PHYSICAL REVIEW B VOLUME 61, NUMBER 6 1 FEBRUARY 2000-II Damage caused to interlayer coupling of magnetic multilayers by residual gases C. H. Marrows* and B. J. Hickey Department of Physics and Astronomy, E. C. Stoner Laboratory, University of Leeds, Leeds LS2 9JT, United Kingdom M. Herrmann, S. McVitie, and J. N. Chapman Department of Physics and Astronomy, University of Glasgow, Glasgow G12 8QQ, United Kingdom M. Ormston and A. K. Petford-Long Department of Materials, University of Oxford, Parks Road, Oxford OX1 3PH, United Kingdom T. P. A. Hase and B. K. Tanner Department of Physics, University of Durham, South Road, Durham DH1 3LE, United Kingdom Received 18 August 1999 The oscillatory interlayer indirect exchange coupling in Co/Cu multilayers is known to be highly sensitive to structural defects. In this paper the dependence of the antiferromagnetic exchange coupling on the background pressure in the vacuum chamber is investigated. Co/Cu multilayers were grown by dc magnetron sputtering in a system equipped with a leak valve to allow the introduction of low levels of air or O2. Below a relatively narrow band of pressure the samples exhibit excellent antiferromagnetic coupling and consequently a high giant magnetoresistance, 40% for Co 8 Å /Cu 8 Å 25 samples. Above this transitional band of pressure no antiferromagnetic coupling, and hence no giant magnetoresistance, is observed. X-ray diffraction measure- ments reveal no change of any significance in any of the layer thicknesses or roughnesses. Whilst the high-field magnetic behavior is found to be isotropic in the sample plane, the reversal of the remanent moment around zero field is found to show a varying degree of uniaxial anisotropy. Lower remanent moments are found to be associated with a more isotropic reversal mechanism. A thorough characterization of the physical and magnetic microstructure by means of various modes of transmission electron microscopy is presented. Cross-sectional images reveal subtle changes in the crystallinity and layer quality of the samples as the background pressure is increased. Plan-view Lorentz microscopy reveals that the isotropic reversal mechanism of the low remanence samples involves a complete domain structure. The reversal mechanism of multilayers with a significant remanent moment varies markedly with field direction and can be dominated by rotation or comparatively simple domain processes. Samples with a significant giant magnetoresistance 20% and remanent fractions 0.7 are found to still show highly anisotropic reversal mechanisms around zero field. Indeed when demag- netized along the easy axis the samples are in a single domain state at remanence. This is compelling evidence for substantial noncollinear ordering of the moments through biquadratic coupling. I. INTRODUCTION low levels of O2 can be beneficial, at least in spin-valve structures where the interlayer coupling is not necessary to One of the most striking properties of the new generation observe GMR.5 of artificial magnetic multilayer materials is the oscillatory We have shown previously that a high level of cleanliness indirect exchange coupling between two magnetic films is necessary for a high GMR in Co/Cu multilayers, as re- separated by a thin nonmagnetic spacer.1 For those thick- sidual gases damage the film in such a way as to reduce the nesses of spacer layer where the coupling is antiferromag- antiferromagnetic coupling.6 Moreover, by selectively dam- etic, the application of a magnetic field to such an artificial aging only certain parts of the sample with gas, we found antiferromagnet can cause the system to undergo a metamag- netic transition. The associated drop in electrical resistivity that the part of the multilayer most susceptible to damage is as the magnetic configuration changes from antiferromag- the bulk of the Cu spacer. This was accomplished by pausing netic AF to ferromagnetic FM is termed the giant magne- growth at a certain point in the multilayer stack and allowing toresistance GMR .2 As the layer thicknesses in such residual gases from the chamber to adsorb onto the film sur- samples are on the nanometer scale, the preparation tech- face. Multilayers where the spacers are only lightly damaged niques for such materials are exacting and few structural de- may still show appreciable GMR, although the remanence fects can be tolerated. In general the consensus is that the may be considerable. The GMR ratio was always found to be best quality structures are prepared under ultrahigh vacuum higher than would be dictated by a simple series circuit of UHV conditions, although there have been few attempts to AF and FM coupled regions in the proportions stipulated by quantify the effects of background gases on the GMR and the remanent fraction. In fact the GMR ratio is found to have coupling of such materials. There is broad agreement that a parabolic dependence on the remanent fraction, suggesting H2O and O2 are particularly damaging.3,4 Other authors find the possibility of noncollinear arrangements of adjacent layer a more complex behavior, with the suggestion that certain moments at zero field.7 Such noncollinear arrangements of 0163-1829/2000/61 6 /4131 10 /$15.00 PRB 61 4131 ©2000 The American Physical Society 4132 C. H. MARROWS et al. PRB 61 layer moments have been observed in a large number of background gas damaging the film, air or O2 was introduced layered magnetic systems, and can be explained by the intro- through a fine leak