1. Introduction

Silicon carbide (SiC) is an attractive semiconductor material for high-quality power devices due to its excellent intrinsic properties such as a high breakdown electrical field, high saturation electron velocity, high thermal conductivity, and chemical inertness^{[1, 2]}. However, these promising properties for power application are largely limited by the quality of the ohmic contact to the substrate. Nickel is the most widely used ohmic contact metal^[3-5]. Though Ni can efficiently react with SiC and demonstrate ohmic behavior after annealing around 1000 ℃, the Ni/SiC interface becomes rough with a large amount of Kirkendall voids^{[6, 7]} and there is a large amount of carbon accumulating on the surface^[8]. Carbon clusters on the surface lead to practical problems in device packaging applications^[9]. Luchowski found that thin NiCr (50 nm) presented better adhesion while thicker NiCr (200 nm) showed poor adhesion due to excess carbon at the surface^[10]. Ohyanagi also reported that the Al-overlaid on Ni-silicide was easily peeled off because of the accumulated carbon^[9].

In order to improve the Ni electrode, some researchers^{[11, 12]} introduced silicon into the Ni system to reduce the reaction of Ni with SiC and the amount of residual carbon. In addition, some other researchers^[13-15] used a Ni-based bi-layer electrode in order to form a stable carbide by reacting carbon with the additional metal. Park and Holloway^[14] introduced a thin Ti bottom layer into the Ni/SiC structure and studied the influence of the Ti thickness as a bottom metal layer. They found that the Ti layer might act as a diffusion barrier, restricting the reaction of Ni and SiC and slightly increasing the specific contact resistivity, however the specific contact resistances does not vary significantly with Ni thickness. Therefore, we proposed a sandwich Ni/Ti/Ni structure which made the Ni as the bottom layer. The multilayer Ni/Ti/Ni ohmic properties with the SiC substrate and the adhesive effect between the Ti/Au overlayer were investigated.

In this paper, we examined the ohmic contacts of the Ni/Ti/Ni sandwich structure on the substrate. We experimentally investigated the effects of Ti in the Ni/Ti/Ni ohmic contact, focusing on its influence on the carbon distribution and the interface quality. The adhesion between the Ni/Ti/Ni ohmic contact and the Ti/Au overlayer was investigated as well.

2. Experiment

The highly doped ($N_{\rm D}$ $=$ 10$^{19}$ cm$^{-3}$, $t_{\rm sub}$ $=$ 350 $\mu$m) n-type 4H-SiC substrate used for sample preparation was purchased from the Tankeblue Semiconductor Company. The resistivity of the substrate is 0.023 $\Omega$·cm. We cut 1 $\times$ 1 cm$^{2}$ square samples from the SiC wafer and used the backside of the substrate to form the ohmic contacts. Before metal deposition, the SiC samples were treated with RCA clean, followed by photolithography to define the electrode patterns. After being rinsed in a 10% hydrofluoric acid solution for 30 s to remove the native oxides and N$_{2}$ blow-dry, the samples were immediately transferred into the vacuum chamber of the electron beam evaporator for contact metal deposition. Then, the ohmic contact patterns were formed by a metal liftoff process. We prepared five samples: four of Ni/Ti/Ni (20/20/100 nm) samples and one of Ni (140 nm) sample for comparison. The Ni/Ti/Ni samples were rapid thermal annealed (RTA) at different temperatures (950, 1000, 1050, and 1100 ℃) for 2 min in ambient nitrogen (N$_{2})$. The Ni sample was rapid thermal annealed at 1000 ℃ for 2 min in nitrogen. We characterized their electrical properties by $I$-$V$ measurement and described the composition by X-ray diffractometer (XRD) and Raman spectroscopy. We characterized the interface using a scanning electron microscope (SEM). The adhesion of the Ni and Ti/Au overlay and the Ni/Ti/Ni and Ti/Au overlay were evaluated by the stud-pull test, using Royce Instruments. In the stud-pull test, the substrates were fixed and the wedge wire-bonding of 25-$\mu$m-diameter Au wire was applied to the Au/Ti/Ni/Ti/Ni/SiC and Au/Ti/Ni/SiC, as shown in Fig. 1. We implemented the test five times at different sample locations. The SEM and EDS were used to identify failure locations by analyzing the surface of the failed electrodes after the wedge wire-bonding test.

