VOLUME 86, NUMBER 6 P H Y S I C A L R E V I E W L E T T E R S 5 FEBRUARY 2001 Interfacial Density of States in Magnetic Tunnel Junctions P. LeClair,* J. T. Kohlhepp, H. J. M. Swagten, and W. J. M. de Jonge Department of Applied Physics and COBRA, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands (Received 25 July 2000) Large zero-bias resistance anomalies as well as a collapse of magnetoresistance were observed in Co Al2O3 Co magnetic tunnel junctions with thin Cr interfacial layers. The tunnel magnetoresistance decays exponentially with nominal Cr interlayer thickness with a length scale of 1 Å more than twice as fast as for Cu interlayers. The strong suppression of magnetoresistance, as well as the zero-bias anomalies, can be understood by considering a strong spin-dependent modification of the density of states at Co Cr interfaces. The role of the interfacial density of states is shown by the use of specially engineered structures. Similar effects are predicted and observed in junctions with Ru interfacial layers. DOI: 10.1103/PhysRevLett.86.1066 PACS numbers: 73.40.Gk, 75.70.­i, 85.30.Mn, 85.70.Kh Since the recent discovery of large magnetoresistance in at interfaces or within intermixed regions. Among Co-3d magnetic tunnel junctions [1], there has been a renewed metal interfaces, this asymmetry is maximal for Cr [16] technological and fundamental interest in the tunneling (as is also the case for 3d impurities in Co [18,19]), while phenomenon. In general, tunneling is considered to be among Co-4d metal interfaces maximal asymmetry occurs an extremely interface sensitive technique [2­5], with the for Ru [16,18,19]. Since the band mismatch is largest in transport properties determined primarily by the density these cases, the resulting interfacial density of states modi- of states at the electrode-barrier interface [6,7]. In view fication is also the largest. In this light, Co-based magnetic of this apparent interface sensitivity, systematically alter- tunnel junctions with Cr and Ru interfacial layers seem to ing or engineering the electrode-barrier interfaces in tunnel be ideal candidates for investigating interface sensitivity structures, e.g., by the use of interfacial layers, is a natu- and the role of interfacial electronic structure. ral way to gain understanding about these devices. Sur- In this Letter, we will present evidence on the genu- prisingly, only a few experiments [4,5,8,9] of this nature ine interface sensitivity of tunneling by the use of spe- have been reported in relation to magnetic tunnel junctions. cially engineered tunneling structures utilizing multiple These experiments have primarily utilized interfacial lay- interfacial layers, designed to strongly modify the inter- ers inserted at the ferromagnetic electrode-barrier interface facial density of states. We will show that, by analyzing ("dusting" layers) in an attempt to clarify the role of the in- conductance-voltage characteristics in these structures, the terfacial density of states. However, some of these experi- TMR decrease may be correlated with the interfacial elec- ments have been difficult to interpret due to growth-related tronic structure. In striking contrast to earlier reported artifacts [9], though in general it is found that the tunneling results for Cu interlayers [8,20], for Cr interlayers the magnetoresistance (TMR) decays rapidly, on a monolayer TMR decays more than twice as fast, near vanishing by scale, as a function of interlayer thickness. 1 monolayer ML Cr. As we will argue, this extremely Theoretically, several models have been advanced rapid TMR collapse can be qualitatively explained in terms [10­14] for magnetic junctions with nonmagnetic interfa- of a strongly modified density of states at the (interdif- cial layers. In contrast to the aforementioned experimental fused) Co-Cr interface, as in magnetic multilayers [17] and data, these models generally predict that sizable TMR dilute Co-based alloys [18,19]. In addition to the strong is maintained for relatively large interfacial layer thick- TMR decrease, we report strong zero-bias anomalies in nesses, and in some cases that the TMR oscillates as a junctions with Cr, with a strong suppression of the con- function of thickness. Thus, it seems that these models ductance about V 0. Utilizing Co M1 M2 Al2O3 Co do not capture the experimentally observed interface (M1,2 Co, Cu, Cr) junctions, we will demonstrate