Forty- and fifty-repeat multilayers have been grown with tFe = 14, 20, 30, 40, and 50Å and t Si = 14 and 20Å. Magnetization curves for 50-repeat (Fe30Å/Si20Å) and (Fe30Å/Si14Å) multilayers grown on glass at nominal RT (about +60°C) are shown in Figure 2. On the y-axis of this plot is the magnetic moment of the multilayer normalized to the moment of an equivalent volume of bulk Fe. The magnetization curve of the (30/20) multilayer looks much like that of an Fe film, while the magnetization curve of the (30/14) multilayer shows the high saturation field and low remanence which characterize AF interlayer coupling. At its saturation field the magnetization of the (30/14) multilayer is about the same as for the (30/20) multilayer. Both of these films have moments only about half as large as an equivalent volume of bulk Fe. Our observation of AF coupling for Si thicknesses between 10 and 20Å and the disappearance of coupling for Si thicker than 20Å confirm previous observations on magnetron-sputtered films.
X-ray diffraction spectra for these multilayers are shown in Figures 3 and 5. Figure 3 shows the small-angle x-ray scattering (SAXS) data with peaks at angles
n2 lambda2 = 4 Lambda2 sin2theta + 2 delta
where lambda is the x-ray wavelength, Lambda == tFe + tSi is the multilayer bilayer period and delta is the index of refraction for x-rays. This grazing incidence data gives information about the quality of the multilayer interfaces. Figure 3 shows four low-angle peaks for both films, indicating a reasonably strong composition modulation along the growth direction. (The higher frequency oscillations between 1° and 4° are finite-thickness fringes from the Ge cap layer.) The most notable difference between the two spectra is that the multilayer peaks are broader for the AF-coupled t Si = 14Å film, indicating more fluctuations in bilayer period and probably more interface roughness. Using the spacing between peak positions to eliminate the unknown delta from Eqn. 1 gives values of the bilayer period lambda for the two films. For the multilayer with nominal layering of (Fe30Å/Si20Å)x50, the derived value for lambda is (41.82 ± 0.07)Å, while for the (Fe30Å/Si14Å)x50 film lambda = (38.10 ± 0.04)Å. lambda is 8.2Å shorter than the nominal value for the t Si = 20Å film, and 5.9Å shorter than nominal for the t Si = 14Å film. Although some of the discrepancy between the nominal and observed bilayer period may be due to calibration inaccuracies, most is undoubtedly due to intermixing of the Fe and Si layers, in keeping with observations in the other metal/Si multilayers.[10,14] Throughout this paper we will continue for convenience to refer to the films in terms of their nominal layer thicknesses.
Comparison of the magnetization data to the x-ray data can give some further insight into the question of intermixing. Because of the presumed interdiffusion of the Fe and Si layers, the magnetic moment of the Fe layers is also reduced from the nominal value. The missing magnetic moment can be expressed as an equivalent thickness of Fe. Figure 4 shows a plot of missing moment in units of Å ngstroms of Fe versus missing bilayer period determined from multilayer peak positions in SAXS for films grown at room temperature (RT). The plot shows that while the diffusion-induced reduction in bilayer period varies between 1 and 8Å, the missing Fe moment per bilayer (for both interfaces) is consistently between 10 and 12Å. The one outlier in Figure 4 is for a film which had t Fe = 20Å, the thinnest Fe for which we have ever observed interlayer coupling. Other groups have previously observed a moment reduction of 12-14Å per bilayer in polarized neutron reflectivity measurements on uncoupled Fe/Si multilayers with thick Si layers.[29,30]
The disparity between the magnetic moment reduction and the bilayer period reduction numbers may at first appear to be puzzling. This disparity occurs because the moment and bilayer period are affected by different aspects of the structure. In calculating the moment reduction in Å the assumption has been made that the Fe layer has the magnetization of bulk Fe. This is equivalent to assuming that there is no Si in the Fe layer, which is undoubtedly false. In calculating the missing bilayer period, the assumption has also been made that the spacer layer is pure Si, also clearly false. The fact that the missing magnetic moment is almost constant irrespective of the reduction in bilayer period suggests that the spacer layer is non-magnetic independent of Si thickness. The lack of variation of the missing moment is then explained by the diffusion of a constant number of iron atoms into the silicon layer, irrespective of its thickness. The wide variation of the measured bilayer period is most likely related to the varying orientation and crystallinity of the spacer layer, neither of which affects the magnetic moment if the spacer itself is non-magnetic.
