Journal of Magnetism and Magnetic Materials 191 (1999) 301-312 Exchange coupling in Co/Cu/Co sandwiches studied by spin-polarized low energy electron microscopy T. Duden, E. Bauer* Department of Physics and Astronomy, Arizona State University, Tempe, AZ 85287-1504, USA Received 24 June 1998; received in revised form 4 September 1998 Abstract The evolution of the magnetic domain structure of Co/Cu/Co sandwiches epitaxially grown on a W(1 1 0) surface under UHV conditions is studied in situ by spin-polarized low energy electron microscopy as a function of the Cu spacer and Co top layer thickness in the thickness range from 2 to 9 monolayers Cu and 1 to 7 monolayers Co. The various coupling modes are correlated with the microstructure of the layers as observed by low energy electron micros- copy. 1999 Elsevier Science B.V. All rights reserved. PACS: 75.70.Cn; 75.70.Kw; 75.30.Et Keywords: Exchange coupling; Sandwiches; Microstructure; SPLEEM; Mean free path 1. Introduction antiferromagnetic (AF). For a given ferromagnetic layer thickness this occurs at certain spacer thick- The exchange interaction between ferromagnetic nesses which are determined by the Fermi surface layers separated by a nonmagnetic spacer layer has of the spacer material. Many theoretical treatments been the subject of intensive study for nearly a dec- of this problem have been published (for references ade, driven in part by its fundamental interest but see Refs. [1,2]), the most flexible being the quantum mainly by its importance for the understanding of well or quantum interference model [1]. the giant magnetoresistance (GMR) which has im- A large GMR is to be expected when the band portant applications in magnetic storage and sens- structure of the majority electrons near the Fermi ing devices. In layered systems GMR can occur surface is similar to the band structure of the spacer when the magnetization in alternating layers is material and the band structure of the minority antiparallel, that is when the exchange coupling is electrons is quite different and vica versa. The sys- tem Co/Cu is particularly favorable from this point of view. Some of the highest GMR effects have indeed been observed in sputtered Co/Cu multi- * Corresponding author. Tel.: #1-602-965-2993; fax: #1- layers. In Co/Cu multilayers grown by molecular 602-965-7954; e-mail: bauer@asu.edu. beam epitaxy (MBE) AF coupling and GMR have 0304-8853/99/$ - see front matter 1999 Elsevier Science B.V. All rights reserved. PII: S 0 3 0 4 - 8 8 5 3 ( 9 8 ) 0 0 3 6 3 - 1 302 T. Duden, E. Bauer / Journal of Magnetism and Magnetic Materials 191 (1999) 301-312 been more elusive, in particular in layer systems in-plane components of the magnetization could be with (1 1 1) orientation. This has led to numerous imaged. The addition of a spin manipulator [13] studies of these systems (for references see Refs. allows now to image all three components so that [3-5]) with the result that AF coupling and GMR the direction of M can be completely determined at are extremely sensitive to the microstructure of the the present lateral routine resolution of several tens layers. No significantly reduced or incomplete AF of nanometers. With this addition we have made coupling have been reported for MBE grown films a systematic study of the dependence of the mag- and only recently completely AF coupled sand- netic domain structure of Co/Cu/Co sandwiches wiches could be grown [6,7] using a surfactant upon the thickness of the spacer layer and the top which improved layer-by-layer growth and sup- Co layer in order to obtain a deeper insight in the pressed stacking fault formation [8]. relationship between exchange coupling and micro- The strong interrelation between exchange coup- structure. ling and microstructure which is evident from nu- merous laterally averaging ex situ studies makes it desirable to correlate on one and the same layer 2. Experimental system in situ the magnetic domain structure - which reflects the coupling - and the microstruc- The experiments were performed in the orginal ture. This can be done best by spin-polarized low LEEM instrument described in Ref. [36] in which energy electron microscopy (SPLEEM) [9]. This the original field emission gun was replaced by method combines the diffraction and interference a spin polarized illumination system with polariza- contrast mechanism of low