Electron energy loss spectroscopy determination of Ti oxidation state at the (001) LaAlO3/SrTiO3 interface as a function of LaAlO3 growth conditions
Jean-Luc Maurice, Gervasi Herranz, Christian Colliex, Isabelle Devos, Cécile Carrétéro, Agnès Barthelemy, Karim Bouzehouane, Stéphane Fusil, Dominique Imhoff, Éric Jacquet, François Jomard, Dominique Ballutaud, Mario Basletic
aa r X i v : . [ c ond - m a t . m t r l - s c i ] D ec epl draft Electron energy loss spectroscopy determination of Ti oxidationstate at the (001) LaAlO /SrTiO interface as a function ofLaAlO growth conditions J.-L. Maurice , G. Herranz , C. Colliex , I. Devos , C. Carr´et´ero , A. Barth´el´emy , K.Bouzehouane , S. Fusil , D. Imhoff , ´E. Jacquet , F. Jomard , D. Ballutaud and M. Basletic Unit´e mixte de physique CNRS/Thales and Universit´e Paris-Sud-XI, Route d´epartementale 128, 91767 Palaiseaucedex, France Laboratoire de physique des solides CNRS UMR8502, Universit´e Paris-Sud-11, 91405 Orsay cedex, France Institut d’´electronique, de micro´electronique et de nanotechnologie, Unit´e Mixte de Recherche CNRS 8520, avenuePoincar´e BP 69, 59652 Villeneuve d’Ascq cedex, France GEMaC-UMR 8635, CNRS, 1 place Aristide Briand, 92195 Meudon cedex, France
PACS – Electron energy loss spectroscopy
PACS – Electron states at surfaces and interfaces
PACS – Electrical properties of specific thin films: insulators
Abstract. - At the (001) interface between the two band-insulators LaAlO and SrTiO , ahigh-mobility electron gas may appear, which has been the object of numerous works over thelast four years. Its origin is a subject of debate between the interface polarity and unintendeddoping. Here we use electron energy loss ’spectrum images’, recorded in cross-section in a scanningtransmission electron microscope, to analyse the Ti ratio, characteristic of extra electrons. Wefind an interface concentration of Ti that depends on growth conditions. Introduction. –
In a pioneering work, Ohtomo andHwang [1] have discovered that growing LaAlO (LAO)by pulsed laser deposition (PLD) onto a TiO -terminated(001) surface of SrTiO (STO), produces a metallic sys-tem, while both materials are insulators with respectiveband gaps of 5.6 eV and 3.2 eV [2]. This observation hasbeen the signal for many studies of this system: amongremarkable results are the superconductivity observed be-low 200 mK [3], the magnetism around 300 mK [4], andthe field effect on conductivity [5].In the ionic limit, the perfect interface carries a globalpositive charge, because the last TiO plane of STO isneutral and faces a first plane of LAO that has composi-tion LaO and carries an uncompensated charge +1/2 persurface unit cell. Such a configuration would lead to aso-called ’polar catastrophe’ [6] if it were not balanced byelectron or ion rearrangements. In the present case, thecapacity of Ti ions to bear a mixed valence may providethe necessary screening, some Ti (valence in bulk STO)becoming Ti . In this most simple scheme – essentiallyconfirmed by calculations [16–18] – 1 / ∼ × cm − ( ∼ ∼ × cm − (5%,the present detection limit, see below) at only 1 nm intothe bulk [12].Several authors have associated the n-type conductiv-ity they have measured with such screening electrons[1, 5, 7]. However, growth by PLD is a complex pro-cess, which will hardly create ’perfect’ materials [8, 9].Thus this scheme of interpretation, based on the inter-face being atomically sharp and chemically stoichiomet-ric, is the object of intense discussion. In our labora-tory, we have inferred from Shubnikov-De-Haas oscilla-tions [10] and cross-section conducting-tip atomic-forcemicroscopy (CT-AFM) [11], that, in the samples grown atlow-pressure, which exhibited the highest mobilities, theelectron gas was 3-dimensional and extended hundreds of µ m into the bulk of the substrate. Moreover, Siemons et p-1.