The tunneling spin polarization (TSP) is directly measured from reactively sputter deposited crystalline MgO tunnel barriers with various CoFe(B) compositions using superconducting tunneling spectroscopy. We find that the Mg interface layer thickness dependence of TSP values for CoFeB/Mg/MgO junctions is substantially different from those for CoFe/Mg/MgO especially in the pre-annealed samples due to the formation of boron oxide at the CoFeB/MgO interface. Annealing depletes boron at the interface thus requiring a finite Mg interface layer to prevent CoFeOx formation at the CoFeB/MgO interface so that the TSP values can be optimized by controlling Mg thickness.
A large tunneling magnetoresistance (TMR) is one of the most important attributes of magnetic tunnel junctions (MTJs). One intriguing possibility illustrated by the Julliere model is that if magnetic electrodes can have spin polarization values of 100%, the TMR ratio of a MTJ with such electrodes would be infinitely large. Much attention has been paid on materials that can lead to a high tunneling spin polarization, thus a large TMR. It was predicted that crystalline tunnel barriers can produce extremely high tunneling spin polarization (TSP) and TMR values. In particular, TMR ratios as high as thousands of percent were predicted from MgO based MTJs1,2 and a giant TMR ratio up to 604% at room temperature has been observed recently with MgO(100) tunnel barriers after intense research activities.3–7 The origin of this giant TMR effect is the active spin filtering of crystalline MgO barrier in addition to ferromagnetic electrodes’ spin polarization, and the measurement of TSP value and sign in electrode/barrier is critical to understand TMR in MTJs.
In this work, the TSP values of reactively sputtered MgO based MTJs using various CoFe(B) compositions are compared to that of CoFe/MgO as a function of Mg interface layer thickness. It is found that the finite Mg interface layer of ∼ 8 Å is required to protect the underlying ferromagnetic electrode against oxidation in case of pre-annealed CoFe/MgO samples whereas the pre-annealed CoFeB/MgO samples do not need the Mg layer, which is due to the boron that reduces the oxidation of cobalt and iron atoms by forming boron oxides at the interface. Annealing diffuses boron atoms away from the barrier and increases the chance of Co and Fe oxidation, thus deteriorating the TSP values when the Mg interface layer is not thick enough. It is also found that the tunnel barrier height increases with thermal annealing in both CoFe and CoFeB cases thus exhibiting the improved barrier condition.
Superconducting tunnel junctions were deposited by dc magnetron sputtering in 3 mTorr argon with a background pressure of ∼2×10-9 Torr using in situ metal shadow masks. Four ferromagnetic electrodes (Co70Fe30, Co61.6Fe26.4B12, Co59.5Fe25.5B15, and Co56Fe24B20) were prepared with various Mg interface layer thicknesses. The layers in the stack are (in Å): 150 Ta/250 Ir22Mn78/15 Co84Fe16/40 CoFe(B)/t Mg/27 MgO/50 Al96Si4. The detailed growth conditions can be found elsewhere.8 A highly textured MgO(100) layer is formed by reactive magnetron sputtering from a Mg target in an Ar-O2 mixture.5 The area of the tunnel junction is ∼240×80 μm2. The measured conductance versus bias voltage curves were fitted using a quasiparticle density of states to extract the superconducting energy gap Δ, the orbital depairing parameter ζ, the spin-orbit interaction parameter b, and the TSP by using the measured temperature and field values.9
Figure 1(a) shows the dependence of TSP on the thickness (t) of the Mg interface layer. The TSP values are sensitive to the Mg layer thickness and the optimum Mg thickness for pre-annealed samples is 8 Å thus showing TSP value of 53% in Fig. 1(c). In general TSP values increase with increasing annealing temperature, irrespective of the Mg thickness, and they become more insensitive to Mg thickness for post-annealed samples. After annealing at 420 °C for 30 min, the TSP value reaches a value of 92% with a 6 Å thick Mg interface layer as shown in Fig. 1(d). This value rivals those previously observed using half-metallic ferromagnetic CrO2.10 The proposed high spin injection scheme can be applied as an efficient room temperature spin injectors for semiconductor spintronics,11 spin-based optoelectronics,12,13 and spin logic devices.14 This high TSP value in the 420 °C annealed sample is likely due to the improved CoFe/MgO interface, the better aligned exchange bias, and the reduced density of impurities/defects in the MgO barrier. The reduced defect density results in the enhancement of the tunnel barrier height (ϕ) estimated from I-V curves as shown in Fig. 1(b).15
(a) TSP of the CoFe/MgO interface versus Mg underlayer thickness for various annealing temperatures. The structure is Ta/IrMn/CoFe/t Å Mg/MgO/AlSi. (b) Extracted barrier height (ϕ) of electrons in MgO barriers with various Mg thicknesses. Conductance versus applied voltage curves of unannealed (c) and annealed sample at 420 °C (d).