energy electron micros- tion manipulator (see also Fig. 3 in Ref. [9]). The copy (LEEM) [10,34] with the magnetization (M) base pressure of the instrument was 2;10\ Torr. sensitivity provided by the exchange interaction During the depositions the pressure stayed in the between the target electrons and the beam electrons 10\ Torr range and was typically around with spin polarization P. The most important 6;10\ Torr. The W(1 1 0) crystal could be LEEM contrast mechanism in the present context heated from the back side by radiation up to 500 K are the quantum size contrast, which allows to and by electron bombardment up to 2000 K. It was determine local thickness variations in the growing precleaned by heating for several hours in an oxy- layers with atomic depth sensitivity, and the step gen atmosphere at a partial pressure of about contrast which allows to image substrate steps 2;10\ Torr. Between the experiments it was down to atomic height. The magnetic contrast cleaned regularly by annealing at approximately - which is usually weak - is proportional to P ) M 1400 K in 5;10\ Torr oxygen for 30 min, fol- and superimposed on the structural contrast but lowed by flashing to 2000 K in UHV. Criteria for can be obtained in pure form by producing the a clean surface were (i) the absence of W carbide difference image between images taken with oppo- segregation at surface imperfections upon anneal- site P. The intimate relation between microstruc- ing at about 1300 K and (ii) step flow growth of the ture and magnetic domain structure has been first Co monolayer during the deposition at 750 K. demonstrated in this manner for example for epi- This growth pattern is very sensitive to surface taxial Co layers on a W(1 1 0) surface [11]. An contamination by segregated or adsorbed impu- important aspect is that these studies can be made rities which cause pinning of the growth fronts and in situ while the film system is growing. nucleation on the terraces. The first monolayer is Preliminary SPLEEM studies of the exchange filled in two steps: first, a pseudomorphic (ps) coupling in Co/Cu/Co sandwiches [12] have al- monolayer is formed in which close-packed (cp) ready shown that at the Cu thickness for which AF islands nucleate and grow until the cp monolayer is coupling has been reported in the past the domain completed. The completion of the ps and the cp pattern of the overlayer was much more complic- monolayer provides a precise rate calibration be- ated than expected in the case of AF coupling. At fore each experiment. After completion of the cp the time when these studies were made only the monolayer the temperature was reduced to about T. Duden, E. Bauer / Journal of Magnetism and Magnetic Materials 191 (1999) 301-312 303 400 K and the deposition continued to the desired 3. Results thickness (7 ML). At this temperature the mobility is high enough and the two-dimensional nucleation The experiments were performed for Cu spacer rate low enough so that large terraces form (several thicknesses ranging from 2 to 9 ML in 1 ML steps 100 nm diameter) which show pronounced thick- with intermediate values of 4.25 and 4.5 ML in the ness dependent quantum size contrast. This con- AF coupling region. The kind of SPLEEM images trast allows the observation of the completion of obtained during a typical experiment is illustrated consecutive layers and the characterization of the in Fig. 1 for a few selected thicknesses of the top Co roughness of the Co film. Once the desired Co film layer. In this example the Cu spacer is 8 ML thick. thickness was reached, the heating was turned off. The three columns show the lateral distribution of After the temperature had dropped to values slight- the three M components obtained with P parallel to ly above room temperature Cu was deposited as W[1 1 0], W[0 0 1] and W[1 1 0] from the left to a spacer layer. The Cu rate was calibrated before the right. The easy in-plane axis of the bottom the Co deposition by the time needed to form Co layer is along the W[1 1 0] direction but there is 1 ML. While the completion of the first monolayer also a weak out-of-plane component (M is well observable the next two layers seem to grow ,/M, +1 : 6 [11]). The images in row (a) are taken from in double layer islands. The top Co layer was de- the bottom Co layer covered with 8 ML Cu. The posited in ML doses. Typical deposition rates were M distributions are completely identical with those 1/8 ML/min both for Co and Cu. of the bare Co layer, only the signal/noise ratio is After each monolayer dose, a measurement cycle reduced by the spin-independent attenuation in the was performed to monitor the resulting magnetic Cu layer. 