PHYSICAL REVIEW B VOLUME 60, NUMBER 13 1 OCTOBER 1999-I Formation of nonmagnetic c-Fe1 xSi in antiferromagnetically coupled epitaxial Fe/Si/Fe G. J. Strijkers,* J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Jonge Department of Physics and COBRA, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands Received 29 March 1999 Low-energy electron diffraction, Auger electron spectroscopy, and conversion electron Mo¨ssbauer spectros- copy have been applied to study antiferromagnetically exchange-coupled epitaxial Fe/Si/Fe 100 . It is shown that a bcc-like 100 structure is maintained throughout the layers after a recrystallization of the spacer layer by Fe/Si interdiffusion. Direct experimental evidence is presented that c-Fe1 xSi (0 x 0.5) is formed in the spacer layer, a nonmagnetic metallic metastable iron silicide phase with a CsCl structure (B2), which supports explanations for the antiferromagnetic exchange coupling given recently. S0163-1829 99 02837-4 Since the discovery of strong antiferromagnetic AF in- which ensures that we mainly probe the surface layers. After terlayer coupling in Fe/Si multilayers1 there have been a deposition of the 60 Å Fe layer, a (1 1) LEED pattern of a number of studies addressing the transformation of the Si bcc-Fe 100 surface is observed with relatively sharp spots, spacer layer into iron silicide and its relation to the observed indicating good single-crystalline growth of the Fe on the interlayer coupling. It is now well established that a metallic Ge 100 substrate. The good crystallinity was further con- iron silicide formed by Fe/Si interdiffusion is responsible for firmed by magneto-optical Kerr effect MOKE measure- the interlayer coupling.2­9 The exact composition of the iron ments that showed hard and easy axes of magnetization for silicides in the spacer layer is considered to be crucial to fields applied along the 100 and 110 directions, respec- understanding the exponential decay of the AF coupling with tively, as expected for single-crystalline bcc-Fe. LEED pat- the interlayer thickness8 in the framework of the Anderson terns b , c , and d show that upon deposition of Si the sd-mixing model10 or the electron-optics model.11 In several spots become more and more faint and completely disappear between 8 and 9 Å. For 12 Å Si only a bright background is studies2,7,9 it is suggested that Fe and Si form an FeSi alloy left, which means that Si grows, at least above 9 Å, in an spacer layer with a metastable CsCl structure (c-FeSi) and amorphous or at least strongly disordered manner onto the an Fe:Si ratio close to 1. Although it has been shown that Fe. c-FeSi can be stabilized epitaxially,12,13 the spontaneous for- When Fe is deposited on this disordered 12 Å Si layer, the mation of c-FeSi in antiferromagnetically coupled Fe/Si- sharp (1 1) LEED pattern with the spots at exactly the based layers has not been directly observed up to now to our same position reappears again at a nominal layer thickness of knowledge. In this paper we present direct experimental evi- about 5 Å Fe, as shown in patterns e and f . This impli- dence for the presence of c-Fe1 xSi with 0 x 0.5 in the cates that by diffusion of Fe into the spacer a recrystalliza- spacer layer of AF coupled Fe/Si/Fe by means of low-energy tion of the disordered Si has taken place into crystalline iron electron diffraction LEED , Auger electron spectroscopy silicide. If this would not be the case one would expect poly- AES , and conversion electron Mo¨ssbauer spectroscopy CEMS . Fe/Si/Fe layers were grown in a molecular-beam epitaxy MBE system VG-Semicon V80M with a base pressure of 2 10 11 mbar. An e-gun source with feedback control of the flux was used for the deposition of natural Fe, whereas 56Fe, 57Fe, and Si were evaporated from temperature- stabilized Knudsen cells. All thicknesses were controlled by calibrated quartz-crystal monitors. The layers were grown at room temperature on Ge 100 substrates, which were cleaned by several Ar sputter (700 °C) and anneal (780 °C) treat- ments until a sharp Ge(100)-(2 1) LEED pattern and no more C and O contaminations were observed. The LEED and AES measurements were performed in situ during sev- eral stages of the Fe and Si growth utilizing wedge-shaped as well as homogeneous layers. The room-temperature CEMS measurements