VOLUME 84, NUMBER 22 P H Y S I C A L R E V I E W L E T T E R S 29 MAY 2000 Real-Space Observation of Dipolar Antiferromagnetism in Magnetic Nanowires by Spin-Polarized Scanning Tunneling Spectroscopy O. Pietzsch, A. Kubetzka, M. Bode,* and R. Wiesendanger Institute of Applied Physics and Microstructure Research Center, University of Hamburg, Jungiusstrasse 11, D-20355 Hamburg, Germany (Received 17 December 1999) We have performed spin-polarized scanning tunneling spectroscopy of dipolar antiferromagnetically coupled Fe nanowires with a height of two atomic layers and an average separation of 8 nm grown on stepped W(110). Domain walls within the nanowires exhibit a significantly reduced width when pinned at structural constrictions. The lateral spin reorientation in the direction perpendicular to the wires has been studied with subnanometer spatial resolution. It is found that the spin canting in the Fe nanowires monotonously increases towards the step edges. PACS numbers: 75.25.+z, 75.30.Pd, 75.70.Kw Driven by the demand for ever higher density in data both along and perpendicular to the nanowires, on a storage media in the past a strong effort has been under- subnanometer scale. In particular, a detailed investigation taken in the preparation and characterization of magnetic of the domain wall width w revealed two different types of nanostructures of the desired size and shape [1]. Recently, walls. While relatively broad domain walls were found in the interest in magnetic nanostructures has focused on per- homogeneous double-layer (DL) wires (w 6 6 1 nm), pendicularly magnetized Fe nanowires grown on slightly much narrower ones appear at structural constrictions miscut single crystals [2­4]. While the structural and elec- (w 2 6 1 nm). The broad domain walls of DL wires tronic properties have frequently been observed by means are consistent with a value of ADL that is very close to the of scanning tunneling microscopy (STM) and spectroscopy Fe bulk value of A 1 3 10211 J m. The reduced width (STS), i.e., with an experimental technique that allows of domain walls that are pinned at structural constrictions real-space imaging with a resolution limit down to the is found to be in good agreement with a recent theoretical atomic level, the magnetic properties have mostly been prediction [8]. investigated by spatially averaging techniques like, e.g., The experiments have been performed in an ultrahigh magneto-optical Kerr effect. In particular, for perpendicu- vacuum (UHV) system with separate chambers for sub- larly magnetized Fe nanowires grown on stepped W(110) strate preparation, sample transfer, metal vapor deposition which exhibit a width of only 4 nm and a periodicity of (MVD), surface analysis, and cryogenic STM equipped 8 nm, a complex magnetic behavior has been found which with a 2.5 T magnet [9]. The base pressure in each was explained by a magnetic anisotropy which changes chamber is in the low 10211 torr range. The W(110) discontinuously on a nanometer scale [3­5]. However, the single crystal is miscut by 1.6± with respect to the (110) minimum distance required to change the magnetization p plane. This substrate is prepared by numerous cycles of direction is determined by the exchange length L A k, long-term heating at 1500 K in an oxygen atmosphere of where A is the so-called exchange stiffness and k the 1027 1026 torr and subsequent flashing up to 2500 K anisotropy constant. It was found that the spin rotation [10]. We used etched W tips which were flashed in vacuo within Fe nanowires on such a narrow lateral scale cannot to remove oxide layers. In the MVD chamber the tips be explained on the basis of bulk properties [4,5]. Instead, were magnetically coated with Gd while held at 300 K, it was assumed that the exchange stiffness A of the first subsequently annealed at T 550 K for 4 min, and and second Fe layer on W(110), i.e., AML and ADL, is 1 then transferred into the cryogenic STM. During the order of magnitude smaller than in bulk Fe-an assump- measurements, the tip and sample were at a temperature tion which has so far not been verified by experimental T 16 K. results. The growth of Fe W(110) has been intensively investi- In this Letter we report on the direct observation of gated in the past [11­14]. Our Fe films were grown at T the magnetic domain structure of Fe nanowires grown 300 K at a rate of 0.6 monolayers per minute (ML min) on stepped W(110) by means of spin-polarized scanning and subsequently annealed at T 520 K for 4 min lead- tunneling spectroscopy (SP-STS) [6,7]. Our results ing to step flow growth. The Fe film grows pseudomorphi- unambiguously show that-in agreement with the expla- cally, i.e., expansively strained by about 10%, as long as nation mentioned