Improved thermal tolerance of silicon-based lateral spin valves

Improvement in the thermal tolerance of Si-based spin devices is realized by employing thermally stable nonmagnetic (NM) electrodes. For Au/Ta/Al electrodes, intermixing between Al atoms and Au atoms occurs at approximately 300°C, resulting in the formation of a Au/Si interface. The Au-Si liquid phase is formed and diffuses mainly along an in-plane direction between the Si and AlN capping layers, eventually breaking the MgO layer of the ferromagnetic (FM) metal/MgO electrodes, which is located 7  m away from the NM electrodes. By changing the layer structure of the NM electrode from Au/Ta/Al to Au/Ta, the thermal tolerance is clearly enhanced. Clear spin transport signals are obtained even after annealing at 400°C. To investigate the effects of Mg insertion in FM electrodes on thermal tolerance, we also compare the thermal tolerance among Fe/Co/MgO, Fe/Co/Mg/MgO and Fe/Co/MgO/Mg contacts. Although a highly efficient spin injection has been reported by insertion of a thin Mg layer below or above the MgO layer, these thermal tolerances decrease obviously.


INTRODUCTION
Silicon (Si) is a suitable material for semiconductor-based spintronic devices because of its long spin lifetime 1,2 , excellent controllability of spin current by gate voltage 3,4 , and availability of mature infrastructures of Si-based electronic devices. Significant advances, including room-temperature demonstrations of Si-based spin devices such as a spin metal-oxide semiconductor field-effect transistor [4][5][6][7][8][9] and a magneto logic gate 10,11 , have been achieved to date by using a magnesium oxide (MgO) tunnel barrier. Recent studies have clarified the spin scattering mechanism in Si based on Elliott-Yafet mechanisms 2,12,13 , and optical phonon emissions suppress the spin diffusion length under high electric fields with applied gate voltage 3 . A remaining problem for the practical application of spin devices is a small signal amplitude in transistor and logic operations due to an insufficient spin accumulation voltage in Si. Thus, exploring a ferromagnetic contact enabling a large spin accumulation voltage in Si is a significant research issue, and half-metallic materials [14][15][16] and the insertion of a thin magnesium (Mg) layer 9,17-20 in a ferromagnetic contact in Si spin devices have been extensively investigated. Among these methods, the insertion of a thin Mg layer above/below the MgO layer is a new approach to enhancing spin signals in spin devices. Indeed, the insertion of the Mg layer above the MgO layer allows for the enhancement of the spin polarization of Fe on Mg/MgO 9,17,18 due to suppression of the formation of a magnetic dead layer, and the insertion of Mg below the MgO layer provides a highly textured MgO tunnelling barrier on the Si channel, which enhances the magnitude of spin signals 20 .
Recently, we established and reported another way to increase spin signals in nondegenerate Si-based spin devices, which involves thermal annealing the Si spin devices at 300°C, and a twofold increase in spin accumulation voltage was realized 21 . This enhancement was attributed to improved crystallinity in the MgO tunnelling barrier and better spin polarization. However, no spin accumulation signals were detected in that study after annealing above 300℃ 21 , which presents a new problem for practical applications. The thermal tolerance of Si spin devices at least above 350°C is strongly desired to obtain compatibility with the infrastructures of Si-based electronic devices since the post fabrication process often employs thermal treatments above 300°C [22][23][24] . Therefore, it is significant to clarify the degradation mechanism of Si spin devices by thermal annealing and to solve the problem of further progress in Si spintronics.
Here, we have investigated the thermal degradation mechanisms of a lateral spin valve (LSV) based on nondegenerate n-type Si by electric and crystallographic methods. It is revealed that the in-plane long-range diffusion of gold (Au) atoms in nonmagnetic (NM) electrodes occurs through the aluminium nitride (AlN)/Si interface. The Au atoms eventually reach a ferromagnetic (FM) electrode that is located 7 m away from the NM electrodes, resulting in the degradation of the MgO tunnelling barrier. To improve the thermal tolerance, the layer structure of the NM electrode is changed from Au/Ta/Al to Au/Ta. The modified device clearly exhibits spin signals even after annealing at 400°C. We also focus on the thermal tolerance of different structures in FM contacts without Mg insertion and Mg insertion below/above the MgO layer, as enhancement of the spin polarization has been reported by insertion of a thin Mg layer 17,18,20 . However, the thermal tolerance obviously decreases down to approximately 300°C due to the existence of the Mg layer.

