The formation of Er-oxide nanoclusters in SiO 2 thin films with excess Si
Annett Thøgersen, Jeyanthinath Mayandi, Terje Finstad, Arne Olsen, Spyros Diplas, Masanori Mitome, Yoshio Bando
TThe formation of Er-oxide nanoclusters in SiO thin films with excess Si Annett Thogersen, Jeyanthinath Mayandi, and Terje Finstad
Centre for Materials Science and Nanotechnology,University of Oslo, P.O.Box 1126 Blindern, N-0318 Oslo, Norway
Arne Olsen
Department of Physics, University of Oslo, P.O.Box 1048 Blindern, N-0316 Oslo, Norway
Spyros Diplas
SINTEF Materials and Chemistry, P.O.Box 124 Bildern, 0314 Oslo, Norway
Masanori Mitome and Yoshio Bando
National Institute of Material Science, Tsukuba, Japan (Dated: October 23, 2018)The nucleation, distribution and composition of erbium embedded in a SiO -Si layer were studiedwith High Resolution Transmission Electron Microscopy (HRTEM), Electron Energy Loss Spec-troscopy (EELS), Energy Filtered TEM (EFTEM), Scanning Transmission Electron Microscopy(STEM) and X-ray Photoelectron Spectroscopy (XPS). When the SiO layer contains small amountsof Si and Er, nanoclusters of Er-oxide are formed throughout the whole layer. Exposure of the oxideto an electron beam with 1.56*10 electrons/nm /sec. causes nanocluster growth. Initially thisgrowth matches the Ostwald ripening model, but eventually it stagnates at a constant nanoclusterradius of 2.39 nm. PACS numbers:
I. INTRODUCTION
For a long time, silicon was considered unsuitable foroptoelectronic applications because of its indirect bandgap and the absence of the linear electro-optic effect.Nanocrystals of silicon doped with rare earth ions havebeen investigated with the prospect of potential use inoptical applications, light-emitting diodes and lasers .It was shown that doping of rare earth ions into the Sistructure is beneficial for Si optoelectronic properties ,especially when using erbium (Er). Er has an unfilled 4fshell surrounded by an external closed shell and intra4f-transitions (4 l / to 4 l / ) will therefore show lumi-nescence at 1.54 µ m . These transitions can be excitedboth optically and electrically .Room temperature light emission from Er-doped Sinanocrystals has been studied extensively . Most ofthe studies are on ∼
300 nm thick silicon oxide films, whilelittle attention has been given for oxide films thinner than50 nm. On the other hand, many studies have been doneon the surrounding atomic environment of Er in Si. Mau-rizio et al. found that the Er atom was surrounded byonly O atoms, and no direct Er-Si bonds were observed.Terrasi et al. reported that heating the sample for threehours at 620 ◦ C created a mixed environment of Si and Oaround the Er atoms. Further heat treatment at 900 ◦ Cremoves the residual Er-Si coordination and produces afull oxygen coordinated first shell with an average of 5 Oneighbors . When the concentration of Er is high, Erforms Er-O bonds in competition with Si. Then Er O forms in addition to ErSi . These studies have beenperformed on Er implanted samples with high Si concen- tration, but few published papers present work on low Siconcentration.In previous studies the nucleation, growth and thecrystal and electronic structure of Si nanoclusters in athin SiO layer have been studied by various microscopyand spectroscopy techniques . In the present work,we studied the nucleation, composition and distributionof Er clusters in SiO with low doses of Si, by meansof High Resolution Transmission Electron Microscopy(HRTEM), Electron Energy Loss Spectroscopy (EELS)mapping, Energy Filtered TEM (EFTEM), ScanningTransmission Electron Microscopy (STEM) and X-rayPhotoelectron Spectroscopy (XPS). II. EXPERIMENTAL