-L. Maurice et al.al. [12], by making measurements before and after oxy-gen annealing, and Kalabukhov et al. [13], by measuringcathodo- and photoluminescence, concluded that the dop-ing was due to oxygen vacancies that had diffused duringthe PLD process. In reality, pulsed laser deposition pro-ceeds with species (atoms, ions and clusters) that have akinetic energy upon landing which depends, among othergrowth parameters, on the pressure in the chamber: inour case, typically several hundreds of eV in the 10 − -10 − Pa pressure range, and decreasing at higher pres-sure [14]. The point defect content of the growing layerthus directly depends on the growth oxygen pressure, notmerely because vacuum creates vacancies, but mainly be-cause landing species have enough energy to create irra-diation damage. This in turn, makes the growing layer apressure-dependent reservoir of vacancies for exodiffusionfrom the substrate. This phenomenon of substrate reduc-tion is however well known in molecular beam epitaxy,where it is utilised for the oxidation of epitaxial layers [15].If bulk substrate properties thus depend on the growthpressure of the epitaxial layer, the question remains aboutthe interface role on conductivity. With our CT-AFM ex-periments [11], we have shown that the interface regiondid exhibit a specific, enhanced, conductivity, but – andthis is a key point – with a low electron mobility. Correl-atively, we also found that this conductivity depended ongrowth conditions [11]. Low-mobility and dependence ongrowth conditions would be both the logical result of anextrinsic source of dopant. It is thus mandatory to charac-terise the doping level at the interface itself, as a functionof growth conditions. This need is further highlighted bya recent result based on X-ray diffraction [9] that showsthe existence of a several unit-cell thick layer of metallicLa − x Sr x TiO at the interface.Our high-resolution transmission electron microscopy(HRTEM) observations in cross section have shown theexistence of a distortion at the interface [19], which couldbe equally consistent with the delta-doping [17, 18] or theLa − x Sr x TiO hypothesis [9]. But they also exhibitednoisy contrasts [18] that would finally back up the lat-ter. Measurements of the Ti ratio were carried outby electron energy loss spectroscopy (EELS) in the scan-ning transmission electron microscope (STEM), by DavidMuller’s group [3, 20] and ours [18]. The LAO/STO inter-faces analysed were in all these cases prepared at relativelylow pressures; Ti was detected at the interface, over awidth reaching 5 nm in ref. [20] and closer to the nm inrefs. [3, 18], thus depending, very likely, on the details ofsample growth and preparation.Here, we bring new STEM-EELS data, obtained thistime from a sample grown at 40 Pa (0.4 mbar, ’high pres-sure’), which was macroscopically insulating, and from anew reference grown at 10 − Pa (10 − mbar, ’low pres-sure’), macroscopically conductive. Our results evidence,for the first time to our knowledge, the role of growthconditions on the microscopic interface atomic structure:oxygen pressure in the growth chamber tends to decrease the apparent interface doping. The latter is thereforenot only due the intrinsic polarity, but also to extrinsicpoint defects. We have additionally performed a com-bined HRTEM, electron-diffraction, and atomic force mi-croscopy study of the two samples, which shows a fun-damental change in relaxation of the epitaxial strain asthe growth pressure increases. At low pressure, relaxationtakes place without dislocations in a way preserving a 2-dimensionnal layer by layer growth, most probably withthe aid of point defects. At high pressure, relaxation oc-curs through a 3-dimensionnal growth mode. Methods. –
