PHYSICAL REVIEW B VOLUME 58, NUMBER 13 1 OCTOBER 1998-I Magnetic behavior of probe layers of 57Fe in thin Fe films observed by means of nuclear resonant scattering of synchrotron radiation L. Niesen, A. Mugarza,* and M. F. Rosu Nuclear Solid State Physics, Materials Science Centre, Groningen University, Nijenborgh 4, 9747 AG Groningen, The Netherlands R. Coehoorn, R. M. Jungblut, and F. Roozeboom Philips Research Laboratories, Box WA-14, 5656 AA Eindhoven, The Netherlands A. Q. R. Baron, A. I. Chumakov, and R. Ru¨ffer European Synchrotron Radiation Facility, BP 220, F-38043 Grenoble, France Received 4 November 1997 The magnetic behavior of epitaxial probe layers of 57Fe down to a thickness of 1 monolayer ML has been investigated with the technique of nuclear resonant scattering by synchrotron radiation NRS in a grazing- incidence geometry. The samples consisted of 10­55 ML Fe deposited onto a Ge 100 substrate and covered with 2 nm Au. Probe layers of 1­10 ML 57Fe were inserted at different depths in the Fe film. The technique yields spectroscopic information, i.e., magnetic hyperfine fields and isomer shifts, as well as structural infor- mation, such as layer thicknesses and interface roughness. The results show the existence of a nonmagnetic Ge/Fe interlayer of at least 10 ML thick after deposition at room temperature. Subsequent conversion electron Mo¨ssbauer spectroscopy CEMS data show that, although the samples were stored at room temperature, the interlayer diffusion proceeds as a function of time. The relative merits of NRS and CEMS for the investigation of ultrathin layers are discussed. S0163-1829 98 03137-3 I. INTRODUCTION opened up the possibility for observation of the hyperfine interaction of minute quantities of 57Fe by nuclear resonant Metallic multilayers in which one of the components is scattering in a grazing-incidence geometry.4 Similar beam magnetic are currently a topic of much interest because they lines are completed, respectively under construction at the show interesting phenomena like ferromagnetic/ American Photon Source APS near Chicago and at the antiferromagnetic coupling as a function of nonmagnetic SPring8 facility near Hyogo, Japan. Earlier, this technique spacer thickness, giant magnetoresistance and perpendicular has already been applied successfully to the study of thicker anisotropy.1 It has become clear that in many cases the mag- multi layers.5,6 A sample of 57Fe nuclei is excited by a short netic behavior is intimately associated with the structure of 100 ps pulse of synchrotron radiation. If a hyperfine inter- the interfaces.1 One of the key issues, therefore, is the de- action is present, the time evolution of the nuclear scattering scription of interface magnetism in terms of different atomic shows a characteristic beat pattern, with frequencies corre- sites at the interface. sponding to the energy level differences, superimposed on a Mo¨ssbauer spectroscopy on magnetic multilayers contain- decay with a characteristic time of the order of the lifetime of ing 57Fe is a well known technique for the study of the mag- the nuclear level 141 ns . In contrast, the electronic response netic behavior of these layers. Similar to the magnetic mo- of the system elastic scattering, photoeffect, Compton ef- ment, the magnitude and direction of the observed magnetic fect is ``prompt'' on this time scale. Even though the elec- hyperfine field is directly related to the local electronic struc- tronic response is normally much stronger than the nuclear ture of the Fe atoms. At the interface, where the local sym- response, the latter can be observed with very low back- metry is broken, one can also observe the interaction be- ground in a suitable time window after the exciting pulse.4 tween the quadrupole moment of the 57Fe nucleus and the Provided one can obtain a sufficiently high counting rate, resulting electric field gradient the quadrupole interaction . this method will be more sensitive than the CEMS technique. Because the magnetic hyperfine field is sensitive to the direct Another advantage is that one can easily study the hyperfine surroundings of the probe nucleus, this quantity can also give