were done ex situ in a spectrometer with a 57CoRh source and a gas-flow detector. FIG. 1. left panel LEED I-V curves of the 00-spot intensity no Figure 1 shows the LEED patterns at 171 eV and LEED background correction and right panel LEED patterns at 171 eV I-V curves of the 00 spot during several stages of the growth of Ge 100 a 60 Å Fe, b 60 Å Fe 6 Å Si, c 60 Å Fe 8.5 of Ge 100 60 Å Fe 12 Å Si 45 Å Fe. The penetration Å Si, d 60 Å Fe 12 Å Si, e 60 Å Fe 12 Å Si 6 Å Fe, f 60 depth of the electrons at this energy is typically 3­4 ML, Å Fe 12 Å Si 45 Å Fe. 0163-1829/99/60 13 /9583 5 /$15.00 PRB 60 9583 ©1999 The American Physical Society 9584 STRIJKERS, KOHLHEPP, SWAGTEN, AND de JONGE PRB 60 of the (1 1) LEED pattern. Apparently, at this point a re- crystallization of the spacer layer takes place. The plateaus can be understood assuming that an equilibrium is reached between Fe deposition on top of the spacer layer and Fe diffusion into the spacer layer. When the Si spacer is satu- rated with Fe above 6.5 Å, an exponential increase and de- crease of Fe and Si AES intensities, respectively, is ob- served, indicating a closed layer growth of Fe. The Fe and Si intensity ratios at the plateaus IFe /ISi 0.56 and 0.83, calcu- lated from the absolute intensities, indicate that Fe1 xSi is formed with x in the range from 0 to 0.5, according to Gal- lego and Miranda.15 However, we have to realize that the interdiffusion of Fe with Si is a complex process and Fe or Si segregation at the surface and the observed recrystallization can seriously alter the AES intensity ratio. Therefore, a defi- FIG. 2. Normalized Si 92-eV LVV open squares and Fe 47-eV nite identification of the formed iron silicide phases cannot MVV solid circles ; AES peak intensities versus the nominal Si be given from the AES intensities alone. and Fe layer thickness for the growth of a Si on 60 Å Fe and b The high sensitivity of CEMS to the local atomic environ- Fe on 60 Å Fe 12 Å Si, respectively. The solid lines in a are ment together with the well-known Mo¨ssbauer parameters exponential fits to the data for tSi 4 Å . for the relevant iron silicides gives us the opportunity to identify the iron silicides formed in our exchange-coupled crystalline rather than single-crystalline growth of the top Fe layers. Furthermore, position sensitive identification of the layer. This reappearance of the LEED pattern can be ob- iron silicides can be obtained using a 57Fe probe layer that served up to a 21 Å Si spacer but disappears for larger thick- easily can be shifted through the multilayer stack. nesses, indicating a limited diffusion depth at room tempera- The measurements were performed on separately grown ture. A Fe/Si-wedge/Fe trilayer prepared in this way shows samples with 6 Å 57Fe probe layers at various positions in an strong antiferromagnetic interlayer exchange coupling, 56Fe matrix, guaranteeing that the Fe in the iron silicide of whose strength varies exponentially with the nominal Si the spacer layer can be clearly discriminated from the rest of layer thickness14 in accordance with de Vries et al.8 for lay- the Fe, in contrast to earlier studies.2,16,9 A first sample was ers grown at 200 °C. designed to give information about the iron silicide spacer In the left panel of Fig. 1 the 00-spot intensity versus the layer only, with the following nominal composition: electron energy is plotted corresponding to the LEED pattern Ge 100 68 Å 56Fe 3 16 Å Si 6 Å 57Fe 19 Å 56Fe) at the right-hand side, except for d in which no LEED spots 30 Å Si, schematically sketched in the inset of Fig. 4. The were found. Upon deposition of Si the I-V curves become relatively thick 56Fe buffer layer prevents that any iron ger- less structured, but regain better pronounced peaks after manide formation distorts the CEMS data and three repeti- deposition of the top Fe layer. Additionally, from the posi- tions were chosen for sensitivity reasons. For a nominally 16 tions of the main Bragg reflections, as indicated with dashed Å-thick Si spacer layer, AES measurements established that lines in Fig. 1, it can be concluded that the perpendicular 6 Å 57Fe will completely react with the Si, ensuring that the lattice constant remains constant at 1.43 Å close to bulk observed CEMS spectrum is only due to the nonmagnetic values for Fe (d100 1.4331 