Journal of Magnetism and Magnetic Materials 191 (1999) 331-338 Magnetic properties of epitaxial Fe(Si\VFeV) films grown on Si(1 1 1) D. Berling , G. Gewinner , M.C. Hanf *, K. Hricovini , S. Hong , B. Loegel , A. Mehdaoui , C. Pirri , M.H. Tuilier , P. Wetzel Laboratoire de Physique et de Spectroscopie Electronique, UPRES. A 7014, 4 rue des Fre%res Lumie%re, 68093 Mulhouse Cedex, France Laboratoire pour l'Utilisation du Rayonnement Electromagne&tique, Universite& Paris-Sud, 91405 Orsay, France LPMS, Universite& de Cergy-Pontoise, Neuville-sur-Oise, 95031 Cergy-Pontoise, France Received 8 June 1998; received in revised form 7 September 1998 Abstract Iron silicide thin films (200 As Fe(Si\VFeV) with 0)x)1 and local cubic CsCl structure) have been grown by coevaporation at room temperature (RT) on Si(1 1 1). X-ray magnetic circular dichroism (XMCD) and magneto-optic Kerr effect (MOKE) measurements indicate that the films are ferromagnetic at RT for x ranging from 1 (pure Fe) to 0.15 (Fe Si). The magnetization is parallel to the film surface, and the magnetic anisotropy is uniaxial, with the easy axis lying along a [1 0 1]1 crystallographic direction and the hard axis along a [1 2 1]1 direction of the substrate. MOKE measurements show that the magnitude of the saturation field increases with increasing Si concentration, while XMCD data indicate that the average local magnetic moment carried by the Fe atoms decreases with decreasing Fe concentra- tion. Models which involve the diminution of the number of Fe nearest neighbors are proposed for the description of the behavior of the Fe moments. 1999 Elsevier Science B.V. All rights reserved. PACS: 75.70.Cn; 78.70.Dm; 75.25.#z; 75.50.Bb; 75.30.Gw Keywords: Iron silicide; X-ray magnetic circular dichroism; Film growth; Kerr effect; Ferromagnetism 1. Introduction solutions with a cubic structure between x"0.46 (Fe Fe silicides have been widely studied in the past  Si) and 1 (pure Fe) [1]. However Si-rich silic- ides, that do not exist in the bulk form, can be years for their fundamental aspect as well as for stabilized as thin layers. Indeed Fe(Si possible applications in microelectronics. Bulk \VFeV) layers with 0)x)1 can be grown pseudomorphically in Fe(Si\VFeV) forms a continuous range of solid the whole composition range on Si(1 1 1) by co- evaporation of Si and Fe at room temperature (RT) [2]. Now the bulk cubic Fe-Si alloys (0.46)x)1) * Corresponding author. Tel.: 33-3-89-33-64-36; fax: 33-3-89- are known to be ferromagnetic, with a Curie tem- 33-60-83; e-mail: m.hanf@univ-mulhouse.fr. perature of 830 K for FeSi [3]. Thus the possible 0304-8853/99/$ - see front matter 1999 Elsevier Science B.V. All rights reserved. PII: S 0 3 0 4 - 8 8 5 3 ( 9 8 ) 0 0 3 6 9 - 2 332 D. Berling et al. / Journal of Magnetism and Magnetic Materials 191 (1999) 331-338 ferromagnetic character of the films would allow the integration of magnetic devices in silicon tech- nology. In this paper we show by means of the magneto- optic Kerr effect technique (MOKE) and X-ray magnetic circular dichroism (XMCD) that FeSi films, like bulk FeSi, are ferromagnetic at RT, and for the first time it is shown that silicide films for 0.15)x)0.5 are ferromagnetic as well. In par- ticular, we show that the magnetic moment carried by the Fe atoms decreases steadily with decreasing Fe concentration and vanishes for FeSi (CsCl, x"0). Different models are proposed to describe the variation of the magnetic moment with Fe content. 2. Experiment Fig. 1. Unit cell of the DO-type FeSi structure. The Fe' are surrounded by eight Fe'', and the Fe'' are surrounded by four Fe' and four Si. The films have been grown in ultra high vacuum (UHV) by the coevaporation of Fe and Si on the substrate kept at RT. The substrate was made of a 10 As thick codeposited FeSi template layer ini- ever, only local DO tially grown on Si(1 1 1). This completely avoids the  order is observed by X-ray photoelectron diffraction, as no superstructure was interdiffusion with the Si substrate that takes place revealed by X-ray diffraction [4]. Local DO even at RT for the Fe-rich films and considerably  order means that one sublattice is actually occupied with improves the structural quality of the