0MP Applied Physics\ ■ |
Low magnetisation alloys for in-plane spin transfer torque devices
K. Oguz, M. Ozdemir, O. Dur, and J. M. D. Coey
Citation: J. Appl. Phys. 111, 113904 (2012); doi: 10.1063/1.4723824 View online: http://dx.doi.org/10.1063/1.4723824 View Table of Contents: http://jap.aip.org/resource/1/JAPIAU/v111/i11 Published by the American Institute of Physics.
Related Articles
Room temperature single GaN nanowire spin valves with FeCo/MgO tunnel contacts Appl. Phys. Lett. 100, 182407 (2012)
Observation of anomalous Hall effect in Cu-Py-crossed structure with in-plane magnetization J. Appl. Phys. 111, 07D307 (2012)
Micro-fabrication process for small transport devices of layered manganite J. Appl. Phys. 111, 07E129 (2012)
Electrical properties of magnetic nanocontact devices computed using finite-element simulations Appl. Phys. Lett. 100, 083507 (2012)
High power and low critical current spin torque oscillation from a magnetic tunnel junction with a built-in hard axis polarizer
Appl. Phys. Lett. 100, 032405 (2012)
Additional information on J. Appl. Phys.
Journal Homepage: http://jap.aip.org/ Journal Information: http://jap.aip.org/about/about_the_journal Top downloads: http://jap.aip.org/features/most_downloaded Information for Authors: http://jap.aip.org/authors
ADVERTISEMENT
ai pAd varices
Special Topic Section:
PHYSICS OF CANCER
Why cancer? Why physics?
View Articles Now
Low magnetisation alloys for in-plane spin transfer torque devices
K. Oguz,1 M. Ozdemir,2 O. Dur,3 and J. M. D. Coey1
1 School of Physics andCRANN, Trinity College, Dublin 2, Ireland
2Department of Physics, Faculty of Science, Marmara University, Istanbul, Turkey
3Calik Holding, Energy, Information and Communication Group, Zincirlikuyu, Istanbul, Turkey
(Received 11 January 2012; accepted 1 May 2012; published online 1 June 2012)
Magnetic properties of thin films of three in-plane alloy systems, namely, Co40Ni40_xCrxB20, Co40Fe40_xNixB20, and Co40Fe40_xCrxB20 are investigated with a view to reducing the critical current density (jc) for spin transfer torque switching in MgO barrier magnetic tunnel junction nanopillars. In all three systems, the saturation magnetisation, Ms, decreases with increasing Cr or Ni substitution. The Co40Fe40_xCrxB20 alloy system is found to be the best in terms of reducing Ms, while maintaining a high tunnel magnetoresistance ratio. Ferromagnetic resonance experiments revealed that the damping coefficient of CoFeCrB alloy remains unchanged with increasing Cr content, but jc is reduced by a factor of four to 4.9 x 105 A/cm2 by using a Co40Fe32.7Cr7.3B20 free layer while maintaining a magnetoresistance of around 60 %. © 2012 American Institute of Physics. [http://dx.doi.org/10.1063/L4723824]
I. INTRODUCTION
Following the prediction of spin transfer torque (STT) associated with spin polarized current in magnetic multilayers by Slonczewski and Berger in 1996,1,2 there has been great interest in the spin torque related effects in both giant magnetoresistive (GMR) spin valves and magnetic tunnel junctions (MTJ) due to their potential application as memory elements and microwave oscillators.3-7 STT can cause precession or complete magnetic reversal of the free layer magnetisation, depending on the magnitude of the spin-polarized current.1,8,9 The critical current density (jc) for magnetisation switching for in-plane devices depends mainly on the square of the saturation magnetisation (Ms) of the free layer, due to the demagnetizing field in the layer.9 Zero temperature macro-spin calculation9 reveals a stability threshold current density of
jAp_p = 2e (h + Hdp + Hk + M,/2). (1)
Here a is the Landau-Lifshitz-Gilbert (LLG) damping coefficient and t is the thickness of the free layer. The spin transfer efficiency, g, is a function of the current polarity, polarization, and the relative angle between the free and pinned layers. H is the external field, Hdip is the dipolar coupling between the free and the pinned layers, and Hk is the uniaxial anisotropy field. For in-plane devices, the Ms/2 term is much larger than the other terms in the expression. Hence, the critical current density is dominated by the saturation magnetisation of the free layer. Typically, jc has been found to be around 3 x 106 A/cm2 for magnesium oxide (MgO) based MTJs.7 It is desirable to reduce it to ^5 x 105A/cm2 for memory applications.10
There are three approaches to achieve this: the magnetisation or the Gilbert damping of the ferromagnet should be reduced or the spin transfer efficiency should be improved. Here, we investigate the effect of reducing the magnetisa-
tion. Amorphous ferromagnetic alloys are good candidates because Ms can easily be controlled by adjusting the transition metal composition or the amount of metalloid in the alloy.11 Amorphous Co-Fe-B re-crystallized on a (001) oriented MgO barrier is proven to be the best way to obtain a high performance MTJ.7,12-16 A Co40Fe40B20 composition is often used, which has a magnetization of ^1200kA/m. It is possible to reduce Ms by substituting iron/cobalt with chromium or nickel. In this study, we prepared and characterized magnetisation and Gilbert damping of three series of thin films with compositions Co40Ni40_xCrxB20, Co40Fe40_x NixB20, and Co40Fe40_xCrxB20. Then, we focused on the latter system and investigated the magnetoresistance and switching current density of MgO based MTJ nanopillars with a Co40Fe40_xCrxB20 free layer.