(a) TSP of the CoFe/MgO interface versus Mg underlayer thickness for various annealing temperatures. The structure is Ta/IrMn/CoFe/t Å Mg/MgO/AlSi. (b) Extracted barrier height (ϕ) of electrons in MgO barriers with various Mg thicknesses. Conductance versus applied voltage curves of unannealed (c) and annealed sample at 420 °C (d).
The prevention of cobalt and iron oxidation at the interface is the key to achieve high spin injector with MgO barriers; the magnetic moment/spin polarization of CoFe oxides is significantly lower than those of CoFe. It is known that a FeOx at Fe/MgO interface is formed.16–19 Theoretical predictions show that even just a monolayer FeO can results in significant reduction of TMR values due to the decreased conductance in the parallel configuration.20,21 In our experiment, the MgO barrier is formed by first depositing a thin layer of Mg metal, ∼0-18 Å thick, followed by the reactively sputtered Mg target to grow MgO layer. The Mg interface layer protects the underlying ferromagnetic electrode against oxidation, while this layer becomes a part of MgO barrier during the reactive sputtering of the MgO layer in our case.22–26
Recent x-ray photoelectron spectroscopy (XPS) on MgO layer to investigate stoichiometry shows that the sputtered films are magnesium deficient while the evaporated films are oxygen deficient.27 In both cases, the XPS spectra clearly exhibit surface oxygen traces that is likely bound by defects in MgO film.28 There are also less chemisorbed oxygen atoms on MgO film when it is grown on a (001) textured underlayer compared to that grown on amorphous one. Thermal annealing of the MgO film results in a significant decrease of the chemisorbed oxygen atoms. The deposition of a thin metal layer on top of the MgO ends up with the displacement of some chemisorbed oxygen atoms into the MgO layer and also with the partial oxidation of the metal layer. Annealing incorporate oxygen atoms from the electrode into the MgO barrier as reported previously.16–18,29,30 Our observation of TSP value increase in samples with no Mg interface layer above 380 °C annealing can be accounted for by the reduction of the cobalt and iron oxides at the interface.
Consistent with these high TSP values, we observe a high TMR value of ∼500% at 5 K, when the top Al96Si4 layer is replaced by 25 Å Co70Fe30/150 Å Co49Fe21B30/100 Å Ta. The advantage of high fields (a few Tesla) in the superconducting tunneling spectroscopy (STS) experiments is that it can completely saturate all the magnetic moments including any paramagnetic moment. However, in MTJ a few hundred Oe fields at most are typically used to switch the devices. Because the TMR ratio is defined by (RAP-RP)/RP, where RP and RAP is the junction resistance in the parallel (P) and antiparallel (AP) alignment of the ferromagnetic electrodes, respectively, a small change in RP substantially affects the TMR ratio. The junction resistances of a MTJ with TMR∼500% at 5 K [see the inset of Fig. 2(a)] were measured while sweeping the magnetic field from positive fields down to zero as shown in each panel of Fig. 2. As expected, the junction resistance decreases with increasing fields; the resistance value 487 Ω at 0.03 T decreases down to 476.8, 451.3, and 451.7 Ω at 1.113, 11.5, and 45 T, respectively. This change in RP value from 0.03 to 11.5 T results in a 30% increase of the TMR. It is clear that a few hundred of Oe is not enough to saturate all the magnetic moments in ferromagnetic electrodes, but a high field of a few Tesla is not necessary in order to saturate all the magnetic moments including any paramagnetic moment.
