Optical and microstructural properties of MgF 2 UV coatings grown by ion beam sputtering process

The optical, mechanical, and microstructural properties of MgF2 single layers grown by ion beam sputtering have been investigated by spectrophotometric measurements, film stress characterization, x-ray photoelectron spectroscopy ~XPS!, x-ray diffraction, and transmission electron microscopy. The deposition conditions, using fluorine reactive gas or not, have been found to greatly influence the optical absorption and the stress of the films as well as their microstructure. The layers grown with fluorine compensation exhibit a regular columnar microstructure and an UV-optical absorption which can be very low, either as deposited or after thermal annealings at very low temperatures. On the contrary, layers grown without fluorine compensation exhibit a less regular microstructure and a high ultraviolet absorption which is particularly hard to cure. On the basis of calculations, it is shown that F centers are responsible for this absorption, whereas all the films were found to be stoichiometric, in the limit of the XPS sensitivity. On the basis of external data taken from literature, our experimental curves are analyzed, so we propose possible diffusion mechanisms which could explain the behaviors of the coatings. © 2000 American Vacuum Society. @S0734-2101~00!00406-1#


I. INTRODUCTION
Fluorides such as MgF 2 are typical materials to make optical coatings for laser applications in the ultraviolet ͑UV͒ spectral range. 1 The coatings, conventionally grown by evaporation, generally exhibit relatively low optical losses and a high packing density, provided that the deposition is performed at high temperature, namely in the range of 250-300 °C.Such deposition conditions lead, however, to rougher films, which can induce optical scattering.
In modern applications, such as semiconductor lithography, the trend to lower wavelengths requires improving the quality of the coatings towards smoother films, in particular.Thus, for many years, various attempts of improvement have been achieved using, for instance, ion assisted deposition techniques to get denser films at low deposition temperature. 2,3Nevertheless, this ion-induced densification was found to create a strong optical absorption in the films, which makes them properly unusable for deep ultraviolet ͑DUV͒ applications.As a more innovative growth technique, the ion beam sputtering ͑IBS͒ of MgF 2 target offers good perspectives to enhance the layer quality.The major advantage of the technique is to produce, at ambient temperature, coatings with higher packing density and thus reduced optical scattering. 4Up to recently, the major obstacle to the use of sputtering techniques for the deposition of MgF 2 coatings has resulted from the high optical absorption of the films.Whatever the sputtering techniques, either cathodic sputtering 5 or IBS, 6,7 the extinction coefficient ͑k͒ of the MgF 2 films at a wavelength ϭ400 nm and lower, never dropped below 10 Ϫ2 .A common explanation for this phenomenon generally is the nonideal stoichiometry of the films attributed to preferential sputtering of lighter atoms by energetic species impinging onto the growing film.Nevertheless, more recently and thanks to a new reactive deposition process, very low absorbing IBS MgF 2 films have been successfully produced with k coefficients in the range of 10 Ϫ6 . 8n this context, the present study deals with the characterization of IBS-grown MgF 2 coatings.It mainly aims at bringing new elements to improve the understanding of the origin of optical absorption in sputtered MgF 2 coatings.Therefore, the deposition conditions of IBS MgF 2 films have been intentionally varied widely.We present here the most significant results regarding the microstructural, mechanical, chemical, as well as optical characteristics of the films.