1 ML Co on top strongly reduces the structure. The images were acquired from the final contrast and at 2 ML Co the magnetic contrast has screen using a CCD camera. For each magnetic disappeared nearly completely (not shown). At image, two images resulting from the average of 64 3 ML Co strong contrast appears in the [0 0 1] consecutive video frames were taken. Between each (90°) image with a completely different M distribu- single image the polarization vector of the incident tion while only very weak contrast is seen in the electron beam was inverted. The magnetic signal [1 1 0] (0°) image (row b). At 4 ML Co (row c) the was then obtained by a normalized subtraction contrast in the 0° image has increased considerably using the formula and continues to do so with increasing Co thick- ness while the contrast in the 90° image decreases A"127#100K (I beyond 4 ML as seen in the 7 ML Co images >!I\)/(I>#I\), (row d). This change of the domain structure with top Co where A is the normalized asymmetry, K a contrast layer thickness, in which the 90° component ap- enhancement factor ranging from 7 to 15 and pears first with increasing thickness is characteristic I>, I\ are the intensities for the images with oppo- for 3, 4, 8 and 9 ML thick spacer layers but at 6 and site spin polarization. The direction of M can be 7 ML Cu the 0° component develops faster than determined to better than $3° by maximizing/ the 90° component. At 4.25 and 4.5 ML both ap- minimizing the magnetic contrast which is propor- pear simultaneously and at 5 ML no clear decision tional to P ) M. The direction of P which is needed can be made. At 2 ML Cu only a contrast minimum for this determination is calibrated using films with was observed at 2 ML Co and the magnetization strong perpendicular M (4 ML Co/Au(1 1 1)) and distribution in the bottom layer was perfectly trans- strong in-plane anisotropy (Co/W(1 1 0)). For ar- ferred to the top layer, presumbly via direct F coup- bitrary angles P is obtained from the P ) M depend- ling through gaps in the Cu spacer. The evolution ence of the contrast in these layers. Initially the of the angular M distribution with top Co layer relationship between the direction of P and the thickness can be quantified somewhat by calculat- settings of the spin manipulator were determined ing pixel by pixel the orientation of M from the grey with a Mott detector [13]. levels in the three component images and plotting it 304 T. Duden, E. Bauer / Journal of Magnetism and Magnetic Materials 191 (1999) 301-312 Fig. 1. Typical SPLEEM image series of a 7 Co/8 Cu/7 Co sandwich. (a) Uncovered bottom Co layer +8 Cu/7 Co. (b) 3 Co/8 Cu/7 Co. (c) 4 Co/8 Cu/7 Co. (d) 7 Co/8 Cu/7 Co. Electron energy +1.2 eV, field of view +6;6 m. T. Duden, E. Bauer / Journal of Magnetism and Magnetic Materials 191 (1999) 301-312 305 Fig. 2. Evolution of the angular M distribution with increasing thickness of the top Co layer (a) from a sandwich just below the AF coupling regime (4 ML Cu spacer), (b) from a sandwich close to the F coupling maximum (8 ML Cu spacer). Center of the projected unit sphere: [1 1 0], left side: [0 0 1], right side: [0 0 1], top: [1 1 0], bottom: [1 1 0]. in a locally orthogonalized projection of the unit for reference, in the center the 0° image of the sphere. This presentation of the angular M distribu- complete sandwich and on the right side the corre- tion is shown in Fig. 2. Series (a) has been taken at sponding 90° image. At 3 ML Cu (a-c) there is only 4 ML Cu just below the AF coupling maximum, very weak fine-grained contrast in the center but series (b) at 8 ML Cu within the F coupling regime. pronounced contrast on the right indicating domi- In both cases, the magnetization reappears prefer- nating biquadratic coupling. At 4 ML (d-f) the con- entially in the W[0 0 1] direction but with increas- trast in the center has increased somewhat but ing Co thickness two completely different M without recognizable F or AF coupling and the distributions develop: just below the AF coupling right image shows even stronger contrast