Microstructure and secondary phases in coevaporated CuInS 2 films : Dependence on growth temperature and chemical composition

The microstructure of CuInS2 ~CIS2) polycrystalline films deposited onto Mo-coated glass has been analyzed by Raman scattering, Auger electron spectroscopy ~AES!, transmission electron microscopy, and x-ray diffraction techniques. Samples were obtained by a coevaporation procedure that allows different Cu-to-In composition ratios ~from Cu-rich to Cu-poor films!. Films were grown at different temperatures between 370 and 520 °C. The combination of micro-Raman and AES techniques onto Ar-sputtered samples has allowed us to identify the main secondary phases from Cu-poor films such as CuIn5S8 ~at the central region of the layer! and MoS2 ~at the CIS2 /Mo interface!. For Cu-rich films, secondary phases are CuS at the surface of as-grown layers and MoS2 at the CIS2 /Mo interface. The lower intensity of the MoS2 modes from the Raman spectra measured at these samples suggests excess Cu to inhibit MoS2 interface formation. Decreasing the temperature of deposition to 420 °C leads to an inhibition in observing these secondary phases. This inhibition is also accompanied by a significant broadening and blueshift of the main A1 Raman mode from CIS2 , as well as by an increase in the contribution of an additional mode at about 305 cm. The experimental data suggest that these effects are related to a decrease in structural quality of the CIS2 films obtained under low-temperature deposition conditions, which are likely connected to the inhibition in the measured spectra of secondary-phase vibrational modes. © 2001 American Vacuum Society. @DOI: 10.1116/1.1329123#


I. INTRODUCTION
CuInS 2 ͑CIS 2 ) is a chalcogenide semiconductor that has potential for high-efficiency solar cell devices.It is characterized by a band-gap value ͑1.5 eV͒ that is well adapted to terrestrial as well as space applications.Recently, a CIS 2 -based solar cell with a total area efficiency of 11.4% was reported. 1However, this is still beyond the efficiencies currently obtained in CuInSe 2 -based devices.Further improvement in the efficiency of the cells requires a better understanding of the microstructure of the polycrystalline CIS 2 layers and the microstructure's dependence on the technological processes.
One of the best-suited techniques for structurally assessing these layers is Raman scattering. 2 The Raman spectrum from CuInS 2 is characterized by the presence of a dominant A 1 mode, at about 290 cm Ϫ1 , which corresponds to the vibration of S anions in the X -Y crystallographic directions of the tetragonal cell. 3The position and width of this peak are very sensitive to structural features, such as disorder of the cation sublattice, stress, and structural defects.Also, films obtained under conditions leading to poor structural quality show an additional mode at about 305 cm Ϫ1 . 3,4The origin of this mode is still not clear.Its relative intensity in relation to that of the A 1 mode correlates with the increase in the shift and broadening of the A 1 peak.The comparison of these features with x-ray diffraction ͑XRD͒ and transmission electron microscopy ͑TEM͒ data show them to be clearly related to the overall structural quality of the films, which supports the assumption of a structural-related origin of the mode.
This work reports on the microstructural characterization of CIS 2 layers obtained by a coevaporation process at different temperatures and with different chemical compositions.For this, Raman scattering data have been correlated with the analysis of the films by Auger electron spectroscopy ͑AES͒, XRD, and TEM techniques.The combination of a Raman microprobe with a scanning AES nanoprobe system has allowed us to develop a detailed study of the films, in terms of structural quality, composition, and presence of secondary phases.