II. EXPERIMENTAL PROCEDURE A. Sample preparation
For this study, different samples were produced in various deposition conditions.For each run, three different substrates were used: UV-grade silica samples, Si͑111͒ 2 in.wafers, and specific bar-shaped Si͑111͒ substrates ͓(50ϫ5) mm 2 and 1 mm thick͔.The silica samples were dedicated to optical characterizations.The different silicon substrates assigned for the other characterizations were first covered with a thin alumina sputtered film to warrant a good adhesion of the MgF 2 layers.The film deposition was performed in a loadlocked IBS chamber equipped with two cryo pumps leading to a base pressure of 4ϫ10 Ϫ8 mbar.During deposition, the a͒ Electronic mail: quesnel@chartreuse.cea.frsubstrate temperature remained lower than 60 °C and the operating pressure was fixed at 1ϫ10 Ϫ4 mbar.The target-tosubstrate distance is equal to 50 cm.
The first deposition run ͑denoted R 0 ͒ was performed by simply sputtering a hot-pressed MgF 2 target with xenon ions accelerated at an energy of 900 eV.As we will see, these operating conditions led to the production of slightly brownish films which exhibit rather high optical losses.Such a phenomenon has been already observed in a previous study 4 we made on other fluoride films, namely IBS YF 3 films, and, on the basis of composition measurements, it was clearly attributed to fluorine deficiency in the films.That is why, for the present work, other MgF 2 samples were prepared using reactive deposition conditions which were gradually improved ͑runs R 1 to R 4 ͒.We based this improvement on two main ideas which consist of the necessity of: ͑i͒ bringing additional fluorine to the sputtered target and to the growing film, ͑ii͒ reducing the ion bombardment of the growing MgF 2 films.The reduction of ion bombardment is obtained by using xenon as sputtering gas instead of argon, limiting that way the flux of reflected ions by the target.The supply of additional fluorine was initially achieved by introducing CF 4 gas in the deposition chamber.For the present purpose and since DUV applications are concerned, diluted fluorine is now preferred in order to limit the carbon contamination.The carbon is indeed known to promote strong DUV optical losses. 9The diluted fluorine we used is a mixture of 10% F 2 in argon.During deposition, the flow rate of diluted fluorine was equal to 5 sccm, which corresponded to a partial pressure of fluorine P F2 of around 2ϫ10 Ϫ6 mbar.From run R 1 to R 4 the operating conditions were thus varied by decreasing the Xe ϩ ion flux arriving at the MgF 2 target relative to the F 2 concentration in the chamber.As a result, the film deposition rate ͑s͒ decreases from 0.35 to 0.25 Å/s.If we assume that the fluorination of the deposited films occurs mainly at the substrate surface, its efficiency would then depend on the flux ratio between the additional fluorine ͑introduced in the chamber͒ which diffuses towards the substrate (v F2 ) and the sputtered Mg and F atoms coming from the target (v Mg ϩ2v F ). Since v F2 depends on the fluorine partial pressure P F2 , and (v Mg ϩ2v F ) depends on the deposition rate s, the fluorination efficiency is directly related to the P F2 /s ratio.This ratio is not dependent on chamber geometry or pumping system and constitutes a relevant process parameter, usable by an experimenter, to define each deposition run ͑see Table I͒.

B. Optical characterizations
In order to evaluate the efficiency of the reactive deposition conditions, a systematic optical characterization was performed on each sample deposited on silica.The reflection ͑R͒ and transmission ͑T͒ spectra were recorded between 200 and 800 nm using a 9 Perkin-Elmer spectrophotometer.This spectrophotometer is equipped with a homemade accessory which enables us to measure the transmission and reflection of the sample at exactly the same irradiated area.As a result, the absolute measurement accuracy is less than Ϯ0.3%.From these measurements, the thickness and optical constants ͑n, k͒ of the films were calculated, in particular their extinction coefficient dispersion curves ͓k()͔ which are related to the optical losses in the films.For this purpose, we used a nonlinear regression technique and assumed a Cauchy law for the refractive index dispersion ͓n()͔.The extinction coefficient dispersion curve was calculated by fitting theory to the measured normalized losses ͑NL͒ of the coating given by the following relation: NL()ϭ͓(1 ϪR()ϪT()͔/T().