II. EXPERIMENTAL METHODS
Thin film stacks and devices in this study were prepared in our Shamrock sputtering system with a base pressure below 2.0 x 10_8 Torr. Thin films of the amorphous alloys were deposited by co-sputtering from a combination of two of the high purity (99.95%) Co40Fe40B20, Co40Cr40B20, and Co40Ni40B20 targets at room temperature (RT) on Si (001)/ SiO2 (500 nm) substrates. Film thicknesses were nominally 50 nm. Co-Ni-B and Co-Cr-B targets were used for Co40 Ni40_xCrxB20 films, similarly Co-Fe-B and Co-Ni-B were used for Co40Fe40_xNixB20 films, and Co-Fe-B and Co-Cr-B targets were used for Co40Fe40_xCrxB20 films. The composition of the films was adjusted by carefully controlling the deposition rates of each target. High purity (99.9999%) Argon was used as process gas and the deposition pressure was kept at 6 x 10_3Torr. Thin films were subsequently annealed at 325 or 350 °C in high vacuum under an in-plane magnetic field of 0.8 T. High quality MgO-based MTJs using a Co-Fe-Cr-B free layer with a Si (001)/Si02 (500)/Ta (5)/ Ru (30)/Ta (5)/Ni81Fex9 (5)/Ir22Mn78 (W^Fe^ (2.5)/Ru (0.9)/Co40Fe40B20 (3)/Mg0 d^/Co^e^-xCixBa) (2.5)/Ta
0021-8979/2012/111(11)/113904/6/$30.00
111, 113904-1
) 2012 American Institute of Physics
(5)/Ru (5) stacking structure were also fabricated. The numbers in parenthesis represent the layer thickness in nm. MTJs were patterned into 90 x 170 nm2 rectangular nanopillars using a combination of electron beam lithography and Ar+ ion milling techniques.17
The films were examined by x-ray diffraction using CuKa radiation both as-deposited and after annealing. The only features seen in the diffraction pattern are due to the substrate. The films are amorphous in both states. Film composition was determined using the energy dispersive x-ray (EDX) technique with 15 keV electrons. Magnetic properties of the as-grown and annealed samples were measured using a quantum design superconducting quantum interference device (SQUID) magnetometer (MPMS XL-5) at both room temperature and 4.2 K. Ferromagnetic resonance (FMR) measurements were carried out by using a Bruker EMX model x-band electron spin resonance spectrometer at a microwave frequency of 9.5 GHz, at room temperature. The damping coefficients of the films were determined from the ferromagnetic resonance linewidth. Magneto-transport properties of MTJs were determined using a conventional 4-point method.
III. RESULTS A. Magnetisation
Figure 1(a) illustrates the RT magnetisation curves of Co4oNi4o_xCrxB2o films grown on SiO2 with different Cr concentrations for the as-grown samples. For x = 0, saturation magnetisation is around 4oo kA/m but it falls dramatically to 3kA/m for x = 13.2, and the films become paramagnetic at higher concentrations. A single step in the magnetization curve around zero field suggests that a single magnetic phase present in all samples. The reduction of the room temperature magnetisation reflects both the effect of Cr on the low temperature saturation magnetisation and its effect on the Curie temperature.
