Fe concentration dependence of tunneling magnetoresistance in magnetic tunnel junctions using group-IV ferromagnetic semiconductor GeFe

Exploiting the spin dependent transport in group-IV semiconductors is particularly important for future semiconductor-based spintronics. Group-IV-based ferromagnetic semiconductor (FMS) Ge_1−xFe_x (GeFe) is one of the most attractive materials for efficient spin injection/detection in Si and Ge^1,2 because GeFe can be epitaxially grown on Si and Ge substrates by low temperature molecular beam epitaxy (LT-MBE) and its conductivity can be varied from metallic to insulating by boron (B) doping.³ The highest Curie temperature T_C ever reported for GeFe is 210 K,¹ which is higher than the highest T_C value of (Ga, Mn)As (200 K).⁴ Recent soft X-ray magnetic circular dichroism measurements have revealed that nm-sized local ferromagnetic regions exist in GeFe even at room temperature,⁵ which makes GeFe promising for spintronics devices operating at room temperature.

Recently, tunneling magnetoresistance (TMR) has been observed in a magnetic tunnel junction (MTJ) composed of Fe/MgO/Ge_0.935Fe_0.065; however, the observed TMR ratio was only 0.27%.⁶ Considering the recent first-principles calculations, which estimated the spin polarization of GeFe to be 70%,⁷ the TMR ratio is expected to be enhanced to 270% in Fe/MgO/GeFe [here, the effective spin polarization of Fe(/MgO) is considered to be 75%].⁸ One of the possible reasons of this low experimental value of TMR is related to the inhomogeneity of the Fe concentration in the GeFe layer. It is known that local high Fe content regions play an important role in the ferromagnetism.¹ This fact also suggests that local low Fe content regions contribute to the decrease in the TMR ratio. By increasing x, the averaged Fe concentration and thus the effective spin polarization of the entire GeFe layer will be increased. Another possible reason of the small TMR ratio is the large mesa size used in the previous study (700 $×$ 700 μm²).⁶ It is well known that MgO layers have pinholes, which cause a leakage current and decrease TMR.⁹ By decreasing the size of the mesa, we may be able to avoid these pinholes. In this paper, by decreasing the mesa size and by increasing x, we successfully increase the TMR ratio up to 1.5%. This is the first observation of the TMR ratio over 1% in MTJs containing a group-IV ferromagnetic semiconductor.

We have grown epitaxial single-crystal trilayer structures composed of Fe (18 nm)/MgO (3 nm)/Ge_1−xFe_x (60 nm, x = 0.065, 0.105, 0.140, and 0.175)/Ge:B (B: $4×1019$ cm⁻³, 80 nm) on p⁺Ge (001) substrates by LT-MBE [Fig. 1(a)]. Before the growth, the Ge substrates were cleaned chemically with ultrapure water, acetone, and ammonia water, and then etched with ultrapure water and buffered HF in a cyclic manner for 1 h. The substrates were introduced in the ultra-high-vacuum MBE growth chamber through an oil-free load-lock system. We annealed them at the substrate temperature T_S of 300 °C for 30 min and then at T_S = 700 °C for 30 min for surface cleaning. We grew the Ge:B buffer layer at T_S = 300 °C, which was followed by the growth of Ge_1−xFe_x at T_S = 240 °C. The MgO layer was grown by electron beam deposition in the MBE growth chamber at T_S = 80 °C with a growth rate 0.02 Å/s. The Fe top layer was grown at T_S = 50 °C, and then we annealed the sample at T_S = 250 °C for 30 min to obtain a flat top Fe surface. The in-situ reflection high-energy electron diffraction (RHEED) patterns of the Ge_0.895Fe_0.105, MgO, and Fe layers show that these layers are epitaxially grown with the epitaxial relationship of Fe[100](001) || MgO[110](001) || GeFe[100](001) [Figs. 1(b)–1(e)]. This epitaxial relationship is the same as that of Fe/MgO/Ge.¹⁰ The spotty pattern of the as-grown Fe layer [Fig. 1(d)] changed to the more streaky pattern by the annealing [Fig. 1(e)], which means that the Fe top layer becomes flat by the annealing. The estimated T_C values of Ge_1−xFe_x are 70 K, 120 K, 160 K, and 170 K for x = 0.065, 0.105, 0.140, and 0.175, respectively.¹¹ 

