PHYSICAL REVIEW B, VOLUME 65, 012412 Interfacially dominated giant magnetoresistance in FeÕCr superlattices J. Santamaria,1,* M.-E. Gomez,1, M.-C. Cyrille,1, C. Leighton,1 Kannan M. Krishnan,2 and Ivan K. Schuller1 1Department of Physics, University of California­San Diego, La Jolla, California 92093-0319 2Materials Sciences Division, National Center for Electron Microscopy, Lawrence Berkeley Laboratory, University of California­Berkeley, Berkeley, California 94720 Received 14 August 2001; published 3 December 2001 We have performed an extensive comparative study of growth, structure, magnetization, and magnetotrans- port in Fe/Cr superlattices. A simple analysis of the experimental data shows that the giant magnetoresistance originates from interfacial scattering in the Fe/Cr system. The saturation resistivity is determined by the roughness lateral correlation length whereas the giant magnetoresistance is determined by the interface width. DOI: 10.1103/PhysRevB.65.012412 PACS number s : 75.70.Cn, 75.70.Pa Studies of giant magnetoresistance GMR in metallic su- structural parameters in a superlattice is rather difficult since perlattices have produced much new physics since its different techniques give information with varying accuracy discovery.1­3 Most studies of magnetotransport in metallic along different directions perpendicular or parallel to the superlattices are performed with the current parallel to the interfaces . In order to obtain a quantitative description of interfaces current in the plane CIP geometry. However, the superlattice it is useful to cross correlate various mea- the geometry in which the current flows perpendicular to the surement techniques on samples made under different condi- interfaces current perpendicular to the plane CPP Refs. tions. The measurement of the magnetoresistance is also 4­10 is much more amenable to theoretical studies, and has complicated by the fact that it is desirable to measure inde- recently produced important applications.11 To the best of pendently the resistivity and the magnetoresistance. The rea- our knowledge there are no experimental studies that connect son for this is that these two quantities may be affected in in a quantitative fashion well-defined structural parameters different ways by structural parameters and, therefore, a and magnetotransport. The reasons for this are as follows: measurement solely of the ratio of the two quantities may not CPP measurements are notoriously difficult experimentally, be sufficient. Moreover, the GMR depends also on the degree the structural complexity of a superlattice requires detailed- of antiferromagnetic AF alignment in the superlattice and, quantitative structural measurements, sample characteristics therefore, measurements of the magnetization are also a key are delicately dependent on preparation conditions, and the ingredient in order to obtain a clear cut answer. magnetic properties are strikingly affected by small changes A key issue in the mechanism of GMR is the relative in preparation conditions and structural parameters. To ad- importance of bulk and interfacial scattering. This is particu- dress all these issues we have performed a detailed experi- larly difficult to clarify since in many cases both the bulk and mental study to investigate the connection between CPP- interfacial scattering are affected when layer or overall thick- GMR and structure. To do this we brought together two well- nesses of the superlattice are varied. Moreover, in the CPP established quantitative structural analysis techniques with a measurements the roughness and interdiffusion are also af- lithography-based CPP measurement technique and magneti- fected by the initial roughness of the electrodes underlying zation on a large set of samples. We find evidence that in the sample. Due to this, whether the GMR is mostly interfa- Fe/Cr superlattices both the CPP resistivity and the CPP- cial or bulk in origin is quite controversial. Measurements as GMR originate mainly from the interfaces. These results pro- a function of layer thickness, analyzed within a particular vide well-defined quantitative results that should be key in- model have claimed that the GMR originates from the bulk gredients in theories dealing with GMR in metallic and that interfacial roughness does not play a crucial role.15 superlattices. Other measurements in which the interfaces were modified Studies of transport in