The samples were made by growing a ∼ on a p-type Si substrate by Rapid Thermal Oxida-tion (RTO) at 1000 ◦ C for six seconds. A 30 nm layerof Si and Er-rich oxide subsequently was sputtered fromSiO :Si:Er composite targets onto the initially formedRTO-SiO film. The area propotion for Si and Er was17 area % (11 at. %) and 1.1 area % (0.1 at. %) respec-tively. Sputtering was followed by heat treatment in aN atmosphere at 1000 − ◦ C for 30-60 minutes.Cross-sectional TEM samples were prepared by ion-milling using a Gatan precision ion polishing system with5 kV gun voltage. The samples were analysed by HRTEMand EDS in a 200 keV JEOL 2010F microscope with aGatan imaging filter and detector, and a NORAN Van-tage DI+ Electron Dispersive Spectroscopy (EDS) sys- a r X i v : . [ c ond - m a t . m t r l - s c i ] S e p tem. When studing the nanocluster growth during elec-tron beam exposure (see section III A), the current den-sity was measured, as the total current density on thefluorescent screen. EFTEM, EELS mapping and STEMwere performed with a 300 keV JEOL 3100FEF micro-scope equipped with an Omega imaging filter. EFTEM-Spectral Imaging (EFTEM-SI) was performed using en-ergy losses from 2 eV to 30 eV and with an energy slitof 1 eV. The EELS mapping images were obtained byplacing an energy slit width of 20 eV for the acquisitionof pre- and postedge images around the Si L , and ErN , edges.Elemental distribution images for Si and Er were dis-played as the difference between two pre-edge imagesand one post-edge. A higher intensity in the experi-mental images reflects a higher elemental concentration.XPS was performed in a KRATOS AXIS ULTRA DLD using monocromated Al K α radiation (h ν =1486.6 eV)on plane-view samples at zero angle of emission (verticalemission) with charge neutralization. The X-ray sourcewas operated at 10 mA and 15 kV. The spectra were peakfitted using Casa XPS after subtraction of a Shirleytype background. III. RESULTS AND DISCUSSIONA. Evolution of nanocluster size
The HRTEM images taken with a 200 keV JEOL2010F microscope, show dark areas of nanoclusters inthe oxide (see Figure 1). The nanoclusters were amor-phous and precipitate throughout the whole oxide thick-ness, except for the RTO-SiO and the SiO top layer. Ata very short electron beam exposure time (less than 20seconds), the Si-Er-rich SiO appears darker than pureSiO observed in previous studies , and only a fewsmall nanoclusters were observed. This is attributed tothe Er atoms being evenly distributed in the oxide beforeexposure to an intense electron beam. Er is a heavy ele-ment and will therefore scatter more electrons. The sam-ple was exposed to an electron beam with a current den-sity of 39.1 pA/cm , measured on the fluorescent screenat 800 000 times magnification. The electron density onthe specimen was calculated as the density on the imagescreen multiplied with the magnification squared, whichresults in an 39.1 pA/cm *800000 = 25.0 A/cm elec-tron density. The electron counts per nm square area for1 second is then 1.56*10 electrons/nm /sec. After only30 seconds of electron beam exposure at an electron den-sity of 25.0 A/cm , the Er atoms form nanoclusters of1.2 nm in radius. Further exposure induces nanoclustergrowth.A plot of the nanocluster radius (nm) versus beam ex-posure time (min) is presented in Figure 2. The visiblenanoclusters start with a radius of 1.2 nm, with a stan-dard deviation of ± ± FIG. 1: HRTEM images of the same area at a) an exposuretime of less than 30 seconds, and b) after long exposure timeto the electron beam. The images were taken with a 200 keVJEOL 2010F microscope.FIG. 2: Figure 2: A plot of the nanocluster radius (nm) versusexposure time (min) with two fitted functions. , R = R + [ 89 V γCDkT ] t, (1)where R is the radius at time t, R is the initial radiusat t=0 minutes, V [cm ] is the volume per atom in thenanocluster, γ [N/m] is the surface tension, C [cm − ] isthe equilibrium concentration and T [K] the temperature.The diffusion term D [cm /s] is of the form D = D e − Ea/kT , (2)with D [cm /s] the diffusion coefficient, E a [eV] theactivation energy and k [1.38*10 − J/K ] the Boltzmannconstant.As there were only a few nanoclusters observed in thebeginning of the experiment, it is reasonable to assumeR = 0 nm. Furthermore, although the sample temper-ature might increase slightly at the start of the exper-iment, it can be assumed to be more or less constantduring the measurements. Thus the Ostwald ripeningmodel can be simplified to R(t)=d*t / where