Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 MoKssbauer-e!ect studies of multilayers and interfaces T. Shinjo , W. Keune * Institute for Chemical Research, Kyoto University, Uji, Kyoto-fu 611-0011, Japan Laboratorium fu(r Angewandte Physik, Gerhard-Mercator-Universita(t Duisburg, D-47048 Duisburg, Germany Received 8 January 1999; received in revised form 10 March 1999; accepted 10 March 1999 Abstract The usefulness of MoKssbauer spectroscopy for the investigation of magnetic multilayer systems is described. By applying Fe MoKssbauer spectroscopy, the behavior of ultrathin magnetic layers, such as FCC-like Fe "lms on Cu(0 0 1), is studied. Position-speci"ed (depth-selective) information is available by preparing samples in which mon- atomic Fe probe layers are placed at speci"c vertical positions, e.g. at interfaces or at the surface. As demonstrated for epitaxial chemically ordered FePt alloy "lms and polycrystalline nanostructured Tb/Fe multilayers, the Fe-spin structure can be determined directly, and a site-selective Fe-speci"c magnetic hysteresis loop can be traced in very-high- coercivity materials. For the studies of non-magnetic layers, on the other hand, hyper"ne "eld observations by Au and Sn probes are worthwhile. Spin polarizations in Au layers penetrating from neighboring ferromagnetic 3D layers are estimated from Au MoKssbauer spectra and are also studied by inserted Sn probes in Au/3D multilayers. In the Sn spectra for Cr/Sn multilayers, it was found that remarkably large spin polarization is penetrating into Sn layers from a contacting Cr layer, which suggests that Cr atoms in the surface layer have a ferromagnetic alignment. 1999 Elsevier Science B.V. All rights reserved. Keywords: MoKssbauer spectroscopy; Hyper"ne "elds; Spin polarization; Interface magnetism; Multilayer magnetism 1. Introduction dependence of the hyper"ne "eld. The MoKssbauer e!ect shows whether the thin-"lm or multilayer MoKssbauer spectroscopy is a useful technique for samples and their interfaces are composed of di!er- the investigation of magnetism in ultrathin "lms, ent metallurgical and magnetic phases. Information interfaces and multilayers. By observing magnetic from MoKssbauer spectroscopy is essentially ele- hyper"ne "elds, B , one is able to con"rm the ment-speci"c and even position-selective if the existence of magnetic order, and estimate the mag- MoKssbauer-isotope probes are located as desired. nitude of the local moment and the direction For instance, by analyzing the spectra for topmost of magnetization. The magnetic transition interface layers in a multilayer, we can provide temperature is determined from the temperature structural information about the interface from a microscopic viewpoint [1]. Since the discovery of giant magnetoresistance * Corresponding author. Tel.: #49-203-3792387; fax: #49- (GMR) [2,3], novel problems have been raised 203-3793601. up concerning magnetism of ultrathin "lms and E-mail address: keune@uni-duisburg.de (W. Keune) interfaces. It is believed that the spin-dependent 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 9 ) 0 0 3 4 6 - 7 T. Shinjo, W. Keune / Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 599 scattering occurs mainly at interface sites, but the relation between interface roughness and the spin- dependent scattering probability is not yet clear. The mechanism of interlayer coupling between magnetic layers across a non-magnetic spacer layer is also a crucial issue, and the studies of magnetism in non-magnetic layers contacting with ferromag- netic layers are of great importance. For the studies of non-magnetic layers, Au and Sn are useful MoKssbauer probes. In this article, several studies using Fe probes are brie#y summarized, emphasizing the unique information which MoKssbauer spectroscopy as a local atomistic technique may provide. Further- more, studies of 3D/Au multilayers using Au and Sn probes are introduced. As the GMR e!ect has been observed for the "rst time in Fe/Cr multilayers, the role of the Cr spacer layer is also of interest. Finally, some preliminary MoKssbauer re- sults on Sn in Cr/Sn systems are described. 2. Survey of 57Fe MoKssbauer studies 2.1. Interfaces It is well known that Fe is the most convenient isotope for MoKssbauer measurements, and there are a large number of publications concerning Fig. 1. Fe MoKssbauer absorption spectra for Fe interfaces MoKssbauer spectroscopic studies of interface mag- covered by various metals: (a) [Fe(100 As)/Fe(3.5 As)/V] at 4.2 K; (b) [Fe(1 0 0 A netism [1]. A recent review of structure and mag- s )/Fe(3.5 As)/Mg] at 4.2 K; (c) [Fe/Fe/Mn] at 300 K; (d) [Mn/Fe/Fe]Mn at 300 K. netism of some Fe mono- and multilayer systems as seen by conversion-electron MoKssbauer spectro- scopy (CEMS) at atomic scale was given by Przybylski [4]. If Fe probes are located selective- ly at interfaces sites, the structure of the interface is tained spectrum is very broad in comparison with conjectured from the hyper"ne "eld observation. a normal 6-line pattern (Zeeman sextet) of -Fe, Several examples of Fe spectra for interface sites suggesting that the hyper"ne "eld is considerably are shown in Fig. 1. Spectrum (a) was taken from an distributed [5]. From the analysis of the hyper"ne interface of BCC( -) Fe contacting with polycrys- "eld distribution P(B talline V metal. The sample was a multilayer pre- ), the range of mixing be- tween Fe