metallic superlattices are affected by the addition of small amounts of interfacial impurities, by many inherent complexities of the material. Many pos- claim that the interfacial scattering plays a dominant role.16 It sible complications arise in these types of artificial materials: may even be possible that the exact mechanism is materials a interfacial roughness and/or interdiffusion at various lat- system dependent. It seems that no experiments are available eral length scales,12­14 b bulk defects, c structural where the role played by ``long-wavelength'' roughness changes as a function of individual layer and/or overall larger than atomic was investigated. thickness, d different length scales affecting the structure, Here we have tackled this problem in a comprehensive magnetism, and transport, and e differences in the magne- fashion. We have made two different types of superlattices totransport along the different directions in the superlattices. by sputtering, where we vary: a the number of bilayers, and Moreover, theoretical treatments of the problem are much b sputtering pressure with a fixed number of bilayers. We more amenable if the current flow is perpendicular to the have characterized the structure of the superlattices using interfaces of the layers CPP . It is, therefore, desirable to quantitative x-ray diffraction and quantitative energy-filtered have a study in which the CPP-GMR is directly related to transmission electron microscopy EFTEM spectra. We structural parameters independently measured using quanti- measured the magnetization in order to obtain a quantitative tative structural probes. The quantitative determination of all measure of the antiferromagnetically aligned fraction. This is 0163-1829/2001/65 1 /012412 4 /$20.00 65 012412-1 ©2001 The American Physical Society BRIEF REPORTS PHYSICAL REVIEW B 65 012412 FIG. 1. Sketch of a rough surface with the various characteristic length scales. The dotted line is the averaged height with respect to FIG. 2. Size-dependent roughness for the individual bilayers the height deviation z. The dashed window of size L is the region of a 20-bilayer Fe/Cr superlattice grown at 8 mTorr. Roughness over which the standard deviation of heights is calculated to obtain increases from lower to upper bilayer index bottom to top in the the size-dependent roughness. The lateral correlation length figure . The key specifies the bilayer index from bottom to top. The is roughly the distance between surface ``bumps.'' lateral correlation length is extracted by fitting the roughness for each bilayer to (L) sat 1 exp( L/ )2 1/2 with 0.7 0.05 see line . Inset: digitized Cr EFTEM profile of the same particularly important when roughness is present because it sample. can cause local changes of the coupling or magnetic shorts. In addition we have measured independently the CPP resis- The inset of Fig. 2 shows a digitized EFTEM map using tivity and magnetoresistance in photolithographically pre- the Cr L3,2 edge of a 20 bilayers Fe/Cr superlattice grown at pared samples of a well-defined geometry. We spent consid- 8 mTorr. Similar maps were also obtained using the Fe L3,2 erable effort to ascertain that the measurement of the edge but no substantial difference was observed between the resistivity is free of nontrivial measurement artifacts16,17 that Fe and Cr profiles. Hence, a representative profile of the are particularly important for CPP measurements. layers constructed with the maxima of the intensities of the To avoid many of these difficulties, in the present experi- Cr L3,2 edge is shown. These profiles show the bilayer modu- ment we have fixed the individual layer thicknesses of Fe 30 lation in cross section and can be assumed to represent the Å /Cr 12 Å this Cr thickness corresponds to a peak in AF bilayer roughness, i.e., we do not account for the difference coupling , and varied other structural parameters such as the in roughness between both layers. A substantial replication of overall thickness i.e., number of bilayers N) and the sput- the long-scale roughness can be readily observed correlated tering pressure P. The conclusions obtained here are based roughness . The study of the lateral correlation of the rough- on a comprehensive analysis of more than 40 samples. ness is done locally by averaging the roughness over differ- Nb-(Fe/Cr)N-Nb multilayers were prepared using high-rate ent window sizes.19 For each bilayer, the