C. Physical characterizations
The physical film characterizations comprise mechanical stress measurements, crystallographic analyses, and transmission electron microscopy ͑TEM͒.The film stress measurements, performed just after deposition, were done on bar-shaped silicon substrates using an interferometric technique.The residual stress ͑͒ was deduced from the change of curvature radius of the samples using the Stoney equation where E s /(1Ϫv s ) is the biaxial modulus ͑229 GPa for the Si substrate͒, e s and e f are the substrate, and MgF 2 film thicknesses, respectively, and r 1 and r 2 are the radii of curvature of the substrate before and after MgF 2 deposition, respectively.
The crystalline structure was determined by x-ray diffraction ͑XRD͒ using grazing incidence of a Cu K␣ line in a Siemens D500 instrument.Concerning TEM, plane-view specimens were prepared by conventional mechanical grinding and dimpling up to 30 m followed by a final Ar ϩ ion bombardment on the substrate side at low angle of incidence ͑6°͒ and at 3 kV in a Gatan precision ion polishing system.For cross-section observations, samples were prepared using the same procedure on previously cleaved and face to face glued samples.The observations were achieved on Hitachi H-800 NA and Philips CM30 electron microscopes working at 200 and 300 kV, respectively.Due to the instability of fluoride layers under electron beam illumination, low dose observation conditions were used.

D. Composition analyses
The composition determination and the detection of contamination levels were carried out by x-ray photoelectron spectroscopy ͑XPS͒ with a Perkin-Elmer Phi 5500 instrument using monochromatic Al K␣ line.In order to test the composition homogeneity across the layer, depth profile spectra were obtained using an Ar ϩ bombardment at low energy ͑3 keV͒.High-resolution scans over the F 1s, Mg 2s, Si 2p, C 1s, O 1s, and Al 2p peaks were taken after each bombardment period with a step size of 0.25 eV.

