PHYSICAL REVIEW B VOLUME 57, NUMBER 21 1 JUNE 1998-I Magnetic moments, coupling, and interface interdiffusion in Fe/V 001... superlattices M. M. Schwickert Department of Physics and Astronomy, Ohio University, Athens, Ohio 45701 R. Coehoorn Philips Research Laboratories, Prof. Holstlaan 4, 5656 AA Eindhoven, The Netherlands M. A. Tomaz Department of Physics and Astronomy, Ohio University, Athens, Ohio 45701 E. Mayo and D. Lederman Department of Physics, West Virginia University, Morgantown, West Virginia 26506-6315 W. L. O'Brien Synchrotron Radiation Center, University of Wisconsin-Madison, 3731 Schneider Drive, Stoughton, Wisconsin 53589 Tao Lin and G. R. Harp Department of Physics and Astronomy, Ohio University, Athens, Ohio 45701 Received 28 October 1997 Epitaxial Fe/V 001 multilayers are studied both experimentally and by theoretical calculations. Sputter- deposited epitaxial films are characterized by x-ray diffraction, magneto-optical Kerr effect, and x-ray mag- netic circular dichroism. These results are compared with first-principles calculations modeling different amounts of interface interdiffusion. The exchange coupling across the V layers is observed to oscillate, with antiferromagnetic peaks near the V layer thicknesses tV 22, 32, and 42 Å. For all films including superlattices and alloys, the average V magnetic moment is antiparallel to that of Fe. The average V moment increases slightly with increasing interdiffusion at the Fe/V interface. Calculations modeling mixed interface layers and measurements indicate that all V atoms are aligned with one another for tV 15 Å, although the magnitude of the V moment decays toward the center of the layer. This ``transient ferromagnetic'' state arises from direct (d-d) exchange coupling between V atoms in the layer. It is argued that the transient ferromagnetism sup- presses the first antiferromagnetic coupling peak between Fe layers, expected to occur at tV 12 Å. S0163-1829 98 06021-4 I. INTRODUCTION simplifying calculations, are experimentally unattainable. In this article, we use both experimental and theoretical Vanadium stands at the edge of magnetism in the 3d tran- studies of 001 oriented Fe/V to characterize the V magne- sition metals. The five elements to the right of V in the tization over the entire spectrum of interface interdiffusion- periodic table are either antiferromagnetic Cr, Mn or ferro- from the random alloy to the perfect superlattice. We show magnetic Fe, Co, Ni near or above room temperature. It is that interdiffusion enhances the V magnetic moments as well known from neutron-diffraction studies1,2 and electronic compared with perfect superlattices. This highlights the im- structure calculations3 that V atoms, when dissolved in Fe, portance of performing calculations on more realistic struc- acquire a sizable induced magnetic moment: 1 tures, as is done here. B in the dilute limit with respect to the Fe moments in the host The V magnetic moments are aligned antiparallel to those metal . Similarly, V is known to acquire a significant mag- in the Fe and decay monotonically, extending 6 Å 4 ML netic moment in close proximity to Fe in thin films,4,5 poly- away from each Fe interface. We term this a ``transient fer- crystalline multilayers,6 or superlattices.7 Here we probe the romagnetic'' state. induced V moment when it is layered with Fe. The thickness Magnetometry reveals exchange coupling between Fe dependence of the V moment provides a measure of its ten- layers that oscillates as a function of V layer thickness. Three dency toward ferromagnetism. antiferromagnetic AF coupling peaks are observed at Generally, magnetic superlattices possess properties dis- 22, 32, and 42 Å. Another AF coupling peak expected for tinct from alloys with the same average composition. This is 12 Å V thickness is suppressed, possibly by the transient easily understood since random alloys have an average trans- ferromagnetic behavior of V in these multilayers. lational symmetry in 3D while superlattices are modulated along the z axis. ``Perfect'' superlattices comprise 2D layers II. THEORY of pure material with abrupt interfaces. Most theoretical Previous theoretical studies of Fe/V 001 superlattices are studies have focused on perfect superlattices which, while not in agreement regarding the magnetic state of V. Whereas 0163-1829/98/57 21 /13681 11 /$15.00 57 13 681 © 1998 The American Physical Society 13 682 M. M. SCHWICKERT et al. 57 TABLE I. Calculated magnetic moments from Fe 5 ML/V n ML superlattices with perfect interfaces. Magnetic moments are in units of B per atom. The layers labeled ``I'' are interface layers, while other layers are labeled with their distance from the interface. Thus, Fe I ­2 is the atomic layer in the center of the Fe layer. As discussed in more detail in the text, the results indicated with n 11* refer to a superlattice in which the Fe sublattice, and not the V sublattice, is tetragonally deformed. n Fe I ­2 Fe I ­1 Fe I V I V I­1 V I­2 V I­3 V I­4 V I­5 Fe Avg. V Avg. 1 2.33 2.46 1.90 1.05 2.21 1.05 3 2.31 2.44 1.76 0.53 0.08 2.14 0.38 5 2.29 2.43 1.79 0.49 0.02 0.05 2.15 0.19 7 2.27 2.39 1.83 0.45 0.08 0.00 0.00 2.14 0.15 9 2.28 2.42 1.77 0.50 0.05 0.02 0.00 0.01 2.13 0.12 11 2.28 2.43 1.80 0.48 0.05 0.02 0.01 0.00 0.01 2.15 0.09 11* 2.24 2.38 1.75 0.52 0.03 0.01 0.02 0.01 0.03 2.10 0.10 an early study8 indicated an induced, transient ferromagnetic to two crystallographically distinct sites within each atomic V state, a later study9 indicated layer antiferromagnetism in layer, one that we call a at a lateral position at which the the V interlayer. Recent first-principles calculations10 indi- Fe layer is locally 6 ML thick, and one that we call b at a cate again a transient ferromagnetic V state. From all studies lateral position at which the Fe layer is locally 4 ML thick. it follows that the interfacial V atoms have their magnetic The numerical accuracy of the Fe and V moments is moments aligned antiparallel to the Fe. Here we follow up 0.02 B and 