II. EXPERIMENTAL DETAILS
Polycrystalline films of CuInS 2 have been grown on Mocoated soda-lime glass substrates by coevaporation of Cu, In, and S. Due to the separation of the indium and copper sources, a layer with a lateral gradient composition is obtained, being Cu rich for about half of the layer and Cu poor for the other half.Four different samples have been grown at different temperatures ͑520, 470, 420, and 370 °C͒ to study the effect of this parameter.
At the Cu-rich region of the samples, segregation of Cu excess occurs at the surface of the layers in the form of CuS.This is removed by chemical etching in a 10% KCN solution.Etching of the whole CuS layer is observed by the disappearance of all CuS peaks from both surface Raman and XRD spectra.Films have an average thickness of 2.5 m, and the dimensions of the samples are 5 cmϫ0.8cm.
Micro-Raman scattering measurements were performed at room temperature with a Jobin-Yvon T64000 spectrometer coupled with an Olympus metallographic microscope, using the green line of an Ar ϩ laser ͑ϭ514.5 nm͒ as excitation light.For the objective used ͑X100, NAϭ0.95͒, the spot size on the sample is slightly submicronic.Spectra were obtained in a backscattering configuration, collecting light from all possible polarizations.In these conditions, the depth investigated by the Raman microprobe is determined by the optical absorption of light in CuInS 2 , with the absorption length being about 100 nm.To avoid thermal effects in the spectra-which are especially conspicuous for these measurements where a relatively high power density can be achieved at the scattering volume spectra-the excitation power of light on the sample has been kept below 1 mW.This corresponds to a power surface density below 0.29 ϫ10 6 W/cm 2 .All the measured spectra have been compared with those obtained at the same conditions on single-crystal stoichiometric CuInS 2 ͑reference sample͒.
Previous Raman scattering measurements performed at the surface of the samples show a dominant contribution of the 305 cm Ϫ1 mode at the Cu-poor region for all the samples. 3This is also accompanied by a significant blueshift and broadening of the A 1 mode.For the Cu-rich side, the spectra show a higher dependence on the temperature of deposition: For the samples deposited at the highest temperatures ͑520,470 °C͒, the spectra are similar to that from the reference CuInS 2 sample.However, for the samples deposited at the lowest temperatures ͑420,370 °C͒, the spectra show the presence of the additional 305 cm Ϫ1 mode, together with the corresponding blueshift and broadening of the A 1 mode.These data indicate the existence of two distinct behaviors, corresponding to high-temperature ͑520,470 °C͒ and low-temperature ͑420,370 °C͒ deposition conditions.Accordingly, the detailed analysis of the samples has focused mainly in the samples deposited at 520 and 370 °C.In the following discussion, these are referred to as the hightemperature and low-temperature cases, respectively.
For Auger measurements, a PHI 670 Scanning Auger Nanoprobe System model has been used.The electron-beam energy was 10 keV, with a current of 10 nA.The beam diameter was below 100 nm, and the scanned area in the samples was about 10 mϫ15 m.To quantify the chemical composition, AES measurements were performed on the single-crystal reference CuInS 2 , and these data were used as standards to define the corrective factors.For the ion sputtering of the sample, an Ar ϩ ion gun was used, with nominal values of energy and current of 2 keV and 50 A, respectively.In-depth Raman measurements were made by focusing the light spot in this region, after equivalent sputter steps of 15 min.These in-depth measurements were performed at equivalent points from both Cu-rich and Cu-poor regions, located at 0.5 cm from the respective edges.Raman scattering measurements performed on the sputtered surface from a single-crystal CuInS 2 sample have revealed the absence of significant damage effects related to the sputter process.
These data were correlated with the analysis of selected layers by XRD.Standard /2 XRD spectra were measured with a Siemens D-500 spectrometer in Bragg-Bentano configuration.Finally, the cross-section TEM analysis of the layers was performed in a Philips CM30 microscope operated at 300 keV and in a Phillips CM200 SuperTwin microscope, equipped with a field emission gun, operated at 200 keV, by observing specimens prepared according to the procedure described in Ref. 5.