layers. Indeed Fe atoms only, while 50% Fe and 50% Si atoms are such films, whose thickness was &200 As, exhibited randomly located on the second. For Fe-rich silic- good (1;1) LEED and IMEED patterns indicat- ides (from Fe ing the epitaxy of the silicides [2]. In the same Si to pure Fe, 0.5)x)1), Fe atoms substitute Si atoms in DO manner 200 As BCC Fe has been grown epitaxially -type FeSi. For films poorer in Fe than Fe as a reference. Finally, all the films were capped Si (0)x)0.5), the Fe' atoms are replaced by Si atoms. As x represents the with a 20 As Si layer to protect them from oxidation Fe concentration on the second sublattice contain- before they were removed from the UHV chamber. ing both Fe and Si species, for x"0, there are no More details about sample preparation and struc- Fe tural characterization can be found in Ref. [2]. ' any more. This corresponds to the metastable FeSi(CsCl) cubic silicide, where Fe atoms have All the films are metallic and present a cubic eight Si NN and six Fe NNN [5,6]. From now on, structure that can be derived from the DO type the denomination FeSi will refer to the epitaxially structure of FeSi. This structure, displayed in Fig. stabilized FeSi (CsCl) phase grown on Si(1 1 1). 1, may be viewed as a CsCl-type lattice made of two The magnetic properties of the samples were simple cubic sublattices. The sites of one sublattice characterized by ex situ MOKE at RT. The MOKE are occupied by Fe atoms only (Fe'' in Fig. 1), while system consists of a polarized laser source, two the other sublattice is made of 50% Si and 50% Fe polarizers, a photo-elastic modulator and a photo- (Fe' in Fig. 1). In bulk FeSi, long range crystallo- diode detector. Measurements were performed in graphic order is present: Fe' atoms have eight Fe'' the longitudinal geometry, i.e. with the magnetic nearest neighbors (NN), and six Si next nearest field applied in the sample plane and in the plane of neighbors (NNN), while Si atoms have eight Fe'' incidence. The incidence of the light was &45° with NN, and six Fe' NNN. In the deposited films how- respect to the sample surface normal. D. Berling et al. / Journal of Magnetism and Magnetic Materials 191 (1999) 331-338 333 The XMCD measurements have been performed 3. Results and discussion at the Laboratoire pour l'Utilistation du Rayonne- ment Electromagne´tique (LURE) on the SU 23 Fig. 2a and Fig. 2b display the Kerr loops for beam line of the Super ACO storage ring using four samples (x"1: pure Fe; x"0.5: Fe circularly polarized light. The rate of circular Si; x"0.33: Fe polarization was &70%. The samples have Si; x"0.15: Fe Si) recorded with the magnetic field oriented along two perpendicu- been analyzed in remanence at RT by absorption at lar crystallographic directions, namely [1 2 1] the Fe ¸ 1   edges. The incidence of the light was and [1 0 1] 60° with respect to the sample surface normal. 1 respectively. The dashed curve in Fig. 2b for pure Fe is an enlarged view of the Dichroism measurements were performed with curve presented in Fig. 2a. The presence of hyster- a constant incident photon spin direction and esis cycles indicates that the films are ferromagnetic by reversing the magnetization by means of a at RT for 0.15)x)1. No ferromagnetism is ob- magnetic field. The latter was applied parallel served for x)0.09, i.e. for iron silicides poorer in to the sample surface and along a [ 1 0 1]1 Fe than Fe crystallographic direction of the substrate, which,  Si. The clearcut square loops ob- tained along [1 0 1] according to the MOKE results that will be 1 indicate that this direction corresponds to a magnetic easy axis, whatever the shown below, corresponds to the magnetic easy value of the Fe concentration. This shows that all axis direction. silicide domains have the same easy axis. Fig. 2a Fig. 2. Kerr loops for x"1, 0.5, 0.33, and 0.15 with the magnetic field applied along the [1 2 1]1 direction (a) and the [1 0 1]1 direction (b), except for the dashed curve (x"1) which has been recorded with the field along [1 2 1]1 . 