The magnetisation curves of Co40Fe40_xNixB20 films grown on SiO2 are shown in Fig. 1(b) for the as-grown samples measured at RT for different Ni concentrations. It seems that Ni substitution is not very effective at reducing the magnetisation of the film. For x = 0, the magnetisation is 984 kA/m, and it falls only to 882 kA/m when x = 21.6. This is a puzzling result because Ni has a much lower moment (0.6 iB) compared to Fe (2.2 iB).
Co40Fe40_xCrxB20 is the best alloy system from the point of view of controlled and effective reduction in magnetisation. Magnetisation curves of different Cr concentrations are shown in Fig. 1(c). The magnetisation of the films decreases as the Cr concentration increases, with a single step close to zero field, indicating that there is only one magnetic phase present in the samples. The saturation magnetisation is 984 kA/m for x = 0, and it decreases to 859 kA/m for x = 4.8 and 235 kA/m for x = 18. A similar behaviour was reported by Kubota et al.18 in (Co75Fe25)80_xCrxB20 films. In their work, magnetisation falls almost to zero for x = 20 at RT because they reduced both Co and Fe in the alloy. Therefore, it is expected to have a faster decay of magnetisation.
Besides the room temperature measurements, we have also measured the magnetisation at 4.2 K, and we have
FIG. 1. Magnetisation curves of (a) Co40Ni40_xCrxB20, (b) Co40Fe40_x NixB20, (c) Co40Fe40_xCrxB20 films on SiO2. Samples were in the as-grown state and measurements were performed at room temperature.
measured it at RT and 4.2 K after field annealing at 350 °C in vacuum. The effect of the annealing, which does not crystallize the films, is just to increase the magnetisation slightly, by about 10 %. For example, the magnetisation of Co40Fe40B20 increases from 984 to 1110 kA/m at room temperature and from 1035 to 1150 kA/m at 4.2 K.
Figure 2 plots the 4.2 K moments of the as-grown films for the three series. The moment in iB/formula is derived from the measured film moment, i = MV in Am2, where V is the film volume using
l = MMW /1000pN01B,
where Mw is the molecular weight of the formula, N0 is Avogadro's number, and density q (kg/m3) is estimated by assuming random dense packing with a packing fraction of
1.2 -1.0 -
S 0.6 -
Concentration (%)
FIG. 2. Magnetic moment per formula measured at 4.2 K versus concentration of Cr/Ni in three alloy systems. Red lines are fits from the magnetic valence model.
0.64 for the transition metal atom. The boron is supposed to occupy voids in the Bernal structure.
B. Ferromagnetic resonance
The FMR spectra were recorded as a function of dc magnetic field applied for different directions with respect to the film normal. In the FMR experiment, the ac (microwave) magnetic field component remains always perpendicular to the dc field in order to record the ac magnetic susceptibility, the imaginary component of which is proportional to the microwave absorption (FMR signal). The microwave frequency was kept fixed at about 9.5 GHz, while the dc field was scanned through the resonance. The spectra were recorded for all three series. The data for Co40Fe40_xCrxB20 in the parallel and perpendicular directions are shown in Fig. 3. The spectra for Co40Fe40_xNixB20 are similar but the line-widths for Co40Ni40_xCrxB20 are about four times greater.
We have developed a computer program to deduce the magnetic parameters of the material from fits of the experimental FMR spectra using the model discussed previously.19
The damping parameter has a major influence on the resonance line shape. Magnetic damping is usually characterized by the phenomenological damping parameter a, which can have both intrinsic and extrinsic contributions. The latter includes surface effects, which may be particularly important in patterned magnetic nanostructures.20 We get satisfactory fits between the experimental and the simulated line shapes of the field derivative FMR spectra for a general direction of external dc field with Gilbert type damping for magnetically homogeneous films. The experimental FMR spectra were fitted using Zeeman, demagnetizing and other anisotropy energies along the film plane and in the perpendicular direction in the same calculation.