A. In-depth Raman AES: Cu poor
Figure 1 shows the AES chemical composition profiles from the samples deposited at the highest ͑a͒ and lowest ͑b͒ temperatures.In both cases, a surface Cu-depleted region is observed, extending down to about 10%-20% of the layer.Below this region, the In content tends to be higher than the Cu, with the ͓In͔/͕͓In͔ϩ͓Cu͔͖ ratio being in the range of 0.50-0.55.The integral ͓In͔/͕͓In͔ϩ͓Cu͔͖ composition ratio in the bulk of the samples is 0.54 ͑high-temperature case͒ and 0.53 ͑low-temperature case͒.The deviations observed for the measurements performed at the deepest regions ͑close to the CIS 2 /Mo interface͒ are related to the high uncertainty in the measurements.This uncertainty is due to overlapping between the Mo and S signals, which explains the sharp increase in the intensity of the S signal.
The whole series of Raman spectra measured after each sputter step in both samples are shown in Figs.2͑a͒ and 2͑b͒, respectively.All these spectra are characterized by the additional mode at about 305 cm Ϫ1 , which in some cases becomes even higher than the CIS 2 A 1 mode.Also, the film deposited at the highest temperature shows the presence of additional modes at higher frequencies, which appear mainly at the central region of the layer.These modes can be fitted with three Lorentzians, centered at about 325, 340, and 360 cm Ϫ1 , respectively.These positions correspond to those reported for the three strongest vibrational modes of CuIn 5 S 8 , 6 which points out the presence of this secondary phase in the central region of the Cu-poor layer.This has also been corroborated by XRD measurements, which show the presence of randomly oriented CuIn 5 S 8 crystals in the layer.The spectrum measured at the deepest position also shows additional peaks at 382 and 408 cm Ϫ1 .These peaks will be discussed later.
Lowering of the temperature of deposition leads to a significant decrease in the spectral contribution from the CuIn 5 S 8 secondary phase ͓Fig.2͑b͔͒.This phase appears mainly at the lower half of the layer and tends to increase with depth.Also, the spectra measured from this sample are characterized by a higher intensity of the 305 cm Ϫ1 mode, which is the dominant one.Dominance of the 305 cm Ϫ1 mode is also accompanied by an increase in both blueshift and broadening of the A 1 mode from this layer in relation to that from the layer deposited at high temperature.This is shown in Figs.3͑a͒ and 3͑b͒, where the Raman shift and full width at half maximum ͑FWHM͒ of the A 1 mode are plotted versus sputtering time for both samples.This behavior indicates a clear decline in structural quality of the layer deposited at low temperature.

B. In-depth Raman AES: Cu rich
In comparison with Cu-poor films, the Cu-rich ones are characterized by a much stronger dependence of chemical composition on processing temperature, as seen in Figs.4͑a͒  and 4͑b͒.For the sample deposited at 520 °C ͓Fig.4͑a͔͒, the in-depth AES profile is similar to that from the corresponding Cu-poor sample ͓Fig.2͑a͔͒: At the surface, there is a thin Cu-depleted region, and below this region the In and Cu contents in the layer are very similar, the ͓In͔/͕͓Cu͔ϩ͓In͔͖ content ratio being close to stoichiometry ͑in the range of 0.47-0.50͒.The integral ͓In͔/͕͓In͔ϩ͓Cu͔͖ composition ratio in the bulk of this sample is 0.49.Decreasing the temperature of deposition to 370 °C leads to drastic changes ͓Fig.4͑b͔͒: In this case, the Cu cation content is higher than the In cation content at the whole layer, except for the first point ͑at the surface͒, due to the surface Cu-depleted region.This leads to a significant deviation from stoichiometry of the integral composition ratio of the layer ͓͑In͔/͕͓In͔ϩ͓Cu͔͖ϭ0.375͒.
The corresponding Raman spectra are shown in Figs.5͑a͒  and 5͑b͒.For the sample deposited at high temperature ͓Fig.5͑a͔͒, the spectra measured at different depths are very similar, always being characterized by a dominant A 1 contribution, and no significant contribution from any secondary phase is observed.Although some of the spectra show a certain contribution at the 300-360 cm Ϫ1 region, this is likely related to sputter-induced damage.Similar sputterinduced effects have been observed from the stoichiometric single-crystal CuInS 2 reference sample.
For the film deposited at low temperature ͓Fig.5͑b͔͒, the most striking feature in the spectra is the presence of the 305 cm Ϫ1 mode, which is detected in the majority of the spectra in the layer.Also, we emphasize the absence of any contribution from CuS vibrational modes in the spectra, in spite of the high Cu content in the layer.The appearance of the 305 cm Ϫ1 mode is also accompanied by a significant increase in the broadening of the A 1 CIS 2 mode in relation to that from the high-temperature case ͓Fig.6͑b͔͒.The position of the mode is also affected, and the mode becomes blueshifted, whereas for the sample grown at high temperature, a redshift is observed ͓Fig.6͑a͔͒, characteristic of layers deposited under Cu-excess conditions.This behavior agrees with that observed from the Cu-poor region of the samples and points out the existence of a general degradation of the structural quality of the CIS 2 layers when the temperature of deposition decreases.

