VOLUME 82, NUMBER 13 P H Y S I C A L R E V I E W L E T T E R S 29 MARCH 1999 Observation of Antiparallel Magnetic Order in Weakly Coupled Co Cu Multilayers J. A. Borchers, J. A. Dura, J. Unguris, D. Tulchinsky, M. H. Kelley, and C. F. Majkrzak National Institute of Standards and Technology, Gaithersburg, Maryland 20899 S. Y. Hsu, R. Loloee, W. P. Pratt, Jr., and J. Bass Department of Physics and Astronomy, Michigan State University, East Lansing, Michigan 48824 (Received 16 December 1998) Polarized neutron reflectivity and scanning electron microscopy with polarization analysis are combined to determine the magnetic structure of Co(6 nm) Cu(6 nm) multilayers. These data resolve a controversy regarding the low-field state of giant-magnetoresistive (GMR) multilayers with weak coupling. As-prepared samples show a strong antiparallel correlation of in-plane ferromagnetic Co domains across the Cu. At the coercive field, the Co domains are uncorrelated. This irreversible transition explains the decrease in magnetoresistance from the as-prepared to the coercive state. For both states, the Co moments reside in domains with in-plane sizes of 0.5 1.5 mm. [S0031-9007(99)08797-9] PACS numbers: 75.70.Cn, 61.12.Ha, 75.60.Ch, 75.70.Pa The combination of polarized neutron reflectivity (PNR) ter demagnetization are both associated with uncorrelated and scanning electron microscopy with polarization analy- domains in adjoining Co layers. In the as-prepared state, sis (SEMPA) represents a powerful tool for studying mag- the antiparallel correlation occurs within small columnar netic order in materials with buried magnetic layers, such Co domains with an average in-plane size of 0.5 1.5 mm. as multilayers composed of alternating layers of ferromag- This domain size is essentially unchanged at HC and after netic and nonmagnetic metals. PNR probes the order of the sample is demagnetized. the entire sample, while SEMPA produces a direct image We focus on a multilayer of nominal composition of the magnetic domain structure within one magnetic layer Co 6 nm jCu 6 nm 20, but supporting PNR results were at a time. In this Letter, we report the successful use of obtained on additional samples. The sample was sputtered PNR and SEMPA to resolve a controversy in giant mag- onto a 1 3 1 cm2 Si substrate as described elsewhere [8]. netoresistance (GMR) in Co Cu multilayers. Specular x-ray reflectivity confirms that the Co and Cu lay- The resistance (R) of a GMR multilayer greatly de- ers are well modulated. The field dependence of the mag- creases when an external field (H) reorients the in-plane netization and magnetoresistance were measured at room magnetizations of the ferromagnetic layers parallel (P) temperature for a "twin" sample grown at the same time. to each other [1]. The magnetoresistance, MR H SQUID (superconducting quantum interference device) R H 2 R P R P , is largest for systems in which magnetometer measurements indicate that the Co moments R H at a low field is associated with antiparallel align- preferentially lie in the layer plane. The magnetization ment of adjacent ferromagnetic layers. Theoretical analy- saturates in an in-plane field of approximately 200 Oe. sis has focused upon this maximum MR [2]. As shown in Fig. 1, the room temperature current-in-plane Increasing the thickness of the nonmagnetic layer, tn, MR(0) is 6.6%, whereas MR HC is only 4.0%. This ratio can lead to an oscillation between antiparallel and paral- of MR 0 MR HC typifies those of sputtered Co Cu lel states with respectively large and small MR [3]. The multilayers with similar Co and Cu thicknesses both at strength of the exchange coupling between the ferromag- room temperature and 4.2 K [4]. netic layers decreases with increasing tn. For weak in- We performed PNR studies at room temperature on terlayer coupling (tn . 