valve. The working pressure of 99.9999% duction of a non-Heisenberg biquadratic term into the indi- purity Ar was 3.0 mTorr, introduced through an ultrahigh rect exchange energy. This is to be compared with the alter- vacuum compatible stainless steel line. Typical deposition native explanation of the nonzero remanance in poorly AF- rates were 2.6 Å/s for Co and 2.9 Å/s for Cu. A magnetic coupled multilayers, where ferromagnetic bridges, or field of 200 Oe was applied in the substrate plane during pinholes in the spacer layers, are said to lead to a ferromag- growth of the whole multilayer stack by a permanent magnet netically coupled volume fraction.8 This would lead to a do- array inside the chamber. To minimize uncontrolled changes main structure at zero field, containing low-total-moment in deposition conditions, groups of samples that are directly AF-coupled and high-total-moment FM-coupled regions. compared were deposited in a single growth run. Biquadratic interactions in multilayered systems continue All the samples discussed in this paper are of the form to attract much experimental and theoretical attention. Al- Co 8 Å /Cu 8 Å 25. The Cu thickness was selected to most all investigated coupled multilayer systems have been match the first AF-coupling peak in the oscillation. No buff- found to exhibit some degree of biquadratic coupling. Re- ers or caps were used as we have found these to be unnec- cently studied examples include Fe/Cr,9,10 Co/Au,11 Co/Ir,12 essary for good quality growth and sample longevity. as well as systems with semiconducting Fe/Si Ref. 13 and Growth was paused for 10 s in the middle of every Cu insulating e.g., Co/Al-O/NiFe Ref. 14 spacer layers. spacer, to allow residual gases to sorb onto the surface. Many more can be found in a recent review by Magnetoresistance was measured by the conventional Demokritov.15 four-probe dc technique. Magnetization loops were measured The aims of the work reported in this article were two- by the Magneto-Optic Kerr Effect MOKE . The field was fold: to determine unambiguously the zero-field moment always applied in the plane of the sample. Low-angle x-ray configuration of a sample showing reduced GMR, and to scattering measurements were performed at station 2.3 at the attempt to find evidence for the residual gas damage in the Synchrotron Radiation Source at Daresbury Laboratory.17 physical microstructure. Towards these goals we employed The wavelength of x rays used was 1.38 Å, close to the Cu K various modes of transmission electron microscopy TEM absorption edge. All of these measurements were performed to image both the physical and magnetic microstructures of at room temperature. samples showing different interlayer coupling, and conse- The samples grown on the window substrates were inves- quently differing GMR ratios. tigated at the University of Glasgow in microscopes highly Sequences of samples were grown on both ordinary Si modified to optimize magnetic-imaging conditions. The substrates and Si3N4 window substrates that are suitable for physical microstructure was studied using conventional plan-view TEM imaging. By application of different imaging bright-field imaging and diffraction techniques. The mag- modes it was possible to investigate the physical and mag- netic microstructure was investigated using the Fresnel mode netic microstructure of the samples without any further of Lorentz microscopy.18 Fresnel imaging shows domain preparation. The layered structure of the sample was exam- walls as black and white lines ``wall contrast'' . In poly- ined by cross-sectional transmission electron microscopy crystalline samples the magnetic dispersion gives rise to XTEM , imaging vertical slices of the samples grown on Si. ripple contrast which appears within domains as fine black As the base pressure of the growth chamber was raised a and white lines running perpendicular to the mean direction reduction in both GMR ratio and AF coupling could be ob- of magnetization. The magnetizing experiments were carried served. These may be linked to changes in the low-field re- out in a modified Philips CM20 TEM with a field-emission versal modes of the remanent moment, observed both mag- gun. In this instrument the primary imaging lenses used netometrically, and by micromagnetic imaging. In the while investigating magnetic samples are so-called Lorentz following section of this paper we will describe the various lenses. By exiting the standard objective lens as appropriate, experimental techniques used. We will then present the mag- a magnetic field is generated that is perpendicular to the netotransport and magnetometric data for the samples, fol- sample plane. Tilting the sample generates a well-known in- lowed by the various electron micrographs. In the final sec- plane field component which has the magnitude H tion we shall draw some conclusions. in-plane Hobjective sin , with being the tilt angle. The highest pos- sible field (Hin-plane 6 103 Oe was applied first to the II. SAMPLE PREPARATION AND EXPERIMENTAL samples before the objective lens field was reduced to the TECHNIQUES value suitable for the investigation of the samples (Hobjective 90 Oe . Therefore while taking the magnetizing sequences Samples were deposited by dc magnetron sputtering in a the samples were only subjected to a very small vertical custom-built vacuum system at the University of Leeds. The field. system is equipped with six sputtering targets and has a base Cross-sectional TEM images