C. CIS 2 ÕMo interface
As already indicated, the Raman spectra measured at depths close to the CIS 2 /Mo interface show, in some cases, the presence of additional modes at 382 and 408 cm Ϫ1 .To clarify the origin of these modes, these spectra were compared with that measured from a sulfurized Mo layer (MoS 2 ). Figure 7 shows the spectra measured from the MoS 2 layer, together with those from the interface region of the Cu-rich and Cu-poor samples deposited at high temperature.As shown in Fig. 7, the spectrum from the MoS 2 layer is characterized by two intense modes at the same position.This clearly indicates the formation of a MoS 2 secondary phase at this interface.This has also been corroborated by XRD measurements, which show peaks characteristic from this phase.In both cases ͑Cu poor, Cu rich͒, decreasing the temperature of deposition leads to the disappearance of these modes, which suggests that formation of the MoS 2 phase is also inhibited at low temperature.Also, the intensity of MoS 2 modes from the Cu-poor sample is higher than that from the Cu-rich one.
The presence of a MoS 2 interficial phase in polycrystalline CIS 2 layers on Mo has already been reported by Scheer and Lewerenz. 7This is similar to CuInSe 2 /Mo structures, where a MoSe 2 phase has been observed by high-resolution TEM. 8 MoSe 2 formation has been observed mainly in ͑In,Ga͒-rich layers from Cu͑InGa͒Se 2 solar cells, and it is not as clear in Cu-rich layers.This also agrees with the behavior reported in Fig. 7 and can be interpreted as being due to a low reactivity between the Mo film and the Cu-rich layers.These compounds are characterized by having a layered structure, where bonding between adjacent layers is done by van der Walls forces.This could lead to a decrease in the adherence of the CIS 2 layers to the Mo backcontact in the solar cell device.Poor CIS 2 adherence to the backcontact constitutes one of the main problems of CIS 2 -based solar cells.However, the role of MoS 2 in this case is not clear, adherence problems occur mainly for Cu-rich layers obtained following a sequential sputtering deposition process, 1,9 when Cu excess tends to inhibit the formation of this secondary phase.