d is nowa new parameter.Fitting this function as above, gives a d value of 1.27 ± / . Figure 2 shows that the fitted functionagrees well with the experimental data at the start of theprocess, but does not agree with the measurements ac-quired at longer time. This behaviour is attributed to thedecrease in Er concentration in the matrix with increas-ing time. Eventually there will be no Er left in the ma-trix, and the nanoclusters will not grow any further. Totake this into account, we propose a model that initiallyagrees with the Ostwald ripening model, but eventuallytends to a certain constant nanocrystal radius, R ( t ) = at / e − bt + c (1 − e − bt ) (3)This is essentially a convex combination of the Ostwaldripening model and the constant function c. Fitting thisfunction to the data as above, yields parameters a=1.21 ± / , b=0.17 ± ± ± . In this way, the system reduces its energyby reducing the relatively high surface potential of thesmaller nanoclusters. B. Composition of nanoclusters
EDS measurements were performed on the nanoclus-ters to study their composition, see Figure reffigure:3.Since the spectra show contributions from Er, Si and O,the nanoclusters will combine Er with either Si or O, orboth.To determine qualitatively the composition of the nan-oclusters, EELS mapping, EFTEM imaging of the mainplasmon peak of Si and STEM were performed on thesamples using a 300 keV JEOL 3100FEF microscope.Figure 4a shows an HRTEM image obtained using thecontrast objective aperture as well as elemental mapping
FIG. 3: EDS spectrum of the nanoclusters and the oxide takenwith a 200 keV JEOL 2010F microscope with a NORAN Van-tage DI+ EDS system of Si (4b) and Er (4c) using the Si L , and Er N , edges.In a previous study of SiO films containing 17 area %Si (11 at. % Si), very small amorphous nanoclustersof Si were detected . Such nanoclusters are, however,very difficult to detect by conventional TEM imaging .Figure 4b shows the Si mapped image. The substrateappeares bright as expected, but no Si nanoclusters arevisible in the oxide. Four Er maps of a similar area show-ing two pre-edge images, one post-edge image and theextracted image (post-edge - pre-edge) with higher res-olution, are presented in Figure 5. Er cluster is markedby an arrow. These results indicate that the dark nan-oclusters seen in Figure 1 contain Er, in agreement withthe EDS analysis.EFTEM of the Si plasmon peak was performed to de-termine if the nanoclusters contain Si. Figure 6a showsan image created from the plasmon peak of pure Si (16.8eV, with 2 eV energy slit). The oxide appears dark andno Si rich nanoclusters are visible. Figure 6b presentsan image created from the plasmon peak of SiO (23 eV,with 2 eV energy slit). As expected, the matrix is bright,and dark nanoclusters are visible. Therefore the darknanocluster areas do not seem to contain any Si. TheEELS spectrum inserted in Figure 6b was taken from themiddle of a dark nanocluster shown by the arrow. As theEELS spectrum has low energy resolution, it is not pos-sible to extract any detailed features. The nanoclusterhas a plasmon peak at 17 eV. Er O has a plasmon peakat 14 eV, and Er at 11.6 eV. Note that the nanoclus-ters are surrounded by SiO , and this can influence thelow loss spectra, since SiO has a plasmon energy of 23eV and is the major contributor to plasmon oscillations.Since the plasmon energy in Figure 6b at 17 eV is closerto the value for Er O (14 eV) this indicates that thenanoclusters contain Er-oxide.STEM images with both Bright Field (BF) and HighAngle Annular Dark Field (HAADF) detector are pre-sented in Figure 7. The STEM HAADF image in Fig-ure 7a shows bright nanoclusters of 3-5 nm in the dark FIG. 4: EDS spectrum of the nanoclusters and the oxide taken with a 200 keV JEOL 2010F microscope with a NORANVantage DI+ EDS systemFIG. 5: Two pre-edge images, one post-edge image and amapped area of the Er N , peak. The images were acquiredusing a 300 keV JEOL 3100FEF microscope. oxide. In this figure there is also a thin bright areanear the SiO -RTO/Si-substrate interface, suggesting thepresence of a thin Er rich layer at the interface. Figure 7bpresents a STEM BF image of the same area as shown inFigure 7a. The nanoclusters that appear bright in Figure7a are dark in Figure 7b. The SiO -RTO/Si-substrate in-terface seems to be more irregular than that seen in Fig-ure 1 and Figure 4a. This could be due to additional Erprecipitation at the interface during prolonged exposureto the electron beam. STEM imaging heats up the sam-ple more than regular TEM, this could lead to diffusionof Er atoms to the SiO -RTO/Si-substrate interface. C. Chemical state of nanoclusters