4 5 nm), the magnetoresistance the NG-1 reflectometer at the NIST Center for Neutron MR 0 for the as-prepared multilayer is often larger than Research. These data are sensitive to the size, in-plane the maximum value at the coercive field MR HC after orientation, and relative interlayer alignment of magnetic saturation [4]. MR(0) usually cannot be restored by field domains in buried layers [9­12]. For specular and diffuse cycling or by demagnetization [5,6]. Because MR HC (i.e., off-specular) experiments, we measured all four cross reproduces upon cycling, most investigators have assumed sections, 22 , 11 , 12 , and 21 . (The 1 and 2 that it approximates the antiparallel state [7]. signs indicate polarizations of the incident and scattered We have performed PNR and SEMPA measurements neutrons parallel or antiparallel to the external field.) The on Co Cu multilayers with Cu layers thick enough (tCu 22 and 11 non-spin-flip (NSF) data depend on the 6 nm) that the exchange coupling between the Co layers is chemical structure, as well as the projection of the in-plane weak. We find that MR(0) originates from strong antipar- magnetization parallel to the applied field. The 12 and allel correlations among the Co domain magnetizations 21 spin-flip (SF) cross sections arise solely from the across the Cu layers. In contrast, MR HC and MR af- projection of the in-plane magnetization perpendicular to 2796 0031-9007 99 82(13) 2796(4)$15.00 © 1999 The American Physical Society VOLUME 82, NUMBER 13 P H Y S I C A L R E V I E W L E T T E R S 29 MARCH 1999 FIG. 1. Current-in-plane magnetoresistance measurements for the Co 6 nm jCu 6 nm 20 multilayer at room temperature. The magnetoresistance of the as-prepared and coercive states are marked. this field [9]. We note that the efficiencies of the NG-1 neutron polarizers were .95% in external fields as small as 1.5 Oe. Figure 2 shows total reflectivity scans along the Qz direction relative to the diffuse scattering for the Co 6 nm jCu 6 nm 20 sample in the as-prepared state (a) and at the coercive field, HC (b). In both cases, the NSF total reflectivity data have a first-order structural FIG. 2. Total PNR (shaded symbols) relative to the diffuse superlattice peak at Q scattering (open symbols) as a function of Q z 0.057 Å21 2p d, where z 4p l sin u for d 11.4 nm is the bilayer repeat distance. Figure 2(a) Co 6 nm jCu 6 nm 20 in the (a) as-prepared and (b) coercive state at H also shows a pronounced magnetic peak in all four cross C 54 Oe. The diffuse scattering was measured by offsetting the angle V by 0.2± and then scanning Qz. sections at the half-order position (Qz 0.031 Å21 The circles and squares correspond to 22 and 11 NSF 2p 2d). The magnetic repeat distance in the as-prepared data, respectively. The up and down triangles mark the 12 state is twice the bilayer thickness d; i.e., a large fraction and 21 SF data. No corrections have been made for the of the Co layer moments are oriented antiparallel along the polarization efficiencies or sample footprint. The insets show the idealized magnetic structures suggested by the scattering in growth axis. The narrow Qz width of the half-order reflec- each state. tion reveals that this antiparallel order is coherent through the entire multilayer thickness. The half-order peak has a substantial diffuse component (open symbols), which In accord with the magnetoresistance data in Fig. 1, ap- originates from discrete domains spread over the layer plication of a field irreversibly destroys the antiparallel plane [12,13]. The in-plane direction of these domains order. The half-order reflection in Fig. 2(a) disappeared within the sample plane is likely random since the diffuse when the sample was saturated in a 2375 Oe field and, as magnetic intensity in the half-order reflection