Spin-dependent tunneling spectroscopy in single-crystal Fe/MgO/Fe tunnel junctions

著者 安藤 康夫 journal or Applied Physics Letters publication title volume 87 page range 142502-1-142502-3 year 2005-10 URL http://hdl.handle.net/10097/34659 APPLIED PHYSICS LETTERS 87, 142502 ͑2005͒

Spin-dependent tunneling spectroscopy in single-crystal Fe/MgO/Fe tunnel junctions ͒ Y. Ando,a T. Miyakoshi, M. Oogane, and T. Miyazaki Graduate School of Engineering, Tohoku University, 6-6-05 Aoba-yama, Sendai 980-8579, Japan ͒ H. Kubota, K. Ando, and S. Yuasab National Institute of Advanced Industrial Science and Technology (AIST), 1-1-1 Umezono, Tsukuba 305-8568, Japan ͑Received 11 April 2005; accepted 9 August 2005; published online 27 September 2005͒ We report a detailed spin-dependent tunneling spectroscopy in single-crystal Fe͑001͒/MgO͑001͒/Fe͑001͒ magnetic tunnel junctions ͑MTJs͒ that show a giant tunnel magnetoresistance effect. Spectra for antiparallel magnetic configurations show asymmetry because of extrinsic electron scatterings caused by structural defects at the barrier/electrode interfaces. Surprisingly, spectra for parallel magnetic configurations exhibit a complex oscillatory structure that has never been observed in conventional MTJs with an aluminum-oxide tunnel barrier. The complex spectra reflect the tunneling process via interface resonant states. These results provide some information that helps to elucidate the physics of spin-dependent electron tunneling and to further enhance magnetoresistance. © 2005 American Institute of Physics. ͓DOI: 10.1063/1.2077861͔

Recent progress of a giant tunnel magnetoresistance scopic study of the single-crystal MTJs is important for clari- ͑TMR͒ effect in magnetic tunnel junctions ͑MTJs͒ with a fying the controversial physical mechanism of spin- MgO͑001͒ tunnel barrier1–4 promises the promotion of in- dependent tunneling. dustrial applications to magnetoresistive random access Single-crystal Fe͑001͒/MgO͑001͒/Fe͑001͒ MTJs with ͑ ͒ memory and magnetic sensors. The giant TMR effect for the MgO barrier thicknesses tMgO of 2.55–3.14 nm were pre- single-crystal Fe͑001͒/MgO͑001͒/Fe͑001͒ MTJs is ex- pared using molecular-beam epitaxy and microfabrication ⌬ plained by coherent tunneling of 1 electrons. The majority techniques. The detailed sample structure is described in the ⌬ 2 spin 1 band has states at the Fermi energy level, EF, literature cited. All measurements were performed using a ⌬ whereas the minority spin 1 band has no states at EF. The lock-in technique. The sample was set in the measurement MgO͑001͒ tunnel barrier preferentially filters the perfectly chamber and cooled to 6 K using He gas. This measurement ⌬ spin-polarized 1 electrons in the case of coherent tunneling, system achieved extremely low fluctuation noise during mea- thereby yielding the giant TMR effect. Moreover, tunnel surement. The modulation frequency of 2.819 kHz allowed magnetoresistance of Fe/MgO/Fe MTJs has shown