Å). Thus, a bcc-like 100 iron silicide in the spacer responsible for the interlayer ex- growth is maintained throughout the whole structure. change coupling. We will refer to this sample as ``reference'' To obtain more insight in the iron silicide formation pro- in the following. A second series was grown with the nomi- cess we followed the growth of a Si wedge on Fe and an Fe nal composition: Ge 100 60 Å Fe 3 10 Å Si 31 Å wedge on Si by AES. In Fig. 2 a the evolution of the Auger Fe 30 Å Si, with the 6 Å 57Fe probe layers deposited 4 Å Si LVV 92-eV and Fe MVV 47-eV peak intensities are below, 2 Å below, at the bottom of, on top of, and 6 Å plotted as a function of the nominal Si thickness deposited above the Si spacer, as schematically sketched in Fig. 5. on a single crystalline 60 Å Fe base layer. For a Si coverage Figure 3 shows the MOKE hysteresis loops for the 16 of about 4 Å a change of slope is observed in both Fe and Si Å-thick spacer, and for one sample of the second series with intensities, a clear sign of interdiffusion between the bottom 10 Å nominal spacer thickness. Both loops show evidence of Fe and the top Si layer up to this thickness, in agreement AF coupling with clear plateaus and high saturation fields. with earlier observations by Gallego and Miranda.15 For cov- The high remanence is mainly caused by the thick Fe buffer erages above 4 Å the AES intensities can be described with layer. We want to emphasize that for all of the samples, for exponentials solid lines in the figure with attenuation which the CEMS results will be presented in the following, lengths in agreement with closed Si-layer growth, excluding AF coupling is present, a necessary condition because we further intermixing. want to investigate the iron silicide responsible for the cou- In Fig. 2 b the evolution of the Si and Fe Auger intensi- pling. ties are presented for an Fe wedge deposited on 60 Å Fe 12 The CEMS spectrum of the reference sample is presented Å Si. A jump and two adjacent plateaus in the Fe as well as in Fig. 4. The spectrum consists of one quadrupole splitted in the Si AES intensities are observed between 3.5 and 6.5 Å line, which can be fitted well with a distribution of quadru- nominal Fe thickness. The jump at an Fe thickness of about pole doublets. The fitting parameters, isomer shift ( ), and 5 Å is accompanied by the already mentioned reappearance quadrupole splitting ( ) in the maximum of the distribution PRB 60 FORMATION OF NONMAGNETIC c-Fe1 xSi IN . . . 9585 TABLE I. Isomer shift ( ), quadrupole splitting ( ), hyperfine field (Bhf), and relative intensities (Id , and Is of doublets and sextets as obtained from fits to the experimental CEMS spectra. Isomer shifts are given relative to -Fe. Doublet Sextet 57Fe position Id Bhf Is mm/s mm/s % mm/s T % Reference 0.24 0.47 100 6 Å above 0.012 32.4 100 On top of Si 0.24 0.43 15 0.076 28.0 85 Below Si 0.24 0.43 29 0.059 29.4 71 FIG. 3. Representative longitudinal Kerr hysteresis loops of the 2 Å below Si 0.24 0.43 17 0.037 31.6 83 Fe/Si samples with 57Fe probe layers and nominal spacer layer 4 Å below Si 0.009 32.9 100 thickness of a 10 Å Si and b 16 Å Si, respectively. The field is applied along the 100 directions of the samples easy axis of the parameters. Furthermore, the perpendicular lattice constant is Fe layers . inconsistent with our LEED results. The remaining candidate is nonmagnetic c-Fe are listed in Table I. There are several possible iron silicides 1 xSi, a metallic metastable phase with a CsCl structure (B2). Sto- reported in the literature12,13 that qualify for the observed ichometric c-FeSi has an isomer shift of 0.26 mm/s, but apparently nonmagnetic iron silicide. no quadrupole splitting due to its cubic symmetry.12,13 How- The first one is -FeSi, a nonmagnetic small-gap semi- ever, when Fe vacancies are introduced (c-Fe conductor with a cubic symmetry B20 . The local Fe sym- 1 xSi) a quad- rupole splitting is observed consistent with our results. For metry is, however, trigonal, which results in a quadrupole example, Fig. 4 shows a remarkable resemblance with the splitted CEMS spectrum with 0.26 mm/s and slightly assymetric quadrupole splitted doublet of c-Fe 0.51 mm/s, close to what we observe.17 However, the for- 0.5Si as reported by Fanciulli et al.12 In their study the spectrum mation