LAO thin films were grown by pulsedlaser deposition (PLD) using a frequency-tripled ( λ =355nm) Nd: yttrium aluminum garnet (YAG) laser onTiO -terminated STO substrates [Surfacenet Gmbh] withgrowth oxygen pressures PO = 10 − Pa and PO = 40Pa, for the low- and high-pressure specimens, respectively,at a deposition temperature of T = 750 ◦ C. The laser pulserate was fixed at 2.5 Hz, whereas the energy density was2.8 Jcm − . The target-substrate distance was 55 mm, re-sulting in a growth rate of about 1 ˚As − . Once the filmdeposition was finished, the samples were cooled down toroom temperature at the growth oxygen pressure (10 − and 40 Pa, respectively). The transport properties of thelow-pressure sample are among the best published to date(mobility > cm V − s − at 4K, see details in ref. [10]).In that specimen, the LAO thickness was 20 nm. Theother sample was insulating, the LAO in this case was 5-nm thick. This choice was made so that point-defect dif-fusion processes would dominate in the low-pressure sam-ple [10, 11] and be negligible in the high-pressure case.The LAO/STO cross-sectional samples for TEM andSTEM-EELS were prepared using standard tripod polish-ing, without water for the last ∼ µ m on each face. It wascompleted by a prolonged, grazing incidence, ion millingwith 2-keV Ar ions, from the backside (STO) only. Thisprocedure resulted, as shown by the profiles below (seefig. 5), in thin foils with fairly parallel faces. This point isvery important as the surfaces of foils are strong sourcesof Ti . We otherwise minimised the influence of sur-faces by selecting relatively thick regions ( ∼ n × m arrays of p -channel spec-tra [21]) acquired by scanning a focused probe step bystep and recording a spectrum at each step. We typicallyrecorded images consisting in 2 to 8 64-pixel lines perpen-dicular to the interface, with a pixel size of 0.33 nm to0.74 nm and a channel size of 0.2 eV. The dwell time var-ied between 0.2 and 1 s. In order to increase signal/noisep-2ELS determination of Ti oxidation state at the LaAlO / SrTiO interface Fig. 1: Evolution of the 3 d electronic structure of Ti as a func-tion of valence and surroundings. Degeneracy in the case ofTi in bulk STO is partially lifted by the octahedral crystalfield (middle), separating 3 d into t g and e g levels. The twocorresponding peaks are well visible in EELS Ti- L and Ti- L edges (see A, B in fig. 4) [22, 24]. In the case of Ti at theinterface (right), both valence and site-symmetry are different,which organises the 3 d levels in quite a different way [18]. Thelevels presented correspond to molecular calculations, an ad-ditionnal spreading will occur in the actual crystal. This willtransform the two peaks into a broad one in the EELS edges. ratio, the profiles finally used, such as those presented be-low, were obtained by summing, parallel to the interface,the 2 to 8 lines. Before this, the drift, measured as theshift of the interface from one line to the next, was cor-rected on pixel position and size. The effective size of theinteraction volume was calculated by deconvoluting themeasured Ti concentration profiles, assuming a step func-tion for the actual profiles. The intensity distribution ofthe probe thus defined was quasi-gaussian with a width athalf maximum of ∼ ratio was carried out usingthe high sensitivity on this ratio of the fine structure ofTi- L , edge [22–24]. Indeed, this edge corresponds totransitions from 2 p discrete levels to the 3 d band, whichis empty for Ti in bulk STO, and contains one half anelectron per site in the simplest model of Ti / at theinterface. Fig. 1 compares the Ti-3 d molecular electronicstructure in the case of the octahedral crystal field of STOand that of the distorted site [18] at the abrupt LAO-STOinterface. The move of energy levels, when going frombulk STO to the interface, will appear in spectra as a finestructure smoothing.More specifically, both the Ti- L , and Ti- L edges inSTO are made of two well defined peaks corresponding tothe e g and t g levels separated by the octahedral crystalfield (see fig. 4 and ref. [22]): a shift a valence towards 3+will show as a progressive fading of these peaks and anincrease of signal in the valley in between (see arrows infig. 