interaction as a function of temperature and/or external mag- structural information. Interface roughness, for instance, can netic field. Finally, because the beam is nearly 100% polar- be studied by inserting the 57Fe probes at different depths ized in the horizontal plane, one obtains more detailed infor- from the interfaces and measuring the resulting distribution mation about the directional distribution of the magnetic of hyperfine fields. By detecting the conversion electrons moments in the sample. emitted by the 57Fe nucleus that absorbed the 14.4 keV pho- In order to test the possibilities of this technique for the ton one obtains a sensitivity of about 1 monolayer. The limit study of thin layer magnetism, we have chosen the system is determined by the nonresonant background. This CEMS Fe/Au 001 , which can be grown epitaxially with good qual- technique has been applied for instance to the system Fe/Ag ity on a Ge substrate.7,8 The results show that it is possible to 100 .2,3 measure the signal of one monolayer of 57Fe. Contrary to The recent construction of a nuclear resonance beamline CEMS, the sensitivity limit is determined by counting rate at the European Synchrotron Radiation Facility ESRF has rather than background. Although in particular Anderson 0163-1829/98/58 13 /8590 6 /$15.00 PRB 58 8590 © 1998 The American Physical Society PRB 58 MAGNETIC BEHAVIOR OF PROBE LAYERS OF 57Fe . . . 8591 et al.8 claim a sharp interface between the S-passivated Ge was very low 0.01­0.02 s 1 . Time spectra were obtained substrate and the Fe overlayer, our results demonstrate a con- using standard fast time electronics.4 siderable interdiffusion, which proceeds as a function of time. III. THEORETICAL DESCRIPTION OF NUCLEAR RESONANCE SCATTERING IN GRAZING INCIDENCE II. EXPERIMENT In a nuclear resonance scattering experiment the sample is Samples were prepared in the metal-MBE set up at Phil- excited by a photon pulse which is so short that its frequency ips Research Laboratories in Eindhoven. Three samples were spectrum is much broader than the typical hyperfine interac- grown on a Ge 001 substrate of 0.3 mm thickness and di- tion spectrum associated with the splitting of the ground and mensions 18 14 mm2. The substrate preparation was done excited nuclear state. Whereas the response of the electrons in the same way as in Ref. 8. Prior to introduction in the is prompt, the nuclear time response is basically governed by UHV chamber the substrate was rinsed with HF and treated the lifetime of the excited state. The response displays a with (NH characteristic beat pattern if the nuclear levels are split by the 4)2S for 20 min at 70 °C. This leads to the forma- tion of a S-passivated surface layer.8,9 The substrates were hyperfine interaction, but it depends also on the geometric annealed at 190 °C for 2 h in UHV before deposition. arrangement of the scatterers and on the incidence angle. Samples A and B consisted of 40 monolayers ML, 1 This is described by the dynamical theory of nuclear reso- ML 0.143 nm natural Fe grown on the Ge substrate, fol- nance scattering.11,12 In this theory, the nuclear resonant layer is described by a frequency-dependent complex refrac- lowed by 10 ML of 57Fe and another 5 ML of natural Fe. Thicknesses were measured by means of a quartz crystal tive index n 1 , where is proportional to the number microbalance. LEED experiments performed right after density of recoilless resonant scatterers and to the nuclear growth showed a good quality Fe 100 pattern. These and all scattering amplitude of an individual 57Fe nucleus. The latter other samples were covered with 2 nm Au to prevent oxi- quantity is a sum of complex Lorentzians describing the in- dation and to provide a well defined interface. For these dividual transitions. The relative amplitudes of these Lorent- samples we use the notation Ge/40Fe/1057Fe/5Fe/Au. The zians depend on the the polarization state of the photons and on the orientation of the photon wave vector k with respect composition of sample C was Ge/5Fe/457Fe/2Fe/Au. Two to the quantization axis used to describe the hyperfine inter- other samples were grown on a 6 mm thick, non S-passivated action. The