Figure 5 shows the high-angle x-ray spectra where peak positions give information about the orientation and crystallinity of the films. The intense peak near 70° in this plot is due to the Si substrate. Included are data for an (Fe40Å/Si14Å)x40 antiferromagnetically coupled multilayer and for an (Fe30Å/Si20Å)x40 uncoupled multilayer, both grown on oxidized Si(001) at RT. The peaks for the (40/14) film are narrower than for the (30/20). The Scherrer formula gives 78Å or about two bilayer periods for the coherence length of the (40/14) film and 34Å or about one bilayer period for the coherence length of the (30/20) film. Coherence lengths in IBS-sputtered antiferromagnetically coupled films are often as long as 200Å. Fullerton et al. have inferred that the spacer layer in thin-Si Fe/Si multilayers must be crystalline based on their observation of coherence lengths longer than a bilayer period. In keeping with its superior crystallinity, the (40/14) multilayer has one superlattice satellite on the low-angle side of the Fe(002) peak. Typically only one satellite on the low-angle side of the Fe (011) or (002) x-ray peak is observed for polycrystalline multilayers grown on glass, in agreement with observations by Foiles et al.
The thin-Si multilayers which have AF coupling usually show a mixed  and  orientation when grown on glass substrates at RT. Occasionally t Si = 14Å films with a pure (011) orientation are obtained at RT. The variation in texture may be due to changes in film stress under slightly different deposition conditions. Stress induced during deposition has been postulated to explain the mixed Mo texture found in Mo/Ge multilayers. In contrast to the thin-Si case, the thicker-Si Fe/Si multilayers which do not show interlayer coupling always have a pure (011) texture. Since the (011) plane is close-packed for the bcc crystal structure, one would expect the (011) orientation to be energetically favored for the Fe in a multilayer with amorphous Si. Films grown at nominal RT on glass or oxidized Si substrates typically had rocking curves about 10° wide indicating a moderate amount of orientation.
Transmission electron microscopy (TEM) has been used to further investigate the morphology of the films. TEM cross-sectional images of an (Fe30Å/Si20Å)x50 multilayer and an (Fe40Å/Si14Å)x50 multilayer grown during the same deposition run are shown in Figures 6a and 6b, respectively. The most salient features of the (30/20) multilayer are the long lateral continuity of the layers and the smoothness of the interfaces. Since there is no interlayer coherence in the (30/20) film, the crystalline grains have a high aspect ratio. The (40/14) multilayer also has long, continuous layer planes but has rougher interfaces, consistent with the SAXS data.
Transmission electron selected-area diffraction patterns for the (30/20) and (40/14) films are shown in parts c and d of Figure 6. The (30/20) films shows only a Fe(011) ring, consistent with the high-angle x-ray diffraction scans. The (40/14) film, on the other hand, displays spots corresponding to the (011) and (002) reflections seen using x-rays. The presence of spots rather than rings in the (40/14) image implies the presence of large, oriented crystallites in the film. Most interestingly, the (40/14) image includes a faint spot near what would be the Fe(001) position were the Fe(001) peak not forbidden by symmetry in the bcc crystal structure. The (001) peak is allowed in the B2 and DO 3 crystal structures. The B2 structure is found in the equilibrium phase diagram only at 10-22% Si range of composition, but workers at ETH have grown this crystal structure throughout the range of composition on Si substrates using MBE. The DO3 phase found in the equilibrium phase diagram is Fe3Si, which is ferromagnetic. Clearly a ferromagnetic spacer phase is not consistent with the observation of antiferromagnetic interlayer coupling, although a non-stoichiometric DO3-structure phase might have different magnetic order. The B2 and epsilon iron silicide phases have both been previously suggested as possible candidates for the spacer layer in AF-coupled Fe/Si multilayers.[5,31,34] The position of the (001) TEM spots is not consistent with the d-spacings of the epsilon phase.