were taken at the University pressure of better than 2 10 8 Torr. This is achieved by a of Oxford. The samples were cleaved, thinned by ion- combination of cryopumping to 1 10 7 Torr, followed milling, and imaged. The cross-sectional high-resolution by the filling of a Meissner trap.16 This is particularly effec- electron microscopy and bright-field images were recorded tive in the pumping of water, the main residual gas remain- using a JEOL 4000EX operated at 400 kV point-to-point ing after cryopumping. The residual gas composition was resolution 0.16 nm . In all cases the sample was aligned so determined using a quadrupole mass spectrometer. Once the that the electron beam was parallel to the 110 direction in trap is filled the residual gas is mainly comprised of N2 or the Si substrate so that the interfaces between the layers CO mass peak 28 . In order to control accurately the level of could be viewed end-on. PRB 61 DAMAGE CAUSED TO INTERLAYER COUPLING OF . . . 4133 FIG. 2. Giant magnetoresistance loops for the large samples FIG. 1. The giant magnetoresistance for samples of the form grown on Si3N4 at three different base pressures; Sample 1, 1.8 Co 8 Å /Cu 8 Å 25 on 001 Si as a function of the base pres- 10 8 Torr circles ; Sample 2, 6.2 10 7 Torr squares ; and sure. Deposition was paused for 10 s in the middle of every Cu Sample 3, 0.9 10 6 Torr triangles . The magnitude of the giant spacer layer. The lowest pressure point corresponds to the lowest magnetoresistance falls as the base pressure is increased. The satu- attainable pressure in the chamber, higher pressures correspond to rated resistivity of all three samples is close to 20 cm. the introduction of air circles or O2 squares . The samples marked with arrows are those which were imaged by XTEM. Note MOKE. We can see that pure oxygen is much more damag- the logarithmic scale on the abscissa. ing than air in this regard. This suggests that N2 is much more inert than O2, as might be expected. Samples were grown on pieces of 001 Si wafer with the In Fig. 2 we show GMR loops for three samples deposited native oxide layer left intact for some magnetoresistance on the nitride-coated Si wafer at different chamber back- measurements and for cross-sectional TEM imaging. For ground pressures. As O2 is introduced into the chamber a plan-view and Lorentz-imaging samples were deposited onto drop in GMR ratio is again observed. As these three samples Si3N4 membranes supported on Si wafer suitable for direct will be extensively discussed we shall refer to them as observation in the TEM.19 These samples were also studied samples 1, 2, and 3. These multilayers are comparable to by MOKE. Meanwhile larger pieces of nitride-coated Si wa- similar samples grown directly onto the Si wafer, as the ni- fer were used for GMR and MOKE experiments. Pairs of tride surface is similarly smooth. It can also be seen in this such samples TEM membrane and larger piece of wafer figure that the shape of the loop is changing. With the leak were mounted side by side on the substrate holders, so that valve closed (1.8 10 8 Torr the GMR peak has a convex the multilayers were grown simultaneously, allowing their top, close to the parabolic GMR response of an ideally AF- properties to be directly compared. coupled multilayer sample 1 . As the pressure rises to 6.2 10 7 Torr the field required to saturate the samples falls III. RESULTS and the GMR peak becomes more pointed sample 2 . These effects are greater still as the pressure is finally raised to In Fig. 1 we show the decrease in GMR ratio in two series 0.9 10 6 Torr sample 3 . Again the sample resistivities are of samples as the base pressure is raised by the introduction all comparable in the magnetically saturated state, with val- of air or oxygen through the leak valve. All these samples ues of about 20 cm. They are, of course, different at were grown on 001 Si wafers. For pressures below a tran- zero field due to the different magnetoresistance ratios. Such sitional band a GMR ratio of 40% was achieved in both changes in GMR loop shape are associated with biquadratic series. This moderate GMR value is due to diffusive bound- coupling.21 ary scattering at the surfaces of the multilayer limiting the Grazing incidence specular x-ray scans for these three spin-allowed mean free path,20 due to the small number of samples are presented in Fig. 3. The data represent the true bilayer repeats. The expected level of impurity introduced specular scatter, determined by subtracting the diffuse scatter with the working Ar is 3 10 9 Torr, well below the pres- measured in another scan with the sample offset from the sure range on the abscissa of the graph. In the high-pressure specular condition by a small amount from the scan mea- regime above the transition zone the GMR is very low. Both sured along the specular ridge. The three samples are very series show a broad transition region, at a higher base pres- similar, in each case there being well-defined Kiessig fringes, sure for air than for O2 alone. The transition pressures differ and a similar rate of intensity fall-off with increasing sample by approximately a factor of three across the whole curve, angle. This indicates that the roughness in the samples is not quite the 5:1 ratio one might expect if the N2 were en- low, and of very similar amplitude. The first-order superlat- tirely inert. The saturated resistivities of all these samples are tice Bragg peaks are of similar intensity and full