above [3­5]-the domain structure is the terrace width remains below the critical Fe double-layer governed by perpendicularly magnetized Fe nanowires width for misfit dislocation formation of 9 nm [3,13]. The with the magnetization pointing alternatingly up and miscut of the W(110) substrate used in this study results down. Additionally, the high spatial resolution of SP-STS in an average terrace width of 8 nm. Indeed, the con- allows the evaluation of details of the spin reorientation stant current topograph of 1.5 6 0.1 ML Fe W(110) as 5212 0031-9007 00 84(22) 5212(4)$15.00 © 2000 The American Physical Society VOLUME 84, NUMBER 22 P H Y S I C A L R E V I E W L E T T E R S 29 MAY 2000 shown in the inset of Fig. 1(a) is dominated by atomically alternating ML and DL Fe stripes. Since in our experi- flat terraces and monatomic steps of a nominal height of ment the substrate step direction deviates from the [001] 2.24 Å. The sample surface is free of any misfit disloca- direction, the width of the nanowires is not homogeneous tions. Schematically, the structure of the sample is repre- but slightly fluctuates along the step edges. sented below the line section of Fig. 1(a). It consists of By using black and white arrows we have also repre- sented the magnetic structure of mono- and double-layer stripes as recently proposed by Elmers, Hauschild, and Gradmann [3­5]. Combining longitudinal and polar Kerr- effect measurements, an onset of perpendicular magneti- zation for Fe coverages Q $ 1.1 ML was found [4]. The coverage range between 1.4 and 1.8 ML Fe W(110) is characterized by magnetic saturation at relatively low ex- ternal perpendicular fields combined with the absence of a hysteresis, i.e., zero remanence [3]. It has been argued [3] that this experimental result is caused by the perpen- dicularly magnetized Fe DL stripes which prefer to occupy a demagnetized ground state by antiferromagnetic (AFM) dipolar coupling, i.e., by periodically changing the mag- netization direction between adjacent DL stripes. Since, however, all available magnetic information was based on spatially averaging experiments, many interesting ques- tions on nanomagnetic details remained unanswered. Before we turn our attention to SP-STS measurements we have to understand the electronic properties of the sample. The upper panel of Fig. 1(b) shows tunneling dI dU spectra measured with a bare (nonmagnetic) W tip taken above Fe ML and DL stripes. As already men- tioned in an earlier STS study performed at 300 K, both the Fe ML as well as the DL exhibit characteristic tunnel- ing spectra [14]. In the present low-temperature experi- ment, we found peaks at U 10.40 V for the ML and U 10.68 V for the DL [15]. We would like to empha- size that-within our measurement accuracy-we found the same spectra above any ML or DL stripe, respectively. In a second set of experiments we investigated the nanomagnetic structure of the Fe DL stripes by means of SP-STS. According to Refs. [3,4] the Fe DL stripes are perpendicularly magnetized. One crucial requirement for obtaining a magnetic contrast in SP-STS measurements is an appropriate magnetization direction of the tip. For the purpose of this study we used W tips which were coated by 8 6 1 ML Gd. Bulk Gd is ferromagnetic below its bulk Curie temperature TCB 292.5 K. It is well known that Gd films with a thickness of 8 ML are ferromagnetic for T , 0.7TCB [16] and exhibit a perpendicular easy axis at T , 0.6TCB [17]. Since both phase transition FIG. 1. (a) Line section and topography (inset) of 1.5 ML temperatures are far above our measurement temperature Fe W(110) showing terraces with an average width of 8 nm. of 16 6 1 K we can safely conclude that the tip is The structure of the sample is schematically represented. ferromagnetic with the magnetization vector pointing (b) The tunneling spectra of ML and DL stripes exhibit peaks at U 10.40 and U 10.68 V, respectively. While no along the tip axis, i.e., perpendicular to the surface plane. differences in the spectra were found between different Fe As already shown in previous publications [6,7], the con- DL stripes with W tips two quantitatively different spectra ductivity between the two magnetic electrodes depends were measured when using Gd-coated tips. (c) A map of the on the electron density of states within a particular energy dI dU signal (U 0.68 V) reveals that the spectra alternate range given by the applied bias voltage and on the sign between adjacent Fe DL stripes being caused by an AFM dipolar coupling. Two ferromagnetically coupled nanowires and the magnitude of the electron spin polarization. and domain walls within single nanowires are marked by black The lower part of Fig. 1(b) shows dI dU spectra which and white