local roughness for magnetron sputtering with the detailed preparation condi- window size L, at a point x0, is obtained as L(x0) tions described elsewhere.17,18 We have conducted this ex- z(x) zav(L) 2 L 1/2 averaging over L. The size- periment with two different sets of samples: set A consists of dependent roughness (L) is then obtained averaging over fixed-low-pressure samples 5 mTorr in which the number each bilayer as (L) L(x0) x see Fig. 1 . Figure 2 of bilayers has been changed between 17 and 40; and set B shows the size L dependence of the roughness of an 8 with samples with a constant number of 20 bilayers, grown mTorr sample for several bilayers. For each interface, rough- at pressures between 5 and 10 mTorr. Superconducting Nb ness increases as a power law and then saturates roughness electrodes serve as electrodes for the CPP measurements. A cutoff . The characteristic length scale over which roughness structural characterization was performed using quantitative saturates is the lateral correlation length ( ), i.e., the dis- analysis of specular x-ray diffraction XRD using the SU- tance over which interface heights ``know about each other.'' PREX Ref. 19 model and quantitative EFTEM. Magnetiza- Many numerical simulations have shown that this corre- tion measurements were performed using a superconducting sponds to the average distance between surface ``bumps,'' quantum interference device magnetometer. The magne- and in polycrystalline samples coincides with grain size.20 totransport measurements were performed in photolito- The correlation length is extracted for each bilayer, fitting the graphically prepared samples of well-defined geometry. This lateral dependence of the roughness to (L) sat 1 allows independent measurement of the resistivity and the exp( L/ )2 1/2 with 0.7 0.05 line in the figure . magnetoresistance. The details of all measurement and We note that the roughness length scale in all samples preparation techniques were described elsewhere together 10­20 nm is comparable or larger than that of the cross- with a detailed discussion of possible measurement section thickness used in the EFTEM measurements. Conse- artifacts.17,18 quently this evaluation is free of artifacts due to the projec- To quantify the roughness we describe the single-interface tion of the two-dimensional 2D roughness pattern on the profile h(x) in terms of the height deviation z(x) with 1D bilayer profiles. It is also worth noting that the lateral respect to an averaged value h(x) see Fig. 1 . The rms correlation length, in the range 10­20 nm, obtained here is in roughness also termed interface width for an in-plane sys- agreement with previous reports on similar samples using tem of size L is defined as (L) z(x) zav(L) 2 L 1/2 diffuse x-ray scattering.21,22 where zav(L) is z(x) averaged over L, and the average is Specular low-angle XRD patterns were refined using the done over all points x within L. SUPREX software19 using a model in which roughness in- 012412-2 BRIEF REPORTS PHYSICAL REVIEW B 65 012412 defined as AP P , with AP being the zero-field resistiv- ity increased with roughness for both sets of samples. This roughness value was obtained averaging the saturation val- ues obtained by EFTEM over the different bilayers, and was in very good agreement with the roughness values obtained from low-angle x-ray fitting. The main panel of Fig. 3 b displays for both sets of samples showing clear depen- dences of GMR with interface roughness. Note that data points from different sets fall into straight lines extrapolating close to zero. It seems that the slope is determined by the degree of ferromagnetic alignment: set A with 1 MR /MS 0.6 has larger slope than set B with 1 MR /MS 0.3. In- terestingly enough, if is corrected for the AF-aligned fraction for both sets of samples dividing by 1 MR /MS , all data points fall into a straight line see inset of Fig. 3 b . This suggests not only a clear dependence of GMR with interface roughness, but also, since this line extrapolates through zero, that roughness is a key ingredient of GMR. It is important to remark that this correction assumes that the main contribution to GMR comes from the AF-aligned por- FIG. 3. a Saturation resistivity P as a function of roughness tion of the sample. Although this is probably the case in the lateral correlation length for set A circles and set B squares . strong coupling limit of thin Cr layers