4). It is not possible, of course, to separate geometryfrom valence contributions in such an evolution. However,as the two are intrinsically related through Jahn-Teller Fig. 2: Atomic force micrograph of the high-pressure sampleexhibiting 3-D features. In the area shown, the largest peak-to-valley depth reaches 6 nm, i.e. the film thickness. type effects, in the analysis presented below, we have con-sidered that all changes of line shape were associated witha valence change.We further assumed that the global Ti- L , cross sec-tion did not depend on the oxidation state of Ti, so thatall experimental spectra be linear combinations of pureTi and Ti spectra. We performed least squares fitsof the experimental Ti- L , edges with reference Ti andTi spectra. The fits were performed after backgroundsubtraction, and after normalising the integral of the re-maining signal. The reference Ti spectrum was pickedfrom the substrate contribution in the same profile wherethe analysed spectrum came from and the Ti spectrumwas taken in ref. [23].In such measurements, sources of error are numerous[25]. In order to increase the signal/noise ratio, we also ap-plied the least-squares analyses to Ti- L , edges obtainedafter summing as many comparable interface spectra aspossible. The summed edges totalised more than 2.5 × counts (background subtracted), in both the cases oflow and high growth pressure (see fig. 4). Even in suchconditions, the error in our measurements, taken as thestandard deviation between data points from the experi-mental and fitting spectra, remained of the order of 3 %,which made in turn our detection limit (3 %: ∼ × cm − ). Point defects in LAO. –
We show in the presentsection that the two samples have undergone different re-laxation processes of the epitaxial strain, which we asso-ciate with different point-defect contents. We focus in thefollowing on the high pressure sample, where growth ap-pears to be quite different from that usually observed inthe LAO/STO system.Figure 2 shows an atomic force micrograph of this sam-ple where the surface appears significantly rough, contraryto what is generally observed with low pressure growth.We associate this roughness with 3-dimensionnal growth.The strucural analysis of this high-pressure sample byTEM in < > cross section (fig. 3) indicates that somep-3.-L. Maurice et al. Fig. 3: HRTEM (a) and corresponding selected area electrondiffraction pattern (b) of the high pressure sample. As can beseen with the lateral shift of 302 spot, the in-plane parameteris different in film and substrate, which indicates plastic relax-ation, a process that does not occur in the low-pressure sample(not shown, see e.g. ref. [19]). The arrows in (a) indicate val-leys between relaxed 3D regions. plastic relaxation has occurred. Selected area electrondiffraction (fig. 3b) shows that the distorted pseudo-cubicparameters a (in-the-plane) and c (out-of-plane) are bothmodified. Taking the internal reference given by STO( a ST O = 0.3905 nm), the lattice pseudo-cubic parame-ters of this sample come out at a LAO = 0.3873 nm and c LAO = 0.3748 nm (+/- 0.001 nm). Correlatively withthe in-plane mismatch, we found some misfit dislocations,but not enough to allow for all the difference measured.Plastic relaxation thus appears to be associated with the 3-dimensional growth mode. When compared to the equilib-rium pseudo-cubic lattice parameter of LAO (0.3792 nm),the parameters found indicate a volume extension, whichmay be attributed to an elastic distortion with a Poissonratio of about 0.22.In similar TEM experiments carried out on the low-pressure sample (not shown), the LAO lattice parameterscame out at a LAO = 0.3905 and and c LAO = 0.3818. Thelow-pressure film thus appeared to be fully strained to thesubstrate, with a relatively small elastic relaxation out-of-plane as has been noticed previously [3, 19]. These valuesindicate a unit cell volume much larger than in the highpressure case, which is consistent with the presence of amuch larger amount of point defects. And indeed, in theobservation of the < > cross section of a sample grownin identical conditions (not shown), we have evidenced theexistence of a superstructure which could not be simulatedusing the exact equilibrium rhombohedral LaAlO unitcell, and had to be due to ordering of large amounts ofpoint defects.Therefore, the two samples would strongly differ inpoint-defect