According to the powder-diffraction files for the B2 and DO3 structures, only the (111) peak of the fcc-family DO3 does not coincide with a B2 peak. The (111) peak would be expected to be very weak in the diffraction patterns formed from cross-sectional specimens of the film. The reason is that a small number of grains contributes to the cross-sectional image, and the probability of sampling a grain with its (111) planes in the observable direction is small because of the random in-plane orientation. Future work will include electron diffraction studies of a (40/14) specimen prepared in the plan-view geometry, where the number of grains which are sampled is considerably larger and the odds of observing the fcc (111) peak are improved.
High-resolution TEM images of the (30/20) and (40/14) multilayers are displayed in Figure 7. The (30/20) film is shown in 7a to have a crystalline Fe layer and amorphous spacer layer, similar to the morphology seen before in Mo/Si[13,14] and Co/Si multilayers. The (40/14) multilayer in Fig. 7b on the other hand is made up entirely of crystalline layers. The coherence between the Fe and silicide spacer is clearly evidenced by the continuity of atomic layer planes from the Fe layer into the spacer. Some crystallites in the (40/14) film extend all the way from the substrate to the surface of the film. The small coherence lengths observed in x-ray diffraction data for the uncoupled thicker-Si films are explained by the presence of the amorphous layers. The lack of crystallinity in the spacer layer of t Si = 20Å films is presumably due to insufficient time for full interdiffusion and ordering in the thicker layers. A kinetic mechanism for the lack of crystallization is supported by experiments which show that intentional placement of Fe in the Si layer allows thicker spacer layers to crystallize.[25,35]
Another striking feature of the image in Figure 7b is the periodic modulation that occurs in the silicide spacer layer. The modulation originates from scattering by inequivalent planes of atoms. Simulation of this image using a multiple-scattering computer calculation may be helpful in positively identifying the crystal structure of the spacer layer phase.
Dark-field images of the (40/14) multilayer can help answer questions about the texture of the film as well. Figure 8a shows the same bright-field image as in Figure 6b. Dark-field images were formed using (001), (002) and (011) spots from the diffraction pattern shown in Figure 6d. The resulting micrographs are shown in Figures 8b, c, and d respectively. Panels a and b of this figure show the same region of the (40/14) multilayer. The brightness of the spacer layers in this dark-field image demonstrates that the (001) reflection does indeed come from the spacer layer and is not the forbidden (001) spot of bcc Fe. Figures 8c and d also show the same region (although a different region than panels a and b). The bright areas in these two images are the complement of one another; where one is bright, the other is dark and vice versa. The dark-field images in panels c and d of Figure 8 demonstrate convincingly that the orientation of the film evolves from predominantly (011) to predominantly (002) as the thickness increases. The reason for the change in orientation with film thickness is not obvious; it may be related to the bilayer-period-number dependence discussed in Section iii C.
The effect of varying the Fe thickness has also been studied. Magnetic properties for films with 20Å <= t Fe <= 50Å are found to change only slightly in keeping with the expected inverse proportionality of the saturation field with t Fe. SAXS peaks tend to broaden and even split with increasing Fe thickness, indicating increased disorder in the layering. The splitting of these peaks may indicate different bilayer periods in areas of the film with the (011) and (001) textures. When the Fe is made less than 20Å thick, the Fe high-angle diffraction peaks disappear and so does the AF coupling. The disappearance of crystalline Fe peaks near t Fe = 20Å is consistent with previous results on evaporated Fe/Si multilayers. Thus poor crystallinity of the Fe layers appears to suppress the interlayer coupling even when the Si thickness is favorable. The lack of AF coupling in films with poorly crystalline Fe may be related to the lack of a template for the crystalline iron silicide spacer to grow on.