334 D. Berling et al. / Journal of Magnetism and Magnetic Materials 191 (1999) 331-338 also shows a magnetic hard axis, at 90° from the XMCD measurements presented in the following easy axis, along the [1 2 1]1 direction. Thus we paragraph. One observes that K increases very observe for the silicides an in-plane uniaxial mag- slowly with Si concentration. Note that K is cal- netic anisotropy, although the surface presents culated for the silicides only, as for the Fe layer the a threefold crystallographic symmetry [4,7]. Sim- [1 2 1] direction is not a magnetic hard axis. ilar uniaxial magnetic anisotropy has been already MOKE measurements have shown the fer- observed for Fe layers deposited on semiconduc- romagnetic character of the Fe silicide films even tors but its origin is not well understood [8,22]. It for composition very close to FeSi, and the pres- may be related for instance to the direction of the ence of an easy axis parallel to the sample surface atomic flux during Fe and Si evaporation with and along a [1 0 1] respect to the sample axis, or to the strain applied 1 crystallographic direction with a twofold rotational symmetry instead of to the substrate by the sample holder during film a threefold one expected from crystallographic growth. Concerning the pure Fe layer, the shape of crystal symmetry. Furthermore, the square-shaped the dashed curve in Fig. 2b recorded along the loops along with a strong H [1 2 1]   dependence with 1 direction indicates that, in contrast to x show that these silicides essentially crystallize in the results obtained on the iron silicides layers, this a simple phase with defined composition. The rela- axis is not a real hard axis. In other words, the tionship between magnetization and film composi- anisotropy is not uniaxial for the deposited Fe film. tion has been investigated by dichroism absorption Experiments are on the way to clarify these differ- measurements performed on five samples, namely ent points. FeSi (x"0), Fe Table 1 displays the values of the magnetic field  Si (x"0.23), FeSi (x"0.33), Fe H Si (x"0.5), and pure BCC Fe (x"1).   at which saturation occurs as a function of Fig. 3 displays the Fe ¸ x for various ferromagnetic samples. Clearly   edge X-ray absorption spectra of Fe(Si H \VFeV) layers recorded at RT ver-   increases with increasing Si content. Apparent- sus x. The data have been normalized to the inci- ly the displacement of the domain walls becomes dent photon intensity. For each concentration two more and more difficult when the Fe content de- curves, labeled creases, probably because of an increasing number > and \, are presented which are recorded after the magnetic field has been applied of defects. This trend is similar to that observed on in two opposite directions along the magnetic easy bulk alloys of Fe with various elements [9]. Also axis [1 0 1] presented in Table 1 are the anisotropy constants 1 . Clearly there is a dichroic signal for all samples with x'0, which confirms the fer- K calculated by minimizing the Zeeman energy romagnetic nature of these compounds. No dichro- when H is applied along the hard axis direction, ism is detectable for FeSi, in agreement with which leads to MOKE measurements. K" Fig. 4 presents the differential absorption curves HM, where H is the applied magnetic field and M the +"( >! \) for various Fe concentrations. These curves have been normalized to the Fe con- magnetization per atom. M is inferred from the tent of each sample by dividing them by I(¸), Table 1 Magnitude of the saturation field H  as a function of x with the magnetic field applied along the [1 0 1]1 (easy axis) and [1 2 1]1 (hard axis) directions. The magnitude of the anisotropy constant for the Fe-Si compounds is also presented (in J/atom) Fe FeSi FeSi Fe Si x 1 0.5 0.33 0.15 H  along the easy axis (Oe) 13 78 164 220 H  along the hard axis (Oe) 19 173 595 830 Anisotropy constant (J/at) 1.35;10\ 1.6;10\ 2.03;10\ D. Berling et al. / Journal of Magnetism and Magnetic Materials 191 (1999) 331-338 335 Fig. 3. Fe ¸  X-ray absorption spectra recorded with the magnetic field applied in the two opposite directions [1 0 1] Fig. 4. XMCD signal for various values of x. The raw data have and [1 0 1]. The dashed and solid lines correspond to the > been normalized to the Fe content of the samples. and \ curves, respectively. where I(¸) is the ¸ peak area of the half sum of signal