III. RESULTS
The main characteristics of the different samples are summarized in Table I.As mentioned in Sec.II A, the different process conditions are described by the P F2 /s experimental parameter.This parameter increases from run R 0 to R 4 , which corresponds to an improvement of the fluorination conditions.The extinction coefficients of the films measured at ϭ351 nm ͑k at 351 nm values͒ are a good indication of their optical quality level.At the same time, whatever the deposition conditions, the films exhibit a low refractive index which remains very close to the one of bulk MgF 2 .As expected, from R 0 to R 4 , there is a drastic decrease of k.It is worth noting that for the R 4 sample, which is a particularly low absorbent ͑lowest k value͒, k was measured using a photothermal deflection technique, which is far more sensitive than spectrophotometry.As shown in Table I, the modification of the optical properties is accompanied by a change in the mechanical stresses.For nonreactive deposition conditions, the stress is slightly tensile, whereas from R 1 to R 4 , for reactive conditions, the stress becomes more and more compressive.The meaning of these stress values must be carefully analyzed.We know, indeed, that different processes contribute to the overall residual stress.The first component th of the stress comes from the thermal mismatch between substrate and film.If we assume that, whatever the sample, the deposition temperature is constant during the whole deposition run (T d ϳ60 °C), we have where E f and v f are Young and Poisson moduli of the MgF 2 film, respectively, ␣ Si and ␣ MgF 2 are the expansion coefficients of substrate and MgF 2 film, respectively.Taking E f ϭ137 GPa, v f ϭ0.3, ␣ Si ϭ2.5ϫ10 Ϫ6 K Ϫ1 , and ␣ MgF 2 ϭ13.7 ϫ10 Ϫ6 K Ϫ1 leads to an estimation of the thermal stress component of ϩ87 MPa.The other stress components are the intrinsic component ( i ), related to the deposition process itself, and the extrinsic stress component ( e ) due to the interaction of the film with its environment ͑in particular water vapor͒.In the case of our sputtered films, the films are dense (dϭ3.14 g cm 3 ) and free of water so that e can be neglected ( e Ϸ0).This assumption is backed up by the fact that after 40 days of aging in room air, the stress level remains perfectly unchanged.So we have Using this expression of the residual stress for the analysis of data from Table I leads to the conclusion that, in the nonreactive deposition conditions, the process does not induce any intrinsic stress except some low tensile thermal ones.On the contrary, for the reactive fluorine conditions, from R 1 to R 4 samples, the thermal contribution decreases in comparison with the compressive intrinsic stress component which becomes more and more dominant.
We must now examine the crystallographic structure of the films.The grazing incidence XRD spectra achieved on different samples are presented in Fig. 1.These spectra are characteristic of well-defined polycrystalline structures.As deduced from the positions of the peaks, the crystallographic nature of the samples corresponds to the tetragonal P4 2 /mnm phase of MgF 2 .Furthermore, the comparison of the diffraction peak intensities with theory values ͑structure factor͒, indicates first, that a slight ͗110͘ texture exists in the samples and second, that the less textured sample is sample R 0 .
In order to complete the structure analyses of the films, TEM observations have been performed.In Fig. 2 typical TEM plane-view observations made on four samples and their corresponding selected area diffraction ͑SAD͒ patterns are reproduced.The polycrystalline structure of the samples is clearly seen in these pictures.It is worth noting that on the pattern of Fig. 2͑a͒, the intense and isolated spots are due to the Si substrate.Statistical measurements of the medium grain size give the results reported in Table I.From sample R 4 to sample R 0 , the grain size increases slightly from 25 to 40 nm as the coating thickness does.On sample R 1 , due to the higher thickness of the layer, the ͗110͘ texture can be locally observed, as indicated by the intensity variation on ring ͑110͒ on the corresponding SAD pattern in Fig. 2͑c͒.This result is in agreement with the previous XRD analyses.Cross-section observations of the samples R 3 , R 1 , and R 0 are reproduced in Fig. 3.The samples exhibit a more or less well-developed columnar structure and an exceptional compactness.Sample R 3 exhibits a very regular structure.Sample R 1 tends to have conic columns.For the R 0 sample, the microstructure appears less regular: close to the alumina interface, it consists of small grains ͑globular growth mode͒ and it evolves to a columnar strongly V-shaped growth mode as the thickness of the layer increases.Finally, we also note increasing surface roughness from sample R 3 to R 0 .
The chemical composition of each sample has been analyzed.As a typical result, Fig. 4 displays the atomic concentration depth profile of elements for the R 3 sample as measured by XPS.The fluorine and magnesium contents are found to be very constant across the whole layer, indicating the very good homogeneity of the coatings and also the reliability of the measurement method.Indeed, any fluorine desorption which could be due to the Ar ϩ bombardment used for the analysis has not occurred during the measurement.Furthermore, Mg and F ratio is in good agreement with the MgF 2 bulk stoichiometry.Finally, the very low level of oxygen confirms the lack of water in the films.The other samples exhibit the same kind of composition depth profiles.In Table II, the average atomic concentrations of the films are summarized for all the samples, including also data on contaminant levels.As not expected, results indicate that, whatever the sample and deposition conditions, all the coatings exhibit a very good stoichiometry, within the experimental error ͑F/Mgϭ2.00Ϯ0.05͒.Moreover, very low levels of contamination in carbon and oxygen are measured in the layers, generally lower than 2% and 1%, respectively.We note, in particular, that sample R 3 exhibits slightly higher C and O concentration levels, due to a higher initial base pressure (ϳ10 Ϫ7 mbar) before deposition.
Finally, the optical properties of the samples have been investigated.We focused our work on the extinction coeffi-cient dispersion curves as presented in Fig. 5. Contrary to the compositional results, a very pronounced change in the optical properties is observed from one sample to another.From R 0 to R 4 , the absorption clearly decreases especially in the DUV part of the spectral region.Moreover, the absorbing coatings exhibit a characteristic absorption peak located at ϭ260 nm.This feature is typical of color centers as frequently reported in literature for MgF 2 crystals. 10In order to check the stability of the absorption observed on the different layers, some of them have been submitted to annealings.Both UV and thermal annealings have been performed.Figure 6 describes the evolution of the k spectrum of the R 0 sample after a first UV annealing during 2 h under a Hg lamp followed by different thermal annealings at different temperatures.After the first UV treatment, the main absorption peak tends to split into two different absorption peaks approximately centered at 225 and 300 nm.Then, subsequent thermal annealings promote the increase of the 300 nm centered peak at the expense of the 225 nm centered peak.Nevertheless, in the DUV spectral part, the overall absorption remains very high, even after annealings up to 350 °C.Decreasing the absorption significantly requires temperatures as high as 400 °C.On the contrary, the behavior of the other samples deposited in reactive deposition conditions is very different.As illustrated by sample R 2 ͑Fig.7͒, whatever the annealing mode, thermally or UV activated, a drastic decrease of the absorption is observed.The low temperature thermal treatment (Tϭ100 °C) makes the absorption decrease but its effect saturates.A subsequent short UV annealing has, however, a strong effect leading to the complete drop of the absorption band.That means that samples deposited in reactive conditions are easily curable.