The measured spectra (black) and calculated spectra (red) are plotted together in Fig. 3. As seen from this figure, the agreement between experimental and fitted values is satisfactory. In some cases, two exchange coupled components were used in the fits. The FMR spectra strictly depend on the composition of the films. The resonance field, the line-width, and even number of the resonance peaks change with the chemical composition. The differences of the resonance fields for parallel and perpendicular geometries can be taken as a rough measure of the dc magnetisation. This means that chromium tends to decrease the magnetisation as we already know from the magnetisation measurements (Fig. 1). The single FMR peak that is observed for samples with low or zero chromium concentration is doubled for the possibly in-homogeneous samples with higher chromium content (x = 18 in Fig. 3). The double peaks can be taken as an indication of compositional inhomogenity across the thickness of the sample. The linewidth and amplitudes of the spectra for parallel and perpendicular geometries are found to be slightly different. This difference implies the presence of extrinsic damping effects in addition to the intrinsic Gilbert type damping. Damping is much larger in Cr-Ni series (a - 0.03-0.07) than in the Co-Fe series (a - 0.01-0.02). Structural inhomogenity, especially at the surface, can
o CoFeCrB
■ CoFeNiB
CoNiCrB
FIG. 3. Ferromagnetic resonance spectra of annealed Co40Fe40_xCrxB20 films. DC magnetic field is applied (a) parallel and (b) perpendicular to the film plane.
TABLE I. FMR data obtained at room temperature on annealed films with different compositions. The data is given for the measurements in parallel direction.
Sample Ms (SQUID) (kA/m) Ms (FMR) (kA/m) l0H0 (T) a
Co40Fe40_ xCrxB20 0% 1110 1029 0.0766 0.020
3.7% 910 825 0.0862 0.025
8.8% 744 760 0.1044 0.028
266 0.1775 0.023
18% 310
420 0.2246 0.023
Co40Fe40- xNixB20 0% 1110 1029 0.0766 0.020
12.8% 978 900 0.0846 0.012
21.6% 924 820 0.0976 0.012
Co40Ni40- xCrxB20 0% 416 468 0.1557 0.070
4.8% 329 390 0.1783 0.030
9% 165 365 0.1818 0.035
13.2% 63 361 0.1837 0.050
contribute to the line-broadening. The deduced magnetic parameters are summarized in Table I. One can see that the linewidth, the magnetization, and the effective damping parameter tend to decrease monotonically as the chromium concentration increases.
Strong angular dependence of magnetic damping (line-width of FMR spectra) is an indication of structural anisot-ropy; otherwise the intrinsic Gilbert type damping would be isotropic. A narrower line-width can be taken as an indicator of sample homogeneity and quality.
C. Magnetoresistance and spin torque switching
The effect of Cr content on the tunnel magnetoresistance (TMR) ratio in nano-MTJs with a 2.5 nm Co40Fe40_xCrxB20 free layer is shown in Fig. 4. A monotonic decrease in TMR ratio is observed for both annealing temperatures. The TMR ratio starts at ~150 % for x = 0 and reduces to ^100 % for x = 7.3%, but there was considerable variation of TMR within a batch of nano-MTJs with the same concentration x of the free layer. This variation reflects the low yield lift-off
process that we employ in our nano-MTJ patterning flow. The non-homogenous Cr concentration in the free layer can also contribute to the TMR variation of the devices for high concentrations. The resistance x area product of the MTJs was ~ 40 X im2.
Typical magnetoresistance curves of a nano-MTJ with Co40Fe40_xCrxB20 free layer for x = 7.3 is shown in Fig. 5(a). The TMR ratio of this particular device was ~ 60% after annealing at 350 °C for 30min. The free layer loop is shifted by 8.2 mT due to dipolar coupling, and a compensatory field is applied for the measurement of the pulsed R-j loop of the same device illustrated in Fig. 5(b). Two distinct changes in resistance are observed at +4.6 x 105 A/cm2 (jACP_P) and _5.1 x 105A/cm2 (jFC_AP). jc(av) is 4.9 x 105A/cm2. In these measurements, positive current corresponds to electrons tunnelling from pinned layer to free layer. Positive current stabilizes the parallel state in this configuration.
As the Cr concentration is increased, the critical current density is progressively reduced as shown in Fig. 6 for devices annealed at 325 °C for 30 min. For x = 0, jc is around 2 x 106 A/cm2 and it falls linearly with increasing Cr content. The lowest value of jc, 4.9 x 105 A/cm2 is found when x = 7.3.