integrated over the ¸ the two ¸  part of the  peaks taken with both polarizations, ¸ namely (   spin-orbit split edge is actually proportional ># \)/2. Here the half summation to the average magnetic moment (x). In doing so ( ># \)/2 is assumed identical to the linear po- (we consider larization absorption spectrum [10]. The statistics     for the whole range of iron silicides) the error is &2% for pure Fe, which is becomes poorer as the Fe content diminishes, how- much smaller than the dispersion of our experi- ever it can be clearly seen in Fig. 4 that the dichroic mental results. As we are interested in trends only signal intensity decreases when the film is richer in and not in the exact value of the magnetic moment, Si, and vanishes for FeSi composition. the different hypothesis and assumptions done are In order to get an idea of the magnitude of the quite reasonable. We proceeded in the following average magnetic moment (x) carried by the Fe way: first we integrated the normalized XMCD atoms, we used results obtained by Alouani et al. signal over the ¸ [11]. They found that the intensity of the calculated  feature. Then we estimated the Fe average moment (x) for the various samples by dichroic signal for Fe, FeN and FeN compounds considering that it is proportional to the XMCD is proportional to the spin magnetic moment at the ¸ iron site. As the orbital moment is much lower than  area, and that its value is 2.22 for the pure Fe layer (which corresponds to the bulk BCC Fe the spin moment for pure Fe (  "0.046 , value). Indeed for the Fe thickness considered here   "2.16 [11]), we assumed that the XMCD the film has reached bulk properties as far as the 336 D. Berling et al. / Journal of Magnetism and Magnetic Materials 191 (1999) 331-338 Fig. 5. Magnitude of the average Fe moment as a function of Fe concentration. The experimental XMCD data are indicated by open circles. The squares correspond to measurements performed on bulk Fe-Si alloys [1,14-16]. Solid and dashed lines correspond to two different models (see text). The triangles give the value of the branching ratio (BR) I(¸)/I(¸). magnitude of the Fe moment is concerned [12] The first model (dashed line) has been proposed Note that with this method we do not need to take by Hines et al. [15]; the Fe into account the degree of polarization of the light. ' atoms, which are surrounded by eight Fe In fact similar results can be obtained by exploiting '' atoms whatever the Fe concentration keep a local moment of 2.22 the sum rules given by Carra et al. [13]: the XMCD , the same moment as bulk Fe. In other words, only signal integrated over the ¸ peak is proportional the NN are taken into account, i.e. the Fe to 31¸ ' atoms X2#21SX2 (if the magnetic dipole operator are screened by the Fe 1¹ '' atoms from an influence of X2 is not taken into account), which is, when the NNN. The Fe '' atoms moment is deduced from  is neglected, proportional to the magnetic mo- Mo¨ssbauer and NMR measurements [17]. These ment (x). data give the value of the hyperfine field at the Fe The different values of (x) are plotted in Fig. 5 '' site, which is found to depend only on the number as a function of Fe concentration x. The multiple of Fe NN (varying from 0 to 8), and not on the circles for each concentration represent several mean Fe concentration. Thus the magnitude of the measurements on the same sample. We can see that Fe the local magnetic moment depends strongly on x, '' moment depends only on the number of Fe' NN, at least for the stable bulk alloys investigated and more precisely increases with increasing Fe in Ref. [17], i.e. for 0.46)x)1. Here we assume concentration. Moreover, for FeSi the value of the that this property is valid over the whole Fe con- moment is zero. Also presented in Fig. 5 by squares centration range, i.e. 0)x)1. Hence the number are the values measured by neutron scattering and of Fe NN for the various Fe concentrations are saturation magnetization for bulk iron-silicon calculated by means of the binomial law: alloys [1,14-16]. These measurements give in FeSi a moment of 2.2-2.4 8! for the Fe' sites, and 1.2-1.35 P(i)" xG(1!x) \G , for the Fe'' sites. We can see that our i!