content, consistently with their growth con-ditions: the low-pressure sample having been hit by fastspecies during growth would contain a much higher levelof point defects than the high-pressure one. Fig. 4: Electron energy loss Ti-L , edge recorded at the inter-face in both the low-(a) and high-(b) pressure samples, com-pared in each case with a bulk spectrum recorded nearby, givingthe reference Ti . These normalised spectra were obtained af-ter summing several recordings, see text. Note the two peaks A and B corresponding to t g and e g levels, and the signal inthe valley (arrows), significantly more important in the low-pressure case. EELS results. –
Fig. 4 presents raw experimentalspectra, of which the treatment has been limited to back-ground subtraction and averaging over several spectrum-images. In fig. 5, the data have been deduced from thespectra by the fitting procedure described in sec. ’Meth-ods’. The experimental conditions were quite similar(probe effective size and current, spectrometer setting),so that the two samples in each figure can be directlycompared.In fig. 4, the mean experimental interface spectra aresuperimposed on bulk spectra, giving the reference Ti .The effect of the interface on the Ti-L , edge is clearlyvisible in the low pressure sample, while it is significantlysmaller in the high-pressure case. The fit with referenceTi and Ti spectra as described in sec. ’Methods’ gives20% Ti in the former case and only 10 % Ti in thelatter.Fig. 5 shows the spatial evolution of Ti ratio, deducedfrom such fits, in two profiles taken in the low- (fig. 5a)and high- (fig. 5b) pressure samples. Error bars were setto twice the standard deviation between experimental andfitting spectra. Variations from one profile to the nextwere negligible in the low-pressure sample, so that the pro-file shown in fig. 5a is quite representative of this sample.In contrast, such variations were quite significant in thehigh-pressure sample, probably due to the 3-dimensionalrelaxation of strain in LAO and its localised effects in thesubstrate. For the profile presented in fig. 5b we have cho-sen a region where the interface Ti level was the lowest.It appears on the figure to be of the order of uncertainty,which is about 5 % in this case.In order to help reading fig. 5, let us recall that the probehas a gaussian shape with a width at half maximum of 2p-4ELS determination of Ti oxidation state at the LaAlO / SrTiO interface Fig. 5: (colour online) Ti profiles in the low-(a) and high-(b) pressure samples. Concentrations obtained after fittingall spectra in each profile with Ti and Ti references asdescribed in text. Superimposed are the corresponding nor-malised Ti profiles in each case. The red curves are guides tothe eye. nm. Thus, it starts to ’count’ Ti and Ti when it is stillcentered in pure LAO. This causes the residual Ti profilein LAO, but also, as the proportion of Ti is the largestat that time, the shift towards LAO of the maximum ofTi ratio in fig. 5a. Because of these probe convolutioneffects, it is necessary to compare the present profiles withmodels to effectively localise the Ti ions, which we doin the next section.In turn, as measurement noise is the largest where the Tisignal is the smallest, the actual uncertainty there (see firstdata point in fig. 5b) is probably larger than our estimatebased on fit quality. Discussion. –
We compare in this section our exper-imental profiles with different models of electron distribu-tions, convoluted with our gaussian experimental response(fig. 6). Let us first note that, once the Ti profiles fromthe literature are convoluted, they only slightly emergefrom the present detection limit (for the present profiles,the sensitivity is worth ∼ ∼ × cm − , of coursenot as fine as that mentioned above for summed spectra).Thus it is not only justified, but even mandatory, to in-clude our high pressure profile, even though it is almostflat, in such comparisons.Quite surprisingly, in both our cases of low- (fig. 6a)and high- (fig. 6b) pressures, the best fit is obtained witha