total frequency dependent reflectivity of the Ge substrate with dimensions 20 45 mm2, sputter-cleaned multilayer, R( ), is calculated by applying the Fresnel co- in situ for 45 min at 700 °C and annealed for 2 h at 780 °C. efficients for reflection and transmission at each interface, Compositions were for sample D: Ge/10Fe/157Fe/4Fe/Au taking into account the propagation through the layers. Inter- and for sample E: Ge/14Fe/157Fe/Au. Sample A was grown face roughness is included in the Nevot-Croce at 200 °C, all other samples at room temperature. Right after approximation.13 In principle this calculation must be per- they were taken out of the MBE set up, the overall magnetic formed for two independent components of the polarization behavior of the first three samples was studied by the vector. However, for the case of 57Fe and the geometry used magneto-optical Kerr effect MOKE . Soft ferromagnetic be- in the present experiment quantization axis perpendicular to havior was observed in all cases when the samples were the beam direction and parallel to the linear polarization vec- magnetized in the 100 easy direction. The magnetization tor we have only one relevant polarization component. In reached saturation already in a field of 3 mT. this case the nuclear scattering amplitude is the sum of the Nuclear resonance scattering was performed at beamline four complex Lorentzians describing the m 1 transi- ID 18 of the ESRF, with the storage ring running in a mode tions between the sublevels of the I 1/2 ground state and with 16 bunches, 176 ns apart. The beam from the undulator was monochromatized in two steps to a bandwidth of 6 meV the I 3/2 excited state of the 57Fe nuclei. After a pulse around the 14413 eV resonance in 57Fe. Typical dimensions excitation, the time dependence of the amplitude of the wave of the photon beam were 0.3 mm vertical and 1.5 mm hori- scattered in the forward direction, G(t 0), is given by zontal. The sample plane was nearly horizontal, with the longest sample dimension 100 making a small, adjustable G t 0 2 1 R Re ei i d , 1 angle with the beam. The scattering plane was vertical and the scattering angle of 2 was defined by slits in front of the where R( ) is the total reflectivity including the nuclear sample and the detector. The scattered photons were detected scattering and Re( ) is the electronic reflectivity. The mea- in a 10 10 mm2 avalanche photodiode EG&G ,10 located sured signal is proportional to G(t 0) 2.14 30 cm downstream from the sample. In order to align the magnetization perpendicular to the beam, a magnetic field of IV. EXPERIMENTAL RESULTS 15 or 44 mT was applied along the horizontal 010 axis, parallel to the linear polarization of the incoming photon 2 scans of both the prompt and the delayed re- beam. Typically, 108 prompt photons/s were collected i.e., flected intensity were obtained for the thicker samples A and 10­20 photons per pulse , while we registered at most 10 s 1 B by varying in an interval 0­25 mrad and adjusting the delayed events from the nuclear resonant scattering process. height of the slit-detector combination. The result for sample Due to the fast time response of the avalanche photodiode A is displayed in Fig. 1. The prompt response Fig. 1 a , APD the delayed counts could be observed without inter- measured with a beam height of 25 m, shows Kiessig ference from the prompt response in an interval 20­160 ns beats15 associated with interference between beams reflected after the exciting photon pulse. The background of the APD from the back and front side of the Fe layer. The solid line is 8592 L. NIESEN et al. PRB 58 FIG. 1. a 2 scan of the specular reflected intensity of 14 403 eV synchrotron radiation with an energy bandwidth of 6 meV impinging on a Ge/Fe/Au layer sample A ; b 2 scan FIG. 2. Time spectra of the nuclear resonant scattering intensity of the delayed reflected intensity due to nuclear resonant scattering, under grazing incidence conditions, for five different Ge/Fe/Au lay- under the same conditions. ers measured at room temperature. a Sample A, 4.7 mrad, b sample B, 4.7 mrad, c sample C, 4.7 mrad, d sample D, a fit using a standard optical formalism, with