RESULTS AND DISCUSSION
Schematics of the device geometry and layer structure of the Si-based LSVs are shown in Fig. 1a. The devices consist of n-type Si, two FM electrodes (F1 and F2), two NM electrodes (N1 and N2) and an AlN capping layer on the Si channel. The details of sample preparation are described in the Methods section. The spin transport properties, such as the spin polarization and the spin diffusion length, were measured by nonlocal four-terminal (NL4T) and local three-terminal (L3T) measurements at room temperature. Crystallographic analyses of the devices were carried out by means of cross-sectional transmission electron microscopy (TEM) with energy dispersive X-ray spectroscopy (EDS).
First, sample A was employed to investigate the thermal tolerance and degradation mechanism of the spin transport properties. The spin signals were measured by means of NL4T measurements, and the I-V characteristics of F1 and F2 are measured at 300 K. Spin accumulation signals, i.e., a nonlocal voltage Vnl, as a function of By before and after thermal annealing, are shown in Fig. 1b. A clear rectangular shape was measured, indicating successful spin transport. The magnitude of the spin accumulation signals was increased after annealing at 300℃, consistent with our previous report 21 . In contrast, the spin accumulation signal completely disappeared after annealing at 350℃. To investigate the electrical properties of FM1 and FM2, three terminal I-V characteristics are measured. Figures 1c and   1d exhibit the three terminal I-V characteristics of F1 and F2 before/after annealing. Whereas the I-V characteristics did not change after annealing at 300℃, they decreased significantly after annealing at 350℃, which was consistent with our previous study 21 . The reduction in the spin signal is attributable to the reduction in the interfacial resistance from 1.5×10 -8 to 1.6×10 -9 m 2 in F1 and from 1.5×10 -8 to 4.5×10 -9 m 2 in F2 due to the conductance mismatch 21,25 .
The one-dimensional spin diffusion model (see methods) also revealed that the spin polarization decreases from approximately 2.8% to less than 1.3% after annealing at 350°C.

Figures 2a and 2b show optical microscopy images of sample A before annealing and after annealing at 350℃.
Although a smooth gold surface was confirmed at the NM electrodes before annealing, a patchy pattern in gold, grey and blue was apparent after annealing. In contrast, no significant changes were observed on the surface of the FM electrodes. To reveal the degradation mechanism in detail, crystallographic analyses using TEM and EDS were carried out. A cross-sectional elemental mapping images obtained by EDS are shown in Figs. 2e, 2g and2h. We the Si channel itself. Therefore, the Si channel is not a probable diffusion path for the Au atoms.
We then clarify the possible mechanism of Au atom diffusion by annealing. The Al and Si binary systems exhibit eutectic alloys, but the temperature of the eutectic point is above 570°C, indicating that the Al-Si liquid phase is not formed at the Al/Si interface during the annealing process. Similarly, Ta-Si, Ta-Al and Au-Al systems are expected to maintain their solid states below 400°C. In contrast, the temperature of the eutectic point of the Au-Si system is approximately 350°C. Considering the above situations, a possible mechanism of Au atom diffusion is as follows: The designed structure is shown in Fig. 3e. First, the intermixing of Al and Au atoms takes place below 300℃, as schematically shown in Fig. 3f, which is confirmed by EDS (Figs. 2g and 2h) despite the 5-nm Ta layer. The interdiffusion length l of Au and Al atoms, estimated from = 5.56 × 10 −10 m at 350℃ 26 , is 2 µm after t = 1 hour.
Therefore, Au atoms easily penetrate the Al layer, which is only 40 nm in thickness, and finally reach the Si channel.
Since the melting point of Au8Si2 is approximately 350°C, Au-Si and/or Au-Al-Si liquid phases are formed at the Au-Si interface (Fig. 3g). Because Au8Si2 is the eutectic point, additional Au and Si atoms rapidly increase the melting point, so most of the Si channel remains without reaction. The Au-Si and/or Au-Al-Si liquid phases diffuse mainly along the AlN/Si interface (Fig. 3h) due to the strong tensile strain 27 , and the interface is more susceptible to destruction than the other areas. Finally, Au reaches the FM electrodes and invades the MgO tunnelling barrier.
To verify the hypothesis of the thermal degradation mechanism, a control experiment was performed using sample B. To circumvent the interdiffusion of Al and Au, we excluded Al from the NM electrodes and employed a thick Ta layer: Au(150 nm)/Ta(40 nm). Ta is often used as a barrier metal for Si against conductive metals such as Cu because the Ta layer is maintained even after annealing at 630℃, which is high enough for any thermal treatments applied to electronic devices 28 . Thermal annealing at 300℃, 350℃, and 400℃ was carried out. The optical microscope image of sample B after annealing at 400℃ showed no salient changes on the surface of the Au/Ta electrode (see Figs. 4a and 4b), which was in contrast to the Au/Ta/Al structure in sample A. Cross-sectional elemental mapping images (see Fig. 4c indicating the examined parts) showed no signal from Au atoms detected at the AlN/Si interface or around FM even after annealing at 400°C (see Fig. 4d). Furthermore, the NM electrode clearly maintains the Ta/Au bilayer structure (see Figs. 4e and 4f). These results are obviously different from those of sample A, which directly supports our hypothesis of the thermal degradation of a Si-based LSV.
To confirm the thermal tolerance of the LSV, NL4T measurements were carried out. Clear hysteresis features were obtained for all conditions, i.e., before annealing and after annealing at 300, 350 and 400°C, as shown in Fig. 5a. The magnitude of the spin accumulation signal before annealing was 9 µV, and the signal was enhanced to 18 µV after annealing at 300℃, similar to that of sample A. In contrast to sample A, the spin signal did not decrease after annealing at 350℃. Furthermore, clear spin accumulation signals were also obtained even after annealing at 400°C.
To corroborate the tolerance of the spin diffusion length, Hanle signals were also measured in the NL4T method for sample B after annealing at 300 and 350℃, as shown in Fig. 5b. To eliminate the spurious effects, the difference in Vnl between the parallel and antiparallel magnetic configurations, nl P − nl AP , was employed for the analyses. The spin diffusion length of the Si channel  was estimated from the curve fitting by using the following equation 13 : where Bz is the magnetic flux density along the z-direction, S0 is a constant value that determines the magnitude of the signal, D is the diffusion constant,  = gBBz/ℏ is the Larmor frequency, g = 2 is the g factor for the electrons, B is the Bohr magneton, ℏ is the Dirac constant and  is the spin life time. The fitting curves obtained using Eq. (1) reproduced the experimental results well, as shown in Fig. 5b Fig. 5a, the spin accumulation signals were strongly suppressed for sample C (Fig. 6a) and not detected for sample D (Fig. 6c) after annealing. The I-V characteristics of their ferromagnetic electrodes are shown in Fig. 6b and 6d. The resistance decreased after annealing, which was consistent with the results observed in samples A and B. More importantly, the resistance decreased at much lower temperatures. Although some improvements in spin polarization can be reported, the thermal tolerance worsens with the insertion of a Mg layer both below and above the MgO.