model of intrinsic polarity. However, the low-pressure –conducting – case appears to fit with a model of localisedinterface electrons (’Intrinsic delta’ in fig. 6a), while thehigh pressure – insulating – case fits on the contrary with Fig. 6: The Ti profiles of fig. 5 in the low-(a) and high-(b)pressure samples compared with the EELS response applied topublished Ti distributions. ’Intrinsic delta’: localised elec-trons that would neutralise the intrinsic polarity in the form ofa delta distribution at the interface; ’Delta + Poisson’: samesurface density of electrons, but spread into the bulk followingthe solution of Poisson equation [12]; Willmott et al. : distribu-tion of Ti given in fig. 3a of ref. [9]. a model of mobile electrons (’Delta + Poisson’ in fig. 6b).As this is precisely the opposite of what is expected, wehave to attribute an extrinsic contribution to at least oneof the two cases.Given that electrons could hardly stay localised in theconducting low pressure specimen, and given that we havemade it so as to increase extrinsic effects, we may appointit to represent the extrinsic case. Thus, we would rathertake the present sharp profile of fig. 6a as the tip of a broaddistribution of mobile carriers such as that we present inref. [11]. If we recall that standard carrier densities are or-ders of magnitude lower [10,11] than the present detectionlimit ( ∼ × cm − ), this hypothesis appears quite con-sistent. Thus this low-pressure profile would be associatedwith donor-like point defects such as oxygen vacancies orLa substituted for Sr. But fast kinetics of oxygen exod-iffusion [15] appear uncompatible with the profile sharp-ness, and allow one to eliminate oxygen vacancies. Con-sequently, we checked cation interdiffusion by secondaryion mass spectrometry on much thicker samples. And wefound indeed an enhanced La-Sr exchange, compared toAl-Ti, that moreover depended on growth pressure [28].The most likely dopant would thus be Lanthanum ratherthan the oxygen vacancy. It can be noticed, by the way,that Sr − x La x TiO chemical solutions are stable over alarge concentration range [26, 27]. Thus, we would finallyhave an extrinsic doping effect, of the type evoked by Will-mott et al. [9] (fig. 6a), though in lower amounts.p-5.-L. Maurice et al. In the high-pressure case (fig. 6b), the Ti profileshould indeed be closer to a model of intrinsic polarity.But it remains quite surprising that it be the mobile-carrier version of the model, as this sample is macro-scopically insulating. An explanation, perhaps difficultto check, would be that the sample is indeed conductivein between the areas where epitaxial stress is concentrated(arrows in fig. 2). The stressed zones would, in turn, actas traps and quench conductivity. Conclusion. –
We have measured the concentrationprofiles of Ti ions in the STO substrate in the vicin-ity of the TiO -terminated (001) interface with LAO us-ing STEM-EELS in cross section. We have studied twogrowth conditions: 200 s and 10 − Pa on the one hand,and 50 s and 40 Pa on the other hand, so that point-defect effects be maximum in the former case, and min-imum in the latter. The concentration of Ti ions ap-pears to depend significantly on growth conditions: in thelow-pressure, thicker sample, Ti surface concentrationcomes out at more than twice that in the high-pressurespecimen. In the former case, there is thus an additionaldoping. Among the different possibly active point defects,cations in substitution appear mor likely candidates thanoxygen vacancies, in qualitative agreement with previousobservations [9].In summary, we have thus shown that Ti ions bearingvalence 3+ are indeed present at the polar interface be-tween LAO and TiO -terminated (001) STO, but we havealso shown, for the first time to our knowledge, that theirconcentration depends on growth conditions, so that theirorigin is, at least partially, extrinsic. ∗ ∗ ∗ We thank M. Bibes (IEF-CNRS Universit´e Paris-Sud-XI) for fruitfull discussions and M. Tenc´e (LPS-CNRSUniversit´e Paris-Sud-XI) for his help with STEM-EELS.