the thickness of 4.0 mrad, and e sample E, 5.2 mrad. The solid lines are the layers and the roughness of the interfaces as parameters. fits using the formalism described in the text. The thickness of the Fe and Au layer turned out to be 6.3 1 nm and 1.8 1 nm, respectively, while the interface rough- intensity were obtained for various rocking angles between ness was taken to be Gaussian with a FWHM of 0.65 10 nm 3.4 and 9.9 mrad. The spectra at 4.7 mrad for samples A Ge/Fe and Fe/Au and 1.2 2 nm Au/air . For sample B we and B are displayed in Figs. 2 a and 2 b . The spectra at obtained a Fe thickness of 6.7 1 nm, a Au thickness of other incidence angles are only slightly different. They are 1.7 1 nm, while the roughnesses were 0.68 10 , 0.74 10 , typical for a magnetic hyperfine interaction. As explained and 1.4 2 nm for the Ge/Fe, the Fe/Au and the Au/air inter- earlier, the beat pattern originates from the interference be- face, respectively. This means that in both samples the Fe tween the four simultaneously excited m 1 transitions. layer is roughly 20% thinner than expected on the basis of The solid curves in Fig. 2 are based on a calculation using the growth data. The fits are insensitive to the thickness of a the theory described in Sec. III. The structural parameters Ge/Fe interlayer see later , because the refractive indices for were taken from the rocking curve fit. It turned out that in the FeGe and Fe are nearly equal. Thus, only the sum of the case of 10 ML thick probe layers the results are only mod- thicknesses of the Fe/Ge and Fe layers can be determined. It erately sensitive to the geometry of the scatterers, whereas should be noted that the roughness parameter measures the for thinner layers the spectroscopic information is nearly uncertainty in the vertical position of the interface over a completely decoupled from the structural aspects of the lateral range from typically 10 10 to 10 6 m. The rather layer. Adjustable parameters are the vertical scale, the back- large values are therefore not incompatible with the expected ground and the hyperfine interaction parameters determining layer-by layer growth on terraces with a typical length of 10 the complex scattering amplitude. Although a fit with one nm. Figure 1 b shows the delayed rocking curve measured magnetic component yields already a reasonable result, a with the full beam , which has a maximum at the angle quite significant improvement is achieved by allowing for a where the electronic reflectivity has the steepest slope. This second, nonmagnetic, component with a random orientation is the point where the nuclear contribution to the refractive of the electric field gradient. This was done by assuming that index has the largest influence on the frequency spectrum of the scattering amplitudes of the 57Fe nuclei in different sur- the total reflectivity of the multilayer.16 The decrease at roundings could be added coherently. We will discuss the lower angles is also caused by the fact that the projection of validity of this assumption later on. the beam becomes larger than the sample size. The magnetic hyperfine fields corrected for the external For samples A and B time spectra of the delayed reflected field are 32.95 3 T for sample A and 32.99 3 T for sample PRB 58 MAGNETIC BEHAVIOR OF PROBE LAYERS OF 57Fe . . . 8593 B, not significantly different from the bulk value at 295 K. The inhomogeneous linewidth of the magnetic component is only 1.30(5) 0 , where 0 is the natural width of the excited state 0.097 mm/s in Doppler velocity units . This suggests a probe layer with a good structural quality. The relative intensity of the nonmagnetic component is 6 1 % for both samples. Also the hyperfine parameters were similar for both samples, with a quadrupole coupling constant eQVzz /h 20(2) MHz quadrupole splitting 0.86 mm/s and an isomer shift 5.8(4) MHz 0.50(4) mm/s versus the magnetic component. The fit yield lines of natural width but the error is large, 0.6 0 . In contrast to the thicker Fe layers, the spectrum of sample C showed no magnetic oscillations Fig. 2 c . It could be fitted reasonably well with a combination of two quadrupole components with roughly equal