Fe and Si appear to be the only known transition-metal/semiconductor combination in which the two elements interdiffuse to form a crystalline spacer layer with coherent interfaces. The reasons why this unusual morphology occurs in the Fe/Si system are unknown but likely involve a high rate of Fe diffusion into a-Si and a low heat of crystallization of the iron silicide compound. A detailed discussion of these issues is beyond the scope of this paper.
Three different crystal structures have been proposed for the crystalline spacer layer of the Fe/Si multilayers. The epsilon phase can be eliminated on the basis of the electron diffraction patterns and TEM dark field images presented here. The B2 and DO3 crystal structures are better lattice-matched to Fe than epsilon-FeSi or alpha- and beta-FeSi2. The lattice constant of the B2 phase was reported by Mäder and coworkers to be 2.77Å, only 3.1% different from Fe. The lattice constant of the epsilon phase is 4.46Å, which matches the Fe(110) plane only in the energetically unfavorable (210) direction.
Recent conversion-electron Mössbauer data are interpreted in support of the B2 crystal structure, although the possibility of the DO3 phase was not considered in that study. It is plausible that the B2 or DO3 structures form in rapid, far-from-equilibrium growth conditions because of their small unit cells. Since silicon deposited at low substrate temperatures is amorphous, the most likely scenario is the following. Silicon deposited on a crystalline Fe layer goes down amorphous and diffuses only slightly into the Fe. Subsequently deposited Fe atoms diffuse rapidly into the amorphous Si, analogous to what happens during the growth of Mo/Si multilayers.[13,14] During the diffusion of Fe into Si, crystallization of the silicide occurs, possibly driven by the heat of mixing or by the kinetic energy of the incident Fe atoms. Growth of the crystalline phase may proceed upward from the lattice-matched Fe template, or downward from the atomically bombarded film surface. If the growth of the crystalline silicide phase proceeds downward from the film surface, one might expect to see some crystalline silicide in the high-resolution TEM image for the t Si = 20Å film (Figure 7b). The lack of any evidence for crystalline silicide in this image suggest that the crystallization proceeds upward from the iron/silicide interface, not downward from the film surface.
It is difficult to determine how realistic this model for growth of the crystalline silicide is since the Fe/Fe-Si and Si-Fe/Fe interfaces appear identical in Figure 7b. In contrast, the Mo/Si and Si/Mo interfaces in Mo/Si multilayers appear quite different from one another.[13,14] In the Mo/Si multilayers, an amorphous MoSi2 region appears which is thicker at the Mo/Si interface than at the Si/Mo interface. Detailed TEM studies of multilayers with t Si larger than 20Å may help to answer whether amorphous silicides can occur in IBS-grown Fe/Si multilayers.
Using the B2 phase lattice constant reported by the Zürich group, we can estimate the expected bilayer period of a nominal Fe/Si multilayer in which Fe atoms diffuse into the Si layer up to a 1:1 stoichiometry. The spacing between the Fe and Fe xSi 1-x layers is taken as the average of the interplanar spacings of the two materials. The result of this rough calculation is that an (Fe40Å/Si14Å) multilayer which interdiffuses up to the 1:1 stoichiometry should form a (Fe33.2Å/FeSi16.3Å) multilayer with a bilayer period of 49.4Å. The missing bilayer period predicted from this model is 4.6Å, in the middle of values on the x-axis of Figure 4. One can also calculate the expected magnetic moment reduction assuming that Fe atoms in the silicide layer have no moment and those in the Fe layer have their full moment. Under this assumption a calculation predicts 8.2Å of missing Fe moment, slightly lower than indicated in Figure 4. This calculation neglects the possibility that some Fe atoms in the Fe layer with Si near neighbors may have reduced magnetic moments.
In the discussion above the possibility has not been mentioned that the missing bilayer period and magnetic moment are due to an inaccurate thickness calibration. This explanation is contradicted by magnetization and x-ray diffraction measurements on Fe/Ge multilayers, where measured magnetic moments and bilayer periods are in much closer agreement with nominal values than for Fe/Si. The improved agreement in the case of Fe/Ge multilayers suggests that interdiffusion is less important in multilayers with Ge spacer layers than in multilayers with Si spacers.
The main point is that the formation of the B2 silicide does qualitatively explain the bilayer period reduction observed in the Fe/Si multilayers. The underlying reason for the bilayer period reduction is that the silicide which forms is denser than both Fe and Si. This situation is similar to that observed in other metal/Si multilayers[10,14] except that in the other multilayers the silicide remains amorphous.
Confirmation that the spacer layer phase has the B2 or DO 3 structure is important for understanding the coupling mechanism in these compounds. Both the B2 and DO 3 phases are known to be metallic for some ranges of composition.[15,16] Thus the present results and those of other workers[5,34] suggest that Fe/Si is really a metal/metal multilayer. The origin of the interlayer coupling is then likely to be described by the same theories as describe coupling in Co/Cu and Fe/Cr multilayers.[40,41] Fe/Si multilayers may therefore not be a good test case for theories which model interlayer exchange coupling across insulators.[8,9]
In the discussion above the possibility has been neglected that the amorphous spacer layer in the thick-Si films may also be metallic. If both the thick amorphous spacers and the thin crystalline spacers are metallic silicides, then it must be the crystallinity that is the essential feature for the existence of AF interlayer coupling. Up to now there have been no reports of AF coupling across amorphous metallic spacer layers. Toscano et al. have reported AF coupling across amorphous silicon spacer layers. These Fe/a-Si/Fe trilayers were prepared at low temperature so as to suppress interdiffusion. The character of AF coupling in the a-Si spacer trilayers is likely quite different than in the multilayers described in this study, where substrate heating increases the strength of coupling.
At the moment there is no direct evidence regarding the metallic or insulating nature of the amorphous spacer layers found in the (Fe30Å/Si20Å) multilayers. Temperature-dependent current-in-plane resistivity measurements suggest that both crystalline and amorphous spacer layers in Fe/Si multilayers are poorly conducting. Fe70Si30 and Fe65Si35 amorphous alloys have a temperature-independent resistivity, suggesting non-metallic behavior. Overall the evidence suggests that the amorphous spacer layers in (Fe30Å/Si20Å) multilayers are not metallic, but spectroscopic measurements like soft x-ray fluorescence are needed for confirmation. The interesting question as to whether there can be AF interlayer coupling across an amorphous metal spacer layer must then be left for another study.