# High-Efficient Spin Injection in GaN at Room Temperature Through A Van der Waals Tunnelling Barrier – Nanoscale Research Letters

#### ByDi Lin, Wenyu Kang, Qipeng Wu, Anke Song, Xuefeng Wu, Guozhen Liu, Jianfeng Wu, Yaping Wu, Xu Li, Zhiming Wu, Duanjun Cai, Jun Yin and Junyong Kang

Aug 15, 2022

H-BN films grown by chemical vapour deposition (see Additional file 1: Fig. S1) were transferred onto the Ga-face GaN as the tunnelling layers. The morphology and structural properties closely associated with the spin injection were characterized. As shown in Fig. 1a, b, the optical micrograph, as well as the AFM image, illustrates a smooth surface of the h-BN tunnelling layer. From the height profile in the inset of Fig. 1b, the thickness is measured as ~ 0.49 nm, corresponding to a monolayer h-BN. Raman spectrum in the illustration of Fig. 1a exhibits a characteristic vibration peak at 1367 cm−1, suggesting a high crystal quality of the h-BN film. Owing to the lack of dangling bond, h-BN provided a clean and stress-free surface for the subsequent growth of spin injection material avoiding the problem of lattice mismatch. For instance, the Fe film displays a nice crystal equality when growing on h-BN/GaN surface, as demonstrated by the cross-sectional TEM image in Fig. 1c. The darker areas in the top and bottom represent the Fe and GaN layers, respectively, which are separated by a brighter BN layer in the middle. Atomically sharp interfaces of the three layers without visible interdiffusion can be clearly identified. The lattice fringes of the Fe film could be recognized from the high-resolution TEM. The interplanar spacings of the crystals are 0.204 nm and 0.167 nm, respectively, corresponding to the (110) and (111) planes of the body-centred cubic α-Fe lattice, and these crystal planes could also be distinguished from the fast Fourier transform (FFT) image in the insert and the inverse fast Fourier transform (IFFT) image in Additional file 1: Fig. S2. Figure 1d depicts the hysteresis loops of the Fe film, where the in-plane and out-of-plane saturation magnetizations Ms are comparable, and the squareness ratio (Mr/Ms) is 0.16 and 0.02, respectively. The in-plane squareness is an order of magnitude higher than that of the out-of-plane, which indicates an in-plane magnetic anisotropy of the Fe film. The smooth and symmetric curves suggest a continuous rotation of the magnetic moment without pinning effect induced by the coarse domains. This also demonstrates a good crystal quality and magnetic property of the Fe film grown on h-BN.

As demonstrated above, h-BN can form a superior sharp and clean interface with both GaN and FM material, which is to the benefit of high-efficient spin injection and transport. Based on the h-BN tunning layer, the three-terminal spin valves were constructed on n-GaN (carrier concentration of 1018 cm−3) to investigate the spin relaxation time and spin diffusion length, as shown in Fig. 2a. During the measurement, a magnetic field is applied in the perpendicular direction of the spin polarization plane. The spin precession at the Larmor frequency will cause spin dephasing, thus making the electrochemical potential difference Δμ decrease to zero with the external magnetic field Bz. By fitting the Hanle curve with Eqs. (1) and (2), the spin relaxation time τsf and spin diffusion length λsf can be obtained [34, 35].

$$Delta mu left( {B_{z} } right) = frac{{Delta mu left( 0 right)}}{{(1 + left( {omega _{1} {tau_{sf}} } right)^{2} )}}$$

(1)

$${lambda }_{sf}={(D{tau }_{sf})}^{1/2}$$

(2)

where Δμ (0) is the electrochemical potential difference under zero magnetic field, ωl = BBz/ħ is the spin precession frequency (g = 1.94 is the Landé factor of GaN) [12], and D is the diffusion constant.

The spin injection is strongly dependent on the ferromagnetic electrodes. To clarify this influence, about 20 nm Fe, Co, and CoFeB injection electrodes with a Ru capping layer were, respectively, fabricated on GaN surface. By detecting the Hanle precession at RT, the performance of devices with the three different ferromagnetic electrodes is compared. Since the spin injection can be current and voltage dependent in a larger bias current, which is especially pronounced for h-BN tunnel barriers[36]. To eliminate this influence, a fixed and relatively smaller bias current of 10 μA is used during the measurement. As shown in Fig. 2b, c, the three-terminal device with Fe (Co) ferromagnetic electrode produces a 65 (52) μV spin accumulation signal. The calculated spin relaxation time and spin diffusion length are 42 (46) ps and 165 (173) nm, respectively, which are significantly increased than the cases of using the Al2O3 tunnelling layer [5, 6]. This is due to the smooth two-dimensional nature of the h-BN film and the sharp van der Waals interface for the Fe (Co)/h-BN/GaN trilayer. With respect to both Fe and Co, the device using CoFeB as a ferromagnetic electrode achieves the expected longest spin relaxation time of 51 ps and the largest spin diffusion length of 182 nm, as shown in Fig. 2d, which can be attributed to its high intrinsic spin polarization and better magnetic property (see Additional file 1: Fig. S3) [13].