(a-d) Junction resistance of a MTJ versus magnetic field at 5 K. Resistance at the highest field is indicated at each figure. For loop measurements the junction voltage is ∼ 5 mV. The Mg underlayer is 8 Å thick and the sample is annealed at 400 °C for 30 mins. The inset in (a) shows a full minor loop of the MTJ.
(a-d) Junction resistance of a MTJ versus magnetic field at 5 K. Resistance at the highest field is indicated at each figure. For loop measurements the junction voltage is ∼ 5 mV. The Mg underlayer is 8 Å thick and the sample is annealed at 400 °C for 30 mins. The inset in (a) shows a full minor loop of the MTJ.
The spin polarization of the top ferromagnet in our MTJ can be less than that of the bottom electrode due to many reasons. For example, for MTJs in which the lower ferromagnetic layer is exchange biased, stacking faults tend to propagate from the antiferromagnetic layer through the lower ferromagnetic layer and the MgO tunnel barrier to the upper ferromagnetic layer. The bias voltage dependence of TMR curve should be symmetric with regard to bias voltage when the magnetic electrodes are identical and the barrier is symmetric in shape. Figure 3 shows typical bias voltage dependences of differential resistance (dV/dI) and TMR for a MTJ at 5 K. The asymmetric TMR and dV/dI curves suggest that barrier profile is asymmetric.30,31 At negative biases electrons are injected from the bottom electrode into the top one. Asymmetry can be ascribed to the poor quality of the upper MgO/CoFe interface presumably due to the propagation of the stacking faults, dislocations, and the lattice distortion. The peak structures at zero and ± 0.3 V of the RP spectra in Fig. 3(a) are not affected by magnetic fields up to 45 T. Note that the resistance reduction as the field increases from 0.9 to 11 T is due to the further saturation of magnetization. The structures around zero bias (< 0.2 V) from both alignments result from magnon excitations as reported before.32,33 It is known that the magnon assisted spin flip process is pronounced in the AP state and sensitive to the magnetic field. However, we could not observe any field dependent magnon process since the AP state is only accessible at low magnetic fields as shown in Fig. 3(b). The two local resistance maxima around ± 0.3 V for P configuration can be explained by the Δ5 majority electron contribution to the total conductivity in the MgO barriers.34
Bias voltage dependences of dV/dI for P and AP states, and TMR of a MTJ at 5 K. TMR is based on dV/dI values at 0.9 and -0.01 T.
Bias voltage dependences of dV/dI for P and AP states, and TMR of a MTJ at 5 K. TMR is based on dV/dI values at 0.9 and -0.01 T.
Amorphous CoFeB ferromagnetic electrodes have been considered to be of great advantage compared to polycrystalline CoFe electrodes in achieving highly homogeneous devices particularly at deep sub-micron dimensions.35 Similarly, in order to obtain low coercivity (Hc) and low parallel coupling fields in MTJ devices, amorphous (or nanocrystalline) ferromagnetic layers are likely to be beneficial.36 Figure 4(a) plots the TSP values versus Mg interface layer thickness for various (Co0.7Fe0.3)1-xBx compositions. In the case of 12% and 15% boron contents, the TSP values of the pre-annealed samples have maxima for thin Mg interface layers. The TSP values gradually decrease with increasing Mg thickness up to ∼ 8 Å, and then suddenly drop above this thickness on the contrary to the report in which TMR value increases with increasing the Mg underlayer.37 The sample with 20% boron content shows a more complicated dependence of TSP values on Mg thickness: two peaks at Mg thicknesses of 0 and ∼ 8 Å. The second peak at ∼ 8 Å Mg is found to coincide with the optimal Mg thickness for CoFe electrode [see Fig. 1(a)]. TSP values increase with thermal annealing, irrespective of the Mg thickness, and they become more insensitive to Mg thickness for post-annealed samples. For example, annealing the MTJ with 12% B content CoFeB electrode at 300 °C for 30 mins shows a broad maximum TMR at ∼3-8 Å Mg thickness as shown in Fig. 4(a). The improved tunnel barrier quality after annealing is inferred from the enhancement of the tunnel barrier height (ϕ) in Fig. 4(b).