is evenly shown in Fig. 2(b), it did not reappear when the sample distributed in all four cross sections. The as-prepared was taken to the coercive field (HC 54 Oe). Instead, state thus has ferromagnetic, in-plane domains that are diffuse magnetic scattering (i.e., in the SF cross sections) oriented antiparallel across the intervening Cu layers, as is distributed over a wide range of Qz values between 0.02 depicted in the inset of Fig. 2(a). and 0.05 Å21. In the coercive state, the Co moments in In Fig. 2(a), the SF scattering at the first-order position all samples seem to order in domains with small in-plane is large relative to that in Fig. 2(b), which originates dimensions that are not magnetically correlated with ad- entirely from the finite efficiencies of the instrumental joining Co layers [inset of Fig. 2(b)]. The consequence polarizers. A fit to the data indicates that the excess is that MR HC is less than MR(0). Moreover, demagne- scattering results from a small fraction (,3%) of the total tizing the sample yielded zero-field PNR data resembling Co moments in the sample that are aligned parallel across the coercive-state data in Fig. 2(b). The initial antiparal- the intervening Cu. Since data for comparable Co Cu and lel configuration was not restored by either field cycling or Co Ag samples [14] showed no excess of SF scattering demagnetization. at the first-order position in the as-prepared state, the The strong antiparallel interlayer ordering in the as- dominant antiparallel Co configuration is undoubtedly prepared state was confirmed by SEMPA with ion milling, responsible for the maximum MR(0) in Fig. 1. which allows direct imaging of the magnetic domain 2797 VOLUME 82, NUMBER 13 P H Y S I C A L R E V I E W L E T T E R S 29 MARCH 1999 structure in successive Co layers of the as-prepared sample. magnetization direction, Df, between the two Co layers. By measuring the secondary electron spin polarization in a The histogram shows that about 60% of the domains scanning electron microscope, SEMPA can sense the sur- are aligned antiparallel (within 20±), while the rest are face magnetization and, simultaneously, the topography of uncorrelated. a magnetic sample [15]. In situ ion sputtering using 2 keV Because SEMPA cannot be used in a field, we could not Ar1 ions was used to clean and depth profile the sample. image the sample at HC. Instead, we examined it after de- Figures 3(a) and 3(b) are SEMPA images of the mag- magnetization, which, as shown by PNR, produces a state netization and topography, respectively, of the outermost analogous to the uncorrelated coercive state of Fig. 2(b). Co layer after removing the protective Cu overlayer. The A histogram of the SEMPA data revealed that the mag- SEMPA topographic image reveals structural grain sizes netizations of the top two Co layers are uncorrelated, as that are about 0.1 mm. In comparison, Fig. 3(a) shows ir- expected. regular magnetic domains with feature sizes generally on Figures 3(a) and 3(c) indicate that the average domain the order of a micron, along with Nèel-like domain walls size in a local region of the sample is on the order of about 0.2 mm wide with random chirality. We note that a micron. This value matches that obtained from PNR, the domain magnetizations in the imaged region are pre- which probes the entire sample. Figure 4 shows SF data dominantly aligned along one direction, but this uniaxial for transverse Qx scans centered at the half-order position anisotropy is not evident in the PNR measurements which (Qz 0.0314 Å21) for Co 6 nm jCu 6 nm 20 in the as- probe the entire sample area. It is thus possible that the prepared, coercive, and saturated states. (The NSF data are anisotropy direction varies with lateral or vertical position. similar.) The as-prepared and coercive data in