0.01 B , respectively. the latter calculations, with special attention given to the ef- The results of these calculations are presented in Tables I fects of Fe-V interdiffusion. It is shown from the calculations and II. Although there is a redistribution of the Fe magnetic that diffusion suppresses the formation of a transient antifer- moments, note that the average Fe moment is hardly changed romagnetic state in the V layer, leading to a transient ferro- from the bulk level. In all cases, the calculations show a net magnetic state like that observed in experiments. negative V moment, indicating that it is aligned antiparallel The calculations were performed using the augmented to the Fe. The V moment is largest close to the interfaces, spherical wave method. Calculational details concerning the and decreases away from the Fe interface with a decay that is treatment of exchange and correlation, atomic sphere radii, quicker for perfect superlattices. Moreover, the layers with basis sets, and Brillouin-zone scanning are as described in perfect interfaces show a slight tendency towards an oscilla- Ref. 10. The calculations give predictions for the magnetic tory spin density that is not present in the calculations with moments at zero temperature, and only ferromagnetic align- diffused interfaces. This is evidence for the suppression of ment between adjacent Fe layers was considered. Two sets of transient antiferromagnetism caused by frustration, as men- calculations were performed, the first on perfect Fe/V 001 tioned above. superlattices of the type 5 ML Fe/n ML V, with n 1, 3, 5, We note that the results for the system Fe 5 ML/V 11 7, 9, and 11, and the second on Fe/V 001 superlattices with ML with a cubic V sublattice and a deformed Fe sublattice an ordered mixed monolayer at the interfaces with an Fe:V indicated in Table I with the label 11* are only marginally concentration ratio of 1:1, of the type 4 ML Fe/ 1 ML different from those for systems in which the V sublattice is Fe0.5V0.5 /(n 1) ML V/1 ML Fe0.5V0.5 , with n 1, 3, 5, 7, deformed and the Fe sublattice is cubic. 9, and 11. The method for choosing the atomic positions of the Fe and V atoms, which have in the elemental metals III. SAMPLE PREPARATION atomic radii that differ by 5.6%, is in both cases identical to the method used in Ref. 10: the Fe sublattice is cubic unde- The superlattices were deposited by magnetron sputter formed , with interatomic distances equal to those in elemen- deposition in an ultrahigh vacuum system base pressure tal Fe, whereas the V sublattice is assumed to be tetragonally 5 10 10 Torr at Ohio University. Sputtering was per- distorted, fitting coherently with the Fe sublattice and with formed in an Ar ambient of 3 10 3 Torr, with deposition the c/a ratio taken such that the volume per V atom is equal rates near 0.5 Å/s. All samples were deposited on MgO 001 to that in bulk V. This implies that the distance between the substrates, which were briefly repolished using 0.05 alu- V layers is 11% larger than in bulk V. In order to check the mina paste and rinsed before insertion into the vacuum sys- sensitivity of the moments calculated to the deformation of tem. The substrates were outgassed for 20 min at 870 K the V sublattice, a calculation was carried out for a Fe 5 prior to deposition, and then coated with a 25 Å buffer layer ML/V 11 ML ideal superlattice for which the V sublattice of either Fe or Cr at that temperature. All samples were was cubic, with lattice parameters equal to those of elemental prepared with 20 Fe/V bilayers except where noted. V, and the Fe sublattice was deformed, with the c/a ratio The samples were then allowed to cool in vacuum. The such that the volume per Fe atom is equal to that in bulk Fe. highest quality Fe/V interfaces were achieved by then depos- For the second set of calculations mixed interfaces the iting a 300 Å Cr-V alloy at 570 K. This alloy has a lattice unit cell was orthorhombic, with a lateral unit cell having constant inbetween those of Fe and V. An Fe 10 Å/ V tV] dimensions aFe 2aFe . Within the mixed interface layer al- superlattice was subsequently deposited at 570 K. In agree- ternate rows of sites parallel to the 100 direction are occu- ment with Ref. 11, we find that this was the optimal growth pied by Fe and V atoms. This corresponds to the structure as temperature. shown in Ref. 10 Fig. 2, case x 12, structure II . This leads The above described samples were compared with similar 57 MAGNETIC MOMENTS, COUPLING, AND INTERFACE . . . 13 683 TABLE II. Calculated magnetic moments from Fe 4 ML/FeV 1 ML/V n-1 ML/FeV 1 ML superlattices, simulating diffused interfaces. Because there are two inequivalent sites in each monolayer depending on nearest or next-nearest-neighbor occupation, there are two magnetic moment values quoted for each layer a, b . See the text for further details. n Fe I ­1 Fe I Fe V V I V I­1 V I­2 V I­3 V I­4 Fe Avg. V Avg. mixed mixed 1 a 2.44 2.11 2.05 2.19 0.88 1 b 2.36 1.99 0.88 3 a 2.38 2.11 2.05 0.28 2.09 0.43 3 b 2.36 1.99 0.72 0.28 5 a 2.38 2.03 1.52 0.25 0.05 2.07 0.27 5 b 2.36 2.03 0.76 0.23 0.04 7 a 2.38 2.03 1.50 0.26 0.08 0.01 2.06 0.20 7 b 2.36 2.03 0.78 0.24 0.04 0.02 9 a 2.39 2.05 1.55 0.24 0.07 0.01 0.02 2.08 0.16 9 b 2.36 2.05 0.78 0.24 0.04 0.01 0.02 11 a 2.39 2.05 1.54 0.24 0.07 0.00 0.01 0.01 2.08 0.13 11 b 2.36 2.05 0.78 0.23 0.04 0.01 0.01 0.01 Fe 7.5 Å/V tV] superlattices deposited directly onto the are displayed in Fig. 2. These provide a qualitative measure 25 Å high-temperature buffer layer, and with a growth tem- of interface roughness and/or interdiffusion. By roughness perature of 370 K. The 370 K samples showed greater inter- we mean long-range ( 10 Å variations of the height of a face roughness or interdiffusion, as determined by x-ray dif- given layer, which may be transmitted throughout the fraction see below . Finally, 001 oriented random Fe-V multilayer correlated roughness . Interdiffusion suggests alloys were prepared by codeposition onto the