that sensitivity of tunneling structures. the Co-Cr interface is specifically responsible for the zero- In order to clearly attribute spin-dependent tunneling bias anomalies and clearly confirm the extreme interface transport properties to an interfacial density of states, a sensitivity of tunneling. We argue that both the conduc- system in which the density of states may be modified tance results, as well as the TMR results, can be explained in a well-known way is needed. The electronic structure in terms of the same strong (spin-dependent) density of of ferromagnetic-nonmagnetic (FM-NM) interfaces has re- states modification at the Cr-Co interface, in analogy with ceived special attention in relation to the giant magnetore- mechanisms for zero-bias anomalies in nonmagnetic tun- sistance effect in metallic multilayers. In this case, the nel junctions with magnetic impurities [7,21]. Finally, degree and spin asymmetry of interfacial scattering can we will validate our conjectures by using another sys- be explained by considering the band matching between tem with a large interfacial density of states modification, FM majority/minority bands and the NM bands [15­17] viz., Co-Ru. 1066 0031-9007 01 86(6) 1066(4)$15.00 © 2001 The American Physical Society VOLUME 86, NUMBER 6 P H Y S I C A L R E V I E W L E T T E R S 5 FEBRUARY 2001 Ferromagnetic tunnel junctions were prepared by Using the relation introduced by Julliere [23] as a simple UHV dc/rf magnetron sputtering (base pressure ,5 3 first-order approximation [3], we may relate the measured 10210 mbar) through metal contact masks onto plasma TMR values to an effective tunneling spin polarization: oxidized Si(100) substrates. The details of this fabrication TMR 2P1P2 1 2 P1P2 where P1 and P2 are the ef- process have been described elsewhere [8,20]. Dusting fective spin polarizations of the first and second tunneling layers were inserted at the bottom Co Al2O3 interface [8] electrodes. As shown by the inset of Fig. 1, for submono- to avoid spurious effects due to clusterlike growth. In situ layer amounts of Cr, the polarization decreases rapidly to x-ray photoelectron spectroscopy and optical techniques near zero values, and if extrapolated, corresponds to a com- were used to confirm that there was no electrode (Co) or plete destruction of the spin polarization at 1 ML Cr. dusting layer (Co, Cu, Cr, Ru) oxidation, with a minimal From studies on Co-Cr multilayers [17] and alloys amount of remaining metallic Al [8,22]. Junction resis- [18,19], it is known that a mismatch between majority tances and conductances dI dV G V or dynamic spin d levels of Co and Cr prevents hybridization of resistances dV dI G21 V were measured using these bands. The resonant scattering of majority spin s-p standard ac lock-in techniques, with the ac excitation kept electrons with Cr d states results in the majority spin well below kBT to avoid modulation broadening. TMR density of states becoming highly localized at Cr sites (DR Rp or DG Ga) was measured using both dc and ac (i.e., the formation of a virtual bound state leads to a high lock-in techniques. majority spin density of states near the Fermi level on Cr Figure 1 shows the normalized TMR at 10 K as sites). The s-p density of states is then suppressed more a function of nominal Cr dusting layer thickness strongly for majority spins than minority spins. Since (Co Cr dCr Al2O3 Co). In contrast to previous results tunneling is particularly sensitive to s-p electrons [2,3], with Cu dusting layers, where an exponential decrease and samples only the interfacial density of states [2,6], with a length scale of j 2.6 Å was found, the magne- we may attribute the strong spin polarization reduction toresistance decay for Cr dusted junctions is considerably to the spin-dependently modified density of states at the faster, giving a length scale of nominally 1.25 Å (1.0 Å) at Co-Cr interface. This may also be viewed in terms of the 10 K (295 K). With the addition of only 3 Å Cr (approxi- magnetism of Co-Cr alloys. We point out that the Co-Cr mately 1.5 ML), the reduced TMR is only 10% of that for interfaces are expected to be significantly interdiffused a control junction. However, by subsequently covering the [24] (few ML's), and for the extremely thin Cr layers Cr with 6.3 or 10 Å Co Co Cr dCr Co dCo Al2O3 Co , used here, we may consider the dusting layer as either the TMR is nearly completely