VOLUME 83, NUMBER 7 P H Y S I C A L R E V I E W L E T T E R S 16 AUGUST 1999 Spin Structure at the Interface of Exchange Biased FeMn Co Bilayers W. J. Antel, Jr., F. Perjeru, and G. R. Harp Department of Physics and Astronomy, Ohio University, Athens, Ohio 45701 (Received 28 October 1998) The origin of exchange biasing in FeMn Co bilayers is elucidated using magneto-optic Kerr effect and x-ray dichroism. It is found that the FeMn spin structure is aligned with the ferromagnetic (FM) moment of the Co layer, indicating that "spin flop" coupling is not the mechanism for exchange bias in this system. Futhermore, the Fe forms an uncompensated surface. It is likely that the FM Fe spins play a key role in the generation of the exchange bias. These results provide new insight to the mechanism of exchange biasing in metallic ferromagnetic antiferromagnetic systems. PACS numbers: 75.70.Ak, 75.25.+z, 75.50.Ee Although Meiklejohn and Bean [1] discovered the ex- [5,6]. Hence the experimental evidence for the origin of change bias phenomenon more than 40 years ago, it is exchange bias is not conclusive. still not well understood. This "locking" of the magne- Furthermore, note that, in the experimental work cited tization direction of a ferromagnetic (FM) layer in contact above, the AF layers are all insulating and have a bulk with an antiferromagnetic (AF) layer manifests as a shift of spin structure consisting of alternating layers of spins the hysteresis loop by a bias field Hb. The earliest theory with 180± alignment. This is not the case with FeMn explained the effect in terms of an uncompensated mono- and other metallic AF layers that are commonly used layer of spins at the surface of the antiferromagnetic layer in magnetoresistive sensors. For example, FeMn has a [2]. However, this model overestimates the observed Hb randomly occupied fcc lattice with a noncollinear (possibly by a factor of 100 [3]. tetrahedral) spin structure [11]. In this arrangement spin In recent years, interest in exchange biasing has inten- flop coupling cannot occur, which would appear to rule sified due to its usefulness in magnetoresistive sensors. out Koon's model. However, it is not known how the Several recent theories give improved predictions of the FM layer affects the spin structure within the AF layer so size of Hb, but do not agree on the physical explanation this conclusion is premature. Here we use x-ray magnetic of the effect. Malozemoff proposed a model with random circular dichroism (XMCD) and x-ray magnetic linear exchange interactions at the AF FM interface. These ran- dichroism (XMLD) to study the spin structure of both the dom interactions, due to surface roughness, lead to the FM and AF layers within Co FeMn bilayers, which serve formation of domains at the interface and give rise to a as a prototype for other systems with metallic AF layers. smaller Hb [3]. Mauri et al. proposed a subsequent model Samples were prepared using magnetron sputter de- with the formation of a domain wall in the AF which position on Si(001) at room temperature. The deposi- also reduces Hb [4]. Finally, Koon performed calcula- tion system has a base pressure of 5 3 10210 Torr and a tions indicating 90± or "spin flop" coupling between the 3.25 3 1023 Torr Ar atmosphere during deposition. The AF and FM layer, which correctly predicted the magni- samples were grown in the presence of a 500 G magnetic tude of Hb [5]. However, the sign of the bias was not field generated by a permanent magnet backing the sub- definitely determined. This was recently addressed by strate. This field serves to "set" the bias field direction Hong [6]. Note that spin flop coupling can occur only of the ferromagnetic layer. This report focuses on three for antiferromagnetic layers with an alternating 180± spin samples, the first incorporating a FeMn "wedge" with structure. the following structure: 50 Å Ta 20 Å Py 0­100 Å Recently Takano et al. performed direct measurements FeMn 17 Å Co 14 Å Al. The Ta and Py layers are of uncompensated spins at the surface in CoO layers in present to ensure that the g phase of FeMn is obtained, CoO MgO and CoO Py Py Ni81Fe19 superlattices while the Al layer is used to prevent oxidation [12]. The [7]. They found approximately 1% of one monolayer second sample is similar but has a fixed FeMn thick- (ML) of Co surface spins were uncompensated, and ness, with structure: 50 Å Ta 20 Å Py 70 Å FeMn 11 Å showed evidence that these spins lead to exchange bias- Co 14 Å