and V at the interface is estimated to be pared in ultrahigh vacuum (UHV) by repeating the not more than three atomic layers [6]. This con- depositions of the following structure n times on clusion is consistent with the analysis of the V hy- a polyimid substrate: [Fe(100 As)/Fe(3.5 As)/ per"ne-"eld distribution observed by NMR [7]. V]xn. If the surface of the Fe layer was ideally #at For comparison, a spectrum of the BCC interface and no di!usion had taken place, the spectrum in contact with Mg metal, [Fe(100 As)/ would indicate the situation of approximately two Fe(3.5 As)/Mg], is shown in Fig. 1(b) [8]. The spec- topmost interface atomic layers. However, the ob- trum of the Fe/Mg interface shows a more clearly 600 T. Shinjo, W. Keune / Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 resolved 6-line pattern than that of Fe/V, sugges- information on the degree of intermixing, that is, ting that the degree of mixing is signi"cantly less. the interface roughness on the most microscopic The condition of Fe/Mg interface preparation was (atomic) scale. In actual interfaces, there exists almost the same as that of Fe/V but the modi"ca- roughness of various scales. For the estimation of tion of the interface structure is moderate in the roughness on a macroscopic scale, X-ray tech- Fe/Mg. Since Fe and Mg are insoluble with each niques are useful. However, if one attempts to other in the equilibrium state, it seems reasonable prepare samples with di!erent roughness, it is un- that the reactivity at the interface is limited. The avoidable that many factors change simultaneously interface reactivity thus depends greatly on the and GMR properties cannot be expressed as combination of elements. Moreover, the structure a function of a single parameter of roughness. depends on the procedure of deposition, if the reac- Therefore the relation between roughness and tivity is relatively high. As an example, the spectra spin-dependent scattering probability remains for two kinds of Fe/Mn interfaces are shown in Fig. a matter of debate. Recently, Schad et al. [13] 1(c) and (d), respectively [9]. The former is a spec- reported that the magnitude of the GMR e!ect in trum for an Fe interface covered by a Mn layer MBE grown epitaxial Fe/Cr(0 0 1) superlattices de- (Mn-on-Fe). The line pro"le is similar to that of the pends on both the vertical roughness amplitude Fe/V interface, but the degree of intermixing is (as obtained from synchrotron X-ray di!raction, supposed to be somewhat less than the Fe/V case. XRD) and the lateral roughness, i.e. the step density In contrast, the spectrum for another kind of n Fe/Mn interface, where the Fe probing layer was   (as determined from CEMS). A linear increase of the GMR with the product, ;n deposited on the surface of Mn prior to the thick  , was found. This result is based on the model by Landes et al. Fe deposition (Fe-on-Mn), was found to be con- [14] and Klinkhammer et al. [15] (which is sup- siderably di!erent. The spectrum includes a central ported by calculations [16,17]), where the hyper"ne line corresponding to a fairly large non-magnetic "elds of the components of the CEMS spectra are fraction of Fe atoms, which means di!usion of Fe characteristic for Fe probe atoms on di!erent atoms into Mn layers for more that several atomic sites at the interfaces or inside the bulk of the layers. layers. The extent of mixing at an interface depends Based on this model, the intensities of the spectral greatly on the procedure of sample preparation. In components observed in Ref. [13] were assigned to a multilayer structure built by successive depos- the abundances of the respective Fe site as fol- itions, it can happen that the structures at the top lows: (i) 33 T to bulk positions (n and bottom interfaces of each layer are di!erent,  ), (ii) 29 and 25 T to sites at interface steps (n and then the compositional modulation has a uni-  ), and (iii) 19 T to sites at atomically #at interface parts (n directional pro"le with respect to the "lm growth ). A re- duced ground-state hyper"ne "eld of 25.8 T for the direction. Thus, the degree of intermixing may be "rst Fe layer at the #at interface was also observed estimated sensitively by analyzing the hyper"ne for Fe(1 1 0)/Cr(1 1 0) sandwiches [18]. The model "eld distribution observed by Fe MoKssbauer by Landes et al. [14] is also supported by CEMS spectroscopy. A MoKssbauer analysis of epitaxially results on sputtered Fe/Cr multilayers [19,20]. grown Fe/Ag interfaces has been reported by In Ref. [13] the type of the interfaces, Cr-on-Fe Schurer et al. [10,11]. (upper) or Fe-on-Cr (lower), could not be distin- An issue derived from GMR studies is the rela- guished, but were assumed to be the same. Our tion between spin-dependent scattering probability CEMS spectra in Fig. 2 demonstrate, however, that and interface roughness. It is indeed interesting to the structure of both interfaces is remarkably di!er- apply the MoKssbauer techniques to this problem, ent. The CEMS spectra in Fig. 2 were measured on and already several arguments have been presented epitaxial Fe/Cr(0 0 1) superlattices including two [12]. However, the results are not consistent with monolayers (ML) thick Fe-interface probe layers each other and the optimum interface condition for grown in UHV on MgO(0 0 1) at ¹ the enhancement of spin-dependent scattering is "433 K. The composition was MgO(0 0 1)/Cr(50 As)/ not clari"ed yet. MoKssbauer spectroscopy gives us [ Fe(12 ML)/Fe(2 ML)/Cr(8ML)];15 (upper) T. Shinjo, W. Keune / Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 601 Moreover, there is a higher P(B ) intensity at lower hyper"ne "elds (below &18 T) for the upper (5%) than for the lower (1%) interface. Such low B values indicate Fe atoms with many Cr- neighboring atoms, i.e. some Fe}Cr intermixing. All of these reasons make us to conclude that the lower interface is atomically #atter and shows much less interdi!usion than the upper interface. This agrees with observations by Heinrich et al. [21] and Davies et al. [22] for growth of Cr on Fe. The interfaces described so far are real interfaces in the sense that they are not ideally sharp but exhibit a certain degree of defects, interdi!usion, etc. From the theoretical side, the magnetic and electronic properties of ideally sharp interfaces have been the subject of advanced ab initio band struc- ture calculations which allow to compute from "rst principles the local spin and charge density in thin "lms of several atomic layers including their surface or interface atomic layer. From the spin density the magnetic moment per atom and the magnetic hf "eld can be obtained. In particular, Freeman and his group [23] have systematically performed com- Fig. 2. Room-temperature CEMS spectra from epitaxial putations of electronic and magnetic properties Fe/Cr(0 0 1) superlattices on MgO(0 0 1), MBE-grown at (including magnetic hf "elds) of thin "lms and ideal ¹"433 K, including 2 ML thick Fe-probe layers located either at the upper (Cr-on-Fe) or at the lower (Fe-on-Cr) interfa- interfaces, based on the full potential linearized ces. Right-hand side: corresponding hyper"ne-"eld distribu- augmented plane wave (FLAPW) method. Parti- tions, P(B ). (b) explains the meaning of uppera and lowera cularly, important is their theoretical discovery interfaces. [23] that the proportionality between local mo- ment, , and hf "eld, B , can fail at ideally #at surfaces and interfaces, contrary to the situation in the bulk. Famous examples are the calculated en- and MgO(0 0 1)/Cr(50 As)/[Fe(2ML)/ Fe(12 ML)/ hancement (relative to bulk values) of the magnetic Cr(8 ML)];15 (lower) ( Fe"Fe metal of moment in the "rst monolayer of free BCC- natural Fe abundance, 2.1%). The CEMS Fe(1 0 0) surfaces (of 31% [24]) and of free BCC- spectra as well as the corresponding distribution of Fe(1 1 0) surfaces (of 19% [25]). By contrast, the hyper"ne "elds, P(B ) of both samples are very calculated hf "elds in the surface layer are reduced di!erent (Fig. 2). The relative areas (relative inten- by 31% for Fe(1 0 0) [24] and by 7.4% for Fe(1 1 0) sities) of the various peaks observed in P(B ) cor- [25]. This discrepancy was explained [24] by respond to the relative abundances of di!erent Fe a change in sign of the hf "eld contribution of the sites at and near to interfaces. The main peak at 33 conduction electron polarization at the topmost T determines n  , which corresponds to 30% surface layer. From thin "lm magnetometry in (22%) in relative intensity for the lower (upper) UHV on free BCC-Fe(1 1 0) surfaces a ground- interface. Thus, there are more Fe atoms sensing state surface moment enhancement of 39% has no Cr atoms in the lower interface. The peak at been measured recently [26], which is about twice 19.6 T (which according to Ref. [13] determines as large as theory predicts [25]. There is a remark- n ) has a much larger relative intensity for the ably good agreement between theoretically pre- lower (21%) than for the upper (16%) interface. dicted and measured reduced surface/interface hf 602 T. Shinjo, W. Keune / Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 "elds (measured by CEMS) for two systems: scopy allowed the determination of the amount of (i) for the free BCC-Fe(1 1 0) surface, investigated Fe coverage of the substrate and the relative by Korecki and Gradmann [27], and (ii) for amount of Fe atoms that participate in the atomic the W(1 1 0)/Fe(1 1 0) interface, measured by Fe-Ag exchange process at the buried Ag/Fe inter- Przybylski et al. [28] and calculated by Hong et al. face [10]. In contrast to the bottom (Fe-on-Ag) [29]. Details are described in Gradmann's review interface, the top (Ag-on-Fe) interface was found to [30]. An excellent extensive review of CEMS in- be very sharp with 90% of the 1-ML Fe-probe vestigations, e.g., on epitaxial BCC-Fe(1 1 0) inter- atoms at the interface making direct contact with faces with di!erent metallic overlayers (Cr, Gd, Y), the Ag overlayer. The magnitude of the hf "eld was was given by Sauer and Zinn [31]. found to be 27.6 T at 300 K for Fe in #at terrace The concept of atomically #at interfaces used in sites [10], which is signi"cantly smaller than the the theoretical work is realized only in very few BCC-Fe bulk value at 300 K. Unfortunately, systems. The theoretically predicted e!ects will be Schurer et al. [10,11] could not measure the hf "eld suppressed by interface roughness and interdi!u- at low temperature; therefore, a reliable compari- sion, which cannot be avoided in many multilayer son with the calculated ground-state hf "eld at the systems. An example is the Fe/Cu(0 0 1) system, Fe/Ag interface [23] is not yet possible. where intermixing (Fe}Cu atomic site exchange processes) within the "rst Cu/Fe interface layer is 2.2. Metastable ultrathin xlms: FCC-Fe