(a) TSP versus Mg underlayer thickness for various (Co0.7Fe0.3)1-xBx/MgO composition. (b) Extracted barrier height (ϕ) of electrons in MgO barriers. (c-d) EELS data from a sample after annealing at 260 °C and 300 °C, respectively.
(a) TSP versus Mg underlayer thickness for various (Co0.7Fe0.3)1-xBx/MgO composition. (b) Extracted barrier height (ϕ) of electrons in MgO barriers. (c-d) EELS data from a sample after annealing at 260 °C and 300 °C, respectively.
It is remarkable that the TSP values of pre-annealed CoFeB samples with no Mg interface layer are comparable with or even larger than those with finite Mg layer thickness, which is significantly different from CoFe electrode. This behavior becomes more pronounced in higher concentration of boron. Inferring from the heat of formation of the BOx (-300 ∼ -1274 kJ/mol), CoO (-238 kJ/mol), and FeO (-272 kJ/mol), the driving force for the formation of a cobalt/iron oxide is reduced for CoFeB compared to alloys with no boron.23 This is consistent with the high TSP values we find from the pre-annealed CoFeB/MgO samples without the Mg interface layer: we use the Mg layer for CoFe electrode to prevent the formation of CoFe oxides at the interface. It was reported that B-oxide was observed in the as-grown CoFeB/MgO samples and suggested that some BOx content can be beneficial to achieve high spin polarization.16,38
The electron energy loss spectroscopy (EELS) data in Fig. 4(d) show that the boron atoms diffuse away from CoFeB/MgO interface after annealing at 300 °C that is consistent with recent studies that boron is not found in high quality MgO barriers,39,40 but contrary to the other reports where boron diffuses into MgO during annealing.41–43 The boron diffusion into MgO in the other studies could be related to the CoFeB crystallization process during annealing,44,45 but in our cases as-grown samples show highly textured crystallinity by reactive magnetron sputtering of a Mg target.5,23 As a result of boron diffusion away from MgO in our samples, Co and Fe atoms become rich and oxidized at the interface, and TSP values decrease with thin Mg interface layer after annealing. However, since some contents of boron atoms still remain at the interface, the optimal Mg layer thickness for CoFeB electrode is smaller than that for CoFe.
In summary, we investigate the role of the Mg interface layer in crystalline MgO tunnel junctions with various CoFe and CoFeB electrodes on the tunneling spin polarization. In case of CoFe the finite thickness of the Mg interface layer is particularly useful in order to reduce the formation of ferromagnetic oxides for the pre-annealed samples. The role of Mg interface layer diminishes with the increasing anneal temperature because annealing reduces the formation of cobalt/iron oxide at the interface. The boron in CoFeB electrode of pre-annealed samples effectively decreases the oxidation of Co and Fe atoms at the interface when the Mg interface layer is very thin (< 3 Å). Upon annealing, the boron becomes depleted at the interface, thus populating cobalt/iron oxide and leading to TSP values reduction with a thin Mg underlayer. Consequently, a finite Mg interface layer thickness is required to optimize the TSP values after annealing in CoFeB electrode like CoFe.
ACKNOWLEDGMENTS
This work was partially supported by the Singapore MOE2008-T2-1-105 and the Singapore NRF-CRP 4-2008-06.