XPS spectra of the different elements present in thesample were acquired during depth profiling with Ar + sputtering. Figure 8 shows the high resolution Er-4d FIG. 6: EFTEM image showing a) the plasmon peak of Si(16.8 eV) and b) the plasmon peak of SiO (23 eV) taken witha 300 keV JEOL 3100FEF microscope. The inset shows anEELS spectrum of a dark nanocluster with a plasmon energyof 17 eV. spectrum at different depths in the oxide. The Er-4dspectrum from the Er rich area (spectrum seven in Fig-ure 8) in the sample is shown in Figure 9a, after a Shirleybackground subtraction. This spectrum was taken froma depth well before reaching the Si substrate and there-fore the signal from pure Si was absent. The Er-4d peakoverlaps with the Si-2s plasmon peak of SiO . To extractthe Er-4d peak from the overlapping spectrum, The Si-2pfrom the SiO (Figure 9d) was used. For spectral cali-bration purposes the Si-2p from the Si substrate was alsoused (Figure 9c). Figure 9b shows the spectrum result-ing from the subtraction of the SiO plasmon peak fromthe overlapping Er-4d and Si-2s plasmon. The spectralpositions were calibrated using the O-1s peak for SiO at 533 eV, the Si -2p peak at 103.6 eV and the Si -2p peak from the substrate at 99.5 eV . The subtractedspectrum in Figure 9b shows three peaks located at 170.3eV, 175.5 eV and 184.9 eV. The literature values of theEr d and Er O -4d binding energy are 168.0 and 168.7 eVrespectively . Since EDS, EFTEM and EELS map-ping showed no Si in the nanoclusters, they consist mostlikely of Er-O oxide. In this respect the peak at 170.3eV arises from Er-oxide, and the peaks at 175.5 eV and184.9 eV are energy loss peaks.The difference in binding energy between the referencevalue for bulk Er O (168.7 eV) and the measured bind-ing energy of the Er-4d peak (170.3 eV) may be due FIG. 7: STEM a) HAADF image and b) BF image using a300 keV JEOL 3100FEF microscope. to quantum size effects, deviation from the Er O sto-ichiometry or in energy referencing issues or a combina-tion of all of these factors. The shift in binding energycan be expressed as∆ E B = K ∆ q + ∆ V + ∆ ϕ − ∆ R (4)In Equation 4, K is the Coulomb interaction betweenthe valence and core electrons and ∆ q expresses thechange in the valence electrons. K∆ q describes there-fore the difference in interaction between core and va-lence electrons. ∆ V is the contribution of the changesin Madelung potential. ∆ ϕ is the changes of the sam-ple work function which may be important in the caseof insulators. These first three terms in Equation 4 referto initial state effects. The fourth term is the contribu-tion of the relaxation energy R, which is the kinetic en-ergy gained (negative sign in binding energy scale) whenthe electrons in the solid respond (screen) to the pho-tohole produced by the photoemission process, and is afinal state effect. If one could neglect relaxation effectsin the above formula, an increased ionicity would leadto a higher binding energy difference (∆ E B ), while anincreased covalence would lead to a smaller difference.The relaxation energy term in Equation 4 reflects thecore hole electron screening efficiency. Lets assume thatin the case of a nanocluster besides to the contribution ofthe Er nanocluster itself, extra atomic screening is likelyto arise from the surrounding matrix. Er O has a higherdielectric constant than the surrounding matrix (SiO =3.9 and Er O = 13), i.e. Er O is more insulating thanSiO . A dielectric material is a non-conducting materialthat can withstand a high