intensities, with eQVzz /h 13.2(6) MHz and 25.5 1.0 MHz, respectively. This corresponds to quadrupole splittings of 0.57 and 1.10 mm/s. The isomer shift between the two components is 0.1 MHz and the linewidths are similar, 2.4(6) 0 . We note that the average quadrupole coupling is very close to that found for the quadrupole component in samples A and B, suggesting a common origin. In order to check the results of this spectroscopic tech- nique we performed CEMS experiments on samples A, B, and C by placing them at the cathode of a parallel-plate avalanche detector.17 The results are shown in Fig. 3. Sample FIG. 3. Conversion electron Mo¨ssbauer spectra of three Ge/ C is indeed nonmagnetic. The two dominant quadrupole dou- Fe/Au samples, measured at room temperature, using a source of blets have splittings of 0.65 and 1.15 mm/s, respectively, and 57Co in Rh. a Sample A, b sample B, and c sample C. the same isomer shift of 0.36 mm/s versus Fe. In addition we observe a third doublet with splitting 2.3 2 mm/s, isomer to avoid complications due to substrate curvature. These shift 1.05 8 mm/s, and 10% relative intensity. The latter samples were placed horizontally on a special Cu tailpiece of component was absent in the time spectra, but for the rest the a closed-cycle refrigerator. Thermal contact was made with results agree nicely. For the two thickest samples we see a crycon grease. Figures 2 d and 2 e display the results on dominant magnetic interaction with Bhf 32.95(5) T. In ad- both samples at room temperature. The fits assume that the dition a quadrupole doublet is observed with large majority of the 57Fe nuclei experience a magnetic in- 0.36(1) mm/s vs Fe and a splitting of 0.86 2 mm/s teraction, with average hyperfine fields of 33.00 15 T for sample A or 0.98 3 mm/s sample B . Its relative intensity sample D and 32.8 3 T for sample E. The linewidths are is 38 1 % for sample A and 23 1 % for sample B. The split- 3.4(5) 0 and 5.5(1.5) 0 , respectively, much larger than in ting is in reasonable agreement with the analysis of the time samples A and B. This points to a distribution of hyperfine spectra but the isomer shift is definitely lower. Moreover, the fields. Such a distribution would imply different widths for relative intensity is up to a factor 4 higher than found from different frequency components see Ref. 25 for a recent the time spectra. discussion of hyperfine field distributions in NRS . However, It turns out that the puzzling differences between the NRS the data are not sensitive to such a refinement. We could data and the CEMS data can be ascribed to the fact that the obtain slightly better fits by allowing for a nonmagnetic com- layers are not stable when stored at room temperature in a ponent with roughly 10% intensity for sample D and 20% dry box. Whereas the NRS measurements were performed intensity for sample E. Although the parameters of this com- 3­4 days after the production of the samples, the CEMS data ponent were badly defined, it was definitely not the nonmag- shown in Fig. 3 were obtained four months later. Subsequent netic component observed in the samples A ­ C. For sample CEMS measurements another three months later showed a D we had a count rate of 0.4 cps, whereas it was only 0.07 clear increase in the nonmagnetic fraction, which now was cps for sample E. This explains the relatively poor quality of 43 1 % for sample A and 30.5 5 % for sample B. Further- the data in Fig. 2 e , an analysis in terms of different hyper- more, the surprising third component observed before by fine fields at the interface was clearly not possible. For CEMS in sample C was absent in this case. sample D, a measurement at 80 K yielded Bhf 33.6(2) T, In view of the results on the first three samples we de- slightly but not significantly lower than the bulk value of cided to grow samples D and E with 1 ML thick probe layers 33.8 T. at a distance of at least 10 ML from the Fe/Ge interface. Moreover, we did not employ S-passivated surfaces because V. DISCUSSION the sulphur atoms may end up at the Fe/Au interface. The substrates were long 45 mm in order to use the grazing- The most puzzling observation in the measurements