restored, saturating at a Co-Cr alloy or an intermixed Co-Cr interface (despite approximately 75% of that for a control junction. This this fact, we will continue to refer to the dusting layers clearly demonstrates not only the dramatic effect of in terms of nominal Cr thicknesses). For bulk Co-Cr Cr interlayers on tunnel spin polarization, but also the alloys, the magnetic moment is strongly reduced, with truly interfacial nature of the spin polarization reduction, the alloy becoming nonmagnetic at 25% Cr [25,26], illustrating that only a few monolayers adjacent to the a composition which may easily be reached at Co-Cr tunnel barrier are important for tunneling [2]. interfaces in the range of thicknesses used. In addition to the rapid TMR decrease, Cr dusted junctions also showed unusual conductance-voltage and 1.0 Co/d Cr Cr conductance-temperature behavior. Figure 2(a) shows Co/d Cr/6.3Å Co 10K Cr conductance vs voltage for a junction with 6.1 Å Cr Co/d Cr/10Å Co Cr 0.8 measured at various temperatures, as well as a control junction at 10 K. Strong zero-bias anomalies are present 1.0 295K 0.6 compared to a control junction, with the conductance vs 10K Co linear fit voltage changing by as much as a factor of 2 in only /P 100 mV. The narrow energy width of the anomaly is 0.4 /Cr 0.5 P Co seen clearly, where the zero-bias conductance changes Normalized TMR much more rapidly than conductance at higher biases. 0.2 0.0 0 1 2 Figure 2(b) shows conductance dI dV vs temperature d (Å) Cr data for a control junction and a junction with 6.1 Å Cr 0.0 0 1 2 3 4 5 6 (measured at V 0). Measurements on many Co-Co d (Å) control junctions routinely show 10%­15% change in re- Cr sistance from 10­300 K, in good agreement with reported FIG. 1. Normalized TMR as a function of Cr interlayer thick- work [27], which has been explained by a reduction of ness for junctions dusted with only Cr, and Cr with Co (6.3 Å, the surface magnetization with temperature [27]. For 10.0 Å), showing the near complete restoration of the original junctions with Cr measured at low voltages, an extremely TMR. Lines are a guide to the eye. Inset: Co Cr spin polariza- tion deduced from the Julliere model (see text) as a function of strong temperature dependence is exhibited relative to Cr interlayer thickness for 10 K (triangles) and 295 K (circles); control junctions, and the temperature dependence is in the line is a linear fit for 295 K. general stronger for thicker Cr interlayers. The zero-bias 1067 VOLUME 86, NUMBER 6 P H Y S I C A L R E V I E W L E T T E R S 5 FEBRUARY 2001 (a) (b) 1.0 Å Cr 1 6.3 Å Co. Compared to a junction with Cr at 1.3 Control 10K No Cr the interface, when the Cr layer is positioned a few ML's 1.0 6.1Å Cr, V~0 (6.3 Å) away from the interface the "anomalous" effects 1.2 0.9 G have nearly disappeared. As pointed out earlier, this is also ) 1.1 accompanied by an almost full restoration of the tunneling -2 p 0.8 (T)/G spin polarization. In addition, Cu interlayers were used cm 1.0 to show that Cr in contact with Co is responsible for the -1 0.7 p ( anomalous behavior. Shown in Fig. 3(b) are G ( 300K p vs voltage p 0.9 20.0K 0.6 characteristics for junctions with 1.8 Å Cr, 3.0 Å Cu, G 15.0K 12.0K ) 1.8 Å Cr 1 3.0 Å Cu, and 3.0 Å Cu 1 1.8 Å Cr dusting 0.8 9.0K 0.5 7.0K layers. For Cu dusting, no anomalies are seen [20], while 4.5K for Cr dusting extremely strong anomalies are observed. 0.7 0.4 -50 -25 0 25 50 0 50 100 150 200 250 300 However, for dusting layers of 3.0 Å Cu 1 1.8 Å Cr, the V (mVolts) T (K) p anomaly strength is reduced by roughly a factor of 10, despite the fact that the Cu thickness is only 1.5 ML. FIG. 2. (a) Conductance vs voltage at various temperatures for a junction with 6.1 Å Cr as well as a control junction at It is clearly seen that when Cr is at the interface but 10 K. The control junction curve has been vertically shifted for backed with Cu rather than Co, the anomaly strength clarity. (b) Conductance vs temperature (normalized to 300 K) is much reduced, indicative of the magnetic nature of for junctions with no Cr and 6.1 Å Cr. the anomalies. Finally, to show that the Co-Cr interface is responsible for the effects, rather than the Cr-Al2O3 conductance minima were present even at 300 K, with a interface (or Cr within the Al2O3), a Co electrode was width of approximately kBT, suggesting that the tempera- dusted with 1.8 Å Cr 1 3.1 Å Cu [Fig. 3(b)]. In this ture dependence