width half comparable, falling in the range 20 2 cm. The height maximum. The Bragg peaks are not all at the same changes in GMR are caused by a decrease in the amount of angle, representing bilayer periods of 16.1 Å, 15.2 Å, and AF alignment in the samples as the pressure is raised, as 16.4 Å in order of increasing background pressure samples borne out by an increase in remanent fraction as observed by 1, 2, and 3 . This represents a sample-to-sample fluctuation 4134 C. H. MARROWS et al. PRB 61 FIG. 4. Transverse diffuse x-ray scans for the extreme samples 1 FIG. 3. Low-angle x-ray specular reflectivity scans for the three open points and 3 solid points . The remarkable similarity in both samples grown on Si intensity and distribution of scatter indicates that the Co/Cu inter- 3N4. Sequential curves have been offset by decades of intensity for clarity. There are no significant differences face structure is almost identical. The slight difference between the between the profiles. The Bragg peak positions correspond to bi- two scans to the left of the figure originate from unequal sample layer periods of 16.1 Å, 15.2 Å, and 16.4 Å. The solid lines are sizes. simulated fits to the data. the field applied along the growth field direction and perpen- dicular to it. We note that the shapes of these loops are of better than 1 Å from the nominal period of 16 Å. The data isotropic in field direction. This is to be expected as any have been modeled using the Bede Scientific REFS MER- anisotropy induced in Co by the growth field is generally CURY code to determine individual layer thickness and found to be weak,23 while the field required to overcome the roughness parameters. Simulated curves are included in Fig. coupling is much larger, of the order of a few kOe. However 3. The rms roughness of all the layers in the samples is 5.5 in the low-field MOKE loops presented in Fig. 6 we can see Å. From the model simulation we have also determined that that some anisotropy is present as the remanent moment re- the fluctuations in the Co layer thickness from sample to verses. For convenience we shall refer to the direction par- sample are much greater than those in the Cu spacer layer. It allel to the growth field as the easy axis, and perpendicular to follows that the changes seen in the GMR in Fig. 2 cannot be it as the hard axis. As we shall see, although the degree of explained by incorrect spacer thickness-a result corrobo- anisotropy varies from sample to sample, the ``easiest'' di- rated by the fact that the AF-coupling peak width and posi- rection is always defined by the growth field. tion are not affected by this type of residual gas damage.7 In Fig. 6 a we see that both the easy and hard axis loops Transverse diffuse scans specimen scan at constant scat- for sample 1 show considerable rounding. There is some tering angle taken through the first-order superlattice Bragg weak anisotropy although the coercivity is 27 Oe in both peak showed a very similar intensity ratio between the inte- easy and hard directions. Both loops are somewhat canted. grated diffuse scatter Id and the integrated specular scatter Is Meanwhile, Figs. 6 b and 6 c are quite similar, with a Fig. 4 . Using the Born wave approximation, we find that squarer easy-axis loop and a canted, lower hysteresis hard- the conformal roughness on the Co/Cu interfaces is 2.5 Å axis loop. The coercive fields are 38 Oe easy axis and 35 for all samples. The overall distribution of the diffuse scatter Oe hard axis for sample 2 and 31 Oe easy axis and 20 Oe with angle is also very similar for all three samples, indicat- hard axis for the slightly more anisotropic sample 3. At ing that the in-plane correlation length is the same. We could first, it is tempting to link the degree of induced anisotropy in find no differences in the interface structure between any of the films with the exposure to oxygen, and there have been the three samples. A more detailed account of the character- attempts to do this in the past.24,25 However, we have found ization of similar Co/Cu multilayers by x-ray diffraction has that the correlation is between the degree of anisotropy and been published elsewhere.22 the remanence. For example, samples grown with 15-Å Cu The changes in GMR amplitude despite the absence of spacers, corresponding to the second FM peak, are always difference in the saturated resistivity is suggestive of a re- highly anisotropic in their reversal mechanism, regardless of duction in AF coupling. This would not affect the saturated the degree of exposure to O2 or any other background gas. resistivity, as the samples are in the same magnetic state with For samples with a very low remanence due to excellent AF all the layer moments aligned parallel to the field. On the coupling the anisotropy at low fields can be very small.26 other hand, the resistivity enhancement at zero field will be In order to gain a more detailed insight into the reversal much smaller than anticipated if the degree of antiparallel mechanisms of the samples, magnetizing experiments were alignment is not high. This is borne out by MOKE loops for carried out, in which the magnetic microstructure was moni- the samples, shown in Fig. 5. The drop in magnetoresistance tored constantly using the Fresnel mode of Lorentz micros- can be explicitly linked with the rise in remanent fraction, as copy. For all investigated samples it was found that the mag- in a previous study.7 netization reversal processes were dependent on the direction Two loops are presented for each sample, measured with of the in-plane component of the applied field. By applying PRB 61 DAMAGE CAUSED TO INTERLAYER COUPLING OF . . . 