IV. DISCUSSION
The first noticeable result of this study is that the properties of the films drastically change with the deposition conditions.This is particularly true for the mechanical and optical properties.Though this result is not surprising, this is, however, not the case as far as the chemical composition is concerned.There is, apparently, a contradiction between the strong evolution of the k-dispersion curves from one sample to another and the unchanged stoichiometry of the films.Moreover, the fact that all the films exhibit a good stoichiometry with F/Mg equal to 2 is quite unusual, since most of the investigations made by different authors on fluorides deposited by sputtering, state that the fluorine to metal ratio is lower than the ideal one with an absolute decrease of the ratio frequently in the range of Ϫ0.1 to Ϫ0.5.For instance, in a previous study we made on IBS YF 3 4 we found, under nonreactive deposition conditions, a F/Y ratio of 2.3 instead of 3. The same trend was observed by other authors on IBS MgF 2 coatings since they measured F to Mg ratio as low as 1.6. 6Finally, on MgF 2 films deposited by cathodic sputtering, 5 Martinu, Biedermann, and Holland reported F/Mg in the range of only 1, or even less.
It is then necessary to check our results by closely analyzing the experimental data, in particular the k() spectra.The specific shape of the k() curves suggests the presence of color centers in the films, which means fluorine vacancies, each one being occupied by one or more electrons.1][12] From this work, we know that many kinds of color centers can be generated  in MgF 2 crystals, the most frequent ones being the F center which induces an absorption peak centered at ϭ250-260 nm and the M center located at ϭ370 nm.In order to clearly elucidate the kind of centers which are responsible for the absorption we found in our film, we have tried to fit our experimental curves using the model developed, for instance, by Dexter. 13According to this model, it is possible to calculate the color center concentration from the ␣(E) curve, where ␣ is the absorption coefficient related to the extinction coefficient by the well known relation ␣ ϭ4k/ and E is the energy ͓E͑eV͒ϭ1.24/(m)͔.If we assume a Gaussian dispersion of ␣(E), the color center concentration N is given by the Smakula formula where n is the refractive index of the material at the centering wavelength of the absorption peak, f is the oscillator strength of the color center, ␣ 0 is the absorption coefficient at the peak, and U is the half-width of the band.Near the absorption peak and at its maximum, we take n equal to 1.41 and it is usual to assume f ϳ1. 14n the case of our IBS films, we fitted their absorption spectra ␣(E) with a Gaussian whose equation is ␣(E) ϭ␣ 0 EXP͓Ϫ(EϪE 0 ) 2 /W 0 2 ͔.W 0 is a fit parameter which is related to U by the relation Uϭ2(ln 2) 1/2 W 0 .The fit has been particularly optimized in the DUV and UV regions of the spectra.Whatever the sample, we found that it was possible to reasonably fit the absorption curve with only one Gaussian which indicates the contribution of only one kind of color center.Figure 8 gives some examples of experiment-theory comparison and Table III indicates the different fit parameters.The agreement between measurement and theory is rather good.We just note, in particular for R 0 , a slight discrepancy in the left side of the peak, near 6 eV.This could result from a displacement of the fundamental absorption edge towards lower energies due to disorder in the coating crystallographic structure as often observed in amorphous or small polycrystallized materials.It is known indeed that disorder can induce new electronic states in the electronic band gap, which forms so-called band tails, that way reducing the optical band gap.From Table III, we see that for as-deposited films the absorption peak is centered at an energy between 5.15 and 5.4 eV, which means around ϭ235 nm.Even if this value is slightly lower than the wavelength of 260 nm generally reported for the F center in MgF 2 crystals, there is no doubt that the absorption of our films is dominated by F-center absorption ͑one electron trapped on a fluorine site͒.In fact, as illustrated by Fig. 9, the centering wavelength of the absorption peak is found to shift, after annealing, to higher wavelengths, up to 253 nm ͑or 4.9 eV͒.The reason for this shift could be the decrease of the absorption near the band edge (EϾ6 eV) and the narrowing of the band tails in the energy band gap leading to a reduced distortion of the ␣ ͑͒ curve.Further optical investigations in the vacuum ultraviolet ͑VUV͒ spectral range down to ϭ120 nm would be required to check this assumption.Furthermore, as indicated in Table III, the half width of the absorption band ͑U͒ was found to be in the range 2.5-3.2eV.Compared to data reported by Sibley and Facey 10 for MgF 2 crystals with values of 0.806 eV at Tϭ300 K, our values are rather high.We have then a broadening of the absorption band related to the polycrystalline nature of our films which, moreover, tends to decrease with the reactive conditions when the film structure is less disordered.