high- atomic scale variations of layer height, with possible inter- temperature buffer layer at 570 K. Some superlattices and change of Fe and V atoms across the interface. Specular alloys were deposited as ``wedged'' samples, where the V layer thickness alloy composition varied with position along the substrate. This permits direct comparison between superlattices alloys grown under identical growth condi- tions. Other samples were prepared with uniform V layer thicknesses to permit detailed x-ray studies. All samples were coated with a final layer of either 20 Å Al, or 20 Å Si3N4 to prevent oxidation after removal from vacuum. The layer thicknesses were controlled with in situ crystal thickness monitors near each sputter source. These monitors had been previously calibrated with the growth of a thick film, whose thickness was independently measured by step profilometry. Additionally, electron stimu- lated x-ray fluorescence was used to verify thicknesses after growth. X-ray-diffraction superlattice features also provided complimentary information regarding layer thicknesses. The results from all these techniques were combined to give the best estimates of layer thicknesses and compositions. IV. X-RAY DIFFRACTION The structural quality of the superlattices was character- ized by x-ray diffraction on samples prepared with uniform layer thicknesses. Figure 1 presents high-angle diffraction data taken from two samples prepared especially for x-ray diffraction having 40 bilayer periods. These scans were taken with a fixed-anode diffractometer with 1° angular resolution and Cu K radiation. The 570 K sample incorporated a 9 Å FIG. 1. High-angle specular x-ray diffraction scans from two V/43 Å Fe]40 superlattice, while the 370 K sample has a 27 Fe/V 001 superlattices. For either 570 K or 370 K growth, only Å V/7.5 Å Fe]40 superlattice. Both spectra show only diffrac- 001 related features are observed, indicating the films have a tion peaks associated with the 001 superlattice, and the single growth orientation. For both growth conditions, numerous MgO substrate, witnessing a single growth orientation. superlattice satellites are observed around the main superlattice fea- Low-angle diffraction scans from the same two samples tures. 13 684 M. M. SCHWICKERT et al. 57 FIG. 3. Upper panel: High-resolution specular x-ray diffraction from an Fe/V 001 superlattice deposited at 570 K symbols . The FIG. 2. Low-angle specular x-ray diffraction from the same main superlattice 002 feature is surrounded by three satellites. The films as in Fig. 1. The film deposited at 570 K shows 6 superlattice Cr-V buffer layer also presents a peak near 63°. The solid line satellites out to 2 12°, while the 370 K film shows no superlat- represents a best fit to the data, as discussed in the text. Lower tice satellites beyond 6°. This indicates more sharply modulated panel: Radial x-ray-diffraction scan through the superlattice 112 electron densities in the 570 K film. feature scattering vector q inclined 35° to surface normal of the same sample. From this scan we deduce that the superlattice film x-ray diffraction cannot distinguish between these two prop- has very little tetragonal distortion, on average. erties and measures a superposition of the two. The spectrum of the sample grown at 570 K shows satel- To characterize the in-plane crystal structure, the super- lite features out to 12° in 2 . This is typical for films lattice reflection corresponding to the bcc 112 lattice con- deposited using the 570 K recipe. By comparison, the 370 K stant was scanned using the four-circle goniometer. A ­2 sample shows no features beyond 6° in 2 , typical for the scan of this reflection is shown in the lower panel of Fig. 3 370 K samples. From this we conclude that the 570 K note that here the x-ray scattering wave vector q is canted samples have more sharply modulated electron densities by 35° with respect to the surface normal . The peak near along the growth direction. 2 80.81° corresponds to the superlattice, while the A sample with 20 bilayer periods consistent with samples smaller peak at 2 79.57° corresponds to the Cr buffer used for magnetic characterization was studied in detail by layer. Less intense peaks may correspond to superlattice sat- x-ray diffraction at West Virginia University, with structure: ellites that are visible due to a limited long-range lateral co- MgO 001 /25 Å Cr @ 870 K/300 Å Cr-V @ 570 K/11.5 Å herence of the superlattice structure. Because only one 112 Fe/ 6.2 Å V/11.5 Å Fe]20 @ 570 K/20 Å Si3N4. This sample peak was observed corresponding to the superlattice, we can was analyzed using a high-resolution four-circle diffracto- assume that both the V and Fe have an identical in-plane meter and Cu K radiation created by a rotating anode gen- lattice parameter of 2.91 Å. This is identical to the lattice erator system. parameter along the surface normal. By comparison, the In the upper panel of Fig. 3, a specular -2 scan reveals Cr-V alloy peak position 79.57°) has an in-plane lattice five film-related peaks. The peak at 62.93°, corresponding to spacing of 2.92 Å, and therefore has a slight tetragonal dis- the 002 reflection, is associated with the Cr-V alloy buffer tortion of 1%. layer, indicating a lattice constant LC of 2.95 Å. This lies A scan corresponding to the 112 superlattice reflec- between the LC's of Cr 2.88 Å and V 3.10 Å in their bulk tion, shown in the upper panel of Fig. 4, reveals a four-fold form, as expected for such an alloy. The rocking curve for in-plane symmetry. A similar scan of the buffer layer this peak is 1° wide. 112 reflection revealed the same four-fold in-plane symme- The main superlattice x-ray peak is located at 63.93° LC try, with the peaks located at the same values of . The 2.91 Å . Its rocking curve is 0.82° wide, which indicates lower panel also shows the MgO 113 peaks, which are high quality epitaxial growth. Additionally, three superlattice offset by 45° with respect to the superlattice and buffer layer satellites are seen Fig. 3 . The positions of the superlattice peaks. These scans confirm that this sample has a well- satellite peaks indicate a bilayer period of 17.7 Å. defined epitaxial relationship with the substrate, with 57 MAGNETIC MOMENTS, COUPLING, AND INTERFACE . . . 