like ours , it is worth Set A shows almost constant values and six data points are over- noting that significant GMR is known to originate from ran- lapping marked with arrow . For set B squares increases with domly oriented magnetization in weakly coupled layers with pressure. Inset: saturation resistivity vs the number of bilayers for thicker nonmagnetic spacers 5­6 nm .23 Additionally this set A. b as a function of roughness for set A circles and set B correction also does not account for possible inhomogene- squares . Inset: as a function of roughness corrected for the ities in the current distribution due to the lower resistance AF-aligned fraction. The AF-aligned fraction 1 MR /MS was 0.6 ferromagnetically aligned portions of the sample.24 for set A samples, and 1 MR /MS 0.3 for set B. We have found a clear dependence of resistivity and GMR creases cumulatively as a power law of the bilayer index.17,18 with long length-scale 10­20 nm roughness . This result The roughness values obtained from x-ray fitting were in should provide a feedback for theoretical calculations assum- very good agreement within 10% with the averaged value ing perfect layers.25 A long-scale roughness may be relevant of the roughness at saturation obtained from EFTEM. not only for its contribution to scattering, but also for its Structurally, the two sets show a markedly different be- influence on the magnetic properties at the interfaces. In fact havior. For set A, the roughness lateral correlation length it has been previously shown from structural probes that ( ) is independent of bilayer index or on the number of magnetic roughness follows the long scale interface bilayers , i.e., constant at a value of 10 nm. For set B, roughness.26­28 It is possible that the roughness lateral cor- increased with the number of bilayers and with pressure, relation length is limited by grain size, which is determined attaining a value close to 20 nm for the 10 mTorr sample. by differences in the growth mechanism at high and low The roughness increased with the bilayer index in both sets of samples and also with pressure in set B. The interface pressures.29 However, this does not necessarily imply that the width hereafter referred as ``roughness'' at saturation aver- saturation resistivity originates from bulk scattering. It is aged over the different bilayers was in the range 0.5­1 nm known that the resistivity of individual Fe and Cr layers is in for set A, and in the range 0.7­2.4 nm for set B. The level of the range 5­15 cm. These values are significantly AF alignment was also different for both kinds of samples: smaller than the saturation resistivities of our samples in the while set A showed an almost constant AF aligned fraction range 35­55 cm), clearly pointing to the interfacial ori- 1 M gin of the saturation resistivity. R / M S 0.6, in set B it attained values close to 0.3, also changing only slightly from sample to sample.17,18 The fact that both and are determined by interface As far as the giant magnetoresistance measurements are morphology ( by interface width and by ) provides an concerned, set A showed a saturation resistivity P indepen- explanation for the CIP magnetoresistance with a reduced dent of the number of bilayers see inset of Fig. 3 a and interface contribution being smaller than the CPP in Fe/Cr circles in main panel , while it increased with pressure for set superlattices. This is in agreement with previous theoretical B see squares in Fig. 3 a . The roughness increased studies30 that propose waveguiding through the paramagnetic smoothly with bilayer index, and also with pressure, as stated layer as a source of reduced GMR in the CIP configuration. above. We have found that the saturation resistivity scales In addition, it is worthwhile noting that both roughness pa- with the roughness lateral correlation length ( ) as illus- rameters affect differently the ratio / , customarily used trated in Fig. 3 a . It is worth noting that set A, with constant to describe GMR. , also shows constant P open circles in Fig. 3 a . In summary, a detailed comparison of structure, magne- 012412-3 BRIEF REPORTS PHYSICAL REVIEW B 65 012412 tism, and transport shows that in Fe/Cr superlattices the re- The authors thank S. Bader, J. Bass, G. Bauer, A. Fert, E. sistivity is mostly dominated by the roughness lateral corre- E. Fullerton, B. J. Hickey, P. Levy, A. M. Llois, and M. lation length, whereas the