of an -FeSi spacer is impossible because our LEED was fitted with three quadrupole splitted doublets, associated results clearly show that an epitaxial relationship is main- to different Fe sites of which the doublet with the highest tained throughout the whole stack of layers, incompatible intensity has a quadrupole splitting of 0.47 mm/s, in with the lattice parameters of -FeSi. Furthermore, no semi- agreement with our data. Furthermore, we have to realize conducting properties of the spacer layer were found from that also strain reduces the local cubic symmetry introducing the temperature dependence of the interlayer coupling.14 an electric field gradient. Thus we might already expect a A second candidate is -FeSi2, the metallic state of iron quadrupole splitting for stoichometric c-FeSi grown coher- disilicide with a tetragonal structure and with Mo¨ssbauer pa- ently on bcc-Fe. From the previous analysis we conclude that rameters 0.23 mm/s and 0.47 mm/s for one doublet our spacer layer exists of, possibly strained, nonstoichomet- and 0.26 mm/s and 0.73 mm/s for a second ric silicon rich c-FeSi with a CsCl structure. doublet.17 Although the parameters of the first doublet match Additional information can be gained from the second perfectly with our results, we do not observe a second maxi- series of samples in which the 57Fe probe layer is shifted mum in the distribution matching the second known doublet through the stack from nominally 4 Å below to 6 Å above the spacer layer. Figure 5 shows the CEMS spectra for the different positions. It is clear from the raw data already that the 57Fe spectra far enough from the spacer are identical and consist of a magnetically splitted Fe sextet, whereas the other spectra are a mixture of magnetic Fe from the magnetic lay- ers and the nonmagnetic c-Fe1 xSi doublet from the spacer layer. All the spectra are fitted with a distribution of hyper- fine fields and, if present, a distribution of quadrupole split- tings for the nonmagnetic doublet. The relative intensity ratio of the sextets is 3:4:1:1:4:3 for all spectra indicating an in- plane magnetization direction in agreement with our magne- tization measurements. The resulting Mo¨ssbauer parameters in the maximum of the distributions are listed in Table I. The hyperfine field of the magnetically splitted part of the spectrum can be related to magnetic iron silicide alloys using FIG. 4. CEMS spectrum of Ge 100 68 Å 56Fe 3 16 Å previous work by Stearns.18 The maximum of the hyperfine Si 6 Å 57Fe 19 Å 56Fe) 30 Å Si. 57Fe deposited directly onto field ranges from 32.9 T, close to pure -Fe for the probe the Si will diffuse completely into the spacer, ensuring that the layer at 4 Å below the Si, to 28.0 T, which can be assigned spectrum is only caused by Fe in the formed iron silicide spacer. to Fe82Si18 for the probe layer directly on top of the Si. For The solid line is a fit with a distribution of quadrupole splitted all positions there is a broad distribution in hyperfine fields, doublets. indicating that there is a composition gradient from pure Fe 9586 STRIJKERS, KOHLHEPP, SWAGTEN, AND de JONGE PRB 60 The same c-Fe1 xSi doublet as for the reference sample is found in spectra b , c , and d , although the quadrupole splitting is slightly lower, which could be a sign of more stoichometric c-FeSi. A thickness-dependent strain effect, however, would result in an increased quadrupole splitting for thinner layers instead of the decrease observed.13 The c-Fe1 xSi doublet is found not only for the probe layer di- rectly below and on top of the Si layer but also for 2 Å below, which confirms the AES results that Fe diffuses also from the bottom into the Si. From the total intensities of the doublets and sextets we estimate that between 3 and 3.3 Å 57Fe and about 8 Å Si contribute to the nonmagnetic doublet, which results in c-Fe1 xSi with an average x in the range between 0.30 and 0.36, using bulk mole volumes of Fe and Si. One might argue that the complex formation of c-Fe1 xSi by diffusion of Fe into the Si spacer layer could strongly depend on the preparation methods and conditions, which might make a universal interpretation of the interlayer ex- change coupling in Fe/Si-based layers impossible. However, as was already shown before,8,9 the thickness dependence and strength of the