REFERENCES[1]
Ohtomo A. and
Hwang H. Y. , Nature , (2004) 423.[2] Robertson J. , Euro. Phys. J. Appl. Phys. , (2004) 265.[3] Reyren N., Thiel S., Caviglia A. D., Fitting Kourk-outis L., Hammerl G., Richter C., Schneider C. W.,Kopp T., Retschi A.-S. , Jaccard D., Gabay M.,Muller D. A., Triscone J.-M. and
Mannhart J. , Sci-ence , (2007) 1196.[4] Brinkman A., Huijben M., Van Zalk M., Huijben J.,Zeitler U., Maan J. C., Van der Wiel W. G., Rijn-ders G., Blank D. H. A. and
Hilgenkamp H. , NatureMater. , (2007) 493.[5] Thiel S., Hammerl G., Schmehl A., Schneider C. W. and
Mannhart J. , Science , (2006) 1942.[6] Noguera C. , J. Phys.: Condens. Matter , (2000) R367.[7] Huijben M., Rijnders G., Blank D.H.A., Bals S., VanAert S., Verbeeck J., Van Tendeloo G., BrinkmanA. and
Hilgenkamp H. , Nature Mater. , (2006) 556. [8] Ohnishi T., Lippmaa M., Yamamoto T., Meguro S. and
Koinuma H. , Appl. Phys. Lett. , (2005) 241919.[9] Willmott P. R., Pauli S. A., Herger R., Schlep¨utzC. M., Martoccia D., Patterson B. D., Delley B.,Clarke R., Kumah D., Cionca C. and
Yacoby Y. , Phys. Rev. Lett. , (2007) 155502.[10] Herranz G., Basletic M., Bibes M., Carr´et´ero C.,Tafra E., Jacquet ´E., Bouzehouane K., DeranlotC., Hamzic A., Broto J.-M., Barth´el´emy A. and
FertA. , Phys. Rev. Lett. , (2007) 216803.[11] Basletic M., Maurice J.-L., Carr´et´ero C., Her-ranz G., Copie O., Bibes M., Jacquet ´E, Bouze-houane K., Fusil S. and
Barthlmy A. , submitted , (see preprint at arXiv:0710.1395v1) .[12] Siemons W., Koster G., Yamamoto H., Harrison W.A., Lucovsky G., Geballe T. H., Blank D. H. A. and
Beasley, M. R. , Phys. Rev. Lett. , (2007) 196802.[13] Kalabukhov A., Gunnarsson R., B¨orjesson J., Ols-son E., Claeson T. and
Winkler D. , Phys. Rev. B , (2007) 121404(R).[14] Khodan A. N., Guyard S., Contour J.-P., Cr´et´e D.-G., Jacquet ´E. and
Bouzehouane K. , Thin Solid Films , (2007) 6422.[15] Uedono A., Shimayama K., Kiyohara M., Chen Z.Q. and
Yamabe K. , J. Appl. Phys. , (2002) 2697.[16] Pentcheva R. and
Pickett W. E. , Phys. Rev. B , (2006) 035112.[17] Park M. S., Rhim S. H. and
Freeman A. J. , Phys. Rev.B , (2006) 205416.[18] Maurice J.-L., Devos I., Casanove M.-J.,Carr´et´ero C., Gachet G., Herranz G., Cr´et´eD.-G., Imhoff D., Barth´el´emy A., Bibes M., Bouze-houane K., Deranlot C., Fusil S., Jacquet ´E.,Domeng`es B. and
Ballutaud D. , Materials Science andEngineering B , (2007) 1.[19] Maurice J.-L., Carr´et´ero C., Casanove M.-J.,Bouzehouane K., Guyard S., Larquet ´E. and
Con-tour J.-P. , Phys. Stat. Sol. (a) , (2006) 2209.[20] Nakagawa N., Hwang H. Y. and
Muller D. A. , NatureMater. , (2006) 204.[21] Jeanguillaume C. and
Colliex C. , Ultramicroscopy , (1989) 252.[22] Maurice J.-L., Imhoff D., Contour J.-P. and
Col-liex C. , Philos. Mag. , (2006) 2127.[23] Ohtomo A., Muller D. A., Grazul J. L. and
HwangH. Y. , Nature , (2002) 378.[24] Abbate M., de Groot F. M. F., Fuggle J. C., Fu-jimori A., Tokura Y., Fujishima Y., Strebel O.,Domke M., Kaindl G., van Elp J., Thole B. T.,Sawatzky G. A., Sacki M. and
Tsuda N. , Phys. Rev.B , (1991) 5419.[25] Colliex C., Maurice J.-L. and
Ugarte D. , Ultrami-croscopy , (1989) 31.[26] Eror N. G. and
Balachandran U. , J. Solid StateChem. , (1981) 85.[27] Tokura Y., Taguchi Y., Yokada Y., Fujishima Y.,Arima T., Kumagai K. and
Iye Y. , Phys. Rev. Lett. , (1993) 2126.[28] Carr´et´ero C., Jomard F., Maurice J.-L., HerranzG., Basletic M., Barth´el´emy A., Bouzehouane K.,Fusil S., Jacquet ´E. and
Ballutaud D. , , (2008) inpreparation., , (2008) inpreparation.