Fig. 4 are Figure 3(c) shows a SEMPA image from the second Co composed of a sharp specular reflection at Qx 0 Å21 layer after removing the top Co and Cu layers. The domain on top of a broad, diffuse peak. Dips are centered at structure of this layer is strongly anticorrelated with that of the sample angles V 0 and V 2u where either the the outermost layer in Fig. 3(a). The anticorrelation even incident or scattered beam is parallel to the sample face and extends to such small features as the domain walls, which is thus reflected (or refracted) out of the sample or detector, preserve chirality in the adjoining layer. [An example is respectively. Since the SF cross section is purely magnetic highlighted by the arrows in Figs. 3(a) and 3(c).] The in origin and the instrumental polarization efficiency is degree of correlation in the area shown is quantified .95%, the diffuse scattering principally originates from in a histogram shown in Fig. 3(d) of the difference in magnetic, rather than structural, features within the sample plane. This conclusion is supported by the decrease of the SF scattering to background levels when the Co moments are aligned in a saturation field of 400 Oe (Fig. 4). We believe that these data are among the best examples of magnetic diffuse scattering from buried layers obtained FIG. 3(color). SEMPA images of the topmost Co layer FIG. 4. Transverse Qx scans at the half-order position (Qz magnetization (a) and topography (b) and second Co layer 0.0314 Å21) for Co 6 nm jCu 6 nm 20 in the as-prepared, magnetization (c) in the Co 6 nm jCu 6 nm 20 sample. coercive (HC 54 Oe) and saturated (H 400 Oe) states. The magnetization direction is mapped into color as indicated The data for each state have been offset by 101.5 for clarity. by the color wheel in the center. A histogram of the difference Only the 12 and 21 SF cross sections are shown (shaded in the magnetization direction between the two layers, Df, is and open circles, respectively). The coercive-state data have shown in (d). been fit to a Lorentz function (solid line). 2798 VOLUME 82, NUMBER 13 P H Y S I C A L R E V I E W L E T T E R S 29 MARCH 1999 by either neutron reflectivity [11,13,16] or x-ray resonant We appreciate discussions with J. F. Ankner, M. D. scattering [17] techniques. Stiles, and R. J. Celotta. Research was supported by NSF The full width of the SF diffuse peak is inversely related DMR-9423795, MRSEC Program DMR-9400417, MSU- to an in-plane magnetic correlation length, which gener- CFMR, and Ford Research Laboratory. ally corresponds to an average domain size [12]. A fit of the peak in the coercive-state data (Fig. 4) to a Lorentz function gives an estimated length of 0.5 1.5 mm in good agreement with the SEMPA data in Fig. 3. Since the over- all width of the diffuse data for the coercive state and the as-prepared state are similar, the magnetic correlation [1] M. N. Baibich, J. M. Broto, A. Fert, F. Nguyen Van Dau, lengths for the Co 6 nm jCu 6 nm 20 sample are nearly F. Petroff, P. Etienne, G. Creuzet, A. Friederich, and the same in these two states. However, subtle differences J. Chazelas, Phys. Rev. Lett. 61, 2472 (1988). in the two line shapes suggest that the details of the in- [2] P. M. Levy, in Solid State Physics, edited by H. Ehrenreich plane magnetic structure are sensitive to field history. For and D. Turnbull (Academic, New York, 1994), Vol. 47, example, the central specular peak at Q p. 367. x 0 Å21 is quite pronounced for the as-prepared state relative to the co- [3] S. S. P. Parkin, N. More, and K. P. Roche, Phys. Rev. Lett. 64, 2304 (1990); S. S. P. Parkin, R. Bhadra, and K. P. ercive state. The coexistence of the diffuse and specu- Roche, Phys. Rev. Lett. 66, 2152 (1991). lar peaks implies that the small, micron-order domains [4] W. P. Pratt, Jr., S.