(8!i)! data for the films are close to the mean Fe moment observed in bulk Fe-Si systems. Thus we tried to where P(i) is the probability of finding i Fe apply to our films the models used to describe the ' NN for an Fe bulk Fe-Si alloys. '' atom in an Fe(Si\VFeV) silicide. Moreover to obtain the value of the magnetic moment we D. Berling et al. / Journal of Magnetism and Magnetic Materials 191 (1999) 331-338 337 have considered that it scales with the measured tude of their moment. This also explains the fact hyperfine fields at the Fe'' sites. Indeed it has been that the Fe magnetic moment is similar for bulk shown that the magnetization of Fe-Si alloys and Fe the average internal magnetic field exhibit the same Si and the FeSi films, although the DO super- structure is not present in the Fe dependence as a function of Fe concentration [17]. Si films. This results are also in line with X-ray photoemission One can see in Fig. 5 that the experimental data spectroscopy (XPS) measurements performed on show a behavior that is rather properly described these systems as discussed in Ref. [2]: the Fe2p and by the model. However for FeSi (x"0.33) the Fe3s core level signals presented multiplet splitting magnetic moment is lower than the value given by effects for the various silicide films grown on this model. Si(1 1 1) except for FeSi, that were attributed to the A second model originates from ab initio calcu- presence of a local magnetic moment on the Fe lations performed by Kudrnovsky et al. [14] in the atoms. However note that for x"0.09 (Fe frame of the so-called local environment model on  Si) XPS data seemed to indicate the presence of a mag- Fe-Si alloys: the local magnetic moment has been netic moment on the Fe atoms, whereas no calculated with the linear muffin-tin orbital method square hysteresis loop was observed with MOKE. and the coherent potential approximation for the Nevertheless, a saturation magnetization effect Fe'' atoms and for 0.5)x)1. In this approach the is observed in both [1 0 1] Fe 1 and [1 2 1]1 . di- '' magnetic moment is found to depend linearly rections. Maybe the number of Fe on the Fe concentration, ranging from pure Fe to ' atoms is too low so that there is no remanence at all, at least FeSi. In the present paper we extrapolate this at RT. linear dependence to lower Fe concentration (for A further confirmation of the diminution of the 0)x)0.5), as was done in Ref. [14]. In this way local magnetic moment with Si content comes from one obtains (Fe'')"0 for FeSi, where the Fe'' the variation of the branching ratio I(¸ atoms are surrounded by eight Si atoms. This is in )/I(¸) in the absorption spectra with Fe concentration. I(¸ line with the Mo¨ssbauer measurements, where the ) and I(¸ hyperfine field is equal to zero when there are no Fe ) are the area under the ¸ and ¸ absorp- tion peaks obtained from the half summation of the NN. In this local environment model, the moment circularly polarized curves ( ># \)/2 which is, as of the Fe' atoms that are surrounded by eight Fe'' pointed above, equivalent to the linear polarization atoms whatever the magnitude of x, remains at absorption spectrum. The areas were estimated 2.22 . One can see in Fig. 5 that this model gives after subtraction of a linear background for each values of (x) quite close to those of the statistical peak, by integrating the curves from 703.9 to model, and that the experimental points exhibit the 717.2 eV for ¸ same trend. Again we note a tendency to observe , and from 717.2 to 734.3 eV for ¸. The branching ratios (BR) obtained in this way are lower moments than those predicted by this model plotted in Fig. 5 with triangles. They decrease from for x)0.5. Possibly this reflects merely the fact 2.26 for pure Fe to 1.98 for FeSi. This last value is that our implicit assumption, that we essentially close to the statistical value of 2.00 [18], and has observe the magnetization at saturation M (¹"0), been already observed for non-magnetic Fe silic- is no longer a good approximation at RT in ides [19]. This result confirms the fact that FeSi is the films with x)0.5 because of the decrease not ferromagnetic and that no moment is carried in Curie temperature and relevant finite temper- by the Fe atoms. On