arrows, respectively. were measured using a Gd-coated probe tip. With the Gd 5213 VOLUME 84, NUMBER 22 P H Y S I C A L R E V I E W L E T T E R S 29 MAY 2000 tip positioned above the Fe ML we found spectra which are qualitatively identical with the spectra measured with bare W tips. In contrast, we found two different types of spectra above the DL stripes which will hereafter be referred to as "" and "#. Both spectra exhibit a peak at the same energeti- cal position already mentioned above, i.e., U 10.68 V, but differ in intensity. While the differential conductivity at the peak position amounts to only dI dU 1.3 nA V for the spectra of type "" it is enhanced by about 40% to dI dU 1.8 nA V for type "#. The relative intensities between both types of spectra invert for U , 0.5 V. These differences in the tunneling spectra of the DL Fe stripes are caused by spin-polarized tunneling between the magnetic tip and Fe DL stripes being magnetized either parallel or antiparallel to the tip. In Fig. 1(c) we have plotted a map of the differential conductivity dI dU at the peak position (U 0.68 V). Different intensities of the dI dU signal show up as different grey levels. Since the dI dU sig- nal at U 0.68 V is much lower for the monolayer than for double-layer stripes the former appears black. Further- more, the data reveal that most double-layer stripes exhibit only one type of spectrum, either "" or "#, and that the type FIG. 2. Line sections showing the change of the dI dU signal when crossing a domain wall being located in a smooth (upper alternates between adjacent stripes. This observation is panel) or constricted (lower panel) Fe DL stripe. Maps of the consistent with the proposed AFM out-of-plane coupling dI dU signal are shown in the inset. The positions at which the of adjacent stripes [3­5]. In Fig. 1(c) we have marked line sections were drawn are marked by solid black lines. some exceptions from this rule by arrows demonstrating the impact of the high spatial resolution of SP-STS. Ap- proximately in the middle of the image one can recognize with the bulk value A 1 3 10211 J m. At structural two adjacent stripes which exhibit the same dI dU sig- constrictions which often serve as pinning centers for do- nal (black arrows). Obviously, these stripes are so close main walls we found, however, narrower domain walls as together (d , 1.5 nm) that the exchange coupling over- can be seen in the line section shown in the lower panel of comes the energy gain due to an AFM coupling. Further- Fig. 2. Typically, the width of domain walls being pinned more, at the very right edge of the image we can find two at structural constrictions amounts to w 2 6 1 nm. stripes which change the type of spectrum from the bot- This behavior has recently been proposed theoretically by tom to the top part of the image, i.e., both stripes exhibit a Bruno [8]. We have applied Bruno's approach by modeling domain wall (white arrows). the width of the DL stripe S x by the quadratic approxi- We have shown so far that the DL peak position (U mation S x S0 1 1 x2d2 (model II in Ref. [8]). Here, 0.68 V) is particularly suited for the imaging of magnetic S0 is the minimum width of the constriction positioned at domains since the contrast between the spectra "" and "# x 0, x is the distance from minimum, and d is a fit pa- is maximum. Another bias voltage which allowed imag- rameter. Indeed, in our case the constriction could well be ing of the magnetic domain structure with high contrast fitted by using S0 0.8 nm and d 1 nm which results was U 20.3 V. In the following we have not taken in a reduced domain width w 8dp 2.5 nm being in full tunneling spectra at every pixel of an image which fair agreement with the experimental observation. requires a measurement time of about 10 h per image. In- Another point of interest is how the spin orientation stead, we have only measured the dI dU signal at the changes when crossing the DL stripes perpendicularly. voltage values given above. This reduces the measurement Figures 3(a) and 3(b) show details of the topography time to about 30 min for an image with (500 3 500) pixel. and dI dU signal (U 20.3 V) of 1.5 6 0.1 ML Figure 2 shows two different types of domain walls which Fe W(110). The AFM coupling between adjacent Fe have been observed within the DL stripes. Relatively DL stripes appears as different grey levels in Fig. 3(b). broad domain walls with a width w0 6 6 1 nm were Since we have never found any magnetic contrast in the found in homogeneous DL stripes [18]. This finding is monolayer stripes with Gd-coated tips, we have adjusted in strong disagreement with a recent publication, in which the z scale of the line section in Fig. 3(c) to allow a the DL exchange length was estimated to