than at maximum the distribution has weak broad maxima 3 ML. Thus biquadratic coupling seems to be near "90°, in the F coupling region the angular strongest between 3 and 4 ML. At 4.25 ML (g-i) M distribution develops a sharp maximum again at and 4.5 ML (j-l) clear AF coupling is evident and "0°. the 90° image (right side) contrast is decreasing. The dependence of the final spatial M distribu- This trend continues into the F coupling region as tion upon spacer thickness is illustrated by the illustrated by the 7 ML spacer images (m-o). Cal- series of images of Fig. 3 which shows in the left culation of the angular M distributions as described coulumn the in-plane M image of the bottom layer above gives a more quantitative picture of the 306 T. Duden, E. Bauer / Journal of Magnetism and Magnetic Materials 191 (1999) 301-312 Fig. 3. Selected SPLEEM images illustrating the evolution of the domain structure of the completed sandwiches with Cu spacer thickness. The left column shows the in-plane images of the bottom Co layer which serve as a reference. The center column shows the images of the top Co layer taken with the same P direction (0° images), the right column the images with P in the perpendicular in-plane direction (90° images). The Cu spacer thickness is 3, 4, 4.25, 4.5, 7, 8 and 4.5 ML from top to bottom. The top Co layer is 6 ML thick in all cases. The last 4.5 ML Cu spacer sandwich was grown at 400 K. Electron energy +1.4 eV, field of view +6;6 m. T. Duden, E. Bauer / Journal of Magnetism and Magnetic Materials 191 (1999) 301-312 307 Fig. 3. Continued. content of Fig. 3. Fig. 4 shows a section through the unit spheres of Fig. 2 along the equator for the complete sandwiches. This section gives the in- plane angular M distribution. Clearly, at 2, 6, 7 and 8 ML Cu the maxima occur at "0° and 180°, at 3 ML Cu at 90° and at 4 ML Cu at 100°. The difference between 3 and 4 ML is outside the limits of error and indicates a small bilinear component admixture at 4 ML to the nearly 100% biquadratic coupling at 3 ML. From Fig. 3g-Fig. 3l it is also clear that there is never pure AF coupling but always a mixture with biquadratic coupling. In order to obtain some insight in how sensitive the coupling is to the deposition conditions the 4.5 ML spacer sandwich was also deposited at elev- ated temperature (+400 K) at which significantly larger Cu crystals grow than close to room temper- Fig. 4. In-plane angular M distributions in the completed (7 ML ture. Fig. 3q-Fig. 3r shows the M distribution of the thick Co top layer) sandwiches. N is the spacer thickness in ML. complete sandwich. There is no relation with the The zeros of the curves have been shifted as indicated on the domain structure of the bottom Co layer (Fig. 3p), right hand side. thus no AF coupling, and both component images show very fine-grained structure. The difference 308 T. Duden, E. Bauer / Journal of Magnetism and Magnetic Materials 191 (1999) 301-312 Fig. 5. Topographic images of 4.5 ML thick Cu spacer layers deposited close to room temperature (left side) and at elevated temperature (+400 K, right side). The corresponding SPLEEM images are shown in Fig. 3j-Fig. 3l and in Fig. 3p-Fig. 3r. 4. Discussion Before the discussion of the results some com- ments on the growth of Co on Cu(1 1 1) and of Cu on Co(0 0 0 1) have to be made. In the thermodyn- amic limit the growth mode of Co on Cu is not clear because the parameters which enter the growth mode criterion are not known accurately enough [14]. In the bulk, Co and Cu are immiscible but this does not exclude the formation of a surface or interface alloy. The experimental results are con- tradictory. Initial growth of bilayer islands in twin position has been reported on the basis of STM studies [15]. An even more three-dimensional Fig. 6. Differentiated Auger electron signals of Co and of the growth has been deduced from low-energy ion scat- substrate material (Mo, Cu) as a function of deposition time. tering measurements from which it was concluded Room temperature depositions with identical rates. ps and cp that only 40% of the Cu(1 1 1) surface were covered refer to pseudomorphic and close-packed monolayers on the by Co at 1.2 ML thickness and only more than Mo(1 1 0) surface. The continuous transition of the signals 80% at 3 ML Co [16]. On the other hand, there is around 15 min show that the second