and known to occur [32]. For ferromagnetic 2 ML Fe FCC-Ni on Cu(0 0 1), a ground-state surface and subsurface \VFeV invar hf "eld of 24.9 and 31.6 T, respectively, is predicted A vast number of reports on the magnetic and [33]. Experimentally, a broad distribution of inter- structural properties of the FCC-Fe/Cu(0 0 1) sys- face hf "elds extending from &12}35 T with an tem is found in the literature [37}39]. FCC-Fe does average hf "eld of &22 T was observed [34] by not exist in bulk form at low temperatures. Epitaxy CEMS in UHV at 55 K (near magnetic saturation), o!ers the possibility to stabilize crystallographic which is very di!erent from the predicted behavior. phases which do not exist as bulk materials under As described in Ref. [31], there is also a discrepancy normal conditions. As FCC-Fe is lattice matched in the hf "elds for the Fe/Ag interface, for which to a Cu substrate with a small mis"t of &0.7% a 4% enhancement (with respect to the bulk values) only, the structure and magnetic properties of has been observed at low temperature, while calcu- metastable epitaxial FCC-Fe ultrathin "lms on lations predict a 7% reduction. Also in this case Cu(0 0 1) have become the subject of numerous interface roughness might play a role. Keavney investigations. The reason for the general interest in et al. have studied [Fe(1 0 0)/Ag(1 0 0)] multi- this system is its delicate interplay between struc- layers by using a 2-ML Fe probe layer [35]. Liu ture and magnetism [40], with magnetism being and Gradmann [36] measured MoKssbauer spectral sensitive to changes in the lattice parameter (or parameters of W(0 0 1)/Fe/Ag structures using one Wigner}Seitz radius). So far the majority of invest- Fe atomic layer as a probe. Schurer et al. [10] igations was performed on MBE grown FCC- performed a careful and very detailed CEMS analy- Fe/Cu(0 0 1); recently, however, intriguing new sis of the MBE growth of a 1 ML thick Fe probe properties of the FCC-Fe/Cu system prepared by layer in BCC-Fe(1 0 0)/Ag(1 0 0) structures on pulsed laser-deposition (PLD) have been reported a single-crystal Ag(1 0 0) surface. Di!erent Fe sites [41,42]. Moreover, FCC-Fe ultrathin "lms have with di!erent nearest-neighbor (Fe, Ag) con"gura- been stabilized on diamond C(0 0 1) [43] and tions at and near the stepped Fe/Ag interface could Cu be observed (e.g., magnetic Fe positions in the cen- Al(0 0 1) substrates [44]. During the 30 years of research on MBE-grown FCC- tral region of terraces, at the edge of the Fe terraces Fe/Cu(0 0 1) "lms after the discovery of epitaxy for (top or bottom sites), in the "rst and second this system [45,46] many seemingly contradictory subinterface layers, and paramagnetic Fe interface- experimental results have been reported in the exchanged with Ag atoms). MoKssbauer spectro- literature. However, in the past years a complex T. Shinjo, W. Keune / Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 603 structural and magnetic phase diagrama (with "lm thickness t and growth temperature ¹ as para- meters) has been suggested [47,48] for MBE grown FCC-Fe/Cu(0 0 1) which (at least for room-temper- ature growth, ¹+300 K) is now widely accepted and has been con"rmed by various experimental methods, including in situ CEMS in UHV [37]. It is generally accepted that such Fe "lms on Cu(0 0 1) prepared at ¹+300 K in the 2}4 ML thickness range are ferromagnetic (FM) with a high Fe magnetic moment of &2 (and with a large satu- ration hyper"ne (hf) "eld of &31}34 T [37], acci- dentally similar to that of BCC-Fe), and posses the fct structure, including some vertical and lateral &buckling' of Fe atoms [49,50]. Moreover, for Fe on Cu(0 0 1) prepared at ¹+300 K in the &5}10 ML range, the interior of the "lms shows the undistorted FCC structure, is paramagnetic at RT and antiferromagnetic (AFM) with a low mag- netic moment (&0.5}0.7 ) and low hf "eld of &1}2 T at low temperature, while ferromagnetism is restricted to the "lm-vacuum surface region, where surface-atom positions relax towards a local FCT structure [49,50]. Fig. 3. Site-selective CEMS spectra at 300 K of room-temper- Site-selective (Fe-probe layer) CEMS provided ature grown Fe probe layers in natural FCC-Fe/Cu(0 0 1): the "rst direct observation of surface-restricted Fe(2 ML)/ Fe(5 ML)/Cu(0 0 1) (top);  Fe(2 ML)/Fe(3 ML)/ magnetic order and paramagnetism in the interior  Fe(2 ML)/Cu(0 0 1) (center); and  Fe(5 ML)/Fe(2 ML)/ of 300 K grown 7 ML Fe "lms on Cu(0 0 1) [37]. Cu(0 0 1) (bottom). [The sharp sextet subspectrum (top) and the This is demonstrated in Fig. 3 where site-selective two lines at !0.9 and #0.7 mm/s (center and bottom) are an artifact due to the -Fe coated sample-holder frame in this case]. CEMS spectra of 2 ML thick Fe-probe layers in Notice the di!erent velocity scale (top). (From ref. [37]). otherwise natural Fe/Cu(0 0 1) "lms (of 7 ML total thickness) are shown: The Fe-probe layer was deposited either on top of the "lm (upper spec- trum), in the "lm center (center spectrum) or at the Fe/Cu(0 0 1) interface (bottom spectrum). At the magnetic (undistorted) FCC-Fe appears at 300 K, surface the dominant spectral contribution involves superimposed to a (less intense) asymmetrical a broad distribution of magnetic hyper"ne "elds, quadrupole-split doublet of splitting E P(B /" ), with an average "eld 1B 2 of &18 T. Ob- 0.63 mm/s. The relative spectral intensity of the viously, the large majority