History dependent domain structures in giant-magnetoresistive multilayers H. T. Hardner and M. B. Weissman Department of Physics, University of Illinois at Urbana­Champaign, Urbana, Illinois 61801-3080 S. S. P. Parkin IBM Almaden Research Center, San Jose, California 95120-6099 Received 1 May 1995; accepted for publication 26 July 1995 Resistance noise measurements of several types reveal the field history dependence of domain structure in sputtered Co/Cu multilayers. We find many smaller domains as the field is decreased from saturation towards zero, but as the field changes sign and is increased in the opposite direction we observe a smaller number of larger domains. Cycling the field without changing its sign preserves the smaller domains, strongly reducing the Barkhausen noise. Discrete fluctuations in resistance due to individual domains yield domain size estimates. © 1995 American Institute of Physics. Transition-metal-nonmagnetic multilayers exhibiting gi- or while H was being ramped Barkhausen . We have previ- ant magnetoresistance GMR are currently of great interest ously reported both Barkhausen noise8 and an increase in for various magnetic field sensor applications including mag- equilibrium noise in regimes with large dR/dH .7 Every netic recording.1 The domain structure in these multilayers is domain in the sample which participates in the GMR and important to a complete theoretical description of the GMR2 which changes its magnetization in large enough steps will and is crucial to understanding the GMR kinetics. Several be observed at some H in the Barkhausen measurement. In techniques have been used to look at domains and domain contrast, large domains with fixed or very slowly fluctuating walls in magnetic multilayers and sandwiches including magnetization are not seen in the equilibrium noise. scanning electron microscopy,3 Kerr microscopy,4 and a re- Figure 1 shows a typical plot of equilibrium noise power cently developed magneto-optical indicator film technique.5 parameter 20 Hz and resistivity versus H, with ( f ) However, determining the domain structure in multilayers is f SR(f)N/R2,9 where SR(f)is the power spectral density of very difficult, in part because most techniques are surface fluctuations in R, and N is the total number of atoms in the sensitive. Domain images produced by transmission electron sample. The spectral slope 1 d ln /d ln f ranged from microscopy TEM which requires extensive sample prepa- 1.04 to 1.19. Two large peaks in H occur for large ration have shown hysteretic domain structures in Co/Cu dR/dH , with the larger peak always found when H has multilayers.6 been reduced from saturation. In this letter, we describe using resistance noise mea- Figure 2 a shows time series corresponding to the top of surements to estimate domain sizes in GMR materials and to the large peak in in Fig. 1 while Fig. 2 b corresponds to study the field history dependence of the domain structure. the top of the smaller peak. Discernible discrete steps are The noise method can be used directly on GMR devices, often apparent in the equilibrium time series around the allowing study of a large number of samples relatively smaller peak, but were not found in some 20 min of data quickly. The technique does not produce spatial images, but from the larger peak. does reveal domain dynamics, and provides spatial informa- Figure 3 shows typical Barkhausen noise, which was tion by comparison of samples with different geometries. always largest as H was increasing. For H near the larger We studied sputtered Co/Cu multilayers with various peak in H as H was reduced , only a small Barkhausen numbers of bilayers from two batches representing the first peak appeared in the large samples, and none were evident in and second peaks in antiferromagnetic coupling between Co layers. Somewhat different results on uncoupled samples, including Co/Ag, will be presented in a future article. Table TABLE I. Sample specifications and estimates of the largest domain area I gives further details on the samples. from Barkhausen data. Area is calculated making the assumption that a Each sample is photolithographically patterned into a domain is one magnetic layer thick. bridge of four parallel legs in one of two different sizes: 3 Sample Pattern area Largest domain area m by 30 m legs or 3 m by 256 m legs. A constant dc No. Description m2 m2 current is applied to the sample and the resulting voltage noise across the bridge is amplified with commercial low- 1 Co/Cu 39 10 Å/10 Å 90 0.25 2 Co/Cu 39 10 Å/21 Å 90 4.8 noise amplifiers, anti-alias filtered, and digitally sampled. 3 Co/Cu 39 10 Å/23 Å 768 60 The resulting digital signal is either Fourier transformed and 4 Co/Cu 39 10 Å/23 Å 90 33 squared to give the noise power spectrum more fully de- 5 Co/Cu 2 10 Å/21 Å 90 20 scribed in Ref. 7 , or is simply recorded for display and 6 Co/Cu 3 10 Å/10 Å 768 73 inspection. 7 Co/Cu 10 10 Å/21 Å 768 10 8 Co/Cu 30 10 Å/21 Å 768 2 When an in-plane magnetic field H was applied, mea- 9 Co/Cu 40 10 Å/21 Å 768 15 surements could be made either at fixed H quasiequilibrium 1938 Appl. Phys. Lett. 67 (13), 25 September 1995 0003-6951/95/67(13)/1938/3/$6.00 © 1995 American Institute of Physics Downloaded¬29¬May¬2001¬to¬148.6.178.13.