monolayer starts to grow before the first monolayer is completed. strong evidence that there are no stacking faults in the first two monolayers in which the Co atoms are in FCC positions [17]. This shows that the growth between the magnetic structures of the low and the mode depends sensitively upon the deposition con- high temperature deposits is clearly caused by the ditions and on the state of the substrate surface structural differences which are evident in the topo- used in the different experiments. In the present graphic images shown in Fig. 5. study the Co islands were too small and the T. Duden, E. Bauer / Journal of Magnetism and Magnetic Materials 191 (1999) 301-312 309 contrast too weak as to allow a clear distinction The second point which needs to be addressed between monolayer versus bilayer growth and before the discussion of the images concerns the twinning versus nontwinning. However, the growth interaction of the spin-polarized electrons with the conditions were very similar to those used in earlier target. The information depth of LEEM/SPLEEM precise Auger electron spectroscopy (AES) studies depends upon the mean free path (MFP) of the which indicate initial monolayer growth followed electrons and the image contrast is strongly in- by a slow transition to flat multilayer growth. Fig. fluenced by quantum size effects (QSE). Both, MFP 6 [18] shows the 51 eV Co AES signal on and QSE are spin-dependent. Very little is known a Mo(1 1 0) surface and on a very flat Cu(1 1 1) about the MPF at the low energies used in layer on this substrate as a function of deposition SPLEEM (1-2 eV). Energy-averaged values for time under identical conditions. Co grows on slow secondary electrons show a linear dependence Mo(1 1 0) up to 2 ML layer-by-layer and continues upon the number of holes in the d band. The value to grow in a flat three level mode [19] which is also for Co is about 8 As [24]. For Cu a MFP of 10 As verified by LEEM on W(1 1 0) [20]. The AES curve has been reported originally [25] which is abnor- for Co on Cu(1 1 1) is very similar with minor mally small for a metal with a filled d shell. A more differences, which are due to differences in the back- recent value is 19 As [24]. Whatever the exact values scattering from the substrate. This is a clear of the MFPs may be, SPLEEM image analysis indication of monolayer-by-monolayer growth, in must take into account the possibility that the agreement with many other studies [21]. sampling depth may be as much as several 10 As. The wetting of the Cu surface by one ML of Co is The spin dependence of the MFP brings an addi- quite plausible: the atomic diameters of Co and Cu tional complication. The MFPs for spin-up and in the bulk are 2.506 and 2.556 As, respectively. The spin-down electrons differ increasingly with de- surface layer wants to have a larger interatomic creasing energy if ineleastic scattering dominates distance which is provided by the Cu substrate so the MFP [26]. At very low energies the MFP for that the misfit favors monolayer formation. In the spin-up electrons may be several times larger than reverse case, Cu on Co, the situation is different: the that for spin-down electrons [26]. For electrons tendency to a larger interatomic spacing than in the 1.5 eV above the Fermi level a MFP of 14.7 As for bulk increases the surface stress and make layer- spin-up electrons and of 5.4 As for spin-down elec- by-layer growth unfavorable. Indeed, LEEM/ trons has been deduced from spin-polarized photo- SPLEEM studies of Co layers with very thin Cu emission measurements of Co on Cu, assuming overlayers indicate that although a monolayer a MFP of 14 As in Cu [27]. forms initially, the next layer grows via bilayer The SPLEEM images shown in Section 3 are islands [22]. This is in apparent disagreement with taken between 1 and 2 eV above the vacuum level, a very recent study of the early stages of epitaxial that is in an energy range in which spin-up elec- growth of Cu on the (0 0 0 1) surface of bulk Co. In trons can couple to allowed states in the volume this study it was concluded that Cu grows approx- but not spin-down electrons which, therefore, can imately layer-like up to 4 ML and only thereafter exist in the crystal only as evanescent waves. How- forms three-dimensional islands [23]. However, it ever, these have a relatively large decay length has to be