IV. DISCUSSION
The magnetic moment per formula unit for amorphous 3d alloys may be estimated from the magnetic valence model.21 The basic assumption of the model is strong
FIG. 4. The effect of Cr content on the TMR ratio of MTJs with a Co40Fe40_xCrxB20 free layer at these annealing temperatures.
FIG. 5. (a) Resistance and TMR ratio curves and (b) R-j loop of a nano-MTJ with a Co40Fe32.7Cr7.3B20 free layer. Pulse width was 1 ms during the measurement. Nanopillar size is 90 X 170 nm2.
FIG. 6. The effect of Cr content on the jc for the nano-MTJs with Co40Fe40_xCrxB20 free layer. Red line is a linear fit to the data.
The benefit of reducing magnetisation of the free layer in spin torque switching is evident (Fig. 6). A similar effect has been observed by Kubota et al. on (Co75Fe25)80_x CrxB20.18 They replaced (CoFe) in CoFeB with Cr and reported a decrease of jc0 by a factor of 2 for 10% Cr. We observed a reduction in jc by a factor of four for 7.3% Cr, mainly due to the decrease in the Ms of Co40Fe40_xCrxB20 free layer. It is important to note that the coercivity of the device is also decreased with increasing Cr content. Therefore, it is not possible to use material with higher Cr concentration (>10%) for spin transfer torque devices due to the lack of a bi-stable state.
The Gilbert damping is essentially unchanged by Cr substitution and the TMR holds up well (Fig. 4). The reduction in TMR ratio may be due to a reduction in tunnelling spin polarization in MTJs with Cr substitution.
ferromagnetism, in other words the majority spin d-band is full. The average magnetic moment per atom is then
(l) = ((Zm)+2N])lB,
where Zm is the magnetic valence of the atom (2 for Fe, 1 for Co, 0 for Ni, _6 for Cr, and _3 for B) and Ns the number of electrons in the s conduction band. The magnetic valence is defined as Zm = 2N\ -Z, where Nd is taken as either 5 for 0, and Z is the valence of the atom (8 for Fe, 9 for Co, 10 for Ni, 6 for Cr, and 3 for B). Taking N" = 0.3, we calculate (i) for the end members as follows:
Co40Fe40B20 _ 1.20 iB, Co40Ni40B20 _ 0.40 iB.
The values of the magnetisation of the as-deposited films deduced from the SQUID measurements and from FMR at room temperature are in reasonable agreement. For these two end members the values are 1.25 iB and 0.57 iB (SQUID) and 1.16 iB and 0.61 iB (FMR). Agreement between these two independent measurements of magnetisation with the zero temperature prediction of the model is satisfactory.
The predicted variation of the magnetic moment with x in the three systems is shown by the red lines in Figure 2. Here, we can see that the trend is reproduced, but the moment falls less rapidly than predicted in all three systems. This may reflect a shortcoming of the magnetic valence model. It takes no account of the local environment of the atoms. This has some influence as shown by the slight increase in magnetisation on thermal annealing. The annealing does not alter the elemental composition of the alloy—the only consideration is the magnetic valence model—but it does relax the local environments while maintaining an amorphous state. The distribution of Cr and the other 3d atoms around the Fe, Co, or Ni sites will influence the moment and its average values, as in the Jaccarino-Walker model.22
The decrease in moment with Cr substitution is better reproduced if we take Zm for Cr as _5 instead of _6. This indicates that the assumption N3d = 0 for Cr is too drastic and that is so better modelled for these alloys with 0.5 electrons in the spin | d-band.
V. CONCLUSIONS
Chromium, and to a lesser extend nickel, substitution for iron is effective at reducing the magnetisation of Co40Fe40B20 with little influence on the Gilbert damping parameter a. The moment and its reduction can be explained semi-quantitatively in terms of the magnetic valence model, especially if the magnetic valence of Cr is taken as _5.
The Co40Fe40_xCrxB20 system is the best of the three we have considered for improving the performance of MgO-based magnetic tunnel junctions. The decrease in tunnel magnetoresistance with Cr substitution is more than offset by the reduction in current density jc required for spin torque switching of the free layer. The current density is reduced by a factor of four to 0.49 x 106A/cm2 when x«7.3. The Co40Fe33Cr7B20 films may be useful for the free layers in memory cells and other spin transfer torque devices.