an oscil- simultaneous recording of 1f and 2f signals with a lock-in lation as a function of the tunnel barrier thickness,2 indicat- amplifier. The modulation amplitude was changed for appro- ing that wave function coherence is conserved across the priate values depending on the samples; the typical value tunnel barrier. Although even higher MR ratios greater than was 1 mV. The bias direction was defined with respect to the 1000% have been predicted,5,6 the magnetoresistance ͑MR͒ top Fe electrode. ratio seems to reach a ceiling at around one-fourth of that Figure 1͑a͒ shows typical bias dependences of differen- ͑ ͒ value. tial magnetoresistance MRdiff as a function of bias-voltage This letter describes our measurement of current-voltage ͑V͒ obtained from the dV/dI-V curve for the Fe/MgO/Fe ͑ ͒ ͑ ͒ I-V curves, dynamic resistance curves dV/dI-V , and de- MTJ with the tMgO of 2.55 nm. The curve is asymmetric with rivatives of the resistance ͑d2V/dI2-V͒ for single-crystal respect to the bias voltage, but it shows a large MR ratio up Fe͑001͒/MgO͑001͒/Fe͑001͒ MTJs. The d2V/dI2-V mea- to 270% at near-zero bias. This characteristic is similar to the surement is well known in inelastic electron tunneling ͑IET͒ bias dependence of MR obtained from the I-V curve.2 Large spectroscopy as a powerful method to detect various sorts of bias dependence of the MR ratio decreases the output voltage 7 ͓ϵ ϫ͑ ͒ ͔ elementary excitations at tunnel-junction interfaces. This Vout V Rap−Rp /Rap for device applications, where Rp spectroscopic technique, when applied to MTJs, detects spin- and Rap, respectively, show tunnel resistances when the mag- dependent excitations, including impurity scattering8,9 and netizations of the two electrodes are aligned in parallel and magnon excitation.10,11 In principle, the spectra reflect not in antiparallel. Therefore, improvement of the bias depen- only inelastic excitations but also direct information of den- dence for negative voltage, defined as biased to the bottom sity of states ͑DOS͒ characteristics. However, it has re- Fe electrode, is an important issue. Corresponding IET spec- mained unclear because an amorphous aluminum-oxide tra are shown in Fig. 1͑b͒. They depict spectra for parallel ͑Al–O͒ tunnel barrier was commonly used in conventional ͑P͒ and antiparallel ͑AP͒ magnetization configurations. MTJs, in which the coherence of tunneling electrons’ Bloch Steep declines of MR between −200 mV and +100 mV are wave functions was not conserved. Therefore, IET spectro- reflected in the sharp peaks in the IET spectrum for AP. The peaks have maxima at ±25 mV because of magnon-assisted 12,13 ͒ a Electronic mail: [email protected] inelastic excitation. Therefore, the initial decrease of the ͒ b Also at: PRESTO, Japan Science and Technology Agency ͑JST͒, Kawagu- MR ratio with respect to the bias originates from the surface chi, Saitama 332-0012, Japan. magnon of both sides of Fe electrodes. In addition, a shoul-