As a consequence of the presence of F centers, there are fluorine vacancies in the films, whose concentration N V can be calculated using the Smakula's formula.Moreover, if we consider that these fluorine vacancies correspond to fluorine atoms missing in the material, it would normally induce a deviation from stoichiometry ⌬͑F/Mg͒ which can be calculated by the following expression: where N V and N Mg are the concentration of F vacancies and Mg atoms, respectively.We have N Mg ͑cm Ϫ3 ͒ϭNa/M , with being the mass density ͑3, 17 g cm Ϫ3 ), N a Avogadro number, and M the molar mass of MgF 2 .The results are reported in Table III.The data clearly evidence that the concentration of fluorine vacancies is low enough not to noticeably affect the F to Mg ratio since ⌬͑F/Mg͒ remains lower than 0.02.This value is far lower than the experimental error of Ϯ0.05 achieved by XPS, which means that such fluorine deficiency is not detectable by XPS.We can conclude that the experimental composition results are not in contradiction with the optical measurements.
The second noticeable result of this study concerns the typical behavior of the coatings after thermal or UV treatments.Two opposite behaviors have been found which can be summed up as follows: ͑i͒ a nonreversible optical absorption for the IBS films when deposited in nonreactive deposi-tion conditions; ͑ii͒ an easily curable absorption for samples deposited in reactive conditions.Moreover, in this second case, the temperature needed for a modification of the absorption is particularly low: 100 °C corresponding to an energy of 0.03 eV.Catlow, James, and Norgett 15 have reported on the energies of activation for different diffusion mechanisms in MgF 2 crystals.Their calculations state that the rutile crystallographic structure of MgF 2 , as a noncompact structure, offers diffusion paths which require low energy of activation.Thus, as far as fluorine vacancies are concerned, among the different available transitions, the easiest diffusion mechanism corresponds to the transition path denoted w 3 ͑as illustrated on Fig. 10͒ which requires an energy of activation of only 0.03 eV.The other mechanisms consume higher energies, 0.31 and 1.53 eV for w 1 and w 2 jumps, respectively.Our results seem then to be consistent with the w 3 low energy mechanism.Nevertheless, since after annealing we observe a clear decrease of the absorption band, that is to say, a decrease of the F-center concentration ͑see Table III͒, a recombination mechanism must occur in the material.Such a recombination which results from interactions between an electron, a F vacancy, and a fluorine atom, implies having some free fluorine in the material.This fluorine could come from F atoms positioned in interstitial sites and/or at grain boundaries.During the first low temperature annealings, the recombination at the grain boundaries by step by  step migration of F vacancies towards the grain surface is probably the dominant mechanism.Such a mechanism would be, indeed, less energy consuming since it does not require supplementary energy for F migration from interstitial to F sites.Nevertheless, the necessity of using UVactivated annealing afterwards to completely bleach the material pleads in favor of a second mechanism as recombinations between F color centers and interstitial F atoms.The Hg-UV lamp exhibits indeed a strong ray emitting at Eϳ5 eV, which corresponds to the energy required to suppress an anion-Frenkel-pair defect. 15Furthermore, the fact that the films deposited under reactive conditions exhibit compressive stresses tends to prove that some supplementary fluorine could lie in the material at wrong positions.The annealing of the films by replacing the F atoms in the right positions would induce a relaxation of the which has not been proved yet.Concerning now the R 0 sample, the various annealings only affected the shape of the absorption curve.Such a change looks quite similar to the photothermal conversion effects reported by Baryshnikov et al. 12 observed on annealed MgF 2 crystals which they had previously made fluorine deficient by electron beam irradiations.Among the different absorption bands which arose in the absorption spectrum after successive annealings, they identified a band at ϭ300 nm whose origin is unfortunately not clear.Our results are too limited to go deeper into the discussion, but the fact that the R 0 sample keeps its absorption high shows that no color center recombination seems to occur in the material.It means that there is either no free fluorine available or the structural disorder in the layer prevents the diffusion mechanisms from happening at low activation energy.Our TEM analyses have clearly evidenced that our deposition conditions, when reactive, promote a far more regular arrangement of the grains in the coatings.Moreover, a slightly preferential orientation of the grains with their ͑110͒ plans parallel to the coating surface is observed.As illustrated in Fig. 10, the activation energy of the F vacancies strongly depends on the crystallographic directions and is much higher along, for instance, the C axis ͑w 2 path͒.In a columnar structure where the column height corresponds to the thickness of the coating ͑200-300 nm͒ and the column width ranges from 25 to 40 nm, if we assume some recombinations to occur at the grain interface, it is clear that diffusion paths perpendicular to the columns would be more favorable.We suggest here that the preferential orientation of the columns observed in R 1 -R 4 samples, by aligning more or less the ''low energy'' w 3 path along the diameter of the columns, may help the recombination mechanisms to happen.