the other hand, BR with ature effects. In other words the mean temperature a magnitude higher than the statistical value are dependent magnetic moment associated with the attributed to high-spin states [18]. In other words, ferromagnetic order measured by means of XMCD the BR reflects the behavior of the local magnetic at RT is substantially lower than the local Fe moment. Note that a long-range ferromagnetic or- moment. der is not necessary to observe the dependence of Yet we can conclude that the local magnetic the branching ratio with the local magnetic mo- moment of the Fe atoms depends essentially on the ment. For example O'Brien et al. have shown that chemical nature of the NN, rather than the magni- the BR for Mn in an ordered Cu(0 0 1)c(2x2)Mn 338 D. Berling et al. / Journal of Magnetism and Magnetic Materials 191 (1999) 331-338 surface alloy phase is higher than for bulk-like References epitaxial Mn on Cu(0 0 1), thus indicating a higher local magnetic moment for Mn within the alloy [1] V. Niculescu, J.I. Budnick, Solid State Commun. 24 (1977) [20]. However no long range ferromagnetic order 631. was detected for both systems. We can see in Fig. [2] S. Hong, P. Wetzel, G. Gewinner, D. Bolmont, C. Pirri, J. Appl. Phys. 78 (1995) 5404. 5 that the BR decreases versus x in a way similar to [3] M. Fallot, Ann. Phys. 6 (1936) 305. that of the magnetic moment. A decrease of the iron [4] S. Hong, C. Pirri, P. Wetzel, D. Bolmont, G. Gewinner, S. magnetic moment with decreasing Fe concentra- Boukari, E. Beaurepaire, J. Magn. Magn. Mater. 165 tion accompanied by a reduction of the BR has (1997) 212. been already observed on Fe-Ge amorphous alloys [5] C. Pirri, M.H. Tuilier, P. Wetzel, S. Hong, D. Bolmont, G. Gewinner, R. Corte s, O. Heckmann, H. von Ka¨nel, Phys. [21]. Concerning this moment diminution, it has Rev. B 51 (1995) 2302. been suggested that in the Fe-Si alloys d-like elec- [6] H. von Ka¨nel, K.A. Ma¨der, E. Mu¨ller, N. Onda, H. Sirrin- trons are transferred from Si to Fe, thus reducing ghaus, Phys. Rev. B 45 (1992) 13807. the number of d-holes and in turn the magnitude of [7] S. Hong, Ph.D. Thesis, unpublished the moment [17]. However, Morrison et al. have [8] J.A.C. Bland, M.J. Baird, H.T. Leung, A.J.R. Ives, K.D. Mackay, H.P. Hughes, J. Magn. Magn. Mater. 113 (1992) shown that for FeVGe\V amorphous alloys the 178. number of holes does not depend on x [21]. In [9] E. Kneller, Ferromagnetismus, Springer, Berlin, 1962, p. contrast, they suggest that the reduction of the 536. magnetic moment is rather due to hybridization [10] W.L. O'Brien, B.P. Tonner, Phys. Rev. B 50 (1994) 2963. causing modifications in spin pairing. [11] M. Alouani, J.M. Wills, J.W. Wilkins, Phys. Rev. B 57 (1998) 9502. [12] G.W. Anderson, M.C. Hanf, P.R. Norton, M. Kowalewski, K. Myrtle, B. Heinrich, J. Appl. Phys. 79 (1996) 4954. 4. Summary [13] Paolo Carra, B.T. Thole, Massimo Altarelli, Xindong Wang, Phys. Rev. Lett. 70 (1993) 694. [14] J. Kudrnovsky, N.E. Christensen, O.K. Andersen, Phys. In conclusion, we have shown that iron silicides Rev. B 43 (1991) 5924. Fe(Si\VFeV) films with a thickness of &200 As are [15] W.A. Hines, A.H. Menotti, J.I. Budnick, T.J. Burch, T. ferromagnetic at RT on the whole range of concen- Litrenta, V. Niculescu, K. Raj, Phys. Rev. B 13 (1976) 4060. tration 0.15)x)1. The magnetic anisotropy is [16] J. Moss, P.J. Brown, J. Phys. F: Metal Phys. 2 (1972) 358. uniaxial, with the easy axis lying along a [1 0 1] [17] M.B. Stearns, Phys. Rev. 129 (1963) 1136. 1 [18] B.T. Thole, G. van der Laan, Phys. Rev. B 38 (1988) 3158. crystallographic direction, and the hard axis along [19] Fausto Sirotti, Maurizio De Santis, Giorgio Rossi, Phys. a [1 2 1]1 direction. The average magnetic moment Rev. B 48 (1993) 8299. on the Fe atoms decreases with increasing Si con- [20] W.L. O'Brien, J. Zhang, B.P. Tonner, J. Phys.: Condens. centration. This reduction is attributed to the dim- Matter 5 (1993) L515. inution of the total number of Fe [21] T.I. Morrison, M.B. Brodsky, N.J. Zaluzec, Phys. Rev. ' atoms, and to the B 32 (1985) 3107. reduction of the Fe'' magnetic moment upon in- [22] J.J. Krebs, B.T. Jonker, G.A. Prinz, J. Appl. Phys. 61 (1987) creasing the number of Si NN. 2596.