of the Fe nuclei at the doublet is clearly enhanced when the Fe-probe "lm surface experiences high B values ranging layer is placed at the Fe/Cu interface as compared from &10 to &35 T at 300 K; hence these Fe to the "lm center (Fig. 3). Therefore, the quadrupole atoms are in a probably ferromagnetic high- doublet was assigned to the Fe/Cu(0 0 1) interface moment (or high-spin, HS) state. The distribution (which is paramagnetic at 300 K), where the elec- P(B ) of the surface does not appear in the center or tronic charge distribution is anisotropic and gives at the Fe/Cu interface (Fig. 3), i.e. the spectrum of rise to an electric "eld gradient at the Fe nucleus. the "lm center and of the Fe/Cu interface do not It is observed that the relative doublet intensity exhibit a magnetic component at 300 K. Instead, amounts to 43% of the total spectral intensity at a dominant sharp central single line due to para- the interface; this means that e!ectively 0.86 ML of 604 T. Shinjo, W. Keune / Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 the (in total) 2 ML thick Fe-probe layer at the Fe/Cu interface sense the in#uence of Cu atoms, implying that the chemical roughness of the Fe/Cu interface is limited to &1 ML Fe only [37]. Since the development of a magnetic hyper"ne "eld upon cooling below the magnetic ordering temperature is independent of the type of magnetic ordering (regardless whether ferromagnetic or antiferromagnetic), MoKssbauer spectroscopy is presently the only technique with monatomic layer sensitivity for determining the NeHel temperature, ¹,, of AFM FCC-Fe "lms in the &5}10 ML range [37,51]. The FCC-Fe single line broadens considerably in the AFM state at low temperatures due to the unresolved magnetic hyper"ne splitting Fig. 4. The FCC-Fe linewidth (FWHM) vs temperature for in the AFM phase [37,51}53]. The average satura- &5 ML Fe grown at 300 K on Cu(0 0 1) (full circles) and on tion hyper"ne "eld observed in the AFM FCC-Fe CuAl(0 0 1) (open circles). (From ref. [34]). state is &1-2 T as compared to the natural linewidth of &0.8 T for Fe. This low hyper"ne "eld indicates that AFM FCC-Fe "lms are in a low moment (low-spin, LS) state. The onset of line broadening upon cooling determines the NeHel tem- perature, as demonstrated in Fig. 4 for a &5 ML thick FCC-Fe "lm on an atomically #at well-or- dered Cu(0 0 1) surface with ¹,&70 K. For com- parison, the single line of a &5 ML FCC-Fe "lm grown at 300 K on a CuAl(0 0 1) single-crystal surface with an expanded lattice parameter (1.6% relative to that of Cu) does not show signi"cant line broadening upon cooling down to 35 K (Fig. 4) [34]. Therefore, ¹, of FCC-Fe/CuAl(0 0 1) is considerably reduced (¹,)35 K), as compared to FCC-Fe/Cu(0 0 1). The reason might be the ex- panded CuAl lattice, although the e!ect of pos- sible Al impurities in the FCC-Fe "lm cannot be excluded [34]. A plot of the saturation hyper"ne "eld B versus the Wigner}Seitz radius r Fig. 5. Hyper"ne "eld near magnetic saturation vs Wigner-Seitz  (in atomic units) for radius, r di!erent FCC-like Fe system is shown in Fig. 5 , for FCC-Fe-like systems [34]: Full square: 300 K grown 5-10 ML FCC-Fe/Cu(0 0 1) measured at 35 K. Full [34] (Note that r"2.67 a.u. for Cu). As B is circle: 300 K or 100 K grown 3 ML thick Fe/Cu(0 0 1) and roughly proportional to the local Fe magnetic mo- 100 K grown 7 ML thick Fe/Cu(0 0 1), measured at 40}55 K. ment, Fig. 5 re#ects also the behavior of the local Full triangle: 300 K grown 5 ML thick Fe/CuAu(0 0 1), moment versus r measured at 30 K [54]. Open circles: 275 K grown . Thus Fig. 5 demonstrates that Fe/Cu there is a transition from an AFM low-spin Fe state \Au(0 0 1) multilayers, measured at 15 K [55]. Open triangles: Fe precipitates in bulk Cu to a FM high-spin Fe state around r \VAlV matrices, measured  &2.69 a.u., at 4.2 K [54]). Open square: Fe precipitates in bulk CuAl which is in agreement with theoretical predictions matrix at 58 kbar pressure, measured at 4.2 K (cited in ref. [54]). for bulk FCC-Fe [40]. In Fig. 5 also data for Full diamond: 475 K grown hcp Fe/Ru(0 0 0 1) superlattice, ultrathin FCC-Fe-like "lms on Cu measured at 4.2 K [56]. Au(0 0 1) [54], T. Shinjo, W. Keune / Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 605 FCC-like Fe/Cu\VAuV(0 0 1) multilayers [55], the low-spin AFM phase that was found to have and for FM HCP-Fe/Ru(0 0 0 1) superlattices [56] a NeHel temperature around 35 K. Apparently, the are included; furthermore, data from AFM FCC-Fe AFM phase is invisible for SQUID magnetometry, precipitates in Cu\VAlV matrices have been ad- and the average magnetization drops, as more of ded in Fig. 5 [54]. These results represent quite well the AFM phase emerges with increasing Fe concen- the trends found in theoretical hf "elds for bulk tration [67]. FCC-Fe by Herper et al. [57], calculated by the It is tempting to investigate FCC-Fe "lms grown method of BluKgel et al. [58]. The hf "elds calculated pseudomorphically on substrates with di!erent ex- by the embedded cluster method for FCC-Fe with panded lattice parameters in order to scan the a simple AFM spin structure [59] deviate consider- in-plane lattice parameter around its value for the ably from those in Ref. [57]. magneto-volume instability (low-spin/high-spin The magnetic and structural properties of transition) described above (Fig. 5). This may be FCC-Fe are related to the Invar e!ect [60], i.e., performed e.g., by choosing Cu anomalies in thermal expansion and other physical \VAuV substrates which show an increasing lattice parameter with properties of Fe-based FCC alloys, for instance, rising Au concentration [55,68,69]. A MoKssbauer- FCC-Ni\VFeV alloys near x"0.65. According to e!ect study by Keavney et al. [55] of FCC- Weiss' phenomenological model [61], the Invar Fe(1 0 0)/Cu e!ect originates from thermal transitions between \VAuV(1 0 0) multilayers with lattice parameters in the range between 3.606 and 3.704 As a ferromagnetic (FM) high-spin Fe state with demonstrated the coexistence of the low-spin AFM a large atomic volume and an energetically higher and the high-spin FM Fe phase. Most interestingly, lying antiferromagnetic (AFM) low-spin Fe state the hyper"ne "elds (and the local Fe atomic mo- with a small atomic volume. Indeed, modern elec- ments) themselves of these two Fe sites do not tronic band-structure calculations of the magnetic change, respectively, with the substrate lattice para- ground state in bulk Ni}Fe Invar alloys predicted meter, but it is the population of the high-spin FM the existence of these two Fe states [62}64]. Experi- phase that was found to increase relative to that of mentally, in bulk FCC-Ni\VFeV alloys, the study the low-spin phase, as the lattice parameter of the of the region of high Fe concentration is hampered substrate is increased [55]. This e!ect causes the by the onset of a structural phase transition to the continuous increase in the average moment per Fe BCC-Ni}Fe phase (martensitic transformation). In atom [55,68,69], as measured by macroscopic mag- epitaxial FCC-Ni\VFeV "lms, however, the mar- netometry. Remarkably, hyper"ne "elds (or Fe tensitic transformation can be suppressed by atomic moments) in FCC-Fe intermediate between pseudomorphic growth on Cu substrates over the those of the low-spin AFM and high-spin FM entire Fe concentration range [65,66]. Walker FCC-Fe phases (see Fig. 5) have never been ob- and his group [67] investigated epitaxial FCC- served convincingly, which probably indicates Ni\VFeV alloy thin "lms on Cu(0 0 1) and Cu(111) a "rst-order transition between these two phases, as utilizing a combination of MoKssbauer spectroscopy predicted by theory [40]. and magnetometry; their results presented clear evidence for the emergence of a low-spin AFM 2.3. Perpendicular spin structures phase in the region of high Fe content (Invar re- gion), which coexists with a high-spin FM phase For the studies of magnetism in thin "lms, the [67]. The emergence of the AFM phase was ob- direction of spontaneous magnetization is a valu- served by magnetometry as a drop in the average able information. If the shape anisotropy is domi- saturation moment, while low-temperature MoKs- nant, the magnetization is preferably oriented in sbauer spectroscopy revealed the emergence of this the "lm plane. However, it is not uncommon to phase by the appearance of a drastically broadened observe an o!-plane component of magnetization. central single-line superimposed to the sextet of the Films having perpendicular magnetizations are of high-spin FM phase. The high-spin FM phase may interest from viewpoints of basic magnetism and exist as superparamagnetic clusters in a matrix of also of technical applications. The average spin 606 T. Shinjo, W. Keune / Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 Fig. 6. Fe MoKssbauer absorption spectra at 4.2 K for [Fe(40 As)/RE(30 As)] multilayers (RE"rare earth). The average direction 1 2 of the Fe magnetic moments relative to the "lm normal is shown for each spectrum. direction is easily estimated from the MoKssbauer angular momentum of interface rare-earth hyper"ne spectra. The intensity ratio of six lines in atoms. a magnetically split Fe spectrum is expressed as Nanoscale chemically modulated Tb/Fe and 3 : x : 1 : 1 : x : 3 and cos "(4!x)/(4#x), other RE/Fe multilayers with rather thin individual where is the polar angle between the direction of layer thickness ((&20 As) have been extensively the Fe magnetic moment and the -ray beam (i.e. investigated, mainly because they may exhibit per- the "lm normal). The measured values of cos pendicular magnetic anisotropy at room temper- are spatial averages of 1cos 2 over the Fe layers. ature [71]. RE/Fe multilayers (in particular Tb/Fe, Usually, a half-cone angle 1 2 is de"ned by arccos Dy/Fe) with thicker individual layer thicknesses (1cos 2). In a powder pattern with random ('&26 As) have crystalline HCP/BCC structure spin orientation, the ratio is 3 : 2 : 1 : 1 : 2 : 3, but in and may show a reversible temperature-driven Fe- the case of perfect perpendicular (in-plane) mag- spin reorientation transition 1 2(T) from parallel netization, the ratio becomes 3 : 0 : 1 : 1 : 0 : 3 to perpendicular spin orientation upon cooling. (3 : 4 : 1 : 1 : 4 : 3). As examples, spectra for Fe/ This phenomenon has been discovered by Fe rare-earth multilayers are shown in Fig. 6 [70]. It is MoKssbauer spectroscopy in HCP-Dy/BCC-Fe observed that the ratio for the most cases is multilayers for the "rst time [72], and was later 3 : 4 : 1 : 1 : 4 : 3, indicating the magnetization being studied in other HCP-RE/BCC-Fe systems [70,73], in the plane. However, the intensity ratios in the and also by Dy MoKssbauer spectroscopy [74]. spectra for Pr, Nd and Tb are di!erent, suggesting Spin-reorientation transitions are of fundamental