of the zero-bias conductance results only case, the anomaly is clearly still present, though approxi- from thermal smearing of a near-singular density of states. mately a factor of 5 weaker than for 1.8 Å Cr alone. The temperature and voltage dependence are roughly The anomaly is approximately a factor of 2 stronger for logarithmic for low bias and temperature, though we note the Co Cr Cu combination compared to the Co Cu Cr that the resistance may be just as convincingly shown to combination, further indicating that the Co Cr interface be logarithmic, as found by previous authors [28]. We will plays the dominant role. return to the origin of the zero-bias anomalies, as well as Returning once again to the underlying physical mecha- their possible relation to the rapid TMR decrease, later on. nisms, large zero-bias anomalies have been extensively Multiple dusting layers can be used to experimen- studied [21] specifically in nonmagnetic junctions where tally establish that the Co Cr interface is specifically magnetic impurities or impurity layers were placed within responsible for these zero-bias anomalies. Figure 3(a) one of the electrodes or within the insulating barrier. For shows Gp vs applied bias for a control junction, a junc- magnetic impurities within a nonmagnetic electrode, the tion dusted with 1.8 Å Cr, and a junction dusted with anomalies were explained by considering the modifica- tion of the interfacial density of states by the impurities (a) (b) [7,29]. Mezei and Zawadowski [7] found theoretically that the tunnel conductance is proportional to the local den- Cr 2.1 Cr 1.8 sity of states at the electrode-barrier interface, which is in 1.6 1.6 turn inversely proportional to the s-d scattering amplitude. Essentially, the logarithmic zero-bias anomalies measure (0) G the energy dependence of the Kondo scattering amplitude. 1.4 Ru 1.2 1.4 p Cr 1.8/Cu 3.0 p (V)/G Although their work may not be directly applicable to the Cr 2.1/Co 6.3 present case (which deals with magnetic junctions), we (V)/G 1.2 p may understand the present experiments based on these Cu 3.0/Cr 1.8 1.2 G p (0) ideas. We feel that the strongly depressed density of states Cu 3.0 at Co-Cr interfaces (particularly for majority spins) in- 1.0 none none 1.0 duced by resonant scattering, as discussed earlier, essen- -400 -200 0 200 400 -400 -200 0 200 400 tially fulfills the requirements of Mezei and Zawadowski V (mVolts) for observing strong zero-bias anomalies. If we further p V (mVolts) p conjecture that Cr moments in Co Cr are spin fluctuat- FIG. 3. (a) Normalized parallel conductance for junctions with ing [18], Kondo-like behavior could be anticipated, and no dusting layer, and dusting layers of (2.1 Å Cr 1 6.3 Å Co), the model of Mezei and Zawadowski would be more ap- 2.1 Å Cr, and 1.2 Å Ru. (b) As in (a) for no dusting layer, 3.0 Å Cu, (3.0 Å Cu 1 1.8 Å Cr), (1.8 Å Cr 1 3.0 Å Cu), plicable. In other words, we probe the energy-dependent and 1.8 Å Cr. Multiple dusting layers clearly demonstrate the scattering of conduction electrons by fluctuating Cr mo- role of the Co-Cr interface. Some curves have been vertically ments. The strong similarities between their model and shifted for clarity. our results for Cr on Co clearly point to an explanation 1068 VOLUME 86, NUMBER 6 P H Y S I C A L R E V I E W L E T T E R S 5 FEBRUARY 2001 related to a strongly modified local density of states. Par- sults of our investigations. We also acknowledge K. Flipse, enthetically, we do not suggest that the zero-bias anomaly P. M. Tedrow, R. Meservey, and R. Coehoorn for their sup- reflects directly the Co Cr density of states, but, rather, port and interest in this work. P. LeClair is supported by energy-dependent electron-electron scattering at the Co Cr the Dutch technology foundation STW. interface. In order to validate our conjectures about zero-bias anomalies, we have also prepared junctions with Ru dusting layers. For 4d-metal interfaces with Co, as well as for impurities in Co, it is Ru which shows the *Corresponding author. maximal scattering cross section as well as the largest Electronic address: pleclair@phys.tue.nl spin asymmetry, and hence the strongest modification [1] J. S. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey, of the interfacial density of states [16,18,19]. Further, Phys. Rev. Lett. 74, 3273 (1995). NMR studies [30] on Co-Ru multilayers indicate strong [2] R. Meservey and P. M. Tedrow, Phys. Rep. 238, 173 (1994). interdiffusion ( 2 ML per interface), and a description [3] J. S. Moodera, J. Nassar, and J. Mathon, J. Magn. Magn. in terms of Ru impurities in Co is reasonable. If an Mater. 