ACKNOWLEDGMENTS
The authors would like to thank Dr. G. Feng for Co40Fe40_xNixB20 samples and Dr. M. Venkatesan and Dr. P. Stamenov for magnetic measurements. Authors also would like to thank Professor B. Aktas for fruitful discussions on ferromagnetic resonance results. This work was supported by Science Foundation Ireland under the MANSE contract 05/IN/1850. It was also conducted under the framework of the INSPIRE programme, funded by the Irish Government's Programme for Research in Third Level Institutions, Cycle 4, National Development Plan 2007-2013.
J. C. Slonczewski, J. Magn. Magn. Mater. 159(1-2), L1-L7 (1996). 2L. Berger, Phys. Rev. B 54(13), 9353 (1996).
3H. Kubota, A. Fukushima, Y. Ootani, S. Yuasa, K. Ando, H. Maehara, K. Tsunekawa, D. D. Djayaprawira, N. Watanabe, and Y. Suzuki, IEEE Trans. Magn. 41(10), 2633-2635 (2005).
4M. Tsoi, A. G. M. Jansen, J. Bass, W. C. Chiang, M. Seck, V. Tsoi, and P. Wyder, Phys. Rev. Lett. 80(19), 4281 (1998).
5J. A. Katine, F. J. Albert, R. A. Buhrman, E. B. Myers, and D. C. Ralph, Phys. Rev. Lett. 84(14), 3149 (2000).
6Y. Huai, F. Albert, P. Nguyen, M. Pakala, and T. Valet, Appl. Phys. Lett. 84(16), 3118-3120 (2004).
7Z. Diao, D. Apalkov, M. Pakala, Y. Ding, A. Panchula, and Y. Huai, Appl. Phys. Lett. 87(23), 232502-232503 (2005). 8Z. Li and S. Zhang, Phys. Rev. B 68(2), 024404 (2003).
9J. Z. Sun, Phys. Rev. B 62(1), 570 (2000).
10J. A. Katine and E. E. Fullerton, J. Magn. Magn. Mater. 320(7), 1217-1226 (2008).
UK. Moorjani and J. M. D. Coey, Magnetic Glasses (Elsevier, 1984).
12S. Ikeda, J. Hayakawa, Y. Ashizawa, Y. M. Lee, K. Miura, H. Hasegawa, M. Tsunoda, F. Matsukura, and H. Ohno, Appl. Phys. Lett. 93(8), 082508 (2008).
13D. D. Djayaprawira, K. Tsunekawa, M. Nagai, H. Maehara, S. Yamagata, N. Watanabe, S. Yuasa, Y. Suzuki, and K. Ando, Appl. Phys. Lett. 86(9), 092502 (2005).
14J. Hayakawa, S. Ikeda, F. Matsukura, H. Takahashi, and H. Ohno, Jpn. J. Appl. Phys., Part 2: Lett. 44(16-19), L587-L589 (2005).
15S. Ikeda, J. Hayakawa, Y. M. Lee, T. Tanikawa, F. Matsukura, and H. Ohno, J. Appl. Phys. 99(8), 08A907-903 (2006).
16Y. M. Lee, J. Hayakawa, S. Ikeda, F. Matsukura, and H. Ohno, Appl.
Phys. Lett. 89(4), 042506 (2006). 17K. Oguz, Ph.D. dissertation, Trinity College Dublin, 2010. 18H. Kubota, A. Fukushima, K. Yakushiji, S. Yakata, S. Yuasa, K. Ando, M. Ogane, Y. Ando, and T. Miyazaki, J. Appl. Phys. 105(7), 07D117 (2009).
19R. Topkaya, M. Erkovan, A. Ozturk, O. Ozturk, B. Akta, and M. Özdemir,
J. Appl. Phys. 108(2), 023910-023919 (2010). 20G. D. Fuchs, J. C. Sankey, V. S. Pribiag, L. Qian, P. M. Braganca, A. G. F. Garcia, E. M. Ryan, Z.-P. Li, O. Ozatay, D. C. Ralph, and R. A. Buhrman, Appl. Phys. Lett. 91(6), 062507-062503 (2007). 21A. Williams, V. Moruzzi, A. Malozemoff, and K. Terakura, IEEE Trans.
Magn. 19(5), 1983-1988 (1983). 22V. Jaccarino and L. R. Walker, Phys. Rev. Lett. 15(6), 258-259 (1965).