13 685 mal'' conditions. The model structure has a first V layer that is actually composed of 40% Fe. The second V layer from the interface still has 10% Fe impurities. The two Fe layers closest to the interface have the same impurity levels. The Fe/V and V/Fe interfaces were not significantly different from one another. In a sense, the experimental multilayers have 4 ML of interdiffusion at the interfaces, although there is a strong compositional gradient in the interdiffused region. Note that the diffusion profile determined here is very different from that modeled in the calculations simulating interdiffusion. The experimental profile is much broader, has a composition gradient, and has no chemical order. The cal- culations assumed a single layer of chemically ordered FeV alloy at the interface. These differences will be important below when we make comparisons between experiment and theory. V. KERR MAGNETOMETRY The bulk magnetic properties of all samples were charac- terized by magneto-optic Kerr effect MOKE magnetom- etry. All samples showed a four-fold in-plane magnetic an- isotropy. The easy axes were aligned with the superlattice 100 and 010 directions. Representative MOKE loops from 570 K Fe/V superlattices are shown in Fig. 5. Easy axis FIG. 4. Top: scan through the superlattice 112 peaks. Bot- magnetic loops, as a function of V layer thickness tV , some- tom: scan through the 113 features from the substrate. These times showed high saturation fields, indicating AF coupling scans demonstrate the epitaxial relationship between the film and substrate, namely, Fe/V 010 MgO 110 . between Fe layers. The optimal 570 K Fe/V superlattices showed a well- Fe/V 010 Cr 010 MgO 110 . defined saturation field for all V thicknesses, which is plotted The x-ray scan with q along the 001 direction was quan- in Fig. 6, along with the zero-field remanence. We observe titatively analyzed using the interdiffusion model developed three peaks in saturation field as a function of tV , at 22, 32, by M. B. Stearns,12 as implemented in the SUPREX computer and 42 Å thickness. These peaks are strongly correlated with program.13 This model assumes that there is a linear change remanence minima. This combination of high saturation field in the lattice constant and the atomic scattering factor at the and low remanence is associated with regions of AF cou- interface of the two materials. The width of this interface pling between Fe layers. The fact that such well-defined AF corresponds to the sum of the roughness and interdiffusion. coupling peaks are observed out to tV 50 Å is another in- The in-plane lattice parameter determined from scans of the dicator of the high quality of these Fe/V superlattices. 112 peak 2.91 Å was used to calculate the in-plane sur- Such AF coupling was previously observed in poly face electronic density for the Fe and V. The number of crystalline Fe/V multilayers,15 but has not been observed monolayers of each material and their lattice constants were previously in epitaxial 001 oriented Fe/V superlattices.16,7 adjusted. Specifically, previous measurements of epitaxial Fe/V The best fit to the experimental data is shown as the solid showed no evidence for AF coupling in the range tV 0 ­18 curve in Fig. 3. The results indicate that the superlattice is Å.16 In line with this, a striking feature in the present data is composed of 11.5 Å of Fe and 6.2 Å of V, with per- the absence of an expected AF coupling peak at 12 Å. It is pendicular monolayer spacings of 1.42 Å for Fe and 1.52 Å surprising that this peak would be absent, since the oscilla- for V. These values are equal to the bulk monolayer spacings tory Ruderman-Kittel-Kasuya-Yosida RKKY coupling is of 1.43 Å for Fe and 1.52 Å for V within the fit uncertainty expected to be strongest for thin V layers. Nevertheless, an- of 0.02 Å. Additionally, this is not far from what is calcu- other type of coupling appears to suppress the AF coupling lated from the bulk Poisson ratios 0.293 and 0.365, respec- in this thickness range. This other effect must be more short tively for Fe and V. For an in-plane lattice spacing of 2.91 ranged than the RKKY coupling, since AF coupling is easily Å, one calculates Fe and V monolayer spacings of 1.43 and observed for greater tV . We hypothesize that this competing 1.56 Å, respectively. The difference between the expected V factor is direct exchange coupling between V atoms in the LC from Poisson's ratio and the actual measured value is not layer. This hypothesis is supported by both theoretical and unusual in metallic superlattices, where deviations from experimental measurements of the V moments, presented Poisson's ratio have been observed in Nb/Cu, Nb/Al, W/Ni, in Sec. VI. As discussed in Sec. VII, it seems unlikely and Mo/Ni, among other systems.14 that direct coupling via pinholes is the origin of the absence The x-ray fit indicates significant interdiffusion even in of the AF peak at 12 Å. the present superlattice, which was prepared under ``opti- We fit the peaks of the saturation field with the function 13 686 M. M. SCHWICKERT et al. 57 FIG. 6. Remanence and saturation field taken from loops as those in Fig. 5. Here the regions of AF coupling are well defined by high saturation fields and low remanence values note negative re- manence values in AF regions . The peaks in the saturation field are fit with an empirical function see text . so-called ``negative remanence'' effect.17­20 The regions of negative remanence are clearly observed in Fig. 6. This is sometimes observed in magnetic multilayers with AF cou- pling. One explanation invokes magnetic moment variations and magnetic anisotropy variations from layer to layer to FIG. 5. Easy-axis magneto-optic Kerr effect loops from