4135 FIG. 5. High-field MOKE loops for the three large samples grown on Si3N4. The remanence rises as the GMR falls. The loops FIG. 6. Low-field MOKE loops for the three samples shown in were measured with the field applied along the growth field direc- Fig. 5. Again, solid points are for loops measured with the field tion solid points and perpendicular to it open points . All three applied along the growth field direction, and open points for those samples were found to be isotropic on this scale. a Base pressure with the field perpendicular to it. A varying degree of anisotropy is 1.8 10 8 Torr Sample 1 ; b 6.2 10 7 Torr Sample 2 ; c exhibited, with greater anisotropy associated with a higher rema- 0.9 10 6 Torr Sample 3 . nent fraction. a Base pressure 1.8 10 8 Torr; b 6.2 10 7 Torr; c 0.9 10 6 Torr. Note that the units of magnetization are fields in situ with the specimen in different orientations the not the same as those for the same sample shown in Fig. 5. hard axis was identified. Application of a field in this direc- tion led to the formation of low-angle walls and magnetiza- 7 . Field reversal leads first to an increase of dispersion and tion rotation within these domains. Wall movement was the development of low-angle domain walls. At higher fields present in all investigated samples and a very common fea- there is rotation of the magnetization in adjacent domains ture is the formation of 360° walls. Extracts of the reversal indicated by an increase in wall contrast. However, signifi- processes of the samples with the field in the orthogonal cant reversal of magnetization does not occur until a field of direction are shown in Figs. 7, 8, and 9. Although this is 30 Oe when a large domain is generated with the walls what was called the easy axis in the discussion of the MOKE almost perpendicular to the array of walls developed at lower loops, what is observed does not clearly resemble an easy- fields and still clearly visible in the areas marked P and Q in axis process in a simple magnetic film. There is a pair of Fig. 7. The newly formed domain was observed to expand in images shown for each sample, one of which is taken at the center of the field of view without any further increase of remanence sample untilted in the vertical field . As the mi- the applied field showing the importance of time-dependent cromagnetic behavior differs most markedly for sample 3, effects in this regime. The overall behavior is consistent with we will begin by discussing this sample. the squarish low-field MOKE loop. It should be noted that At remanence considerable magnetic dispersion is present the domain walls (A and B) are very irregular but that the although the clear directionality of the magnetization ripple degree of dispersion within the reversed region is quite simi- indicates that the mean direction of magnetization lies paral- lar to that observed over the whole field of view at rema- lel to the direction in which the field had been applied Fig. nence. Hence the mean direction of magnetization can once 4136 C. H. MARROWS et al. PRB 61 FIG. 7. Fresnel images of sample 3; field direction hard-axis- FIG. 9. Fresnel images of sample 1; field direction hard-axis- like direction, field values given, mean direction of magnetization like direction, field values given, mean direction of magnetization as indicated, A and B denote domain walls. as indicated. again be deduced as indicated in Fig. 7. Further increase of the reverse field to 44 Oe results in an almost single do- The plan-view bright-field images of all investigated main state of the sample. Samples 2 and 1 also show direc- samples are indistinguishable. An example is shown in Fig. tionality of the magnetic dispersion in the remanent state 10. The samples are of micropolycrystalline structure with an with the mean direction of magnetization being parallel to average grain size in the range of 15­20 nm with no texture the applied field Figs. 8 and 9 . It was, however, not pos- present as can be seen in the diffraction pattern of the tilted sible to determine a direction of magnetization that leads to sample Fig. 10 . The majority of the crystallites have fcc the sharp switch in magnetization of sample 2 corresponding structure, but there is evidence for a fraction of hcp crystal- to the low-field MOKE loop of this sample. Instead, samples lites from the diffraction patterns. 2 and 1 look much more alike in these TEM experiments As we have previously reported,6,26 we have been unable with their magnetization-reversal processes being in better to find any differences in the physical microstrucure between agreement with the kind of hysteresis loop of sample 1, clean AF-coupled samples, and those so heavily damaged by where the magnetization reversal in the easy-axis-like direc- residual gases as to be entirely FM coupled. Techniques used tion is more gradual. In both of these samples we observe the include synchrotron x-ray analysis, 59Co nuclear magnetic formation of highly complex domain structures without resonance,6 and plan-view TEM Ref. 26 and this work, Fig. dominant directionality. 