0003-6951/2005/87͑14͒/142502/3/$22.5087, 142502-1 © 2005 American Institute of Physics Downloaded 11 Jun 2008 to 130.34.135.158. Redistribution subject to AIP license or copyright; see http://apl.aip.org/apl/copyright.jsp 142502-2 Ando et al. Appl. Phys. Lett. 87, 142502 ͑2005͒

FIG. 1. Bias-dependence of ͑a͒ MR ratio and ͑b͒ the corresponding IET spectra. The MR ratio was obtained from the dV/dI-V curve, MRdiff, ͓ϵ͑ ͒ dV/dIa −dV/dIp /dV/dIp, where dV/dIp and dV/dIa, respectively, rep- resent dynamic resistance with parallel and antiparallel magnetic configura- FIG. 2. IET spectra in an antiparallel magnetization configuration. Spectra ͔ ͑ ͒ ͑ ͒ ͑ ͒ tions for Fe 001 /MgO 001 /Fe 001 MTJ with tMgO of 2.55 nm. IET are measured for various thicknesses of MgO tunnel barriers, tMgO. Base spectra for the parallel and antiparallel magnetization configurations are lines for respective spectra are shifted to clarify the spectrum shape. For shown. MTJs with thinner barrier thickness, the maximum applied voltage was re- duced to avoid junction breakdown. der is visible only in the negative bias ͑indicated by an ar- 1 row͒. This additional component of intensity, which is calcu- sufficiently below EF. Consequently, conduction channels ⌬ lable by subtracting the spectrum for positive bias from that between the majority spin and the minority spin 1 bands for the negative bias, is almost identical for all MTJs with would open when the applied bias is higher than the band different tMgO. It had a broad maximum around −180 mV. edge for AP magnetic configuration, thereby creating the This extra structure, which was observed only for negative high-energy peaks. The observed peak positions are consis- bias, reflects an asymmetry in the MTJs. Although the tent with the band structure calculated for Fe͑001͒. Tunnel- Fe/MgO/Fe MTJs’ structure is basically symmetric, a strong ing current is dominated by electrons with momentum vec- asymmetry was apparent in the density of misfit dislocations tors that are exactly normal to the barrier because the at the barrier/electrode interfaces.2 The dislocation density tunneling probability decreases rapidly when the momentum for the lower interface is about four times higher than that for vectors deviate from the barrier-normal direction when the the upper interface. These results imply that the strong bias tunnel barrier is thick. On the other hand, even electrons with dependence of TMR for negative bias originates from spin- momentum vectors deviating from the barrier-normal direc- flip scattering caused by the interface dislocations. There- tion have a finite tunneling probability when the tunnel bar- ⌬ fore, it might be improved by reducing the dislocation den- rier is thin. Therefore, the influence of the 1 band edge on sity. The IET spectrum intensity is small for P, indicating the IET spectrum can become more prominent with increas- 12 that the spin-flip inelastic excitation is dominant for AP. ing tMgO, resulting in enhancement of the high-energy peaks. Figure 2 shows IET spectra in the AP magnetic configu- The tunnel resistance for the MTJ with tMgO =3.14 nm is two ration for various tMgO. Sharp peaks at low bias voltages orders of magnitude higher than that for the MTJs with tMgO attributable to magnon excitation tend to decrease with in- of 2.55 nm. Such high resistance allows clear peak detection. creasing tMgO. Surprisingly, along with these peaks, broad One might claim that the visible broad peaks for tMgO peaks appear and grow around ±1000 mV ͑indicated as ar- Ͻ3.14 nm seem to shift to lower voltages. Although this ͒ rows with increasing tMgO. It is noteworthy that peaks at could be related to interface resonant states discussed below, such high bias voltages have never been observed in the there is no clear explanation at the present moment. MTJs with Al–O tunnel barrier and are usually not encoun- Figure 3 shows IET spectra in P magnetic configuration tered in ordinary practice because such high-energy inelastic for various tMgO. The intensity for P configuration is very excitation usually does not exist. IET spectra generally rep- small. Therefore, the vertical axis scale is enlarged to afford resent spin-dependent inelastic excitations on both interfaces. a better view of the detailed structure. Broad peaks around Energy of the hot electrons biased at 1000 mV is much ±1000 mV, which closely resemble the spectra for AP con- ⌬ higher than the Curie temperature of the ferromagnetic elec- figuration, were also observed. The 1 band edge effect trodes. Therefore, the magnon excitation should cease.13 The might also be reflected on IET spectra for P configuration if high-energy peaks in the d2V/dI2-V spectra are considered to spin flip scatterings occur during tunneling. It is noteworthy include information on spin-dependent conductance channels that complex structures were observed between −1000 mV that is related to the two electrodes’ DOS. The minority spin and +1000 mV. Furthermore, it is surprising that the spectra ⌬ ͑ ͒͑ ͒ 1 band has no states at the EF for the 001 barrier-normal exhibit a clear oscillatory behavior as a function of the bias direction, but a band edge exists at approximately 1.2 eV voltage. As a guide for the eyes, thin broken lines are in- ⌬ ͑ ͒ ͑ ͒ above EF, whereas an edge of the majority 1 band exists cluded in Fig. 3 at the first ±100 mV , second ±350 mV , Downloaded 11 Jun 2008 to 130.34.135.158. Redistribution subject to AIP license or copyright; see http://apl.aip.org/apl/copyright.jsp 142502-3 Ando et al. Appl. Phys. Lett. 87, 142502 ͑2005͒