Fe 10 explain this effect.20 More specifically, the layers carrying Å/V tV] superlattices. Three regions of antiferromagnetic coupling the larger moment must have the smaller magnetic anisot- are observed at tV 22, 32, and 42 Å. Interestingly, these loops ropy. Hence the AF coupling causes them to switch away show negative remanence in the AF coupled regions. Note the from the applied field direction as the magnetic field is re- change of vertical scale between left and right panels, and the duced to zero, leaving the low-moment layers aligned with change of horizontal scale for the tV 22 Å loop. the field.20 The moment/anisotropy variations need not be large for negative remanence to occur, and we do not believe exp t such variations impact the other results presented here. Hpeak V / Finally, note that in Fig. 5 the AF coupled loops sat A C 1 t2V corresponding to 22 and 32 Å thickness show a step roughly halfway between remanence and saturation. This is caused dashed line curve in Fig. 6 . Here A 6.5 106 Oe, 7.9 by the four-fold anisotropy within the multilayer film. As the Å, and C 20 Oe. The first term has the form of the envelope film switches between AF and ferromagnetic alignment with of an RKKY-type oscillatory coupling, damped by an expo- increasing field, it pauses at the point where alternating lay- nential factor that represents the effect of lattice incoherency. ers have 90° alignment. At 90° alignment, all magnetic lay- The small offset field may be viewed as representing ef- ers may have their moments aligned along an easy axis, rep- fects other than interlayer exchange coupling that determine resenting a minimum in the anisotropy energy. In this state, the saturation field observed, such as coercivity. From this fit half the layers are aligned with the field, and half are aligned it would follow that in the absence of the ferromagnetic 90° to the field, hence this point has a magnetization short-range direct exchange coupling the saturation field at value equal to half the saturation value, as is observed. the first AF coupling peak, which would then be observable, would be 10 kOe. VI. XMCD As an interesting aside, we point out that in the AF X-ray magnetic circular dichroism XMCD measure- coupled regions many easy-axis loops Fig. 5 displayed the ments were performed at the Synchrotron Radiation Center, 57 MAGNETIC MOMENTS, COUPLING, AND INTERFACE . . . 13 687 at the University of Wisconsin-Madison. Approximately 85% circularly polarized x radiation allowed direct measure- ment of both Fe and V magnetic moments. This radiation was incident with an angle of 45° with respect to the surface normal, and the plane of incidence was parallel to the mag- netic easy axis. An electromagnet switched the magnetiza- tion direction along this easy axis at each photon energy, with measurements taken in remanence. As seen in Figs. 5 and 6, the remanent state of these films is still fully saturated for tV 18 Å. The total electron yield of each sample was normalized to the yield from a Cu or Ni mesh, resulting in x-ray-absorption spectra. The difference in the x-ray- absorption spectrum for the two magnetization directions is the XMCD. It is now well established that the magnitude of the XMCD is nearly proportional to the magnetic moment for a given element, independent of the magnitude of that mo- ment. This was first elucidated for the spin moment domi- nant in transition metals in Ref. 21. This can be understood from the fact that the shape of the band structure of an ele- ment changes only slightly with a variation of the occupation number of spin-up and spin-down bands. In the present study, experimental proof of the latter point is found in the fact that the complicated shape of the V XMCD spectrum is exactly the same to within experimental error for all the samples discussed here, regardless of magnetic moment. Thus, the relative size of the V or Fe magnetic moment FIG. 7. X-ray absorption solid lines and dichroism spectra is easily extracted for comparison between samples. With an symbols from an Fe/V superlattice. The magnitude of the dichro- additional measurement of a ``standard'' sample, where the ism effect is a direct measure of the Fe and V magnetic moments. magnetic moment is known, it becomes possible to extract For each element, the dichroism data is overlaid with a scaled ``standard'' dichroism spectrum dashed line that is used to deduce absolute magnetic moments. To objectify the XMCD magni- the absolute Fe and V magnetic moments. tude measurement, we compare the XMCD spectrum of each sample to that from the standard. The standard spectrum is tionality constant between the V XMCD and its magnetic scaled to achieve the best fit with that of the sample. This moment. scaling factor then represents the magnitude of the moment In Fig. 7 we present examples of XMCD spectra taken at in the sample, in terms of the moment in the standard. For a the Fe and V absorption edges in an Fe 10 Å/V 5 Å super- more complete explanation of this process, see Refs. 22 and lattice film. By comparison with standard spectra see below 23. we deduce average Fe and V moments of 2.32 The XMCD-determined magnetic moments contain at B and 0.65 least two independent sources of error. One is a statistical B , respectively. Each XMCD spectrum symbols is overlaid with a standard spectrum solid curve that has been error, which can be estimated from the quality of fit between scaled to match the data. the sample and standard spectra. This statistical error esti- mate is used to generate the error bars shown in the figures below. This error quantifies the reproducibility of the XMCD A. Fe-V alloy moments measurement itself. A second, larger, source of error comes To set the proportionality constant between V XMCD and from the assumed proportionality between the XMCD and V moment, we have pursued a study of Fe-V alloys, for magnetic moment. This ``systematic'' error has been dis- which previous experimental2 and theoretical3 studies have cussed previously24,22 