Given the superior performance of the three-terminal device with CoFeB ferromagnetic electrode, the GaN-based four-terminal non-local spin valves with CoFeB/h-BN spin injector were further studied to reveal the spin injection efficiency. Furthermore, h-BN has been shown to be a preferred tunning layer, and its thickness was also optimized. The devices were constructed through the EBL followed by a magnetron sputtering of 20/50 nm CoFeB/Ru layer, as shown in Fig. 3a. Figure 3b displays the I-V curves detected on the central two terminals of the devices, and the results using monolayer and bilayer h-BN tunning barrier are compared. Both the I-V curves display clearly nonlinear properties, indicating the presence of a Schottky barrier at the tunnelling interface. The inset depicts the corresponding differential conductivities, where the parabolic shape is typical of tunnel junctions. It is worth noting that the device with a bilayer tunnelling layer shows a lower slope, indicative of a larger tunning barrier and a higher resistance than that with monolayer h-BN. During the magnetoresistance measurement, by applying a constant bias current to the left ferromagnetic electrode and reference electrode, the spin-polarized electrons are injected through h-BN into GaN. Since the middle injection and detection electrodes are designed with different aspect ratios (Fig. 3a), the magnetization orientation of the middle two electrons will be tuned between parallel and antiparallel when an in-plane magnetic field sweeps from + y to − y direction. Accordingly, the magnetic field-dependent voltage signals can be measured. To eliminate the influence from the background resistances, a high-precision constant current source and a lock-in amplifier are employed to collect the data during the measurement of magnetic resistance, to avoid the influence of spurious effects and make sure the accuracy of the results. With a 10 μA injected current (Iinject), the measured voltage changes (ΔV) under various in-plane magnetic fields are provided in Additional file 1: Fig. S4a, b, and the deduced non-local magnetic resistance signals (ΔR = ΔV/Iinject) are shown in Fig. 3c, d for the two devices, respectively. In the case of monolayer h-BN tunnelling barrier, the peak ΔR is about 5 Ω, while the device with bilayer h-BN exhibits a higher peak ΔR of 13 Ω, indicating a superior spin injection and transport performance of the latter.

The spin-dependent magnetoresistance and spin polarization PJ comply the following relationship [11]:

$$mathrm{Delta R}= {P}_{J}^{2} rho frac{{lambda }_{N}}{A}{mathrm{ e}}^{-L/{lambda }_{N}}$$

(3)

where ρ is the resistivity of GaN, λN is the spin diffusion length, A is the cross-sectional area of the transport current in GaN, and L = 600 nm is the distance between the spin injector and detector. Since the GaN thickness is far larger than the carrier diffusion length, we perform a theoretical simulation to reveal the effective cross-sectional area of the transport current (see Additional file 1: Fig. S5). By employing the above parameters (a effective cross-sectional area of 1.2 × 10–12 m2, a resistivity of 0.11 Ω·m for ~ 1018 cm−3 n-GaN measured through the four-probe method), the spin polarization is estimated to be about 9.0% and 14.5% for the two devices, respectively. To our knowledge, the spin polarization value with bilayer h-BN is higher than any previous report for the spin injection in GaN [5, 11,12,13]. There are two possible reasons for the better performance of bilayer h-BN. Firstly, the bilayer h-BN has a higher tunnelling potential barrier compared with the monolayer, which can well overcome the conductivity mismatch between the ferromagnetic electrode and semiconductor. Secondly, owing to the weak van der Waals interaction, the bilayer can better alleviate the roughness of the original GaN, providing a smoother surface for the subsequent growth of the ferromagnetic electrode. As a result, spin depolarization caused by the magnetostatic fringe field of the ferromagnetic electrode can be further reduced [23]. If further increased the thickness of h-BN layer, unavoidable breakage and contamination can be produced during the transfer process, which could induce more spin scattering in the interface and even suppress the spin polarization instead.

In addition to the significant role played by ferromagnetic electrodes and tunnelling barrier, the carrier concentration of GaN is critical to the spin transport. In this consideration, n-GaN with lower carrier concentrations was employed as the substrate, to compare the spin polarization with that of the ~ 1018 cm−3 n-GaN. Based on the four-terminal non-local spin valve device, the non-local ΔR as a function of in-plane magnetic field for CoFeB/ML h-BN/GaN (~ 1017 cm−3) is shown in Fig. 4a (the as-measured magnetic resistance curve is shown in Additional file 1: Fig. S4c), where the peak ΔR is about 9 Ω. By introducing the related parameters for the ~ 1017 cm−3 n-GaN, including the four-probe measured resistivity of 0.57 Ω·m, the effective cross-sectional area of 2 × 10–12 m2 (see Additional file 1: Fig. S5), and the spin diffusion length of 195 nm (derived from the Hanle measurement), the yielding spin polarization is about 5.9%. This value is 3.1% lower than the aforementioned ~ 1018 cm−3 n-GaN with monolayer h-BN tunnelling barrier. If the carrier concentration is further reduced, the spin transport and the polarization are intensively suppressed instead, leading to an undetectable signal (not shown). Normally, a lower carrier concentration has a weaker spin relaxation caused by the doping atoms, which should be beneficial to a higher spin polarization. However, the decreased carrier concentration can also reduce the conductive channels accessible for the spin transport, resulting in a lower spin polarization instead. Figure 4b displays the ΔR signal for the spin injection in free-standing single-crystal n-GaN (~ 1018 cm−3), with the measured ΔV provided in Additional file 1: Fig. S4d. The peak ΔR is increased to be as large as 37 Ω, which is much larger than that in GaN film grown on sapphire surface. This is attributed to the high quality of the free-standing GaN crystal, in which the defect-induced spin scattering is significantly restrained during the spin transport.

Figure 5a summarizes the mechanism of the spin injection and transport dependence on the h-BN tunnelling barrier, and the carrier concentration as well as crystal equality of GaN. The systematic understanding and the maximal spin polarizations obtained at RT in MOCVD GaN film could be considered as a critical move for the practical applications of semiconductor spintronics.