the magnetization is oriented close to the "lm nor- interest because of the competing magnetic inter- mal. Although all the samples with the same nom- actions involved [75,76]. Since the interface rough- inal structure were prepared in the same procedure, ness plays an important role in the origin of and have similar crystallographic qualities, the perpendicular magnetic anisotropy, Tb/BCC-Fe e!ective anisotropies of Fe layers are di!erent, multilayers have been ion-beam irradiated with depending on the counterpart rare-earth ele- heavy ions in order to modify the interface and ments. The reason is attributed to the orbital BCC-Fe structure by demixing, mixing and T. Shinjo, W. Keune / Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 607 amorphization [77}79]. Depending on the type of ions and the irradiation dose employed enhance- ment or reduction of the perpendicular Fe-spin orientation was observed by CEMS. 2.4. Magnetization process and domain structure Fe-site-selective detection of magnetic domains and calculation of an Fe-speci"c magnetic hyster- esis loop generally may be achieved by MoKssbauer spectroscopy in an external "eld, B , in such cases where the coercivity B is larger than the experi- mental linewidth +0.8 T, and the system shows a strong uniaxial magnetic anisotropy. This applies in particular to Fe-containing modern hard- magnet systems either as thin-"lm/multilayer or bulk materials. This is demonstrated for HCP- Tb/BCC-Fe multilayers at 50 K, where they exhibit strong perpendicular magnetic anisotropy and B +1.2 T [80]. Similar results have been obtained at 4.2 K [80]. Polycrystalline HCP-Tb/BCC-Fe multilayers of composition [ Fe(17.5 As)/ Fe(5 As)/ Fe(17.5 As)/Tb(26 As)];30 and capped by Al(50As) were prepared by thermal evap- oration of high-purity elements in UHV (p)3;10\ mbar during deposition, rate 0.2 As/s) onto liquid-nitrogen cooled polyimid (Kapton) substrates for MoKssbauer transmission experi- Fig. 7. Fe(5 As)-probe layer MoKssbauer spectra of [Tb(26 As)/ Fe(40 As)];30 measured at 50 K along the hysteresis loop in ments. The MoKssbauer signal from the natural Fe various perpendicular "elds, B layers ( Fe) is negligible, and the signal arises , in the sequence indicated on the right-hand side. The Fe-probe layers are in the center of essentially from the center of the BCC-Fe layers. the 35 As thick natural Fe layers. Thus we avoid possible spectral contributions from intermixed interface regions [81,82]. Low-angle XRD patterns of our multilayers typically exhibited four or more superstructure Bragg peaks [83] in- of 263 is obtained at remanence, i.e. the average dicating good layer quality. MoKssbauer transmis- Fe-spin orientation is strongly (but not completely) sion experiments were performed at 50 K in perpendicular to the plane in the center of the external "elds of !5 T)B )#5 T parallel or BCC-Fe layers. Even at saturation (#5 T) a "nite antiparallel to the -ray direction and to the "lm angle 1 2 of 223 remains. At B normal. "#1.5 T, a splitting into two sextets with sharp lines and Typical MoKssbauer spectra measured along with di!erent &e!ective' "eld values, "B! a hysteresis loop at di!erent "elds from B ", are ob- "#5 served, representative of &up' domains (#) and to !4 T and back to #1.5 T are shown in Fig. 7. downa domains (!) respectively, of the perpen- As expected, at remanence (B "0) and at satura- dicular multidomain state. The reason for the split- tion (B "#5 T) only one magnetically split sex- ting is the vectorial relation B tet with sharp lines typical for BCC-Fe is observed; "B #B  between the measured e!ective "eld at the Fe the sample is in a single-domain state in this case. nucleus, the intrinsic hyper"ne "eld, B From the line-intensity ratio a half-cone angle 1 2 , and B . As B is related to $ by B "A $ (where 608 T. Shinjo, W. Keune / Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 A"!15 T/ for BCC Fe), two di!erent $ ori- which changes with B entations (corresponding to &up' or &down' do- . This provides the basis for calculating an Fe-speci"c magnetization hysteresis mains) with respect to B  result in the two "B! " loop [80] from measured values of A!, "B! values observed. The relative spectral area (relative " and cos1 2 intensity) A>(A\) of the sextet corresponding to !. The result is shown in Fig. 8, data (a), together with the perpendicular magnetization upa ( downa) domains is equal to the Fe-speci"c loop obtained at 50 K from the same sample by volume fraction of these perpendicular domains, SQUID magnetometry (data (b)). Comparison of both curves provides important information: (i) they have about the same coercivity, and (ii) at saturation, the average magnetization in (b) is only 0.7 /Fe-atom, compared to 2.1 /Fe-atom in (a); this demonstrates that 1.4 /Fe-atom are compen- sated at the Tb/Fe interfaces at 50 K by antifer- romagnetic coupling with Tb moments. If domain structures are formed in a magnetic "lm in zero external "eld, the direction of magneti- zation in a "lm is not uniform. For example, the direction of magnetization in a closure-domain structure will vary depending on the depth from the surface. Depth-dependent analysis of the magnetiz- ation direction using Fe probe layers was carried out for a [Co(30 As)/Au(20 As)]xn system [84]. For MoKssbauer absorption spectroscopy, Fe probe Fig. 8. Perpendicular magnetization loops at 50 K: (a) full