After the growth, an Al layer was deposited on the sample to prevent the surface oxidization. For the tunneling transport measurements, we fabricated circular-shaped mesa MTJ devices whose diameters ϕ are 5.54, 16.7, 56.3, and 169 μm, using conventional photolithography and Ar-ion milling. We note that these mesa sizes are much smaller than that used in the previous study (700 $×$ 700 μm²).⁶ We deposited SiO₂ for the passivation of the mesa diodes and evaporated Al as the top electrode. In the tunneling transport measurements, we grounded the substrate and applied a bias voltage V to the top Al electrode.

Figure 2(a) shows the resistance area product (RA) as a function of the magnetic field μ₀H measured at 3.7 K for the MTJ with x = 0.105 and ϕ = 16.7 μm when V = 100 mV and H is applied along the [100] axis of GeFe in the film plane. Note that the [100] axis corresponds to the easy magnetization axis of the Fe layer.¹² The GeFe layer has only very weak magnetic anisotropy in the film plane, and the magnetization response hardly depends on the magnetic-field direction in GeFe.⁶ We obtained the TMR ratio of 1.5% (⁠$=(RAP−RP)/RP×100$⁠), which is higher than that obtained in the previous study.⁶ Here, we define R_P as the RA value at μ₀H = 0.3 T and R_AP as the highest RA value when H is negative. Though R_P was defined as the value at μ₀H = 0.2 T in the previous study,⁶ the difference in the magnetic fields used for deriving R_P is not so important because the MR is nearly saturated at both magnetic fields. This improvement of the TMR ratio is attributed to the smaller mesa size and larger x than those used in the previous study (700 $×$ 700 μm², x = 0.065).⁶ The jumps of RA at $±$ 24 mT correspond to the reversal of the magnetization direction of the Fe layer. We note that the small minor loop [see the red curves in the main graph and the inset of Fig. 2(a)] originates from the small hysteresis of GeFe. The minor loop will open when the ferromagnetic order in GeFe is enhanced and the hysteresis of GeFe becomes larger, for example, by annealing the GeFe layer.¹ The V dependence of the TMR ratio of our MTJs shows a typical feature of TMR;^8,13 the TMR ratio increases with decreasing |V | to 0 V [Fig. 2(b)].

To confirm that our data originates from TMR, we performed the same vertical transport measurement on a reference sample composed of Ge_0.895Fe_0.105 (60 nm)/Ge:B (B: $4×1019 cm−3$⁠, 80 nm)/p⁺Ge, which was grown with the same conditions as those of the trilayer sample. Figure 3 represents the observed magnetoresistance (MR) curves of this reference sample when applying the magnetic field in the same direction as that used in Fig. 2(a) (i.e. the [100] axis). Here, we do not see any clear feature at all, which indicates that anisotropic magnetoresistance (AMR) of GeFe does not affect the TMR curves obtained for Fe/MgO/GeFe. Also, because the magnetization curve of GeFe is very rounded as shown in Fig. 4(a), tunneling anisotropic magnetoresistance (TAMR) of GeFe cannot explain the sharp jumps of RA in our data shown in Fig. 2(a). The influence of (T)AMR of Fe can be also neglected as discussed later in Fig. 5.

The TMR curves obtained for the MTJ are well reproduced by the Julliere model,¹⁴ by which the TMR ratio is expressed as⁶ 

