why is the CA ordering preferred to form at first as the dominating CIS ordering independent from the precursor composition.Second, which mechanism induces the transformation into CH.
The first point might be related to the temperature.The difference in formation energy between CH and CA ordering for CuInS 2 has been calculated to be small ͑Ϫ7.8 meV/four atom͒. 25The results presented here show, however, that the metastable CA ordering is energetically favored to form at low temperatures, while the CH formation requires higher energies.This has been found as well for coevaporated samples. 15For III-V semiconductor alloys it has been theoretically shown that CuPt ordering occurs due to surface reconstructions. 26The formation of the CA ordering may be promoted by similar effects.
In order to clarify the second question, we compare the different stoichiometries ͑Cu rich: Cu/Inϭ1.8 and Cu poor: Cu/Inϭ0.8͒.As soon as the temperature is high enough, the CH formation takes place as well as a recrystallization of the CA ordering.But for the recrystallization the presence of a Cu-S binary seems to be promotive: while the Cu-rich film undergoes the CA→CH transformation, no transformation here is observed for the undoped Cu-poor film.The preferred CH ordering in the presence of a Cu-S binary has also been shown for samples prepared via coevaporation.Therefore, a model has been developed, explaining the transformation by the solubility of the CA ordering in Cu 2 S at a certain temperature promoting the formation of the stable CH ordering. 27he influence of sodium on the growth of chalcopyrites is still discussed. 9,28,29But, as Na has a similar electronic configuration to copper, it appears possible that Na partially takes on the function of Cu.For instance, Na might imitate a Cu-rich environment by filling Cu vacancies or might form Na 2 S initiating the CA→CH transformation.However, with the top temperature chosen the transformation is not complete.So far conceivable possibilities for the completion of the transformation might be on the one hand higher temperatures or on the other hand an increased amount of sodium.Further experiments have to detail this point.Another effect of sodium is that both the beginning of the formation of the CuAu phase and of the transformation into the chalcopyrite require higher temperatures.This is in accordance with results obtained from calorimetric experiments on the selenide system. 30

V. CONCLUSIONS
We introduced Raman spectroscopy as a method for quasireal-time experiments on the chalcogenization of Cu-In layers.Annealing experiments show a shift towards lower wave numbers of the CIS main mode with increasing temperature.The temperature dependence of the peak shift has been calculated and an approximative agreement with the experimental value has been found.The growth experiments showed that the start of growth of CIS is dominated by the CA ordering.The CH formation takes place at elevated temperature and is promoted by the presence of Cu excess or additionally supplied sodium.

Figure 2
FIG. 1. Raman scattering intensities vs wave numbers for the annealing experiments ͑a͒ Cu-rich ͑Cu/Inϭ1.8͒sample, ͑b͒ Cu-poor ͑Cu/Inϭ0.8͒sample with sodium, and ͑c͒ Cu-poor ͑Cu/Inϭ0.8͒sample without sodium.Raman spectra for growth experiments: ͑d͒ Cu-poor sample without sodium in the heating up period, ͑e͒ Cu-poor sample without sodium in the cooling down period, and ͑f͒ Cu-rich sample in the cooling down period.

FIG. 3 .
FIG. 3. The Growth experiments for samples: ͑1͒ Cu rich ͑Cu/Inϭ1.8͒,͑2͒ Cu poor ͑Cu/Inϭ0.8͒with Na, and ͑3͒ Cu poor ͑Cu/Inϭ0.8͒without Na are presented as the peak shift of the CIS main mode vs the temperature.The CIS main mode obtained from a single Lorentzian fit is given as a function of the substrate temperature.The peak shift is corrected by the temperature dependence of the annealing experiments.The arrows indicate the process sequence and the round markers the beginning of the CIS formation.

FIG. 4 .
FIG. 4. ͑a͒ Temperature profile and areal intensity of characteristic Raman modes for the sulfurization process of Cu-rich ͑Cu/Inϭ1.8͒prepared samples vs process time.͑b͒ Temperature profile and areal intensity of characteristic Raman modes for the sulfurization process of Cu-poor ͑Cu/Inϭ0.8͒prepared samples without sodium vs process time.