kept in mind that in these experiments Cu (small imaginary part of k) because the band gap in was deposited without interruption and at much the [0 0 0 1] direction is wide and the energy of the higher rate than here. Under these conditions growth electrons is close to the band edge [9]. As a conse- is much more determined by kinetics than in the quence, the spin-down electrons travel far enough present work which was done closer to equilibrium, to loose energy - to be `absorbed' - which reduces favoring the Stranski-Krastanov growth noticed also their reflected intensity I in Ref. [23]. The subsequent discussion is, therefore, \ and, thus, the magnetic contrast. For the MFP in Co mentioned above based on the model of a relatively flat bottom three- (5.4 As) I level Co/Cu interface with large terraces and a some- \ is reduced to 47, 22 and 5% of the intensity in the absorption free case in a 1, 2 and what rougher top Cu/Co interface with small terraces. 4 ML thick film due to inelastic scattering on the 310 T. Duden, E. Bauer / Journal of Magnetism and Magnetic Materials 191 (1999) 301-312 way in and out. In Cu, on the other hand, absorp- F coupling which can be deduced from these mac- tion is much weaker so that Cu overlayers weaken roscopic measurements occurs at about 7.6 ML Cu. the magnetic contrast much less than Co over- A zero of J layers, in qualitative agreement with the SPLEEM  at 2.5-3 ML is suggested by some experiments. The theoretical value for the J observations.  oscillation period is 4.50 ML [31] which would Finally, the QSE which has long been used in give a zero of J LEEM for the determination of film thickness dif-  at about 3.4 ML. The fact mentioned earlier that at 2.5 ML the Cu ferences, has a major influence on the image con- layer consists of double layer islands on top of the trast in SPLEEM. This is due to the fact that the first Cu layer explains why the F contrast is not k values, that is the wavelengths for a given energy, reduced as much at 2 ML Co by inelastic scattering differ for spin-up and spin-down electrons due to as in the case of thicker Cu spacers before the the exchange splitting of the band structure E(k). As contrast increases again with increasing Co thick- a consequence the interference conditions in the ness: there is some ferromagnetic coupling through quantum well differ and the asymmetry shows the first monolayer in the gaps between the double strong oscillations as a function of energy and film layer islands. The weak contrast which becomes thickness. This has been well demonstrated experi- visible in the 90° image above 5 ML Co shows, mentally for Co layers on W(1 1 0) [9,12,28] however, that there is already a biquadratic contri- and theoretically by spin-polarized LEED calcu- bution to the coupling. At 3 ML Cu biquadratic lations [29]. coupling is dominating as seen in Fig. 3a-Fig. 3c. Considering all these aspects, it is obvious that The domains in the parallel (0°) image are so SPLEEM images have to be analyzed with con- small that they are hardly recognizable in the noise siderable caution as long as no theory is available after it become visible at 3 ML Co and are uncor- which takes proper account of these aspects or related to the domain structure of the bottom Co as long there is not enough empirical experience layer which suggests that the bilinear coupling in the analysis. While the local direction of M parameter J can be determined unambiguously - except +0. The large domains in the 90° image interestingly show a certain correlation possibly for the sign due to QSE contrast reversals which is determined by the substrate step structure. at certain energies with increasing thickness [30] At 4 ML Cu there is still no clear correlation with - the magnitude of M cannot be determined at the domain structure in the substrate in the 0° present. In spite of these limitations valuable in- image but the image is less fine-grained (Fig. 3e). formation can be extracted from the images shown The contrast in the 90° image (Fig. 3f) is still larger in Figs. 1-5. The image interpretation will be based than that in the 0° image but the domain size is now on the assumption - based on the strong absorp- smaller than at 3 ML. This indicates that biquad- tion of spin-down electrons in Co discussed above ratic coupling is still dominating but weaker than - that all images with