and arises from 1 variations of the been performed. spin moment relative to the orbital magnetic moment, and Figure 8 displays the XMCD-determined Fe and V mag- 2 variations of the spin moment relative to the magnetic netic moments in a series of Fe-V random alloys symbols . dipole correction term. In the worst case, such systematic Error bars are shown only for V and indicate the statistical errors may amount to 20% of the moment determination.24,22 reproducibility of the XMCD measurement see above . The A third source of error, important only for the V mo- statistical error bars for the Fe moments were smaller than ments, comes from the standard sample. Since V is not nor- the symbols shown. The ordinate axis scale for the Fe data mally magnetic, the measurement of a standard with a was set by comparison to a thick Fe 001 film, capped with a known magnetic moment is problematic. To provide the best 20 Å Al layer. To make contact with previous results,7 the calibration of V moment, we present data taken on Fe-V standardization for V is initially chosen to be the same as in alloys over a range of compositions. By comparing these that paper. data with previous neutron-scattering measurements and Overlaying the XMCD data are the results of spin- with calculations, we obtain our best estimate of the propor- polarized neutron-scattering measurements by Mirebeau and 13 688 M. M. SCHWICKERT et al. 57 FIG. 8. Magnetic moments in Fe-V alloys as determined by FIG. 9. Magnetic moments in Fe/V 001 superlattices as deter- XMCD. The Fe moments are hardly changed from bulk values. mined by XMCD. Again, the Fe magnetic moments are hardly This graph sets the y-axis scale, i.e., the proportionality constant changed from the bulk Fe value. The experimentally determined V between the measured XMCD and magnetic moment for V. By moments are relatively large, and they decay monotonically with comparison with previous experiments Mirebeau and Parette Ref. increasing tV . The experiments are compared with calculations of 2 and theory Johnson and co-workers Ref. 3 the XMCD scaling perfect and diffused superlattices. Interdiffusion leads to higher V factor is adjusted to achieve agreement. Here we find that the best moments for most calculations. The experimental moments are still scaling factor is not significantly different from that used in a pre- higher, partly due to greater interdiffusion and disorder in the ex- vious publication Ref. 6 . perimental multilayers. Parette MP Ref. 2 dashed line . Note that the MP data Å , the Fe XMCD is substantially enhanced. For the thinnest presented here supersede data presented by the same group interlayers, we have observed a similar Fe XMCD enhance- in Ref. 1 . The data here are in good agreement with the MP ment for Fe/Cr Ref. 22 and Fe/Rh Ref. 23 , but not for data, without any change of scaling factor applied to the Fe/Ru.25 present results. Beyond 1 ML V, the Fe XMCD varies only a little from Additionally, the results of theoretical calculations of that of bulk Fe. Moreover, the Fe moments from the rougher Johnson and co-workers, taken from Ref. 3, are overlaid as a 370 K films are very similar to the moments in the optimized solid line. The calculations show good agreement with the 570 K films, in spite of the different Fe layer thicknesses for present data, though the agreement is slightly improved these samples. This is essentially in agreement with calcula- when the experimental XMCD results are scaled down by tions of the Fe moments both for perfect and diffused inter- 15%. These two comparisons suggest that our choice of propor- faces, see Fig. 9 . tionality constant relating the V XMCD and V moment is Moving now to the V moments, we find that the V atoms essentially correct, especially considering the other, system- carry a relatively large magnetic moment, especially for atic, errors mentioned above. Therefore, we choose to use small V thicknesses Fig. 9 . The growth recipe is an impor- exactly the same proportionality constant that was applied in tant factor for the V moments only when the V is thin. This Ref. 7. This proportionality constant is used in the next sec- is sensible, since the interface makes up a larger fraction of tion for the determination of the V moments in Fe/V super- the V film for small tV . The largest average V moment we lattices. have ever measured ( 1.5 B) is observed in superlattice films with tV 1.5 Å deposited at 370 K. Similarly large moments are observed for analogous films deposited with B. Superlattice moments 211 and 110 orientations.7 Here the Fe/V interdiffused Figure 9 displays the Fe and V moments in the 001 region is larger than the total V thickness. Yet these V mo- oriented superlattices. When the V layers are 1 ML 1.5 ments are larger than in dilute V alloys ( 1 B), and also 57 MAGNETIC MOMENTS, COUPLING, AND INTERFACE . . . 13 689 larger than in more perfect Fe/V superlattices. Somehow, the 370 K growth conditions have led to a maximization of the V moment. Before turning to the calculational results, we emphasize that the experimental ones cannot be explained based on a simple model assuming alloyed interfaces. Using the x-ray- diffraction results for the interface composition, one could assign a V magnetic moment to each interface layer depend- ing on its alloy composition. However, it was found that the average V moment in this model decays much too quickly to be of use. Furthermore, such a model could never predict the 1.5 B moment mentioned above, since it is larger than the observed moments for any alloy composition. The closest comparison of experiment and theory is be- tween the 570 K superlattices and the diffused calculation. We observe that the experimental V moments are in fair agreement with the calculated values for thin V layers, where all of the V is alloyed, but that they are enhanced by more than 100% for the largest V thicknesses considered. Recall that the experimental diffusion profile is broader than that used for the