D. Transmission electron microscopy
Figure 8 shows the cross-section images of the Cu-rich and Cu-poor samples deposited at 520 and 370 °C.The comparison between the four images clearly shows the structural differences from the Cu-poor and Cu-rich sides, as well as the influence of the deposition temperature: Cu-rich layers present a much larger grain size and roughness, whereas Cupoor areas show a smaller grain size, with a very smooth surface and a very high density of planar defects inside the grain.
The grain size in the Cu-rich side of the samples ranges from about 0.8Ϯ0.2m for the sample deposited at the lowest temperature of 370 °C up to about 1.6Ϯ0.5 m for the highest temperature, as can be seen in Figs.8͑a͒ and 8͑c͒.This behavior agrees with the structural degradation already suggested by both the spectral features of the A 1 CIS 2 mode  and the appearance of the 305 cm Ϫ1 mode at low temperature of deposition.Furthermore, the roughness of these samples is of the order of the grain size.
For the Cu-poor areas, the grains present a rather columnar shape, but their size is difficult to determine due to the extremely high density of defects which are at their inside.From Figs. 8͑b͒ and 8͑d͒, which correspond to the Cu-poor areas of both samples deposited at the highest and lowest temperatures, respectively, one would say that the grain size is slightly larger for the higher temperature.This can be seen in the images, where some grains, which are almost perfectly oriented along low-index zone axis, appear dark in the images.The size of one of these dark zones is larger for the higher temperature.Consistent with this result is the roughness, which is larger ͑about 150 nm͒ for the higher ture sample.These results are further confirmed by electron diffraction patterns, which are characterized by the presence of streaks, a clear indication of the existence of high densities of planar defects inside the grains, well aligned along determined crystallographic directions.In principle, these planar defects can contribute to the observed broadening of the A 1 Raman mode from Cu-poor samples, due to the decrease of the phonon lifetime induced by the defects.Defects can also induce a shift of the mode, due to phonon confinement effects.However, for this the average distance between defects which break the translational symmetry of the crystal and confine the phonons has to be of the order of few nm.Then, the observed blueshift of the A 1 mode from Cu-poor samples is more likely related to residual stress in the Cupoor CIS 2 lattice.
On the other hand, in these samples it has been impossible to identify the presence of the CuIn 5 S 8 additional phase observed by Raman scattering, probably a consequence of the small grain size.At the CIS 2 /Mo interface region, the presence of the MoS 2 phase has not been detected, which is probably due to its reduced thickness.Furthermore, energy dispersive x-ray measurements cannot help in this case, as the S K peak overlaps the Mo L one.
Previous observation of polycrystalline layers deposited by both coevaporation and sequential sputtering processes under Cu-excess conditions has revealed additional spots in the electron diffraction pattern that correspond to CuAu order.The coexistence of both chalcopyrite and CuAu ordering has been observed in epitaxial CuInS 2 films by Su and Wei. 10 This has been interpreted as related to the small difference in the energy of formation of both phases, which is 2 meV/atom, according to their calculation.The presence of CuAu ordering from Cu poor samples is not clear in this experiment, and has still to be clarified.