-F. Lee, J. M. Slaughter, R. Loloee, P. A. are mixed with larger domains (i.e., in-plane dimensions Schroeder, and J. Bass, Phys. Rev. Lett. 66, 3060 (1991). $100 mm) that are aligned antiparallel across the Cu lay- [5] P. A. Schroeder, S.-F. Lee, P. Holody, R. Loloee, Q. Yang, ers in the as-prepared state. The latter disappear upon field W. P. Pratt, Jr., and J. Bass, J. Appl. Phys. 76, 6610 cycling and cannot be restored, even after demagnetizing (1994). the sample. [6] Ch. Rehm, F. Klose, D. Nagengast, B. Pietzak, H. Ma- As to the origin of the magnetic ordering in the as- letta, and A. Weidinger, Physica (Amsterdam) 221B, 377 prepared state, we speculate that dipolar interactions aris- (1996). ing from the fields at the edges of the micron-sized domains [7] See, e.g., S. S. P. Parkin, A. Modak, and D. J. Smith, in one Co layer may be strong enough [11] to induce local Phys. Rev. B 47, 9136 (1993); B. Doudin, A. Blondel, and J.-Ph. Ansermet, J. Appl. Phys. 79, 6090 (1996); antiparallel alignment of the growing domains in the next L. Piraux, S. Dubois, C. Marchal, J. M. Beuken, L. Fili- Co layer. The resulting structure consists of columns of pozzi, J. F. Despres, K. Ounadjela, and A. Fert, J. Magn. domains within which the Co layer magnetizations, includ- Magn. Mater. 156, 317 (1996). ing the domain wall directions [Figs. 3(a) and 3(c)], are [8] J. M. Slaughter, W. P. Pratt, Jr., and P. A. Schroeder, Rev. aligned antiparallel. Once a layer is complete and covered Sci. Instrum. 60, 127 (1989). by additional layers, the dipolar forces become secondary [9] C. F. Majkrzak, Physica (Amsterdam) 221B, 342 (1996). relative to intralayer exchange coupling and domain wall [10] S. K. Sinha, E. B. Sirota, S. Garoff, and H. B. Stanley, pinning due to structural defects. After field cycling, these Phys. Rev. B 38, 2297 (1988); V. Holý and T. Baumbach, interactions inhibit the development of antiparallel align- Phys. Rev. B 49, 10 668 (1994). ment as the Co magnetization relaxes. Alternately, mag- [11] J. A. Borchers, P. M. Gehring, R. W. Erwin, J. F. Ankner, netic anisotropy might trap the multilayer in local energy C. F. Majkrzak, T. L. Hylton, K. R. Coffey, M. A. Parker, and J. K. Howard, Phys. Rev. B 54, 9870 (1996). minima [18]. [12] S. K. Sinha in Neutron Scattering in Materials Science II, In summary, we have characterized the magnetic struc- edited by D. A. Neumann, T. P. Russell, and B. J. Wuen- ture of a Co 6 nm jCu 6 nm multilayer with weak in- sch, Materials Research Society Symposium Proceedings terlayer coupling using the complementary techniques of Vol. 376 (Materials Research Society, Pittsburgh, 1995), PNR, which simultaneously probes all Co layers, and p. 175. SEMPA, which examines one Co layer at a time. To- [13] G. P. Felcher, Physica (Amsterdam) 192B, 137 (1993). gether, these measurements reveal that the average in-plane [14] J. A. Borchers, J. A. Dura, C. F. Majkrzak, S. Y. Hsu, domain sizes in the as-prepared, coercive, and demagne- R. Lolee, W. P. Pratt, Jr., and J. Bass (unpublished). tized states are all approximately 1 mm. These domains [15] M. R. Scheinfein, J. Unguris, M. H. Kelley, D. T. Pierce, have strong antiparallel ordering across the Cu layers in and R. J. Celotta, Rev. Sci. Instrum. 61, 2501 (1990). [16] M. Takeda, Y. Endoh, H. Yasuda, K. Yamada, A. Kamijo, the as-prepared state, whereas the ordering is more nearly and J. Mizuki, J. Phys. Soc. Jpn. 62, 3015 (1993). random in the coercive and demagnetized states. Antipar- [17] J. F. MacKay, C. Teichert, D. E. Savage, and M. G. allel ordering of the Co is thus responsible for the large Lagally, Phys. Rev. Lett. 77, 3925 (1996). magnetoresistance MR(0) in the as-prepared state, relative [18] H. Holloway and D. J. Kubinski, J. Appl. Phys. 83, 2705 to that of the coercive state MR HC . (1998). 2799