Al. The final sample used a Co wedge: 50 Å ing when CoO is used in FM AF bilayers (supporting Ta 20 Å Py 70 Å FeMn 0­15 Å Co 14 Å Al. Malozemoff's model). In contrast to this, Ijiri et al. used The samples were studied using magneto-optic Kerr neutron diffraction and found spin flop coupling within effect (MOKE) loops. The upper inset of Fig. 1 is a Fe3O4 CoO superlattices [8], in support of Koon's model. hysteresis loop for a FeMn thickness tFeMn of 70 Å Another experiment, examining Fe FeF2 bilayers, discov- along the FeMn wedge. MOKE has a sampling depth of ered a positive exchange bias [9] and spin flop coupling 200 Å, thus it is possible to see both the Py underlayer [10], both of which are explained within Koon's model and the Co overlayer. Py, with a smaller magnetization 0031-9007 99 83(7) 1439(4)$15.00 © 1999 The American Physical Society 1439 VOLUME 83, NUMBER 7 P H Y S I C A L R E V I E W L E T T E R S 16 AUGUST 1999 normally incident linearly polarized light and switches M in 90± steps between parallel and perpendicular to the po- larization axis. Sequential measurements at 0±, 90±, 180±, and 270± are taken at each photon energy. The 0± and 180± are averaged to give ak, while the 90± and 270± measurements are averaged to give a . The difference spectrum a 2 ak gives a measure of m2i . Note that XMLD gives a maximal signal when the sublattice of a given element has a collinear spin arrangement (e.g., FM or 180± AF), but gives zero signal for many AF spin struc- tures including that of bulk FeMn [11]. In most FM samples flipping the magnetization by 180± using an applied field is equivalent to rotating the sample by 180± in zero applied field. But in exchange biased samples the former experiment leads to the formation of a domain wall parallel to the sample surface, whereas the FIG. 1. Plot of Co Hb vs tFeMn. These values are extracted latter does not. To probe the difference between these two from MOKE hysteresis loops. The upper inset is a hysteresis states a second set of XMCD (and XMLD) measurements loop for tFeMn 70 Å. The lower inset shows Qsat K as were made by mounting the sample on a computer- measured from the Co wedge film, indicating that 2­3 ML of Co are ferromagnetically "dead." controlled motorized rotary feed through. The XMCD spectra were obtained by rotating the sample normal and a signal further reduced by attenuation, is seen in the back and forth between 145± and 245± (with regard to top part of the hysteresis loop (marked with an arrow). the photon helicity) at each photon energy. Similarly, The larger, bottom part of the loop is due to the Co XMLD data were collected using linearly polarized x rays overlayer. From loops such as these, values of Hb are and rotating the sample about the surface normal in 90± extracted for both the Co and the Py. steps. To distinguish them, data taken while rotating the Concentrating on the Co, a plot of Hb vs tFeMn is shown sample are given the subscript R, while data taken with in Fig. 1. The bias begins to have an effect when tFeMn an external magnetic field are subscripted H. 30 Å, and saturates at tFeMn 60 Å. This is due to finite The XMCD spectra of the 70 Å FeMn 11 Å Co sample size effects. Layers of FeMn thinner than 30 Å have a are shown in Fig. 2. The data are for the L absorption blocking temperature below room temperature. As tFeMn edge of each of the three elements. For the XMCDH increases, the blocking temperature increases also. The (and XMLDH) measurements, a static field was applied coercivity (not shown) shows a sharp rise coinciding with to cancel the effect of Hb. Then a dynamic field flipped the bias turning on. With increasing tFeMn the coercivity the magnetization symmetrically about this compensated value decreases slightly. The coercivity at t state. A method has been developed to normalize and FeMn 70 Å is 250 Oe. The behavior of the coercivity is consistent with then compare XMCD [17,18] to "standard" spectra taken that seen previously in a FeMn Py bilayer system [13]. from samples with a known moment. In this way we The interfacial spin structure is determined using obtain quantitative measure of the average moment per XMCD [14] and XMLD [15,16] performed at the Syn- atom within the sample. However, not all atoms carry a chrotron Radiation Center (University of Wisconsin). In moment. To count spins it is necessary to assume a given XMCD, 85% circularly polarized x rays are incident at value of the moment within those atoms that contribute