IV. DISCUSSION
The high Cu content in the Cu-rich layer grown at low temperature gives indirect evidence on the inhibition in the segregation of excess Cu at the surface of the layer in the form of CuS.Segregation of Cu excess at the surface in the form of CuS or Cu 2 S allows the formation of stoichiometric CIS 2 layers with improved crystallinity ͑higher grain sizes and lower defect densities͒. 7The inhibition in the formation of the CuS secondary phase at the surface of the film for low-temperature deposition conditions could be related to a strong decrease in the mobility of cation ͑Cu,In͒ atoms during growth of the films.Excess Cu atoms in the CIS 2 grains cannot diffuse toward grain boundaries, avoiding Cu segregation at the surface.However, accommodation of such a high Cu content in the CIS 2 grains is not clear, and the presence of a Cu-rich secondary phase in the layer cannot be excluded, although no experimental evidence has been obtained from Raman scattering and XRD measurements.
Decreasing the temperature of deposition also leads to a significant inhibition in the observation of other secondary phases in the Raman spectra, such as MoS 2 at the interface region from both Cu-rich and Cu-poor layers and CuIn 5 S 8 from Cu-poor layers.This is also accompanied by the presence of the 305 cm Ϫ1 mode in the Raman spectra, which correlates with an increase in both blueshift and broadening of the A 1 Raman line.The comparison between the Raman spectra and the AES data shows that the additional 305 cm Ϫ1 mode is not directly related to the In content.All this gives support to a structural-related origin of the mode.
However, the correlation between the intensity of the A 1 mode and the spectral features of the A 1 peak does not stand for the central region of the Cu-poor sample deposited at high temperature, where a significant CuIn 5 S 8 contribution is found.This can be seen in Fig. 9, where the relative intensity of the A 1 CIS 2 mode, the additional 305 cm Ϫ1 mode, and the CuIn 5 S 8 modes are plotted versus sputter time.In this case, the relative intensity of the 305 cm Ϫ1 additional mode decreases in the region where a maximum contribution of the CuIn 5 S 8 secondary phase is detected, which also corresponds to the region with maximum blueshift and broadening of the A 1 CIS 2 mode ͓Figs.3͑a͒ and 3͑b͔͒.This suggests that the shift and broadening of the A 1 mode in this heterogeneous region with both CIS 2 and CuIn 5 S 8 phases is due to additional features such as stress.
Similar data have been obtained by Winkler et al. 11 from the Raman scattering and structural analysis of Cu-In-Sbased layers.In this work, in-depth Raman spectra were measured from samples beveled by ion etching, and the observed blueshift of the A 1 mode in the region where the CuIn 5 S 8 phase is detected was attributed to built-in internal FIG. 9. Relative intensity of the mode A 1 , the mode at 305 cm Ϫ1 , and the CuIn 5 S 8 modes in the Raman spectra vs sputtering time from the Cu-poor sample deposited at 520 °C ͓the spectra are plotted in Fig. 2͑a͔͒.
stress.The origin of this internal stress has still to be clarified.As already indicated, XRD shows the existence of a random orientation of CuIn 5 S 8 crystals.For CuInS 2 , the XRD spectra show a higher intensity of ͑112͒ peaks, which indicates a partial preferential orientation of the CuInS 2 crystals from Cu-rich sample growth at high temperature.This suggests the absence of a coherent relationship between both phases, which are more likely separated by well-defined grain boundaries without any preferential relative orientation of both CIS 2 and CuIn 5 S 8 crystals.According to Winkler et al., 11 compressive stress in the CIS 2 grains could originate at the CIS 2 /CuIn 5 S 8 interfaces, related to different thermal properties of both materials during cool-down of the film.This would also be accompanied by a corresponding tensile stress in the CuIn 5 S 8 phase.From Fig. 3͑a͒, a stress-induced shift in the layers is in the range of 2.5 cm Ϫ1 , which corresponds to an average compressive stress level of about 0.5 GPa.Then, the decrease in the intensity of the 305 cm Ϫ1 mode could be related to an enhancement of lattice ordering in the CIS 2 phase, likely being the excess of cations in this region accommodated in the CuIn 5 S 8 phase.
In relation to the additional mode in the Raman spectra at 305 cm Ϫ1 , this could be related to changes in the arrangement of Cu and In atoms in the cation sublattice.In principle, CuInS 2 can be arranged in three different crystalline phases, chalcopyrite being the most stable phase at 0 K.At high temperatures ͑above 1200 K͒, the most thermodynamically stable phase is sphalerite, where Cu and In atoms are randomly distributed in the tetragonal cation sublattice. 12At intermediate temperatures, the CuAu cationic ordering becomes metastable.
To clarify the possible correlation of the 305 cm Ϫ1 mode with the sphalerite phase, XRD measurements have been carried out in the samples.The distinction between the chalcopyrite and the sphalerite ͑zinc-blende-like͒ structures by XRD is compromised by the similarity of the spectra.In principle, there are two main differences between them.First, reflections allowed for the chalcopyrite structure corresponding to (hkl) planes with k even and (hl) odd, or with h even and (kl) odd, are not allowed for the zinc-blende structure.The most intense reflections are the following: The other characteristic difference is the splitting of some of the peaks associated with equivalent planes in the sphalerite structure, which correspond to different atomic planes in the chalcopyrite structure.The most intense splittings are ͑ 200͒ sphalerite ⇒͓͑004͒ϩ͑200͔͒ chalcopyrite , at 32.2°/32.4°,͑ 220͒ sphalerite ⇒͓͑220͒ϩ͑204͔͒ chalcopyrite , at 46.4°/46.5°,͑ 311͒ sphalerite ⇒͓͑312͒ϩ͑116͔͒ chalcopyrite , at 54.8°/55.1°.
Figures 10͑a͒ and 10͑b͒ show the XRD spectra from the Cu-poor samples deposited at high temperature and low temperature, respectively.The first spectrum is characterized by the presence of the different peaks related to the chalcopyrite phase, and in principle, no evidence of the sphalerite phase is given.In the second case, all reflections forbidden for the sphalerite structure are absent in the spectrum, and splitting of the modes characteristic of the chalcopyrite phase does not seem to occur.This suggests the presence of the sphalerite phase at the Cu-poor region from the layer deposited at low temperature.However, the differences between both spectra could also be related to the poorer structural quality of the CIS 2 layer growth at low temperature, due to the decrease in the intensity of the XRD peaks and the increase in their FWHM.In any case, the absence of any evidence related to the sphalerite phase in the Cu-poor sample deposited at high temperature does not support the assignment of the 305 cm Ϫ1 mode in the Raman spectra with this phase.This agrees with recent results on Raman polarization measurements performed in epitaxial CuInS 2 layers, which show both the CIS 2 A 1 mode and the additional 305 cm Ϫ1 mode to be totally symmetric. 13This suggests this mode to be related to CuAu ordering or to local arrangements of atoms giving rise to a local vibrational mode.The presence of CuAu ordering in the polycrystalline layers is still not clear, and further measurements are in progress to clarify this issue.
Finally, we have to remark that coevaporation has been chosen as a preparation method for this work because of the possibility to grow samples with a lateral gradient in composition.We note that Raman spectra as well as cell characteristics are quite similar when a sequential preparation process 9 is used.The best cells have an active area efficiency in the range of 12%-13%. 14,15There is a clear correlation between the results presented here and the cell performance: Cu-poor films, due to their high defect density even if grown at high substrate temperatures, are heavily compensated and semi-insulating.Without additional efforts ͑doping, flux assisted growth 16 ͒, only films grown under Cu excess are suitable for solar cells.In the latter case, the minimum substrate temperature that yields good devices is on the order of 500 °C, i.e., in the same temperature range where we found complete segregation of the Cu excess to the surface and where the 305 cm Ϫ1 additional mode disappears from the Raman spectrum.Significantly higher substrate temperatures are not considered because they would not be compatible with the soda lime glass substrates.