Here, P_Fe(P_GeFe) is the spin polarization of Fe (GeFe), and m_Fe (m_GeFe) is the magnetization of Fe (GeFe) normalized by the saturation magnetization. We fitted Eq. (1) to the observed TMR curves with the least squares method [Fig. 4(b)] using the m_Fe − H and m_GeFe − H curves shown in Fig. 4(a). Here, for deriving the m_GeFe − H curve, we used the magnetic circular dichroism (MCD) signal divided by that at μ₀H = 0.7 T measured for a Ge_0.895Fe_0.105 sample [Fig. 4(a)], which was obtained by etching the upper Fe and MgO layers of the Fe/MgO/Ge_0.895Fe_0.105 sample. For the m_Fe − H curve, we used a square hysteresis with the coercivities of $±$ 24 mT [Fig. 4(a)], which correspond to the jumps of RA in the observed TMR curves [Fig. 2(a)]. From the fitting, the product of P_Fe and P_GeFe, which is used as a fitting parameter, is estimated to be 6.0 $×10−5$⁠. As shown in Fig. 4(b), the fitting curves well reproduce the experimental result, which means that our tunneling transport data purely originate from TMR.

Figure 5 summarizes the x dependence of the TMR ratio. At x = 0.065, the TMR ratio is increased to 0.55% from 0.27% obtained in the previous study⁶ by the reduction of the size of the mesa diode (from 700 $×$ 700 μm² to ϕ = 56.3 μm). Similarly at x = 0.105, the TMR ratio is increased from 1.2% to 1.5% due to the reduction of the size of the mesa (from ϕ = 169 μm to ϕ = 16.7 μm). These improvements are thought to be caused by the reduction of the number of pinholes in the MTJs. Furthermore, we have successfully increased the TMR ratio up to 1.5% by increasing x from 0.065 to 0.105. On the other hand, the TMR ratio decreased when $x>0.105$ (this reason is described later). The result that TMR greatly changed at $x≤$ 0.105 (Fig. 5) excludes the influences of AMR and TAMR of Fe, which do not depend on x, in our TMR curves. The observed MR curves in Fig. 2(a) are not affected by AMR or TAMR of Fe (although AMR and TAMR are present in the Fe layer).

As shown in Figs. 6(a)–6(d), the decrease in the TMR ratio when $x>0.105$ (Fig. 5) can be attributed to the degradation of the crystallinity of the MgO layer with increasing x; the RHEED pattern of the MgO layer becomes broader and darker as x increases. This is probably partly because the lattice constant of GeFe decreases from 0.5648 nm (at x = 0.065) to 0.5640 nm (at x = 0.175)¹¹ and becomes away from the intrinsic lattice constant of MgO (0.4212 $×2=$ 0.5957 nm) and thus the lattice mismatch is larger with increasing x.¹⁰ In addition, atomic force microscope (AFM) measurements for reference GeFe/Ge:B/p⁺Ge samples show that the roughness of the surface of the Ge_1−xFe_x layer tends to be larger with increasing x [Figs. 6(e)–6(h)]. The obtained root mean square (RMS) values of the roughness of the Ge_1−xFe_x layer are 0.23, 0.26, 0.25, and 0.30 nm when x = 0.065, 0.105, 0.140, and 0.175, respectively. Also, the roughness of the Fe top layer of the trilayer sample tends to become larger with increasing x [Figs. 6(i)–6(l)]; the RMS values are 0.27, 0.37, 0.44, and 0.31 nm in ascending order of x. These are the possible origins of the TMR decrease when $x>$ 0.105 (Fig. 5). To solve the problems of the lattice mismatch and the roughness of GeFe, interlayer insertion between the MgO and GeFe may be useful. For example, inserting a thin Ge layer will relieve the lattice mismatch problem and decrease the roughness of the MgO surface. By improving the crystallinity of the MgO layer, it is expected that TMR is largely enhanced in Fe/MgO/GeFe MTJs.

In summery, we have systematically studied TMR in Fe/MgO/Ge_1−xFe_x with various x and successfully increased the TMR ratio from 0.27% to 1.5% by decreasing the mesa size from 700 $×$ 700 μm² to ϕ = 16.7 μm and by increasing x of Ge_1−xFe_x from 0.065 to 0.105. This is the first observation of the TMR ratio over 1% in MTJs containing a group-IV FMS. While the TMR ratio increased when $x≤0.105$⁠, it decreased when $x>0.105$⁠. We have found that the MgO layer becomes rougher with increasing x, which is though to be the cause of the upper limit of the TMR ratio. The quality of the MgO layer on GeFe is an important factor for further improvement of TMR in Fe/MgO/GeFe.