de- incidence beam as effectively as possible and thick 6 mm scribed here is the presence of a nonmagnetic component in 8594 L. NIESEN et al. PRB 58 the 57Fe probe layers. There is only one sensible explanation: more severe for sample A deposited at 200 °C than for we see interdiffusion of Ge and Fe to such an extent that sample B produced at room temperature . each Fe in the probe layer of sample C is surrounded pre- At this point we want to make a comparison with the data dominantly by Ge atoms. Literature values for in crystal- of Anderson et al.,8 who claim that the Ge/Fe interface is line FeGe compounds are 0.48 mm/s for the cubic B20 sharp after deposition of Fe on a S-passivated Ge 100 sub- phase18 and about 0.30 mm/s for the hexagonal and mono- strate. Although we disagree on this conclusion, their experi- clinic phases.19 Only the cubic compound is not ordered at mental data are not necessarily in conflict with ours. The 295 K. Amorphous Fe/Ge layers are also nonmagnetic at this claim is based on Auger intensities of the various elements temperature, showing an isomer shift of 0.35­0.40 mm/s and during deposition as a function of the thickness of the Fe a quadrupole splitting ranging from 0.5 to 0.9 mm/s.20,21 layer. Due to the limited depth resolution of this method, an These values are sufficiently close to those measured here to interdiffusion over a depth of 5 ML or less cannot be ex- conclude that we have indeed a Ge/Fe interlayer of at least 9 cluded. The second point is that we observe interdiffusion in ML thick. This result contradicts the claim of Anderson a time interval of days to months, whereas Anderson et al. et al.8 that S-passivation prevents intermixing. We will probe the profile in situ during deposition. Obviously, the present a detailed comparison with the data from those au- interdiffusion on a long time scale, observed here ex situ in thors later on. Au-capped Fe layers, is relevant for possible applications of We observe a similar nonmagnetic component in the NRS Fe/Au multilayers grown on Ge substrates. A complication may arise in the interpretation of NRS experiments on samples A and B, although only with 6% measurements on thin probe layers, because the assumption relative intensity. At this point we have a problem, because a that one may add the contribution of the two phases coher- correct analysis requires knowledge about the position of the ently is not necessarily correct. Analysis of the coherence nonmagnetic phase with respect to the Fe layer. If we assume properties of the synchrotron beam has shown that although that this component is due to an Ge-Fe interlayer with a the longitudinal coherence length is very long, the effective homogeneous thickness, this layer would include the begin- transverse coherence length is not.22 For the present geom- ning of the 57Fe probe layer, because a fit using such a model etry, in which the solid angle of the detector slit is rela- shows that the 57Fe atoms in a 40 ML thick Ge-Fe interlayer tively large, we estimate a vertical coherence length of only with natural Fe contribute only 4% to the intensity of the 4 nm and a horizontal coherence length one order of magni- nonmagnetic component, whereas the relative intensity of tude smaller. With a typical incidence angle of 4 mrad, the the nonmagnetic component in the probe layer is 3.5%. Be- coherence length projected onto the surface is 1 m. This cause the deeper layers are less illuminated by the photon means that one should add the contribution of the two phases beam than the probe layer, the fraction of 7.5% nonmagnetic coherently only if a 4 nm wide section of the beam crosses 57Fe atoms yields a relative intensity of only 6% in the spec- both of them. For those sections where the interlayer is so trum. However, there are two reasons to rule out such a thick that it reaches the probe layer this is not necessarily the thick homogeneous interlayer. First, from the reflectivity case. Adding the two contributions noncoherently leads to a data of Fig. 1 it was deduced that the total thickness of different picture, in which the information about the relative FeGe Fe is only 6.5 nm. This is already smaller than