an angle of 45± with respect to the surface normal. Two to the dichroism signal. Here we assume the Fe has a x-ray absorption spectra are taken concurrently by mea- moment close to 2mB since this is the case for many suring the total electron yield (TEY). For the first, the FeCo and FeMn alloys. Likewise, we assume Co atoms sample is magnetized such that the projection of the have their bulk moment (1.6mB) for the discussion below. magnetization M is parallel to the photon helicity. These choices are necessarily rough estimates, and any The magnetization is then switched 180± and a point in error in our choice is propagated into our estimates of the second spectrum is recorded. The difference be- magnetic thicknesses below. tween the two spectra is the XMCD which is proportional In Fig. 2, observe the lack of a Mn XMCDH signal. A to the average magnetic moment per atom mi (i Mn, similar spectrum was observed for XMCDR. This indi- Fe, Co). This lets us characterize ferromagnetism within cates that the Mn spins are almost perfectly compensated the Co and near the FeMn interface, but the relatively with no more than a few percent of 1 ML residual ferro- short probing depth of TEY eliminates any contribution magnetic spins. from the Py layer. One might expect a similar result for the Fe. How- In a perfect AF the XMCD is zero. Thus to charac- ever, substantial Fe XMCDH is observed amounting to terize the FeMn layer we employ XMLD. XMLD uses 4.1 6 0.4% of that from a thick Fe standard. Using 1440 VOLUME 83, NUMBER 7 P H Y S I C A L R E V I E W L E T T E R S 16 AUGUST 1999 Co layer, just as is usually assumed for exchange biased systems. Comparatively, clear Fe and Mn XMLDR signals are observed (Fig. 3). Recall that in ferromagnetic transition metals XMLD is typically 10 times smaller than XMCD [16]. Thus, the Fe XMLDR in Fig. 3 is much too large to be accounted for by those few uncompensated Fe spins which give XMCD in Fig. 2. This indicates a considerable signal from Fe atoms in the AF layer. Likewise, Mn showed zero XMCDR, so all of the Mn XMLDR signal must originate from within the AF layer. The presence of XMLDR indicates that the AF spin structure near the Co FeMn interface is not the same as that of bulk FeMn. The XMLD spectrum of a bulk Fe sample [16] is shown in Fig. 3 (top XMLD spectrum). By comparing the shape of the Fe XMLDR with that of the bulk Fe we determine that the unique spin axis within the AF Fe is parallel to the exchange bias of the Co film. This definitively rules out spin flop coupling in the present system since a 90± orientation of the Fe spin axis would invert the Fe XMLD spectrum, and this is not observed. No suitable calibration spectrum for Mn XMLD was available. However, the Mn XMLDR is similar in magni- FIG. 2. X-ray absorption and circular dichroism vs photon tude to that from the Fe, suggesting that roughly the same energy at the L edge of Mn, Fe, and Co. The Mn and Co number of Mn spins contribute to the signal as in the Fe spectra were taken by flipping the magnetization in the Co layer case. Furthermore, the similarity in the absorption line (i.e., XMCDH). For Fe, both XMCDH and XMCDR are shown. shapes suggests that the Mn spin axis is parallel to that of the Fe. mFe 2mB, we estimate about 0.4 of one fcc(111) ML It has been theorized [5] and seen experimentally in of Fe spins flip with (and are parallel to) the Co [19]. the Fe3O4 CoO system [8] that the AF layer moments These Fe atoms are effectively part of the FM layer. In- can align perpendicular to the FM layer. The XMLDR terestingly, the Fe XMCDR signal is twice as large. Thus data indicate that this is not the case with the FeMn Co the surface of the AF layer has about 0.8 ML of uncom- system. Here we observe a net ferromagnetic moment pensated Fe spins aligned with the Co in the remanent state, but only half of which switch with the Co. Finally, the Co shows substantial XMCDH as antici- pated. Yet the measured XMCDH is only one-fourth as large as a thick Co standard. With the assumed Co mo- ment, this indicates a net "loss" of 2­3 ML of Co spins. Since this film was capped [20] we conclude that the lost spins are at the Co FeMn interface. To verify the num- ber of lost spins, MOKE loops were acquired from the Co wedge film and the saturation Kerr effect Qsat K is plotted in the lower inset of Fig. 1. It is seen that Qsat K rises linearly from the Py base-line level beginning at tCo 5 Å, in excellent agreement with the XMCD result. We