tunneling is the interface resonant states, whose electronic structure is probably much more complex than those of the ⌬ 5 1 states. The observed oscillatory IET spectra might there- fore originate from the minority-spin tunneling at the hot spots. The spectra might contain intrinsic information related to the tunneling process via the hot spots if this were the case. In conclusion, we investigated the spin-dependent spec- troscopy in single-crystal Fe/MgO/Fe tunnel junctions that showed a giant TMR effect. The asymmetric IET spectra for AP magnetic configurations showed that the extrinsic spin- dependent electron scattering occurred at the bottom barrier/ electrode interface. In addition, we observed the oscillatory structure for the spectra for a P configuration. This might constitute experimental evidence for hot-spot tunneling. Tun- neling via the interface resonant states could be understood if rigorous theoretical calculations on the tunneling spectros- copy were performed and carefully compared with the present experimental results. The authors would like to thank Y. Suzuki, T. Katayama, FIG. 3. IET spectra in a parallel magnetization configuration. Spectra are T. Nagahama, A. Fukushima, and M. Yamamoto for their measured for various thicknesses of MgO tunnel barrier, tMgO. Base lines for assistance in sample preparation. This study was partially respective spectra are shifted. Thin broken lines represent approximate peak supported by the IT-program of Research Revolution 2002 positions: first ͑±100 mV͒, second ͑±350 mV͒, and third ͑±800 mV͒. ͑RR2002͒, Grants-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science, and Tech- nology of Japan, and NEDO Grant Program. and third ͑±800 mV͒ peak positions. We confirmed that these oscillations were consistently reproducible and that no peaks 1S. Yuasa, A. Fukushima, T. Nagahama, K. Ando, and Y. Suzuki, Jpn. J. existed that were attributable to molecular vibrational states. Appl. Phys., Part 2 43, L588 ͑2004͒. 2 The peak positions are independent of tMgO. A peak for the S. Yuasa, T. Nagahama, A. Fukushima, Y. Suzuki, and K. Ando, Nat. Mater. 3, 868 ͑2004͒. IET spectrum implies a stepwise increase of the correspond- 3 ing tunneling conductance curve at the peak position. An S. S. P. Parkin, C. Kaiser, A. Panchula, P. M. Rice, B. Hughes, M. Samant, and S. H. Yang, Nat. Mater. 3,862͑2004͒. oscillation in the IET spectrum produces a stair-shaped I-V 4D. D. Djayaprawira, K. Tsunekawa, M. Nagai, H. Maehara, S. Yamagata, curve. N. Watanabe, S. Yuasa, Y. Suzuki, and K. Ando, Appl. Phys. Lett. 86, Heretofore, oscillatory IET spectra have not been pre- 092502 ͑2005͒. 5W. H. Butler, X.-G. Zhang, T. C. Schulthess, and J. M. MacLaren, Phys. dicted theoretically. The main tunneling channels for the P ͑ ͒ ⌬ Rev. B 63, 054416 2001 . configuration are between the majority-spin 1 states with 6J. Mathon and A. Umerski, Phys. Rev. B 63, 220403 ͑2001͒. 5,6 7 kʈ =0 in the two Fe electrodes. Therefore, such a complex P. K. Hansma, Phys. Rep., Phys. Lett. 30C,147͑1977͒. IET spectrum is not expected from the simple parabolic char- 8J. Murai, Y. Ando, N. Tezuka, and T. Miyazaki, J. Magn. Soc. Jpn. 22, 573 ͑1998͒. acteristics of the ⌬ bands for kʈ =0. Although the quantum 1 9M. Hayashi, Y. Ando, M. Oogane, H. Kubota, and T. Miyazaki, Jpn. J. well effect, or interference effect of tunneling states in the ͑ ͒ 2,5 Appl. Phys., Part 1 43, 7472 2004 . MgO layer, might cause an oscillatory behavior, this effect 10J. S. Moodera, J. Nowak, and R. J. M. van de Veerdonk, Phys. Rev. Lett. cannot be the origin of the observed IET spectrum oscillation 80, 2941 ͑1998͒. 11 because the observed oscillation is independent of the thick- Y. Ando, J. Murai, H. Kubota, and T. Miyazaki, J. Appl. Phys. 87,5209 ͑2000͒. ness of MgO. Contrary to the majority-spin tunneling be- 12 A. M. Bratkovsky, Phys. Rev. B 56, 2344 ͑1997͒. ⌬ 13 tween the 1 states, minority spin electrons tunnel at certain S. Zhang, P. M. Levy, A. C. Marley, and S. S. P. Parkin, Phys. Rev. Lett. 5,6 kʈ ͑0͒ points ͑so called hot spots͒. The origin of hot-spot 79, 3744 ͑1997͒.

Downloaded 11 Jun 2008 to 130.34.135.158. Redistribution subject to AIP license or copyright; see http://apl.aip.org/apl/copyright.jsp