isomer shift is lost. However, it turns out that our NRS data expected on the basis of the growth data if this layer is pure cannot be fitted satisfactorily in this noncoherent picture. Fe; if the layer would be largely GeFe the discrepancy would Whereas 2 decreases with roughly a factor of 2 if we add be unacceptably large. Secondly, fits of the time spectra as- the complex nuclear scattering amplitudes of the two com- suming such a thick Ge-Fe interlayer are clearly worse than ponents, 2 only increases if a nonmagnetic GeFe time spec- those assuming a natural Fe layer of equal thickness, posi- trum is simply added to a pure Fe time spectrum. We con- tioning the nonmagnetic component in or close to the probe clude that the regions in which the nonmagnetic layer layer. We think, therefore, that the thickness of the inter- reaches the 57Fe probe atoms have lateral dimensions smaller mixed layer is varying locally, in some places reaching the than 1 m. 57Fe probe layer. Samples D and E, which were grown on nonpassivated For samples A and B, the combined NRS and CEMS re- Ge surfaces, do not show indications for the formation of an sults clearly show a progress in the Ge/Fe interdiffusion as a FeGe interlayer. Since the natural Fe buffer layer is only 10 function of time. We also have an indication that interdiffu- ML for sample D, this suggests that S-passivation favors the sion took place after sample C was produced, because this interdiffusion process. On the other hand, the large linewidth sample showed a clear magnetic Kerr signal right after depo- for sample D points to a lower structural quality of the sition, but no magnetic signal in the 57Fe probe layer during evaporated layer than in the case of samples A ­ C. This is in the NRS measurements three days later. It is highly improb- agreement with other reports concerning the growth of Fe on able that the magnetic Kerr signal originated only from the 2 Ge 100 .8,23 Apparently the sulphur atoms on the surface act ML natural Fe covering the probe layer. The increase of the as surfactants, i.e., they promote two-dimensional growth of intensity of the GeFe component in samples A and B is cor- the Fe overlayer. In the absence of S atoms, the initial growth related with a decrease of the isomer shift. The isomer shift is in the form of three-dimensional Fe islands. The coales- in the NRS data is consistent with the formation of crystal- cence of these islands gives rise to many structural defects. line cubic GeFe, whereas the isomer shift in the CEMS data Unfortunately, an increase in linewidth directly leads to points to the existence of an amorphous GeFe phase. We an increase of the damping of the NRS signal and an accom- have no explanation for this puzzling behavior, nor for the panying decrease of the time integral of the delayed counts. fact that the interdiffusion on a long time scale months is When comparing the total delayed count rate of sample D PRB 58 MAGNETIC BEHAVIOR OF PROBE LAYERS OF 57Fe . . . 8595 with samples A and B, we also have to consider the fact that ties of thin Fe layers. The samples showed a number of sur- the NRS signal for these thin samples is roughly proportional prising features, most of which are related to the formation to the square of the total number of resonant scatterers, be- of a nonmagnetic GeFe interlayer. The current sensitivity cause we sum the amplitudes of the individual contributions limit of the method determined by counting rate is 1 ML. rather than the intensities. These two effects explain the big Because the signal is quadratic if the scatterers respond in decrease in the total delayed count rate of samples D and E, phase, one can get good-quality spectra already from 3 although these samples are much bigger than samples A ­ C. interfaces each 1 ML thick in a thin multilayered sample. The NRS spectrum of 1 monolayer of 57Fe is still easily By taking more layers, one could even study submonolayer observable, but a detailed analysis is difficult. Nevertheless amounts of 57Fe at the interfaces. the results on sample D are surprising when compared to With the present incident photon flux, the NRS technique those on Fe/Ag 100 ,2,3 where