magnetoresistance is determined Weissman for their fruitful comments. This work was sup- by interface width in the Fe/Cr system. These results should ported by the US Department of Energy and the New del provide the quantitative connection between structural mea- Amo Program. J.S. thanks the Fundacion Jaime del Amo and surements and transport for the development of a quantita- Fundacion Flores Valles for its support. M.E. Gomez thanks tive theory of GMR. Universidad del Valle and COLCIENCIAS. *On leave from U. Complutense, 28040 Madrid, Spain. Email ad- 17 M. C. Cyrille, S. Kim, M. E. Gomez, J. Santamaria, K. M. Krish- dress: jacsan eucmax.sim.ucm.es nan, and I. K. Schuller, Phys. Rev. B 62, 3361 2000 . On leave from Universidad del Valle, A.A.25360 Cali, Colombia. 18 M. C. Cyrille, S. Kim, M. E. Gomez, J. Santamaria, C. Leighton, Present address: IBM Almaden. K. M. Krishnan, and I. K. Schuller, Phys. Rev. B 62, 15 079 1 P. M. Levy, Solid State Phys. 47, 367 1994 , and references 2000 . therein. 19 I. K. Schuller, Phys. Rev. Lett. 44, 1597 1980 ; W. Sevenhans, 2 M. A. M. Gijs and G. E. W. Bauer, Adv. Phys. 46, 285 1997 , and M. Gijs, Y. Bruynseraede, H. Homma, and I. K. Schuller, Phys. references therein. Rev. B 34, 5955 1986 ; E. E. Fullerton, I. K. Schuller, H. 3 A. Barthelemy, A. Fert, and F. Petroff, in Handbook of Magne- Vanderstraeten, and Y. Bruynseraede, ibid. 45, 9292 1992 ; D. tism, edited by K. H. J. Buschow Elsevier, New York, 1999 , M. Kelly, E. E. Fullerton, J. Santamaria, and I. K. Schuller, Scr. Vol. 12, and references therein. 4 Metall. Mater. 33, 1603 1995 . W. P. Pratt Jr., S.-F. Lee, J. M. Slaughter, R. Loloee, P. A. 20 A. L. Barabasi and H. E. Stanley, in Fractal Concepts in Surface Schroeder, and J. Bass, Phys. Rev. Lett. 66, 3060 1991 . 5 M. A. M. Gijs, S. K. J. Lenczowski, and J. B. Giesbers, Phys. Growth Cambridge University Press, Cambridge, 1995 . 21 Rev. Lett. 70, 3343 1993 . R. Schad, P. Belien, G. Verbanck, C. D. Potter, H. Fischer, S. 6 M. A. M. Gijs, M. T. Johnson, A. Reinders, P. E. Huisman, R. J. Lefebvre, M. Bessiere, V. V. Moshchalkov, and Y. Bruynseraede, M. van de Veerdonk, S. K. J. Lenczowski, and R. M. J. van Phys. Rev. B 57, 13 692 1998 . 22 Gansewinkel, Appl. Phys. Lett. 66, 1839 1995 . R. Schad, P. Belien, G. Verbanck, V. V. Moshchalkov, Y. 7 T. Ono and T. Shinjo, J. Phys. Soc. Jpn. 64, 363 1995 . Bruynseraede, H. E. Fischer, S. Lefebvre, and M. Bessiere, 8 L. Piraux, J. M. George, J. F. Despres, C. Leroy, E. Ferain, R. Phys. Rev. B 59, 1242 1999 . Legras, K. Ounadjela, and A. Fert, Appl. Phys. Lett. 65, 2484 23 J. Bass and W. P. Pratt, Jr., J. Magn. Magn. Mater. 200, 274 1994 . 1999 ; J. A. Borchers, J. A. Dura, J. Unguris, D. Tulchinsky, M. 9 A. Blondel, J. P. Meier, B. Doudin, and J. Ph. Ansermet, Appl. H. Kelley, C. F. Majkrzak, S. Y. Hsu, R. Loloee, W. P. Pratt, Jr., Phys. Lett. 65, 3019 1994 . and J. Bass, Phys. Rev. Lett. 82, 2796 1999 . 10 K. Liu, K. Nagodawithana, P. C. Searson, and C. L. Chien, Phys. 24 P. M. Levy private communication . Rev. B 51, 7381 1995 . 25 P. Weinberger, J. Phys.: Condens. Matter 8, 7677 1996 ; A. M. 11 J.-G. Zhu, Y. Zheng, and G. A. Prinz, J. Appl. Phys. 87, 6668 Llois and M. Weissman private communication ; P. Zahn et al., 2000 . Phys. Rev. Lett. 80, 4309 1998 ; 75, 2996 1995 . 12 E. E. Fullerton, D. M. Kelly, J. Guimpel, I. K. Schuller, and Y. 26 J. W. Cable, M. R. Khan, G. P. Felcher, and I. K. Schuller, Phys. Bruynseraede, Phys. Rev. Lett. 68, 859 1992 . Rev. B 34, 1643 1986 . 13 J. M. Colino, I. K. Schuller, V. Korenivski, and K. V. Rao, Phys. 27 M. J. Pechan, J. M. Ancher, D. M. Kelly, C. F. Majkrzak, and I. Rev. B 54, 13 030 1996 . K. Schuller, J. Appl. Phys. 75, 6178 1994 . 14 M. Velez and I. K. Schuller, J. Magn. Magn. Mater. 184, 275 28 J. F. Mac Kay, C. Teichert, D. E. Savage, and M. G. Lagally, 1998 . Phys. Rev. Lett. 77, 3925 1996 . 15 J. E. Mattson, M. E. Brubaker, C. H. Sowers, M. Conover, Z. Qiu, 29 E. E. Fullerton, J. Pearson, C. H. Sowers, S. D. Bader, X. Z. Wu, and S. D. Bader, Phys. Rev. B 44, 9378 1991 ; Z. Q. Qiu, J. E. and S. K. Sinha, Phys. Rev. B 48, 17 432 1993 . Mattson, C. H. Sowers, U. Welp, S. D. Bader, H. Tang, and J. C. 30 W. H. Butler, X. G. Zhang, D. M. C. Nicholson, T. C. Schulthess, Walker, ibid. 45, 2252 1992 . and J. M. MacLaren, Phys. Rev. Lett. 76, 3216 1996 ; I. Mertig, 16 S. S. P. Parkin, Phys. Rev. Lett. 71, 1641 1993 . J. Appl. Phys. 79, 5276 1996 . 012412-4