¬Redistribution¬subject¬to¬AIP¬license¬or¬copyright,¬see¬http://ojps.aip.org/aplo/aplcr.jsp FIG. 3. Barkhausen noise, V/V as a function of a constantly swept field, in sample 4 Co/Cu 10 Å/23 Å 39 layers . FIG. 1. The dimensionless noise parameter 20 Hz and resistivity as a function of field for sample 2 Co/Cu 10 Å/21 Å 39 layers . vation of stacks only 2­4 layers thick in a 14-layer sample.6 Given the large areas, the domain sizes must be partly lim- the small samples. Individual steps such as those shown are ited by the 3 m pattern width. most often visible in the smaller samples and are presumably The data shown so far involve cycling H from saturation due to discrete reorientations of individual domains. The through zero and out toward the opposite saturation. Figure largest Barkhausen steps are nearly four times larger than the 4 a shows versus H when cycling between zero and satu- largest equilibrium steps seen in Fig. 2 b from the same ration without changing the sign of H. The increasing-H peak sample. is larger than when sweeping from the opposite sign. The The resistance step sizes seen in both equilibrium time Barkhausen noise is nearly eliminated, as shown in compar- series and Barkhausen noise can be used to get a simple ing Fig. 4 b with Fig. 3, taken with a full cycle on the same estimate of domain volumes (VD) sample. The single-sign sweep produces noise more like that found on reduction from saturation than like that found on V sweeping through H 0. D VS R/ R, 1 The field history dependence of the Barkhausen noise in these antiferromagnetically coupled multilayers indicates the where VS is the sample volume, R is the step in R, and R is the total GMR change in R. This estimate yields a mini- mum VD because the R step ordinarily comes from a region making less than a full GMR change in its resistivity. Table I shows such VD estimates, reported in the form of the domain areas which would correspond to domains one magnetic layer thick for the largest individual step observed in each sample. VD shows little or no dependence on the number of bilayers, from 2 to 40 bilayers, indicating that the large rela- tive domain reorientations do not occur coherently over many layer boundaries. This is in agreement with the obser- FIG. 4. a as a function of H for sample 7 Co/Cu 10 Å/21 Å 10 layers . The solid line is data taken starting from about 1500 Oe and stepping all the way to 1500 Oe while the solid squares are data points taken starting at FIG. 2. Equilibrium time series V/V vs t for sample 2 taken at a 686 1500 Oe going down to zero and then going back up to 1500 Oe. b Oe corresponding to the top of the large peak in Fig. 1 and b 774 Oe Barkhausen data identical to Fig. 3 same sample, 4 except the H sweep is corresponding to the top of the smaller peak in Fig. 1 . reversed near zero. Appl. Phys. Lett., Vol. 67, No. 13, 25 September 1995 Hardner, Weissman, and Parkin 1939 Downloaded¬29¬May¬2001¬to¬148.6.178.13.¬Redistribution¬subject¬to¬AIP¬license¬or¬copyright,¬see¬http://ojps.aip.org/aplo/aplcr.jsp field history dependence of the domain structure. Larger do- In conclusion, we have studied the hysteresis of the do- main rearrangements produce more Barkhausen noise than main structure in Co/Cu multilayers using a simple technique smaller ones. Therefore, the consistent strong asymmetry of suitable for any small magnetoresistive element. The sub- the Barkhausen effect shows that there are more large do- stantial Barkhausen noise allows us to observe individual main motions when H has passed through zero than when a domains and estimate their rather large sizes. Comparison single sign of H has been maintained after saturation. of different sample geometries shows that the lateral coher- A change of the distribution of domain sizes will also ence extends over microns, while there is little coherence in affect the equilibrium noise. The spectral slope 1 indicates GMR changes between layers. The Barkhausen noise can be that there are more domains a bit too slow, i.e., too large, nearly eliminated by cycling with only one sign of H. than there are ones a bit fast i.e., too small to show up in The UIUC work was supported by the NSF DMR 93- the equilibrium noise near 20 Hz. A shift of the size distri- 05763 and, through the facilities and staff of the Materials bution toward larger slower domains would then reduce the Research Laboratory, by NSF DMR 89-20538. number within our frequency window. Thus, we would ex- pect that after H sweeps through zero, 20 Hz would be reduced and it would be easier to pick out discrete steps which are obscured when the number of fluctuators in- creases . The reduction of was observed in all samples, and 1 the increased detectability of steps was found in the one J. Heremans, J. Phys. D 26, 1149 1993 . 2 S. Zhang and P. M. Levy, Phys. Rev. B 50, 6090 1994 . small sample for which it was checked. 3 J. Unguris, R. J. Celotta, and D. T. Pierce, Phys. Rev. Lett. 67, 140 1991 . Thus both the Barkhausen noise, which probes the larger 4 M. Ruhrig and A. Hubert, J. Magn. Magn. Mater. 121, 330 1993 . domains, and the equilibrium noise, which probes an inter- 5 L. H. Bennet, R. D. McMichael, L. J. Swartzendruber, S. Hua, D. S. mediate range of domain sizes, show that the domain size Lashmore, A. J. Shapiro, V. S. Gornakov, L. M. Dedukh, and V. I. Ni- kitenko, Appl. Phys. Lett. 66, 888 1995 . distribution in all the tested antiferromagnetically coupled 6 L. J. Heyderman, J. N. Chapman, and S. S. P. Parkin, J. Appl. Phys. 76, samples is shifted toward larger sizes after H sweeps through 6613 1994 . zero. A similar result has been obtained via TEM images.6 7 H. T. Hardner, M. B. Weissman, M. B. Salamon, and S. S. P. Parkin, Phys. An explanation of this history dependence of the domain size Rev. B 48, 16156 1993 . 8 distribution may lie in the predicted production of metastable H. T. Hardner, S. S. P. Parkin, M. B. Weissman, M. B. Salamon, and E. Kita, J. Appl. Phys. 75, 6531 1994 . parallel layered Neel domain wall configurations on reducing 9 M. B. Weissman, Rev. Mod. Phys. 60, 537 1988 . H from saturation.10 10 H. Fujiwara, IEEE Trans. Magn. 29, 2557 1993 . 1940 Appl. Phys. Lett., Vol. 67, No. 13, 25 September 1995 Hardner, Weissman, and Parkin Downloaded¬29¬May¬2001¬to¬148.6.178.13.¬Redistribution¬subject¬to¬AIP¬license¬or¬copyright,¬see¬http://ojps.aip.org/aplo/aplcr.jsp