10 . All these findings are consistent with those reported The differences in magnetic contrast for the different here. samples is consistent with their GMR values. Good antifer- Figure 11 shows in-focus high-resolution XTEM images romagnetic alignment results in high GMR and little mag- of the two samples marked in Fig. 1. On the far left of the netic contrast. Perfect antiferromagnetic alignment in alter- images the highly ordered crystalline Si substrate can be nate layers would result in no overall Lorentz deflection of seen, with a thin ( 15 Å amorphous native oxide layer the electron beam while passing through the layer stack. visible as a bright band between it and the sample. Epoxy Therefore the high contrast of sample 3 is in agreement with the low GMR of this sample ( 7%), whereas the much lower contrast of sample 1 is consistent with better antifer- romagnetic alignment and higher GMR ( 44%). It is pos- sible to identify all the switching behavior observed in the TEM with the coercive fields measured from the low-field MOKE in Fig. 6. FIG. 8. Fresnel images of sample 2; field direction hard-axis- like direction, field values given, mean direction of magnetization FIG. 10. Plan-view bright-field image of sample 2; inset shows as indicated. diffraction pattern of tilted sample. PRB 61 DAMAGE CAUSED TO INTERLAYER COUPLING OF . . . 4137 FIG. 11. In focus high-resolution XTEM images for two samples of the form Co 8 Å /Cu 8 Å 25 grown on 001 Si wafer. The Si is visible on the left of the micrographs as a highly ordered crystalline layer. Next to this is the native SiOx layer, which is seen to be amorphous. The contrast between individual Co and Cu layers is not visible in this image due the close proximity of these two elements to each other in the Periodic Table. used in sample mounting can be seen at the right of each with a small number of ferromagnetic bridges across the Cu image. The contrast between Co and Cu atoms is poor due to present. Although the differences are very subtle, these their close proximity in the Periodic Table. As a result the XTEM images represent the only observed differences in the layered structure of the samples is not evident in these im- physical microstructure of such samples to date. This state- ages. We can see that the samples consist of columnar grains. ment carries with it the caveat that it would prove impossible In the low-background-pressure sample Fig. 11 a the to infer the level of background gas that a sample had been grains extend throughout the height of the entire multilayer exposed to only from XTEM images of this sort. film in places. This does not always appear to be true in the high-background-pressure sample Fig. 11 b . Some strain IV. DISCUSSION AND CONCLUSIONS is evident. Most of the fringes visible in the film are the 111 fringes of the fcc Co or Cu structure. The lack of The rise in remanence and reduction in GMR in samples texture is evident, with 111 planes making a wide variety which are gas damaged clearly indicate that the AF ordering of angles with the film normal. In certain areas of the images is far from perfect. The presence of GMR in these samples overlapping grains give rise to Moire´ fringes. The crystal means that some degree of nonparallel alignment exists be- structure seen in the image of the higher-background- tween the moments at zero field however, and this can be pressure sample is slightly less good, but even in the cleaner quantified by the remanence displayed by the samples in Fig. sample the degree of crystallographic perfection is not at all 5. If we wish to describe the smooth transition from AF to high. FM coupling with increasing base pressure by simply in- For the correct level of defocus the chemical periodicity creasing the FM-coupled volume fraction then we should of the multilayer is seen, as shown in Fig. 12. These images expect to see regions in the Lorentz images with very differ- are best viewed at a shallow angle looking along the layers. ent contrast, corresponding to variations in the resultant mag- We can see that for the first few bilayers the layering struc- netic moment. In fact, the images reveal overall uniform con- ture is poor, resulting in a ferromagnetic block of CoCu al- trast in all three samples, moreover in the wide anhysteretic loy. This is consistent with the lack of GMR in multilayers regimes from about 50 Oe to saturation the samples are with very low numbers of bilayer repeats (N 5),20 and with single domain. It is possible to explain this apparent contra- the results of fitting polarized neutron reflectivity spectra for diction by the introduction of the possibility of noncollinear such samples.27 It must also account for at least some of the ordering, as suggested by previous results.7 This would allow remanence observed in MOKE loops, such as Fig. 5 a , al- the angle between the moments to vary smoothly from to though that is a different sample to that imaged in Fig. 12 b . 0 as the base pressure in the sputtering system crosses the The lighter bands are the Cu spacer layers. For the low- transition zone. background-pressure sample the layers are mostly continu- Suppose the moments in a Co/Cu/Co trilayer take angles ous, if somewhat wavy in places. In the higher-background- 1 and 2 with respect to an applied field H. We may write pressure samples the layers show a few more discontinuities, the energy per unit area of such a trilayer as 4138 C. H. MARROWS et al. PRB 61 FIG. 12. Out-of-focus high-resolution XTEM images of the same areas as displayed in Fig. 11. In these images the layer structure is visible, while the defocus is such that the individual atoms can no longer be discerned. The lighter bands are the Cu layers. 