V. CONCLUSION
This study shows that the advanced IBS fluoride process we developed enables the production of very low UV absorbing MgF 2 films which exhibit a bulk-like low refractive index.Contrary to previous works, the fluorine deficiency, even in nonreactive deposition conditions, remains extremely low.Microstructure differences between the samples, depending on the deposition conditions, have been evidenced: the samples deposited in reactive conditions exhibit regular columnar microstructures ͑slightly ͗110͘ textured͒, whereas the sample deposited in nonreactive conditions exhibits a more irregular and randomly orientated microstructure.Moreover, the results show that the improvement of the optical properties is connected with a change of the mechanical stress in the films from tensile to more and more compressive.
The origin of the residual absorption has been identified as due to F centers in the material.On the basis of a simple color center model, the concentration of these centers has been calculated and found to be very low.Such a result emphasizes the particularly high sensitivity of MgF 2 films to fluorine vacancies, as far as optical absorption is concerned.Moreover, the reactive deposition conditions, by greatly improving the arrangement of the structure of the films, seem to make the optical recovery of the films easier provided that we assume free atomic fluorine to be available.On the contrary, the disturbed structure of films deposited in nonreactive deposition conditions by promoting the high stability of the strong optical absorption could explain the fact that these films are not curable.

FIG. 4 .
FIG. 4. Atomic concentration depth profiles of the sample R 3 as measured by XPS.

FIG. 7 .
FIG. 7. K-spectrum evolution of a reactively deposited sample after various post-treatments: example of sample R 2 .

FIG. 9 .
FIG.9.Evolution of the experimental absorption spectra and their corresponding Gaussian fitting curves ͑solid line͒ after various thermal annealings: example of sample R 2 .

TABLE I .
Typical characteristics of deposited samples.
a See definition in text.

TABLE II .
Average layer atomic concentrations ͑in percentage͒ of C, O, F, and Mg in the layer as measured by XPS for all the samples.

TABLE III .
Fit parameters for the absorption curves and deduced concentrations of F-vacancies and F/Mg ratio deviations.