LDL 0.5 nm high sensitivity on the dI dU signal of the DL stripes based on Kerr-effect measurements [4]. Instead, using the p only. Obviously, the dI dU signal of the dark DL stripes definition w0 2L 2 A k [8] our results suggest that monotonously decreases when moving away from the LDL 3 nm. With kDL 1 3 1026 J m3 [4] this leads ML-DL transition while the dI dU signal of the bright to ADL 9 3 10212 J m which almost perfectly agrees DL stripes monotonously increases. Since the electronic 5214 VOLUME 84, NUMBER 22 P H Y S I C A L R E V I E W L E T T E R S 29 MAY 2000 are canted by about 30± to the surface plane. (iii) The canting increases monotonously towards 90± with respect to the surface plane when approaching the step edges. We can conclude that in spite of the fact that ADL Avol the AFM coupling between adjacent DL stripes being separated by ML stripes of only 4 nm width becomes possible because the spin rotation occurs close to the ML-DL transition. In summary, we have imaged the domain structure of an- tiferromagnetically coupled Fe nanowires with a thickness of only 2 ML. Two different types of domain walls were found within the wires: though smooth nanowires exhibit walls with a width of about 6 nm, structural constrictions lead to a much smaller wall width of about 2 nm. The lateral spin reorientation in the direction perpendicular to the wires has been investigated in detail and compared to a theoretical model. Our SP-STS results reveal that the spin canting within the Fe nanowires is nonsymmetric and mo- notonously increases towards the step edges. Financial support from the Deutsche Forschungsge- meinschaft (Grant No. Wi 1277/3), from the GIF (Grant No. I-550-184.14/97), and from the BMBF is gratefully acknowledged. *Email address: bode@physnet.uni-hamburg.de [1] F. J. Himpsel et al., Adv. Phys. 47, 511 (1998). [2] J. Shen et al., Phys. Rev. B 56, 2340 (1997). FIG. 3. (a) Topography and (b) dI dU signal at U 20.3 V [3] J. Hauschild, U. Gradmann, and H. J. Elmers, Appl. Phys. as measured with a Gd-coated probe tip. Because of the fact Lett. 72, 3211 (1998). that at this particular sample bias the dI dU signal of the "" DL [4] H. J. Elmers, J. Hauschild, and U. Gradmann, Phys. Rev. stripes is almost equal to the ML, both appear with a similar grey level [cf. Fig. 1(b)]. (c) Line section drawn across the B 59, 3688 (1999). nanowires. Schemes of the lateral spin reorientation are shown [5] H. J. Elmers, J. Magn. Magn. Mater. 185, 274 (1998). in (d) (taken from Ref. [5]) and (e) (new model proposed). [6] M. Bode, M. Getzlaff, and R. Wiesendanger, Phys. Rev. Lett. 81, 4256 (1998). properties of all double layers are identical (cf. Fig. 1), [7] M. Bode, M. Getzlaff, and R. Wiesendanger, J. Vac. Sci. this behavior can be explained only by a magnetic effect. Technol. A 17, 2228 (1999). In this context it is worthwhile to compare the result of [8] P. Bruno, Phys. Rev. Lett. 83, 2425 (1999). Fig. 3(c) with a detailed model of ultrathin films with [9] O. Pietzsch et al., Rev. Sci. Instrum. 71, 424 (2000). [10] M. Bode, R. Pascal, and R. Wiesendanger, Surf. Sci. 344, laterally modulated anisotropies which has been proposed 185 (1995). by Elmers and co-workers [4,5]. As schematically rep- [11] H. J. Elmers et al., Phys. Rev. Lett. 73, 898 (1994). resented in Fig. 3(d), it has been suggested that for a [12] H. J. Elmers et al., Phys. Rev. Lett. 75, 2031 (1995). coverage of 1.5 ML Fe W(110) and a DL stripe width of [13] C. Jensen, K. Reshöft, and U. Köhler, Appl. Phys. A 62, 4 nm a continuous spin rotation occurs from in plane in 217 (1996). the middle of the ML stripe to out of plane in the middle [14] M. Bode, R. Pascal, and R. Wiesendanger, J. Vac. Sci. of the DL stripe. The line section of Fig. 3(c) reveals, Technol. A 15, 1285 (1997). however, that the maximum contrast is not obtained in the [15] The peaks which appear at U , 0.3 V are partially tip middle of the DL stripe but at the descending step edges. induced and not relevant in the framework of this work. Based on the available data we propose the following [16] M. Farle et al., Phys. Rev. B 47, 11 571 (1993). picture of the spin structure which is also schematically [17] G. André et al., Surf. Sci. 326, 275 (1995). [18] The given error bar is estimated on the basis of the noise represented in Fig. 3(e): (i) The Fe ML stripes exhibit no within the magnetic signal. Meanwhile we have achieved out-of plane magnetization even very close (,3 Å) to the atomic spin resolution using ferromagnetically coated transition to the DL stripes. (ii) Even very close to the probe tips; therefore we can exclude the fact that spatial transition to the ML (,3 Å) the spins in the DL stripes averaging due to a blunt tip plays an important role. 5215