hypothe- size that these Co atoms have interdiffused with the FeMn layer and are participating in the AF state. This is per- haps not surprising since CoMn alloys are antiferromag- netic even for Mn concentrations down to 35%. The Co XMCDR is not significantly different from the XMCDH. To characterize the AF state within the FeMn layer the XMLD spectra for Mn and Fe were collected [21]. Not surprisingly, XMLD FIG. 3. X-ray absorption and linear dichroism vs photon H from both Fe and Mn showed a energy at the L edge of Mn and Fe. For Fe, the top XMLD negligible signal. This indicates that the AF spins are spectrum is that of a thick Fe standard, while the bottom bound to one orientation and are not free to rotate with the spectrum is from the exchange biased sample. 1441 VOLUME 83, NUMBER 7 P H Y S I C A L R E V I E W L E T T E R S 16 AUGUST 1999 in the Fe spins amounting to almost one ML of atoms. [1] W. H. Meiklejohn and C. P. Bean, Phys. Rev. 102, 1413 Only half of these atoms "belong" to the FM layer while (1956). the other half are strongly coupled to the AF layer. The [2] W. H. Meiklejohn and C. P. Bean, Phys. Rev. 105, 904 coupling between Fe atoms on either side of the interface (1957). may be the cause of exchange biasing in this system [3] A. P. Malozemoff, Phys. Rev. B 35, 3679 (1987); A. P. (i.e., Malozemoff's model applies). Co atoms near the Malozemoff, Phys. Rev. B 37, 7673 (1988). interface might play the same role, but such a small [4] D. Mauri et al., J. Appl. Phys. 62, 3047 (1987). change in the relatively large Co XMCD signal was below [5] N. C. Koon, Phys. Rev. Lett. 78, 4865 (1997). [6] T. M. Hong, Phys. Rev. B 58, 97 (1998). our sensitivity limits. [7] Kentaro Takano et al., Phys. Rev. Lett. 79, 1130 (1997). Also keep in mind that in the interface region, 2­3 ML [8] Y. Ijiri et al., Phys. Rev. Lett. 80, 608 (1998). of Co atoms appear to participate in the AF layer. This, [9] J. Nogués et al., Phys. Rev. Lett. 76, 4624 (1996). and the presence of the nearby FM Co layer, may explain [10] T. J. Moran et al., Appl. Phys. Lett. 72, 617 (1998). the deviation of the top few ML of the AF layer from [11] Magnetic Properties of Metals, edited by H. P. J. Wijn the bulk FeMn spin structure. Moreover, it is known (Springer-Verlag, Berlin, 1991), pp. 49­53. that AF CoMn alloys have a 180± spin structure [11] [12] V. S. Speriosu et al., Phys. Rev. B 47, 11 579 (1993). which supports this interpretation. It seems likely that [13] R. Jungblut et al., J. Appl. Phys. 75, 6659 (1994). the unique spin axis of the AF layer is important to the [14] G. Schütz et al., Phys. Rev. Lett. 58, 737 (1987). mechanism of exchange biasing in the present system. A [15] G. van der Laan et al., Phys. Rev. B 34, 6529 (1986). similar mechanism might also be at work at the Py FeMn [16] See M. M. Schwickert et al., Phys. Rev. B 58, R4289 (1998), and references therein. interface since NiMn alloys are also known to have a 180± [17] M. A. Tomaz et al., Phys. Rev. B 55, 3716 (1997). spin structure [11]. [18] Tao Lin et al., Phys. Rev. B 59, 13 911 (1999). Finally, the determination of the FeMn spin structure [19] The thickness of the magnetic layer t R R M is calculated from using the above measurements can be summarized by re- C tM 0 exp 2t l dt 0 exp 2t l dt, where C is iterating the two most important results. First, the FeMn the ratio of the measured XMCD signal to that of the bulk. spin axis is aligned parallel to the ferromagnetic Co mo- l 20 Å is the sampling depth of the TEY measurement. ments. Thus the spin flop mechanism is not relevant to [20] We have previously tested Al capping layers and have this system. Second, the Fe is found to form an uncom- never found any suppression of magnetic moments associ- pensated surface. The ferromagnetic Fe spins, which are ated with them, for either Fe or Co. divided evenly between the FM and AF layers, probably [21] The XMLD spectra from the Co showed measurable play a key role in the generation of the exchange bias. signals but their interpretation is difficult, because the FM component of the Co is incompletely magnetized during The authors acknowledge support from the National the XMLD Science Foundation CAREER Award No. DMR- R measurement. In the XMCD it is possible to correct for this using MOKE loops, but since the XMLD 9623246. The Synchrotron Radiation Center is supported contains contributions from both FM and AF Co, making by the NSF under Award No. DMR-9531009. such a correction is impossible. 1442