V. CONCLUSIONS
The detailed microstructural characterization of CuInS 2 polycrystalline films by combining in depth micro-Raman scattering and Auger electron spectroscopy measurements as a function of the chemical composition and temperature of processing has allowed us to identify the main secondary phases in the layers as CuIn 5 S 8 for Cu-poor samples and CuS for Cu-rich ones.CuIn 5 S 8 mainly appears at the central region of the layers, whereas CuS segregates at the surface.Moreover, in both cases a MoS 2 secondary phase is also observed at the CIS 2 /Mo interface region, although the comparison between the Raman spectra suggests a lower occurrence of this phase for Cu-rich layers.
The presence of these secondary phases is strongly related to the temperature of processing: these secondary phases are inhibited when the growth temperature decreases to 420 °C.This is also accompanied by a significant degradation of the structural CIS 2 features, as reflected by the increase in both shift and broadening of the A 1 CIS 2 mode in the spectra, and by the decrease of the grain size estimated by cross-section TEM.Also, Raman spectra measured from samples grown at lower temperatures are characterized by the presence of an additional mode at about 305 cm Ϫ1 , which gives additional support to its previously proposed structural origin.The complementary analysis of the layers by TEM and XRD suggests that the sphalerite structure is not directly related to this additional mode, which could be more likely related to either CuAu ordering or to local arrangements of cations ͑maybe rich in In atoms͒ vibrating around S anions.

FIG. 2 .
FIG.2.Raman spectra from sputtered Cu-poor samples deposited at ͑a͒ 520 and ͑b͒ 370 °C.The spectra are normalized to the intensity of the dominant mode.The spectrum at the bottom corresponds to the surface of the layers, and spectra measured after consecutive sputter steps are shifted vertically.

FIG. 6 .
FIG. 6. Position ͑a͒ and FWHM ͑b͒ of the A 1 Raman peak vs sputtering time from the sputtered Cu-rich etched samples deposited at 520 and 370 °C.Dotted lines indicate the spectral features of the A 1 mode as measured from the reference sample.

FIG. 8 .
FIG. 8. TEM cross-section images of the samples deposited at 520 °C ͑a͒ Cu rich and ͑b͒ Cu poor and at 370 °C ͑c͒ Cu rich and ͑d͒ Cu poor.

FIG. 10 .
FIG. 10.XRD spectra from Cu poor etched samples deposited at ͑a͒ 520 and ͑b͒ 370 °C.The position of peaks corresponding to sphalerite ͑zinc blende͒, chalcopyrite (CIS 2 ), and Mo structures are indicated in the spectra.