calculation. Moreover, the interdiffusion in the experiment has a composition gradient and is chemically dis- ordered, while the calculation assumed 1 ML of ordered al- loy. This may partly explain the difference between the ex- perimentally and theoretically determined V moments. However, it is not expected that more interface roughness FIG. 10. The layer integrated V magnetic moment in the Fe/V would lead to dramatically factor of 2 higher V moments in superlattices. For calculations of perfect superlattices, this quantity the calculations. As the interdiffused region becomes thicker, saturates already at 4.5 Å 3 ML V. This indicates that beyond this it begins to resemble an Fe-V alloy that has lower Fe and V thickness, no additional V moment is added with increasing thick- magnetic moments. Theoretically, we should consider other ness. For calculations of diffused superlattices, this quantity does ways in which the structures assumed for calculations may not saturate so quickly. The experimental measure of the quantity be different from those in the experiments. Recall that we saturates even more slowly only data from the 570 K multilayers find that deformation of the V sublattice does not have a are shown . This suggests that in the experiments, even the fourth V strong influence on the V moments calculated Table I . monolayer from the Fe interface possesses a significant moment, Thus we have no explanation for the larger than expected and that below 15 Å 10 ML , all V moments in the layer are ferromagnetically aligned. differences between experimentally and theoretically deter- mined V moments. point, as the V moment subsequently goes down for the ran- As a final comparison, we observe that the calculations of dom alloy infinite diffusion . A point by point comparison diffused superlattices typically show larger V moments than of the superlattices discussed here with alloys having the the calculations of perfect superlattices. For tV 3 ML, in- same average composition readily verifies this statement. terdiffusion enhances the V moment by 30­40%. The only Hence, there is an intermediate diffusion level that maxi- case where the perfect superlattice shows a higher V moment mizes the induced V magnetic moment in Fe/V superlattices. is for tV 1 ML. This can be understood by considering that the induced V moments are strongly correlated to the number of Fe nearest- and next-nearest-neighbor atoms, as discussed VII. DISCUSSION in Ref. 10 see Fig. 9 b in that paper . At perfect interfaces, That the induced V moment can be so large is an inter- V atoms have 5 nearest- and next-nearest-neighbor Fe atoms, esting result, yet how is this moment distributed through the whereas V atoms in the mixed interface layers for n 3 have V layer? One way to obtain more information is to plot the 7 nearest- and next-nearest-neighbor Fe atoms. As the V mo- total magnetic moment of the V layer i.e., the average V ment of these atoms directly at the interfaces contribute moment multiplied by the equivalent number of V monolay- strongly to the average V moment, the average moment is ers versus tV . This is done in Fig. 10. larger for systems with mixed interfaces than for systems Beginning with the calculation of perfect interfaces, we with perfect interfaces. For the case n 1, the V atoms in the find that the total V moment saturates at about 4.5 Å 3 ML . perfect superlattice as well as those in the mixed interface Indeed, Table I shows small but oscillatory V moments in layers have 10 nearest- and next-nearest-neighbor Fe atoms. the interior of the V layer i.e., transient antiferromag- The observed relatively small difference between the V netism . moments for the two n 1 structures can in this case appar- For diffused interfaces, the calculations show a gradually ently be understood only from a consideration beyond that increasing V moment with increasing tV Fig. 10 , out to at on the total number of nearest neighbors. least 10.5 Å 7 ML . As compared with calculations having To summarize, we find that the V magnetic moments typi- perfect interfaces the individual layer moments extend to cally increase with increasing interdiffusion. But only up to a greater distances from the Fe/V interface, and are generally 13 690 M. M. SCHWICKERT et al. 57 larger in magnitude Table II . Here the V is clearly in a pinholes within the V layers, which could cause direct ferro- transient ferromagnetic state. magnetic bridging between the Fe layers. If present, such This effect is even more apparent in the experiments Fig. bridging could suppress the first AF coupling peak. How- 10 , which have greater interdiffusion. It is clear that the total ever, the x-ray-diffraction results indicate essentially zero Fe V magnetic moment is not saturated even at tV 12 Å 8 content in the third and fourth V monolayers away from the ML . Recall that at 12 Å thickness, we believe there are Fe/V interface. We therefore believe that pinholes do not about 4 ML of pure V in the layer interior. Although we do play a significant role in determining the magnetic properties not know the V moment profile, Fig. 10 suggests that at 12 of the present films. Å, the innermost V monolayers still possess a non-negligible Still, interdiffusion between the Fe and V creates some magnetic moment, and that this moment is aligned ferromag- ambiguity regarding the thickness of the ferromagnetic and netically with all the other V monolayers. This magnetic mo- nonmagnetic layers. This ``magnetic roughness'' causes a ment arises from direct exchange coupling between adjacent kind of frustration, which works to weaken the RKKY spin- V atoms throughout the layer. The present results are distinct density wave. At the same time, we have shown by calcula- from results of thin V overlayers on Fe 001 , which dis- tions that interdiffusion destroys a tendency toward antifer- played oscillating V moments with increasing tV .4,5 Note romagnetism in the V layer. The experiments, which have that here the V layers are bounded on both sides