0tCoMH cos 1 cos 2 J1 cos 1 2 tCo 8 Å. We have used this value for M, which is signifi- cantly below the bulk value (1.422 MA m 1), as we have J2 cos2 1 2 , 1 found previously that there is significant reduction in mo- where the Co layers are of magnetization M and thickness ment in very thin Co layers at finite temperature.28 Recent t vibrating sample magnetometer measurements29 on these Co . The coupling is parameterized by two constants J1 and J samples confirm that this is a reasonable value. 2, which are the bilinear and biquadratic coupling energies, respectively. It is the presence of a substantial biquadratic We have argued that the samples have a uniform resultant term that leads to noncollinear ordering of the moments at magnetic moment horizontally from the uniformity of con- low fields. To transfer this model to a multilayer we have trast observed in the Lorentz TEM. Since these images rep- only to note that each layer will be coupled to two neighbors, resent an integration of the magnetic moment of the sample and adjust the coupling constants by an appropriate factor of in the vertical direction they are not directly sensitive to ver- 2. tical inhomogeneities. We do have evidence to suggest that It is possible to determine the coupling constants and the samples are magnetically uniform in the vertical direc- equilibrium angles from fitting the high-field MOKE and tion apart from the small number of FM-coupled layers near GMR curves by numerically following the path of minimum to the substrate. Some information can be deduced from the energy in Eq. 1 as H is varied. As an alternative in simple Lorentz images themselves, in particular around the domain systems, some features may be determined analytically. Set- walls. Only on rare occasions is any splitting of the walls ting visible, meaning that the walls are coherent throughout the 1 2 and setting partial derivatives / 1 0 and / height of the stack. Moreover, in recent polarized neutron 2 0 to find the angles of minimum energy, we find that reflectometry measurements the vertical coherence length of the magnetic scatter was always found to be the full height of 4 J the multilayer stack, even in much more weakly coupled 2nd H 1 2J2 sat AF peak Co/Cu samples.30 Finally, the effects we describe in 0Mt , 2 the present article have been found to be independent of the number of bilayer repeats deposited. The dependence of J 1 0 cos 1 2J . 3 TABLE I. The parameters determined for the three samples sub- 2 jected to magnetic analysis. The remanant fraction may be determined directly from Eq. 3 as cos( J1 (mJ m 2) J2 (mJ m 2) 0 0/2). Measurement of the remanent fraction and saturation field hence leads directly to a determination of Sample 1 0.064 0.035 156° the coupling constants J1 and J2 and the equilibrium angle Sample 2 0.013 0.038 100° between the moments . These results are presented for our Sample 3 0.033 0.036 60° three samples in Table I, assuming M 1.0 MA m 1 and PRB 61 DAMAGE CAUSED TO INTERLAYER COUPLING OF . . . 4139 GMR and remanence on base pressure has been found to be A brief remark on the growth of similar films by the same for multilayers with 10, 25, and 50 bilayer repeats. molecular-beam epitaxy MBE is in order at this point: the We therefore believe that the degree of uniformity of the work of Miranda suggests that the presence of pinholes is a resultant magnetic moment is high in both the vertical and characteristic feature of the growth of 111 Co/Cu by this horizontal directions. technique,36 unless suppressed by the use of surfactants. This The Lorentz TEM studies of the low-field reversal mecha- is due to the possibility of twinned growth domains forming nisms reveal some anisotropy in all three samples, as borne in this system, giving rise to relatively open grain boundaries out by the low-field MOKE. In all three cases the easy axis when they meet. Because of this problem there has been was defined by the growth field direction in the sputtering some controversy over whether37 or not38 it is possible to chamber. However, the degree of anisotropy seems to be obtain any AF coupling in Co/Cu multilayers grown by this related to the quality of the AF coupling, with good AF technique. On the other hand, there have been many reports samples showing a more isotropic response to the applied of near-perfect AF coupling and very high GMR in sputtered field. In the Lorentz TEM, it was not possible to find an Co/Cu layers of many different textures, including 111 .39 obvious easy axis in samples 1 and 2 where the magnetiza- Reasons for this are not clear-sputtering is a much more tion switches sharply. The details of why a sample with a energetic process, and it has been shown that heavily re- larger remanent moment should show a more pronounced flected neutral bombardment can be correlated with dense anisotropy are unclear at this stage. Another interesting fea- compressive grain boundaries and high GMR.40 It may be ture of these experiments is that samples 2 and 3 seem more that impurities in the Cu change the surface energies suffi- similar in the low-field MOKE, while in the Lorentz images ciently to maintain the structural imperfections that give rise samples 1 and 2 are more alike. It should be noted however to the pinholes during the sputtering process. For the pur- that the samples all exhibit a continuous rise in GMR, AF poses of this article it seems wise to terminate this line of alignment, and isotropy of