more than 3 ML Co in at 3 ML Cu. the top layer show only the magnetization of the The equal contrast in the two images at 4.25 ML top layer. Cu (Fig. 3h and Fig. 3i) and the somewhat weaker For comparison with the present data, some re- contrast in the 90° image than in the 0° image at sults from the literature should be recalled. The 4.5 ML Cu (Fig. 3k and Fig. 3l) show the decreasing average value - obtained from many studies of the influence of the biquadratic coupling near the AF Co/Cu/Co system - of the Cu thickness at which maximum believed to be between these two thick- the first maximum of the magnetoresistance or of nesses. The AF coupling is evident from the com- the saturation field occurs is 4.15 ML, that of the parison of Fig. 3g and Fig. 3h and of Fig. 3l and second maximum 9.2 ML. These maxima are at- Fig. 3m. At 5 ML Cu (not shown) there is already tributed to AF coupling, that is to negative extrema weak F coupling in the 0° image with good correla- of the bilinear coupling parameter J. AF coupling tion to the bottom layer domain structure but weak is believed to occur only in part of the layer system, contrast which is comparable again with the 90° the majority being F coupled. The maximum of the contrast. Thus J is positive but small. At 6 ML Cu T. Duden, E. Bauer / Journal of Magnetism and Magnetic Materials 191 (1999) 301-312 311 (not shown) there is already strong F coupling images during the growth. Just as in the case of the again which persits up the the largest spacer thick- 2.5 ML Cu sandwich grown at room temperature, ness studied (9 ML) as illustrated in Fig. 3n and the original domain structure is still well visible Fig. 3o for 7 ML Cu. Interesting are the changes in at 2 ML Co where it is barely noticeable in the 90° images: at 6 ML Cu the contrast is still the other sandwiches. At 3 ML Co there is practic- increasing with Co layer thickness in the same ally no contrast in the 0° image while in the manner as in the 0° images but at 7 ML CU and in 90° image the first signs of the contrast seen in particular at 8 ML Cu it passes through a max- Fig. 3r already develop. This can be understood imum at 3-4 ML Co while the 0° contrast is still in the same manner as the 2.5 ML images: the weak and increasing. This trend is still noticeable Cu layer is broken up into large crystals which though weaker at 9 ML Cu. At 7 and 8 ML Cu the allows close proximity of the top layer to the bot- 90° contrast is nearly vanishing which is compatible tom layer in between them, attenuating initially the with a maximum of J close to these thicknesses. domain structure of the bottom layer only partially. This dependence of J is in good agreement with As soon as the top layer becomes magnetic (at that derived in the past from the macroscopic 3 ML Co) it develops a fine grained magnetization measurements but there is a major difference in the which is determined by the roughnees of the inter- sign of the coupling: The macroscopic measure- face, with locally strongly varying coupling. ments which all involve an external field indicate The magnetoresistance in this case is now mainly always some AF coupling above 3 ML Cu while due to the fine grains with varying magnetization the microscopic zero field data presented here show directions. only F coupling from 6 ML Cu onwards in the 0° images. Apparently, the GMR is not due to AF coupling but caused by the breakup of the domains 5. Summary in subdomains with wrinkled magnetization due to the presence of biquadratic coupling. This sugges- The microscopic view of the effects of interlayer tion is supported by the observation that the bi- coupling in zero field presented here shows how quadratic contribution is smallest at the minimum complex the resulting domain structure is. The do- of the magnetoresistance between 7 and 8 ML Cu main structure of the bottom layer is certainly and is in partial agreement with the conclusion that strongly influenced by the topography of the sub- the GMR is due to 90° coupling [32,35]. Interesting strate as shown elsewhere [11] but the domains are and not understood is the fact that the top layer large enough so that the local coupling may be develops at 6 ML Cu and above initially, that is at considered to be representative also for the more 3 ML Co, predominantly 90° magnetization while ideal case of very large domains. The results