coupling are generally the same for layers prepared with initially different FeSi spacer and magnetic layer compositions. Apparently, the interlayer exchange cou- pling does not depend crucially on the exact spacer layer composition, as long as a crystalline FeSi spacer is formed with the CsCl structure. This is confirmed by recent calcula- tions by Moroni et al.,20 who have shown that the density of states near the Fermi level for stoichometric and defective c-FeSi are almost identical, including a sharp peak in the density of states about 0.2 eV above the Fermi level. Within FIG. 5. CEMS spectra of Ge 100 60 Å Fe 3 10 Å Si 31 the framework of the Anderson sd-mixing model, this peak Å Fe 30 Å Si with a 6 Å 57Fe probe layer deposited at various is believed to mediate the coupling in Fe/Si-based layers. positions in the multilayer stack as indicated in the figure. The solid In conclusion we have systematically studied iron silicide lines are fits to the experimental spectra as explained in the text. formation in AF coupled Fe/Si/Fe 100 layers. With LEED and AES it was confirmed that a crystalline iron silicide is in the bulk of the magnetic layers towards c-Fe1 xSi in the formed in the spacer layer, which was identified as spacer layer. Furthermore, an asymmetric iron silicide profile c-Fe is observed in spectra b and c . More Fe diffuses from the 1 xSi from CEMS measurements. The formation of bottom than from the top into this 10 Å Si layer. It is clear c-FeSi corroborates recent explanations for the observed an- that the Fe/Si and Si/Fe interface are inequivalent with re- tiferromagnetic exchange coupling in Fe/Si based layers.8 spect to the iron silicide formation, an observation earlier The work of G.J.S. was supported by the Foundation for made by photoemission studies by Kla¨sges et al.19 Fundamental Research on Matter FOM . *Author to whom correspondence should be addressed; electronic 5 A. Chaiken, R. P. Michel, and C. T. Wang, J. Appl. Phys. 79, address: Strijkers@phys.tue.nl 4772 1996 . 1 E. E. Fullerton, J. E. Mattson, S. R. Lee, C. H. Sowers, Y. Y. 6 J. A. Carlisle, A. Chaiken, R. P. Michel, L. J. Terminello, J. J. Jia, Huang, G. Felcher, and S. D. Bader, J. Magn. Magn. Mater. 117, T. A. Callcott, and D. L. Ederer, Phys. Rev. B 53, R8824 1996 . L301 1992 . 7 A. Chaiken, R. P. Michel, and M. A. Wall, Phys. Rev. B 53, 5518 2 E. E. Fullerton, J. E. Mattson, S. R. Lee, C. H. Sowers, Y. Y. 1996 . Huang, G. Felcher, S. D. Bader, and F. T. Parker, J. Appl. Phys. 8 J. J. de Vries, J. Kohlhepp, F. J. A. den Broeder, R. Coehoorn, R. 73, 6335 1993 . Jungblut, A. Reinders, and W. J. M. de Jonge, Phys. Rev. Lett. 3 C. L. Foiles, M. R. Franklin, and R. Loloee, Phys. Rev. B 50, 78, 3023 1997 . 16 070 1994 . 9 J. T. Kohlhepp, J. J. de Vries, F. J. A. den Broeder, R. Coehoorn, 4 J. Dekoster, A. Vantomme, S. Degroote, R. Moons, and G. R. M. Jungblut, A. Reinders, G. J. Strijkers, A. A. Smits, and W. Langouche, in Structure and Properties of Multilayered Thin J. M. de Jonge, in Magnetic Ultrathin Films, Multilayer and Films, edited by T. D. Nguyen et al., MRS Symposia Proceed- Surfaces­1997, edited by J. Tobin et al., MRS Symposia Pro- ings No. 382 Materials Research Society, Pittsburgh, 1995 , ceedings No. 475 Materials Research Society, Pittsburgh, p. 253. 1997 , p. 593. PRB 60 FORMATION OF NONMAGNETIC c-Fe1 xSi IN . . . 9587 10 Z.-P. Shi, P. M. Levy, and J. L. Fry, Europhys. Lett. 26, 473 1991 . 1994 . 16 J. Dekoster, H. Bemelmans, S. Degroote, R. Moons, J. Verhey- 11 P. Bruno, Phys. Rev. B 52, 411 1995 . den, A. Vantomme, and G. Langouche, J. Appl. Phys. 81, 5349 12 M. Fanciulli, G. Weyer, H. von Ka¨nel, and N. Onda, Phys. Scr. 1997 . T54, 16 1994 . 17 O¨. Helgason and T. I. Sifu¨sson, Hyperfine Interact. 45, 415 13 M. Fanciulli, G. Weyer, A. Svane, N. E. Christensen, H. von 1989 . Ka¨nel, E. Mu¨ller, N. Onda, L. Miglio, F. Tavazza, and M. 18 M. B. Stearns, Phys. Rev. 129, 1136 1963 . Celino, Phys. Rev. B 59, 3675 1999 . 19 R. Kla¨sges, C. Carbone, W. Eberhardt, C. Pampuch, O. Radar, T. 14 G. J. Strijkers, J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Kachel, and W. Gudat, Phys. Rev. B 56, 10 801 1997 . Jonge unpublished . 20 E. G. Moroni, W. Wolf, J. Hafner, and R. Podloucky, Phys. Rev. 15 J. M. Gallego and R. Miranda, J. Appl. Phys. 69, 1377 B 59, 12 860 1999 .