electric field. If a material with a high dielectric constant is placed in an electric field,the magnitude of the field will be measurably reducedwithin the dielectric. Because of the comparatively largedistance between the atoms in the dielectric, none of theatoms interact with one another. A material with highdielectric constant has low screening efficiency. If an Er-4d core hole in the Er O nanoclusters is screened bythe SiO matrix (which has a lower dielectric constant),the screening in the nanoclusters would be superior tothe screening in bulk Er O . This would be reflected asa reduction in binding energy compared to bulk Er O ,however, this is not the case. The universal screeninglength is expressed as a u = 0 . Z . + Z . a (5)where Z is the atomic number for atoms 1 and 2 and a is the Bohr radius (52 . ∗ − nm). Calculationsusing equation 5 for the Si-Si, Si-O, O-O, Er-Er, Er-O and Er-Si interactions, resulted in a O − Ou =0.0145 nm, a Si − Ou =0.0136 nm, a Si − Siu = 0.0128 nm, a Si − Eru =0.0105nm, a Er − Ou = 0.011 nm and a Er − Eru = 0.0089 nm. Theabove values are aligned with the dielectric nature ofthe two oxides, SiO and Er O . The average screeninglength in SiO is larger than that in Er O . However,the screening length of SiO is very short compared tothe nanocluster size, therefore, the core hole screening inthe “bulk” of the Er-oxide nanocluster will not be largelyaffected by the surrounding matrix. The binding energyof the Er oxide in the nanoclusters found in this work ishigher than what is reported for bulk. This increase istherefore likely to be due to initial state effects, ratherthan final state effects.Er has a low solubility in SiO . This is due to themismatch in size and valence between the Er ions andSi ions in SiO . In Er O , Er ions are bonded to sixO atoms, with bond lengths around 0.22-0.23 nm . As-suming purely ionic bonding, an Er ion would donate halfan electron to each of its O neighbours. A tetrahedrallycoordinated Si ion , which is known to be the commonSi state in amorphous SiO , would donate one electronto each O neighbour. Pure Er in SiO , in the form ofErO , fits poorly into a pure SiO network, even whenallowing for local reconstructions . Energy is thereforegained by forming clusters in which several Er ions canshare nearest neighbours .The valence spd levels of Er are higher in energy thanthe valence sp levels of Si, with respect to the oxygenvalence band level. Charge transfer from Er to O willtherefore be favoured over charge transfer from Si to O.The amount of charge transfer from Er to O will dependon the amount of charge that O recieves from the otherneighbours (Si in this case). The replacement of Si forEr in the next-nearest-neighbour shell of an Er atom willtherefore increase the charge transfer away from the Ersites in accordance with numerical results (since Si to Ocharge transfer is less favourable) . Laensgaard calcu- FIG. 8: XPS spectra of the Er-4d peak, acquired during Ar + sputtering. The top spectrum is from SiO and the bottomfrom the Si-substrate. The spectra in the middle are fromEr-oxide in SiO . lated the charge transfer (in units of e) for Er atoms inseveral compounds, with crystalline Er Si O and Er O amongst them. It was found that Er in Er Si O showsa charge transfer of 0.290 e − , while Er in Er O showsa charge transfer of 0.284-0.276 e − , a positive numbermeans that electrons are transferred away from the atom.Guittet et al. found the charge transfer for Si and O inSiO to be 2.05 e − and -1.02 e − respectively.In the case of Er-oxide nanoclusters, initial state ef-fects seem to play a profound role as the cluster sizedecreases. In the context of the above discussions, atthe interface between Er O nanoclusters and SiO ma-trix, charge transfer from Er towards O is expected tobe larger as compared to the “bulk” of the nanoclusters.This additional charge transfer from the Er sites, ∆ q maybe the reason for the increased binding energy of the Er-4d peak in the nanoclusters as compared to bulk Er O .The Pauling electronegativity values for Si, O and Er are1.9, 3.44 and 1.24 respectively. These values support theargument for an increased charge transfer from the Ersites towards the O neighbours in the presence of Si.It is noticeable that the binding energy shifts in the FIG. 9: XPS spectrum of a) the Er-4d peak overlapping withthe plasmon peak of SiO , b )the Er-4d peak (after subtract-ing the SiO -2s plasmon spectrum), c) the SiO -2s plasmonpeaks. case of Er O nanoclusters occur in the opposite direc-tion of those of amorphous Si nanoclusters in a SiO matrix . In both cases it seems that initial state effectsgovern the shifts. IV. CONCLUSION