CONCLUSION
In conclusion, we investigated the degradation mechanism of nondegenerate Si-based LSVs by thermal annealing and found two different mechanisms: Au diffusion from the NM electrode to the FM electrode and degradation of the Mg layer employed for FM electrodes. In the sample using Au/Ta/Al electrodes, interdiffusion of Au and Al atoms occurred, followed by the melting of Au and Si alloy at approximately 350°C due to the eutectic alloy properties. The Au-Si liquid phase mainly diffused through the AlN/Si interface and finally broke the FM electrodes. By changing the structure from Au/Ta/Al to Au/Ta with a 40-nm-thick Ta layer, the thermal tolerance was improved, and clear spin signals were obtained even after annealing at 400°C. The insertion of a thin Mg layer below or above the MgO layer was also examined, and their thermal tolerance was less than that without the Mg layer. The maximum thermal tolerance of 400℃, which was sufficiently high for the post fabrication process of electronic devices was provided by the combination of Au/Ta for NM and Fe/Co/MgO for FM electrodes. The thermally tolerant Si-based LSV certified the greater compatibility of Si-based spin devices with the electronics industry.
Phosphorus (P) atoms were doped into the Si layer by the ion implantation technique to form the n-Si and n + -Si (only for sample A) layers. Then, rapid thermal annealing was carried out for their activation. The dopant concentration in the Si channel was confirmed by secondary-ion mass spectrometry, which showed a small distribution perpendicular to the plane in the range of 1×10 17 to 2×10 18

Thermal annealing
The temperature was increased to the set value within 30 minutes, kept for 60 minutes and decreased naturally for 300 minutes using a Futek K-A102HP furnace. All the annealing processes were conducted in a vacuum (~1×10 -4 Pa).

Non-local four terminal (NL4T) method
A schematic of the current-voltage configuration is shown in the insets of Figs. 1b and 5a. A direct current was applied from N2 to F2. Voltage was measured at F1 and N1 under the magnetic field swept along the y-axis in Fig.  1a. Spin polarization β was estimated by applying the one-dimensional spin drift-diffusion equation 29,30 to spin signal ΔVnl. In this model, ΔVnl is expressed as: Where J is the charge current density, Si is the spin diffusion length of Si,  is the conductivity of Si, RA1 and RA2 are the resistance area products of F1 and F2, N = Si is the spin conductance of Si and i1(2) = 1 1(2) is the spin conductance of the tunnel barrier of F1(F2). We assumed identical  values for F1 and F2 and negligible spin resistance of the ferromagnetic metal. The experimentally measured values of RA1, RA2, and  were used for the analysis.

Local three terminal (L3T) method
A schematic of the current-voltage configuration is shown in the insets of Figs. 6a and 6c. A direct current of 1.0 mA was applied from F2 to F1. The external magnetic field was swept along the y-axis, and the voltage difference was measured at F2 and N2.