200, 248 (1999). explanation based on a strongly altered interfacial density [4] J. S. Moodera, T. H. Kim, C. Tanaka, and C. H. de Groot, of states and spin fluctuations is correct, it is expected Philos. Mag. B 80, 195 (2000). [5] J. M. D. Teresa et al., Science 286, 507 (1999). that junctions with Ru interlayers should behave similarly [6] H. Itoh, J. Inoue, S. Maekawa, and P. Bruno, J. Magn. Soc. to those with Cr interlayers. Figure 3(a) shows Gp V Jpn. 23, 52 (1999). for a control junction, a junction with 2.1 Å Cr, and a [7] F. Mezei and A. Zawadowski, Phys. Rev. B 3, 167 (1971); junction with 1.2 Å Ru. As with Cr, junctions dusted 3, 3127 (1971). with Ru indeed also exhibit large zero-bias anomalies, [8] P. LeClair et al., Phys. Rev. Lett. 84, 2933 (2000). with the conductance changing by more than a factor of 2 [9] J. J. Sun and P. P. Freitas, J. Appl. Phys. 85, 5264 (1999). within 300 mV, supporting our explanation. Further, the [10] J. Mathon and A. Umerski, Phys. Rev. B 60, 1117 (1999). TMR decrease observed for Ru interlayers is analogous to [11] W. Zhang, B. Li, and Y. Li, Phys. Rev. B 58, 14 959 (1998). that for Cr interlayers, viz., j [12] V. Vedyaev et al., Europhys. Lett. 39, 219 (1997). Ru 1 Å, with a near zero effective spin polarization for Ru thicknesses greater than [13] A. Vedyayev, M. Chshiev, N. Ryzhanova, and B. Dieny, 1 ML. We emphasize here again that Cu interlayers Phys. Rev. B 61, 1366 (2000). [14] S. Zhang and P. M. Levy, Phys. Rev. Lett. 81, 5660 (1998). show no zero-bias anomalies [20], and exhibit a decay [15] J. Inoue and S. Maekawa, Prog. Theor. Phys. Suppl. 106, length more than a factor of 2 longer than for either Cr or 187 (1991). Ru interlayers [8]. For Cu on Co, the resonant scattering [16] H. Itoh, J. Inoue, and S. Maekawa, Phys. Rev. B 47, 5809 condition is not fulfilled (i.e., the virtual bound state is far (1993). from the Fermi level), and thus the strong suppression of [17] C. Vouille et al., Phys. Rev. B 60, 6710 (1999). the local density of states is not expected as for Cr or Ru [18] P. L. Rossiter, in The Electrical Resistivity of Metals and on Co [16,17]. One can also view this in terms of the more Alloys (Cambridge University Press, Cambridge, 1987). drastic effect of Cr and Ru on the interface magnetism [19] V. S. Stepanyuk, R. Zeller, P. H. Dederichs, and I. Mertig, compared to Cu. Further, for Co-Cu, a relatively sharp Phys. Rev. B 49, 5157 (1994). interface was observed [8], and thus a description in terms [20] P. LeClair, J. T. Kohlhepp, H. J. M. Swagten, and of Cu impurities in Co is less valid. W. de Jonge, Appl. Phys. Lett. 76, 3783 (2000). [21] E. L. Wolf, in Principles of Electron Tunneling Spec- In conclusion, we have experimentally established the troscopy (Oxford University Press, London, 1985), dramatic role of the interfacial density of states in (mag- Chap. 8, for an extensive review. netic) tunnel junctions. We have also, utilizing TMR [22] P. LeClair et al., J. Appl. Phys. 87, 6070 (2000). and conductance-voltage characteristics, given experimen- [23] M. Julliere, Phys. Lett. 54A, 225 (1975). tal indications for the underlying mechanisms. A more [24] Y. Henry, C. Mény, A. Dinia, and P. Panissod, Phys. Rev. complete theoretical picture is clearly needed -the model B 47, 15 037 (1993). of Mezei and Zawadowski [7] needs to be extended to the [25] E. Kneller, in Ferromagnetismus (Springer-Verlag, Berlin, case of magnetic electrodes, or alternatively, the model of 1962), Chap. 10. Itoh et al. [15,16] needs to be extended to tunneling struc- [26] R. M. Bozorth, in Ferromagnetism (D. van Nostrand Com- tures. Once the role of interfacial density of states effects pany, Inc., Princeton, 1961). is more completely understood, magnetic tunneling struc- [27] C. H. Shang, J. Nowak, R. Jansen, and J. S. Moodera, Phys. Rev. B 58, R2917 (1998). tures may perhaps be engineered for improved magnetore- [28] L. Y. L. Shen and J. M. Rowell, Phys. Rev. 165, 566 (1968). sistive properties. [29] T. Ivezic´, J. Phys. C 8, 3371 (1975). The authors are indebted to A. Fert and J. S. Moodera for [30] H. Wieldraaijer, J. T. Kohlhepp, and P. LeClair their comments and stimulating discussions about the re- (unpublished). 1069