reversal with falling base pressure speculation at this point. during deposition. Sample 2 simply appears more like 1 or 3, To summarize, we have observed a monotonic decrease in depending on the measurement technique used. GMR and increase in remanence in Co 8 Å /Cu 8 Å multi- The XTEM images appear to show the presence of re- layers as the base pressure of the sputtering chamber in gions where there are discontinuities in the Co and Cu layers, which they were prepared rises. To explain these changes and there are more such regions in the gas-damaged sample. consistently with magnetizing sequences of Lorentz TEM However, the presence of pinholes is not disastrous for the images, we must invoke a substantial biquadratic term in the noncollinear model. The work of Bobo shows significant exchange coupling between neighboring Co layers. This re- twisting of the moments around a pinhole, leading to sub- sults in configurations of the moments that are not collinear. stantial noncollinear ordering over a large area of film.31,32 If the pinholes are sufficiently close together laterally then the ACKNOWLEDGMENTS preferred interlayer angle is ``averaged'' over the area of the film.33 In NiFe/Cu, a substantial J2 and the anomalous C.H.M. was supported by the Royal Commission for the change in sign of J1 with temperature were directly attrib- Exhibition of 1851. The work of M.H. was supported by a uted to the presence of pinholes.34 First AF peak Co/Cu mul- TMR grant of the European Commission. A.K.P.-L. was tilayers from another deposition system with comparatively supported by the Royal Society. We would also like to thank large remanence have also been recently found to exhibit the Engineering and Physical Sciences Research Council for dominant biquadratic coupling.35 funding. *Electronic address: c.marrows@leeds.ac.uk 11 T. Duden and E. Bauer, Phys. Rev. B 59, 474 1999 . 1 Ultrathin Magnetic Structures, edited by B. Heinrich and J. A. C. 12 H. Yanagihara, K. Pettit, M. B. Salamon, E. Kita, and S. S. P. Bland Springer, Berlin, 1994 , Vol. 2, Chap. 2. Parkin, J. Appl. Phys. 81, 5197 1999 . 2 B. Dieny, J. Magn. Magn. Mater. 136, 335 1994 . 13 Y. Endo, O. Kitakami, and Y. Shimada, Phys. Rev. B 59, 4279 3 F. Yoshizaki and T. Kingetsu, Thin Solid Films 239, 229 1994 . 1999 . 4 K. Kagawa, H. Kano, A. Okabe, A. Suzuki, and K. Hayashi, J. 14 C. C. Yu and A. K. Petford-Long, J. Appl. Phys. 85, 5753 1999 . Appl. Phys. 75, 6540 1994 . 15 S. O. Demokritov, J. Phys. D 31, 925 998 . 5 W. F. Egelhoff, P. J. Chen, C. J. Powell, M. D. Stiles, R. D. 16 C. R. Meissner, Transactions of the 7th National Vacuum Sym- McMichael, J. H. Judy, K. Takano, and A. E. Berkowitz, J. posium, AVS Pergamon Press, London, 1960 , p. 196. Appl. Phys. 82, 6142 1997 . 17 http://www.dl.ac.uk/srs/XRD/2.3.dir/ 6 C. H. Marrows, B. J. Hickey, M. Malinowska, and C. Me´ny, 18 J. N. Chapman, J. Appl. Phys. 17, 623 1984 . IEEE Trans. Magn. 33, 3673 1997 . 19 B. Khamsehpour, C. D. W. Wilkinson, J. N. Chapman, and A. B. 7 C. H. Marrows and B. J. Hickey, Phys. Rev. B 59, 463 1999 . Johnston, J. Vac. Sci. Technol. B 14, 3361 1996 . 8 G. Rupp and H. A. M. van den Berg, IEEE Trans. Magn. 29, 20 C. H. Marrows, Ph.D. thesis, University of Leeds, 1997. 3102 1993 . 21 C. C. Kuo, M.-T. Lin, and H. L. Huang, J. Appl. Phys. 85, 4430 9 B. Heinrich, J. F. Cochran, T. Monchesky, and R. Urban, Phys. 1999 . Rev. B 59, 14 520 1999 . 22 C. H. Marrows, N. Wiser, B. J. Hickey, T. P. A. Hase, and B. K. 10 S. M. Rezende, C. Chesman, M. A. Lucena, M. C. de Moura, A. Tanner, J. Phys.: Condens. Matter 11, 81 1999 . Azevedo, F. M. de Aguiar, and S. S. P. Parkin, J. Appl. Phys. 23 J. C. Slonczewski, in Magnetism, edited by G. T. Rado and H. 85, 5892 1999 . Suhl Academic, New York, 1963 , Vol. 1, Chap. 5. 4140 C. H. MARROWS et al. PRB 61 24 M. S. Cohen, E. E. Huber, G. P. Weiss, and D. O. Smith, J. Appl. No. 313 Materials Research Society, Pittsburgh, 1993 , p. 467. Phys. 31, 291S 1960 . 32 J.-F. Bobo, M. Piecuch, and E. Snoek, J. Magn. Magn. Mater. 25 R. D. Heidenrich and F. W. Reynolds, in Structure and Proper- 126, 440 1993 . ties of Thin Films, edited by C. A. Neugebauer, J. B. Newkirk, 33 J. C. Slonczewski, Phys. Rev. Lett. 67, 3172 1991 . and D. A. Vermilya Wiley, New York, 1959 , p. 402. 34 K. Pettit, S. Gider, S. S. P. Parkin, and M. B. Salamon, Phys. Rev. 26 C. H. Marrows, B. J. Hickey, M. Herrmann, S. McVitie, and J. N. B 56, 7819 1997 . Chapman, J. Magn. Magn. Mater. 198, 408 1999 . 35 J. N. Chapman, J. Rose, P. R. Aitchison, H. Holloway, and D. J. 27 C. H. Marrows and S. Langridge unpublished . Kubinski, J. Appl. Phys. 86, 1611 1999 . 28 C. H. Marrows, R. Loloee, and B. J. Hickey, J. Magn. Magn. 36 R. Miranda, Phys. Scr. T49, 579 1993 . Mater. 184, 137 1998 . 37 D. Greig, M. J. Hall, C. Hammond, B. J. Hickey, H. P. Ho, M. A. 29 C. H. Marrows, F. E. Stanley, and P. W. Mitchell unpublished . Howson, M. J. Walker, N. Wiser, and D. G. Wright, J. Magn. 30 S. Langridge, J. Schmalian, C. H. Marrows, D. T. Dekadjevi, and Magn. Mater. 110, L239 1992 . B. J. Hickey, cond-mat/9906145 unpublished . 38 W. F. Egelhoff and M. T. Kief, Phys. Rev. B 45, 7795 1992 . 31 J.-F. Bobo, H. Fischer, and M. Piecuch, in Magnetic Ultrathin 39 See, e.g., Refs. 2 and 22, and further citations therein Films, Multilayers and Surfaces/Interfaces and Characteriza- 40 R. J. Highmore, W. C. Shih, R. E. Somekh, and J. E. Evetts, J. tion, edited by B. T. Jonker et al., MRS Symposia Proceedings Magn. Magn. Mater. 116, 249 1992 .