When the SiO -Si-Er layer contains low Si concen-trations, Er-oxide (possible Er O ) clusters nucleatethroughout the oxide. Under the exposure to an electronbeam with 1.56*10 electrons/nm /sec., the nanoclusterradius grows initially according to the Ostwald ripeningmodel, but eventually grows asymptotically towards ananocluster radius of 2.39 nm. The increased Er-4d bind-ing energy of Er-oxide in nanoclusters apart from energyreferencing reasons could be attributed to initial stateeffects dominated by increased charge transfer from Ertowards O in the presence of Si at the nanocluster-SiO matrix interface. V. ACKNOWLEDGEMENT
Financial support by FUNMAT@UiO, the University ofOslo, Kristine Bonnevie’s and SCANDEM’s travellingscholarship is gratefully acknowledged. One of the au-tors (A.T.) would like to thank Georg Muntingh for manyhelpful comments and discussions. H. Ennen, J. Schneider, G. Pomrenke, and A. Axmann,Appl. Phys. Lett. , 943 (1983). Meldrum, A. Hryciw, A. MacDonald, C. Blois, T. Clement,R. Decorby, J. Wang, and Q. Li, J. Lumin. , 199 (2006). C. Zhang, H. Xiao, Y. Wang, Z. Cheng, X. Cheng, andF. Zhang, Physica B , 208 (2005). M. Fujii, M. Yoshida, Y. Kanzawa, S. Hayashi, and K. Ya-mamoto, Appl.Phys. Lett. , 1198 (1997). G. Franzo, F. Priolo, S. Coffa, A. Polman, and A. Carnera,Appl- Phys. Lett. , 2235 (1991). J. Michel, J. Benton, R. Ferrante, D. Jacobson, D. Eagle-sham, E. Fitzgerald, Y.H.Xie, J. Poate, and L. Kimerling,J. of Appl. Phys. , 2672 (1991). J. H. Shin, M. Kim, S. See, and C. Lee, Appl. Phys. Lett. , 1092 (1998). P. Kik and A. Polman, j. of Appl. Phys. , 534 (2002). C. Chryssou, D. Pacifici, V. Vinciguerra, F. Priolo, andF. Iacona, Appl. Phys. Lett. , 2011 (1975). A. Kenyon, C. Chryssou, C. Pitt, T. Shimizu-Iwayama,D. Hole, N. Sharma, and C. Humphreys, J. of Appl. Phys. , 367 (2002). D. M. Zhigunov, O. Shalygina, S. Teterunkov, V. Y. Tim-oshenko, P. Kashkarov, and M. Zacharias, Semicond. Sci.Technol. , 1193 (2006). F. Xu, Z. Xiao, G. Cheng, Z. Yi, T. Zhang, L. Gu, andX. Wang, Thin Solid Films , 94 (2002). C. Maurizio, F. Iacona, F. Acapito, G. Franzo, and F. Pri-olo, Phys. Rev. B , 205428 (2006). A. Terrasi, G. Franzo, S. Coffa, F. Priolo, F.D’Acapito,and S. Mobilio, Appl. Phys. Lett. , 1712 (1997). M. Cerqueira, M. Stepikhova, M. Losurdo, M. Giangrego- rio, A. Kozanecki, and T. Monteiro, Opt. Mater. , 836(2006). A. Thøgersen, S. Diplas, J. Mayandi, T. Finstad, A. Olsen,J. Watts, M. Mitome, and Y. Bando, J. of Appl. Phys. ,024308 (2008). A. Thogersen, J. Mayandi, T. G. Finstad, A. Olsen, J. S.Christensen, M. Mitome, and Y. Bando, J. of Appl. Phys. , 94315 (2008). J. Verhoven,
Fundamentals of Physical Metallurgy (JohnWiley and sons, 1975). W. Ostwald, Z Phys. Chem. , 495 (1900). J. M. Lifshift and V. V. Slyozov, J. Phys. Chem. Solids ,35 (1961). C. Wagner, Z. Electrochem. , 581 (1961). J. F. Moulde, W. F. Stickle, P. E. Sobol, and K. D.Bomben,
Handbook of X-Ray Photoelectron Spectroscopy (Perkin-Elmer Corporation, 1992), p. 201. S. Gokhale, S. Mahamuni, S. Deshmukh, V. Rao, A. Ni-gavekar, and S.K.Kulkarni, Surf. Sci. , 127 (1990). J. F. Moulde, W. F. Stickle, P. E. Sobol, and K. D.Bomben,
Handbook of X-Ray Photoelectron Spectroscopy (Perkin-Elmer Corporation, 1992), p. 201. Y. Y. Zhu, R. Xu, S. Chen, Z. B. Fang, F. Xue, Y. L. Fan,X. J. Yang, and Z. M. Jiang, Thin Solid Films , 86(2006). S. Tian, J. Appl. Phys. , 5893 (2003). J. Lgsgaard, Phys. Rev. B , 174114 (2002). M. J. Guittet, J. P. Crocombette,, and M. Gautier-Soyer,Phys. Rev. B63