Polycrystalline ferromagnetic (FM) films with an fcc(111) preferred orientation have been used often as bottom electrodes in magnetic tunnel junctions (MTJs). It was previously shown that initial plasma oxidation of a thin Al film progresses through grain boundaries. Inhomogeneity of the insulating layer can be one reason for the decrease of tunnel magnetoresistance (TMR) ratio with increasing bias voltage. This is fatal in development of MRAM with large capacity of Gbit order. In this study, the high quality MTJ with epitaxial FM bottom electrodes was fabricated. It showed TMR ratio of 50.7% after annealing at 250℃ for 1 hr. The applied bias voltage dependence of the TMR ratio was also remarkably improved and the $V_{half}$ was about 750 mV.
Si (111) substrate was cleaned in $H_{2}SO_{4}/H_{2}O_{2}$ solution to remove organic impurities, and rinsed in de-ionized water. Subsequently, It was etched in $NH_{4}F$ solution to remove the native oxide layer and to obtain hydrogen-terminated flat surfaces. The layer structure of a buffer layer is Si (111) substrate/Ag (111)/Cu (111). We used $Ni_{80}Fe_{20}$ as FM bottom electrode. The stacking sequence of the MTJ was Si (111)/epitaxial Ag 3 nm/epitaxial Cu (50 and 100 nm)/epitaxial $Ni_{80}Fe_{20}$ $50 nm/Al-O 1.6 nm/Co_{75}Fe_{25}$ $4 nm/Ir_{22}Mn_{78}$ $20 nm/Ni_{80}Fe_{20}$ 20 nm/Ta 5 nm. All the layers in the junction were prepared using inductively coupled plasma (ICP) assisted magnetron sputtering with base pressure below $8\times10^{-7}$ Pa without breaking vacuum. The insulating barrier was formed by depositing Al, using ICP assisted rf magnetron sputtering and oxidizing by ICP oxidation in Ar/$O_{2}$ gas mixture. The buffer layers of Ag, Cu and NiFe were deposited at RT with rates of about 0.06 nm/s, 0.06 nm/s and 0.03 nm/s, respectively. Junctions were patterned using a micro-fabrication method including photolithography and the area patterned was in the range of $3\times3$ through $100×100\mu m^{2}$.
The θ - 2θ scans of the following layer structure, Si(111)/Ag/Cu/NiFe, shows only the {111} peaks of NiFe and Cu, indicating 111-orientation. The rocking curve of the NiFe peak had a full width at half-maximum (FWHM) of 0.77℃ inferring a very small dispersion. The Φ -scan of {111} planes of NiFe revealed three peaks at the same Φ positions with those of Si, verifying epitaxial growth. The other three peaks were also observed at angles of 180℃ translated from the former three peaks ascribing to the existence of twin epitaxy.
The normalized TMR - V curves measured at RT were obtained from I - V curves measured for anti-parallel (AP) and parallel (P) alignment states of the magnetization of top and bottom electrodes. The $V_{half}$, the bias voltages at which the TMR ratio is reduced to half near the zero bias, was measured about +750 mV and -700 mV, much higher value than those of conventional MTJs.
The surface roughness, $R_{a}$, was determined about near 1.0 nm. Although the interface between the FM bottom electrode and the insulating layer is rather rough in both samples, it did not exert any significant effect to the bias dependence itself except that $V_{half}$ and TMR ratios decreased with a reduction in insulating layer thickness.
The peaks by excitation of magnon and Al-O phonon were observed around ±20 mV and ±120 mV in inelastic electron tunneling spectroscopy (IETS), similar to previous studies of conventional MTJs. However, the result in this study revealed remarkable features. First, the intensity of spectra near the zero bias is small. The peak observed at small bias of several mV is induced by inelastic tunneling due to impurities at the interface and in the tunnel barrier, and clearly separated with the peak of magnon excitation in case of conventional MTJ. On the other hand, the intensity of the peak near the zero bias is smaller compared with magnon excitation peak in this study on epitaxial MTJ, representing that the density of impurities is small. Thus, we could say that the clear interface structure was obtained for epitaxial MTJs, compared with conventional MTJs. Second, the spectrum of P state reveals remarkably small intensity, suggesting that inelastic excitation due to spin scattering is small. Finally, the spectral intensity of positive bias is larger than that of negative bias and a hillock appears around -60 mV, exhibiting asymmetry in the bias voltage. The electronic structure and the spin-dependent density of states (DOS) of the FM electrodes must be responsible for this. The band structure and DOS of FM electrodes should also be considered to explain the bias dependence of TMR. Dynamic conductance measured simultaneously with IETS revealed an obvious asymmetry and, for P state, had a local minimum at about -100 mV and +230 mV. This elastic component must have affected the shape of IETS.
In FM electrodes of conventional MTJs, there existed many high-angle grain boundaries. They served as sites of defects or impurities both at the Al-O/FM interface and the insulating layer. These localized defects in the insulating layer increase the inelastic tunnel process, resulting in a strong bias dependence of TMR according to the two-step tunneling model. On the other hand, the grains in the epitaxial NiFe layer do not have high angle grain boundaries but only twin boundaries. It was also reported that Al grew epitaxially on epitaxial NiFe even if the lattice parameter difference between Al and NiFe is as large as 12%. The absence of high angle grain boundaries in the Al precursor metallic layer and the NiFe layer would lead to better uniformity of the insulating layer and well-defined sharp interface of Al-O/FM. They could reduce the number of trap sites through which spin-independent tunneling occurs in the insulating layer and affect the interfacial DOS, resulting in a small bias dependence of TMR in this epitaxial work in spite of large interface roughness.