by Fe, and still greater interdiffusion, exhibit transient ferromagnetism, this may explain the difference for V in the two cases. suggesting strong ferromagnetic V-V direct exchange cou- We now have an explanation for the suppression of the pling. The coexistence of a weakened RKKY coupling and first AF coupling peak between Fe layers at 12 Å V thickness strengthened V-V direct exchange coupling leads to the sup- Sec. V . The direct exchange coupling between adjacent V pression of the first AF coupling peak. Thus interdiffusion monolayers competes with the indirect RKKY coupling be- plays a key role in determining the magnetic properties of tween Fe layers. Direct coupling favors the ferromagnetically these superlattices. aligned state observed , while the RKKY coupling favors AF alignment for this V thickness. From the analysis given VIII. CONCLUSION in Sec. V it follows that at tV 12 Å the ferromagnetic cou- pling field direct exchange interaction is larger than about We have performed experimental and theoretical studies 10 kOe. Because the direct coupling is expected to fall more of Fe/V 001 superlattices and alloys. Our main results are quickly with increasing thickness as compared with the summarized as follows: 1 Finite interface interdiffusion fa- RKKY coupling, the RKKY coupling dominates at the sec- vors an increased average V magnetic moment, as compared ond, third, and fourth coupling peaks, so that these are ob- with either no interdiffusion perfect superlattices or infinite served. interdiffusion alloys . 2 The measured V moments are We liken the behavior of the present V layers to the be- characterized by a transient ferromagnetic state, with strong havior of Pd layers in Fe/Pd superlattices. Pd is well known V-V direct exchange coupling. 3 This direct exchange cou- to be nearly ferromagnetic, and recent XMCD measurements pling suppresses the first AF coupling peak between Fe lay- indicate significant Pd moments extending up to 4 ML from ers at tV 12 Å. However, because the direct exchange cou- the Fe interface.26 In that system, the oscillating RKKY cou- pling decays more quickly than the indirect RKKY coupling pling is dominated by a ferromagnetic bias, which suppresses between Fe layers, three other AF coupling features are ob- the first two AF coupling peaks at about 6 and 10 ML , served for greater tV . although a remnant of these peaks is visible as oscillations of the ferromagnetic coupling strength.27 The ferromagnetic ACKNOWLEDGMENTS bias is certainly related to the long-ranged nature of the in- duced Pd moments. In the particular Fe/V superlattices under The authors gratefully acknowledge support of the Na- study here, interdiffusion supports a transient ferromagnetic tional Science Foundation CAREER Contract No. DMR- state, and we argue that the long-ranged nature of the in- 9623246. The Synchrotron Radiation Center is supported by duced V moments is related to the suppression of the first AF the NSF under Contract No. DMR-9212658. One of the au- coupling peak in the present study. thors M.M.S. was partly supported by the Condensed Mat- It is important to consider the possibility of Fe-containing ter and Surface Science Program of Ohio University. 1 K. Adachi, in 3d, 4d, and 5d Elements, Alloys, and Compounds, 6 G. R. Harp, S. S. P. Parkin, W. L. O'Brien, and B. P. Tonner, Vol. 19a of Landolt-Bo¨rnstein, New Series, Group 3, edited by Phys. Rev. B 51, 3293 1995 . H. P. J. Wijn, Springer-Verlag, Berlin, 1986 , pp. 330­331. 7 M. A. Tomaz, W. J. Antel Jr., W. L. O'Brien, and G. R. Harp, J. 2 I. Mirebeau, G. Parette, and J. W. Cable, J. Phys. F 17, 191 Phys.: Condens. Matter 9, L179 1997 . 1987 . 8 N. Hamada, K. Terakura, and A. Yanase, J. Phys. F 14, 2371 3 D. D. Johnson, F. J. Pinski, and J. B. Staunton, J. Appl. Phys. 61, 1984 . 3715 1987 . 9 A. Vega, A. Rubio, L. C. Balbas, J. Dorantes-Davila, S. Bouarab, 4 T. G. Walker and H. Hopster, Phys. Rev. B 49, 7687 1994 . C. Demangeat, A. Mokrani, and H. Dreyesse´, J. Appl. Phys. 69, 5 P. Fuchs, K. Totland, and M. Landolt, Phys. Rev. B 53, 9123 4544 1991 . 1996 . 10 R. Coehoorn, J. Magn. Magn. Mater. 151, 341 1995 . 57 MAGNETIC MOMENTS, COUPLING, AND INTERFACE . . . 13 691 11 L.-C. Duda, P. Isberg, S. Mirbt, J.-H. Guo, B. Hjo¨rvarsson, J. 20 H. Theuss, T. Becker, D. Weller, and H. Kronmuller, J. Appl. Nordgren, and P. Granberg, Phys. Rev. B 54, 10 393 1996 . Phys. 81, 3914 1997 . 12 M. B. Stearns, Phys. Rev. B 38, 8109 1988 . 21 Paolo Carra, B. T. Thole, Massimo Altarelli, and Xindong Wang, 13 E. E. Fullerton, I. K. Schuller, H. Vanderstraeten, and Y. Phys. Rev. Lett. 70, 694 1993 . Bruynseraede, Phys. Rev. B 45, 9292 1992 . 22 M. A. Tomaz, W. J. Antel Jr., W. L. O'Brien, and G. R. Harp, 14 A. Fartash, M. Grimsditch, E. E. Fullerton, and I. K. Schuller, Phys. Rev. B 55, 3716 1997 . Phys. Rev. B 47, 12 813 1993 ; A. Fartash, I. K. Schuller, and 23 M. A. Tomaz, D. C. Ingram, G. R. Harp, D. Lederman, E. Mayo, M. Grimsditch, in Evolution of Surface & Thin Film Microstruc- and W. L. O'Brien, Phys. Rev. B 56, 5474 1997 . tures Symposium, edited by H. A. Atwater, E. Chason, M. H. 24 D. Weller, J. Sto¨hr, R. Nakajima, A. Carl, M. G. Samant, C. Grabow, and M. G. Lagally, MRS Symposia Proceedings No. Chappert, R. Megy, P. Beauvillian, F. Viellet, and G. Held, 280 Materials Research Society, Pittsburgh, 1993 . 15 Phys. Rev. Lett. 75, 3752 1995 . S. S. P. Parkin, Phys. Rev. Lett. 67, 3598 1991 . 25 16 Tao Lin, M. A. Tomaz, M. M. Schwickert, and G. R. Harp, Phys. P. Granberg, P. Nordblad, P. Isberg, B. Hjo¨rvarsson, and R. Wa¨p- pling, Phys. Rev. B 54, 1199 1996 . Rev. B to be published . 26 17 C. Gao and M. J. O'Shea, J. Magn. Magn. Mater. 127, 181 Jan Vogel, A. Fontaine, V. Cros, F. Petroff, J.-P. Kappler, G. 1993 . Krill, A. Rogalev, and J. Goulon, Phys. Rev. B 55, 3663 1997 . 18 27 M. J. O'Shea and A.-L. Al-Sharif, J. Appl. Phys. 75, 6673 1994 . Z. Celinski, B. Heinrich, and J. F. Cochran, J. Appl. Phys. 70, 19 X. Yan and Y. Xu, J. Appl. Phys. 79, 6013 1996 . 5870 1991 .