con- the F coupling is still building up, and then is firm in general the conclusions drawn from macro- increasingly replaced by F coupling with increasing scopic measurements in an external field regarding thickness. Apparently the ratio of interface rough- the coupling but differ also in several aspects. In ness to average Co layer thickness plays an impor- particular, the strong dependence of the type of tant role for the magnitude of the biquadratic coupling - bilinear or biquadratic - upon the thick- coupling. ness of the spacer layer and of the top magnetic Finally, the results obtained during the growth of layer is new. It appears that biquadratic coupling the sandwich with 4.5 ML Cu at the elevated tem- builds up faster with increasing top magnetic layer perture have to be discussed yet. Fig. 3p-Fig. 3r thickness than bilinear coupling, except for spacer shows that the final domain structure has no rela- thicknesses close to maximum F coupling, and dis- tionship to that of the bottom layer and that 0° and appears with increasing bilinear coupling. Below 90° image have the same contrast without preferred the thickness with maximum AF coupling biquad- contrast levels, indicating that there is no preferred ratic coupling dominates even at larger top layer magnetization direction. The cause for this loss of thicknesses. That these effects have not been magnetic order is evident from the evolution of the noticed before may be due to differences in the film 312 T. Duden, E. Bauer / Journal of Magnetism and Magnetic Materials 191 (1999) 301-312 structure but at least also in part due to the fact [10] E. Bauer, Rep. Prog. Phys. 57 (1994) 895. that the zero magnetization configuration in [11] T. Duden, E. Bauer, Phys. Rev. Lett. 77 (1996) 2308. measurements in an external field (H"H [12] E. Bauer et al., J. Magn. Magn. Mater. 156 (1996) 1. ) is quite different from that of the virgin state (H"0). An- [13] T. Duden, E. Bauer, Rev. Sci. Instrum. 66 (1995) 2861. [14] J.H. van der Merwe, E. Bauer, Phys. Rev. B 39 (1989) other possible reason is that biquadratic coupling 3632. produces clear domain patterns in SPLEEM while [15] J. de la Figuera et al., Phys. Rev. B 47 (1993) 13043. Lorentz microscopy shows only wavy boundaries [16] A. Rabe et al., Phys. Rev. Lett. 73 (1994) 2728. with relatively low contrast [33]. An additional [17] M. Hochstrasser et al., Phys. Rev. B 50 (1994) 17705. advantage of SPLEEM is that it can be done on [18] M. Tikhov, E. Bauer, Unpublished. [19] M. Tikhov, E. Bauer, Surf. Sci. 232 (1990) 73. bulk substrates and during film growth. [20] T. Duden et al., Unpublished [21] Ch. Rath et al., Phys. Rev. B 55 (1997) 10791. [22] T. Duden, E. Bauer, Phys. Rev. B, in press. Acknowledgements [23] J.E. Prieto et al., Surf. Sci. 401 (1998) 248. [24] H.C. Siegmann, Surf. Sci. 307-309 (1994) 1076. The authors acknowledge the loan of the [25] D.T. Pierce, H.C. Siegmann, Phys. Rev. B 9 (1974) SPLEEM equipment by the TU Clausthal, Ger- 4035. many. [26] F. Passek, M. Donath, K. Ertl, J. Magn. Magn. Mater. 159 (1996) 103. [27] J.C. Groebli, D. Oberli, F. Meier, Phys. Rev. B 52 (1995) R13095. References [28] K. Wurm, M.S. Thesis, TU Clausthal, 1994. [29] T. Scheunemann et al., Solid State Commun. 30 (1997) [1] P. Bruno, Phys. Rev. B 52 (1995) 411. 1. [2] B.A. Jones, IBM J. Res. Dev. 42 (1998) 25. [30] H. Poppa et al., MRS Symp. Proc. 313 (1993) 219. [3] B. Heinrich, J.F. Cochran, Adv. Phys. 42 (1993) 523. [31] P. Bruno, C. Chappert, Phys. Rev. Lett. 67 (1991) 1602. [4] B. Heinrich, J.C.A. Bland (Eds.), Ultrathin Magnetic Struc- [32] Z.J. Yang, M.R. Scheinfein, Appl. Phys. Lett. 66 (1995) 236. tures II, Springer, Berlin, 1994, Ch. 2. [33] J.D. Kim et al., J. Appl. Phys. 76 (1994) 6513. [5] R.F.C. Farrow, IBM J. Res. Dev. 42 (1998) 43. [34] E. Bauer, in: S. Amelinckx, et al., (Eds.), Handbook of [6] J. Camarero et al., Phys. Rev. Lett. 76 (1996) 4428. Microscopy, VCH, Weinheim, 1997, p. 487. [7] W. Kuch, et al., J. Magn. Magn. Mater. 170 (1997) L13. [35] Z.J. Yang, M.R. Scheinfein, Phys. Rev. B 52 (1995) 4263. [8] J. Camarero et al., Phys. Rev. Lett. 73 (1994) 2448. [36] E. Bauer, W. Telieps, in: A. Howie, U. ValdreŽ (Eds.), [9] E. Bauer, in: S. Amelinckx, et al., (Eds.), Handbook of Surface and Interface characterization by Electron Optical Microscopy, VCH, Weinheim, 1997, p. 751. Methods, Plenum, New York, 1988, p. 195.