circles: Fe site-selective loop calculated from MoKssbauer data layers with a nominal thickness of 1.5 As were obtained from Fig. 7. (b) open circles: SQUID magnetization located in the center of 30 As Co layers at di!erent loop [80]. depths in di!erent samples. As shown in Fig. 9, the Fig. 9. Concept of magnetic structure in the multilayer system [Co(20 As)/Au(20 As)];31. Examples of MoKssbauer spectra from position- selectively located Fe probes are also shown. T. Shinjo, W. Keune / Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 609 line-intensity ratios in the MoKssbauer spectra indi- cate clearly the di!erence of magnetization direc- tions between the interface and the middle of the multilayer system. The magnetization is closer to the "lm plane at interface layers and turns towards the perpendicular direction at the inner layers. This result provides evidence for closure domain forma- tion. More details related to the dependence of MoKssbauer spectra on the depth from the interface, the number of magnetic layers and on externally applied magnetic "elds will be presented in a forth- coming paper. 2.5. Artixcial layered structures: FePt(0 0 0)L1 xlms Thin "lms of the chemically ordered face- centered tetragonal (FCT) phase of FePt, FePd [85] or (metastable) FeAu [86] with the L1 struc- ture (or CuAu(I) structure) have become a topical subject. This structure consists of alternating mon- atomic layers of the elements with the c-axis of the FCT unit cell perpendicular to the layers. Epitaxial Fig. 10. CEMS spectra of L1 growth of L1 -ordered FePt systems at -ordered FePt thin "lms on 300 K: (a) 250 As epitaxial FePt(0 0 1) alloy film. (b) Epi- MgO(0 0 1) with the c-axis along the "lm-normal taxial (0 0 1)-oriented [Fe(1.5 As)/Pt(2.0 As)];60 superlattice. direction by molecular-beam epitaxy (MBE) [87] (c) (1 1 1)-textured (non-epitaxial) [Fe(1.5 As)/Pt(2.0 As)] or magnetron sputtering [88,89] results in high multilayer grown without Fe(3 As) seed layer. perpendicular magnetic anisotropy. Because of this property and its large magneto-optical Kerr rota- tion [87], FePt is considered to be a potentially useful storage medium for high-density magneto- grow epitaxially, but showed a clear FePt(1 1 1) optical perpendicular recording. texture. MoKssbauer spectroscopy o!ers a convenient Fig. 10 shows the CEMS spectra of the Fe}Pt way to detect the presence of the L1-phase and samples measured at room temperature. The inci- of other Fe}Pt phases in this system. For this dent gamma radiation was perpendicular to the purpose we have studied 250 As thick chemically "lm plane. All spectra in Fig. 10(a)}(c) could be ordered (L1) FePt alloy "lms and analyzed in terms of a dominant magnetically split [Fe(1.5 As)/Pt(2.0 As)] superlattices grown by Zeeman sextet with slightly asymmetric line posi- MBE at 5003C on MgO(0 0 1). Initially, the tions which is assigned to the epitaxial (FCT) FePt MgO(0 0 1) surface was covered at 5003C by a 3 As phase with L1 Fe seed layer followed by a 150 A  structure, and a less-intense central s Pt(0 0 1) bu!er single line due to some cubic non-magnetic (minor- layer. All samples were capped with 10 As Pt at ity) Fe}Pt phase. The relative spectral intensity of 5003C for protection. High-angle XRD patterns this non-magnetic phase amounts to 3.2% of the exhibited clear FePt(0 0 1)- and (0 0 3) Bragg re#ec- total spectral intensity in Fig. 10(a), 9.3% in tions from the L1 superstructure, in addition to Fig. 10(b) and increases to 23.2% in Fig. 10(c). The the basic FePt(0 0 2) and (0 0 4) Bragg peaks, non-magnetic Fe}Pt phase was found to order typical for epitaxial growth. Similar multilayers magnetically at low temperature with a hyper"ne prepared without Fe-seed layer, however, did not "eld di!erent from that of the L1 phase. 610 T. Shinjo, W. Keune / Journal of Magnetism and Magnetic Materials 200 (1999) 598}615 Spectral analysis yields an average hyper"ne "eld of 27.5 T for the FCT (L1-ordered) FePt phase at RT. The outer sextet lines are slightly broader than the inner lines, in particular in the case of the (1 1 1) textured "lm (Fig. 10(c)). This is the result of a dis- tribution of hyper"ne "elds (which is relatively narrow and sharply peaked, though). The P(B ) distribution re#ects small deviations from the ideal short-range-order parameter of the perfect L1 phase. The latter has one unique Fe-lattice site only, and thus should show a sharp B value and no distribution. From the relative line intensities of the spectra in Fig. 10(a) and (b) a tilting angle 1 2 of &193}213, i.e. nearly complete perpendicular Fe-spin orientation, was obtained for the epitaxial "lms. The angle 1 2 deduced from Fig. 10(c) is 603 for the (1 1 1)-textured multilayer. This is close to the angle of 54.73 between the preferred cry- stallographic 11 1 12 direction along the "lm normal and the tilted c-axis directions in this tex- tured system. The non-cubic (FCT) L1 structure with its an- isotropic electronic charge distribution manifests Fig. 11. Au MoKssbauer spectra of [Au(10 As)/3d-metal] itself in a strong electric quadrupole interaction multilayers at 16 K: (a) [Au(10 As)/Fe(8 As)]; (b) [Au(10 As)/ with the Fe nucleus. This is re#ected by asymmet- Co(20 As)]; (c) [Au(10 As)/Ni(20 As)]. rical line positions in the Zeeman sextet [90]. Spec- tral analysis combined with the assumption that the main component of the electric "eld gradient,