it was observed that the room applied to thin layers has a sensitivity that is comparable to temperature hyperfine field in the interior of the layer is ap- CEMS. Already now it is worthwhile to perform this type of preciably lower than in the bulk for Fe layers thinner than NRS experiments in case one wants to measure in large mag- 40 ML. This is attributed to a change in the spin wave netic fields and at various temperatures, a situation in which spectrum. For 15 ML we expect a decrease of 0.3 T with CEMS is difficult to apply. Although this feature is not ex- respect to the bulk value. We do not observe this behavior, plored here, the NRS technique is also promising for deter- but in view of the large linewidth we cannot draw definite mining the direction of magnetic moments in a probe layer, conclusions. The average field at the interface sample E is especially when combined with a polarization analyzer be- also higher than in the case of Fe/Ag 100 . Unfortunately, hind the sample.24 the statistics of the NRS spectrum on this sample prohibit an analysis in terms of several 57Fe probe atom sites at the interface. ACKNOWLEDGMENTS VI. CONCLUSIONS AND OUTLOOK The authors would like to thank C. R. Laurens and F. C. Voogt for help during the measurements. They also thank the The data on Ge/Fe/Au 100 structures presented here ESRF staff, in particular J. Ejton and Z. Hubert, for provid- show that nuclear resonant scattering NRS can provide de- ing the excellent conditions which made these experiments tailed information about the magnetic and structural proper- possible. *Present address: CEIT, Departamento de Materiale, 200009 San ler, R. Ru¨ffer, and H. Winkler, Phys. Rev. B 32, 5068 1985 ; J. Sebastian, Spain. P. Hannon, G. T. Trammell, M. Mueller, E. Gerdau, R. Ru¨ffer, Present address: SPring8/JASRI, Kamigoro-cho, Ako Gun, Hyogo and H. Winkler, ibid. 32, 6363 1985 . 678-12, Japan. 13 L. Nevot and P. Croce, Rev. Phys. Appl. 15, 761 1980 . 1 For a recent review, see, e.g., Ultrathin Magnetic Structures, ed- 14 The computer code written for this calculation is based on A. Q. ited by J. A. C. Bland and B. Heinrich Springer, Berlin, 1994 , R. Baron, thesis, Stanford, 1995. Vols. 1 and 2. 15 H. Kiessig, Ann. Phys. Paris 10, 769 1931 . 2 G. Liu and U. Gradmann, J. Magn. Magn. Mater. 118, 99 1993 . 16 A. Q. R. Baron, J. Arthur, S. L. Ruby, A. I. Chumakov, G. V. 3 P. J. Schurer, Z. Celinski, and B. Heinrich, Phys. Rev. B 48, 2577 Smirnov, and G. S. Brown, Phys. Rev. B 50, 10 354 1994 . 1993 ; 51, 2506 1995 . 17 4 G. Weyer, in Mo¨ssbauer Effect Methodology, edited by I. J. Gru- R. Ru¨ffer and A. Chumakov, Hyperfine Interact. 97/98, 589 verman Plenum, New York, 1976 , Vol. 10, p. 301. 1996 . 18 5 R. Wa¨ppling and L. Ha¨ggstro¨m, Phys. Lett. 28A, 173 1968 . A. I. Chumakov, G. V. Smirnov, A. Q. R. Baron, J. Arthur, D. E. 19 Brown, S. L. Ruby, G. S. Brown, and N. N. Salashchenko, Phys. R. Wa¨ppling, L. Ha¨ggstro¨m, and E. Karlsson, Phys. Scr. 2, 233 Rev. Lett. 71, 2489 1993 . 1970 . 20 6 T. S. Toellner, W. Sturhahn, R. Ro¨hlsberger, E. E. Alp, C. H. O. Massenet and H. Daver, Solid State Commun. 21, 37 1977 . 21 Sowers, and E. E. Fullerton, Phys. Rev. Lett. 74, 3475 1995 . J. Patankar, Y. V. Bhandarkar, S. M. Kanetkar, S. B. Ogale, and 7 W. Folkerts, W. Hoving, and W. Coene, J. Appl. Phys. 71, 362 V. G. Bhide, Nucl. Instrum. Methods Phys. Res. B 7/8, 720 1992 . 1985 . 8 G. W. Anderson, P. Ma, and P. R. Norton, J. Appl. Phys. 79, 22 A. Q. R. Baron, A. I. Chumakov, H. F. Gru¨nsteudel, H. Gru¨nsteu- 5641 1996 . del, L. Niesen, and R. Ru¨ffer, Phys. Rev. Lett. 77, 4808 9 K. Shintaku, Y. Daitoh, and T. Shinjo, Phys. Rev. B 47, 14 584 1996 . 1993 . 23 G. A. Prinz, in Ultrathin Magnetic Structures Ref. 1 , p. 35. 10 A. Q. R. Baron, Nucl. Instrum. Methods Phys. Res. A 353, 665 24 D. P. Siddons, U. Bergmann, and J. B. Hastings, Phys. Rev. Lett. 1994 . 70, 359 1993 . 11 Y. Kagan, A. M. Afanas'ev, and V. G. Kohn, J. Phys. C 12, 615 25 Yu. V. Shvyd'ko, U. van Bu¨rck, W. Potzel, P. Schindelmann, E. 1979 . Gerdau, O. Leupold, J. Metge, H. D. Ru¨ter, and G. V. Smirnov, 12 J. P. Hannon, N. V. Hung, G. T. Trammell, E. Gerdau, M. Muel- Phys. Rev. B 57, 3552 1998 .