3. Results and discussion

After RTA annealing, all Ni/Ti/Ni/SiC samples demonstrate ohmic behaviors, and the lowest contact resistance is achieved at 1050 ℃ as shown in Fig. 2. Taking into account the ohmic contacts made on the thick bulk backside, the modified four-point method was used to measure the specific contact resistivity^[13]. The specific contact resistivity of Ni/Ti/Ni/SiC at 1050 ℃ is 3.16 $\times$ 10$^{-5}$ $\Omega$·cm$^{2}$. It is found that this value is of the same order as the specific contact resistivity of Ni (3.7 $\times$ 10$^{-5}$ $\Omega$·cm$^{2})$ measured in this work for reference.

Then XRD measurements were used to examine the composition of the contacts. In Fig. 3, the X-ray diffraction peaks corresponding to Ni$_{2}$Si are observed for all of the annealed Ni/Ti/Ni samples. The Ni$_{2}$Si comes from the reaction between Ni and SiC during the high temperature annealing treatment.

The Raman scattering measurement was performed at room temperature with a wavelength of 473 nm. According to Fig. 4, there is a strong peak at 777 cm$^{-1}$ for the SiC surface^[16]. For the Ni/SiC sample, there are distinct carbon-related peaks of D-band (1356-1360 cm$^{-1})$, G-band (1584-1593 cm$^{-1})$, and 2D-band, indicating that elementary graphite is formed during the reaction between the Ni and the SiC substrate. The distinct peaks around 100-102 and 137 cm$^{-1}$ indicate the existence of Ni$_{2}$Si, while the peaks at 190-195 and 211-214 cm$^{-1}$ indicate the existence of NiSi^[16]. For the Ni/Ti/Ni/SiC sample, there are no significant peaks related to carbon. Moreover, the TiC peak around 605 cm$^{-1}$ shows that the C reacted with Ti to form TiC. At 1100 ℃, the Ti and C reacted with Si to form Ti$_{3}$SiC$_{2}$^{[17, 18]} as indicated by the 590 and 674 cm$^{-1}$ shift in the Raman spectra. The Raman results indicate that carbon clusters exist on the Ni electrode surface, but not on the Ni/Ti/Ni electrode surface.

We deposited a Ti/Au (20/300 nm) overlayer on two samples using a metal liftoff process: the Ni/SiC sample annealed at 1000 ℃ and the Ni/Ti/Ni/SiC sample annealed at 1050 ℃. Then, an Au wire was bonded on the overlayer (Fig. 1). The bonding tests of the metal contacts were implemented by Royce Instruments and summarized in Table 1. The Ni/Ti/Ni/SiC sample has a higher bonding strength than the Ni/SiC sample, with average strengths of 9.76 gf and 7.52 gf for Ni/Ti/Ni/SiC and Ni/SiC, respectively. For the Ni/Ti/Ni structure, all the metal pads remain attached to the substrate during the test, even the bond wire is broken by the large pull strength. However, for the Ni/SiC structure, the metal pads and Ti/Au-coating are peeled off of the substrate during the bonding strength test, and the metal of the lower contact is exposed. Some of the specific bonding strengths of Ni/SiC are close to those of Ni/Ti/Ni/SiC ($F_{\rm max}$(Ni/SiC) $=$ 9.289 gf and $F_{\rm min}$(Ni/Ti/Ni/SiC) $=$ 9.254 gf), which is attributed to the non-uniform distribution of the carbon on the surface^[19] of Ni/SiC and the C-rich region^[10]. The failure site of the Au/Ti/Ni/SiC sample was investigated via SEM and EDS. Figure 5(a) shows the EDS spectrum from the Au/Ti/Ni/SiC before wire-bonding, and a strong Au peak is observed. Figure 5(b) shows the EDS spectrum of the failed wire-bonding region. Furthermore, the spectrum shows strong peaks of Ni and Si and a weak peak of C. There is no peak of Au and Ti in Fig. 5(b), indicating the delamination of the Au/Ti overlayer at the Au/Ti/Ni/SiC interface. The carbon at the interface of Au/Ti/Ni/SiC is regarded as the main cause of the poor adhesion^[10]. Clearly, the Ni/Ti/Ni/SiC ohmic contact structure is preferable to Ni/SiC for device packaging applications considering their different adhesion strengths.

Compared to the Ni/SiC structure, the Ni/Ti/Ni structure shows an improved interfacial roughness as shown in the cross-sectional SEM images (Fig. 6). The interface of the Ni/Ti/Ni structure becomes smoother by Ti participating. As shown in Fig. 6(a), a uniform alloy layer is formed and there are only a few voids at the interface for the Ni/Ti/Ni structure. While for the Ni/SiC structure in Fig. 6(b), the interface is rough with a large amount of voids. The intermediate Ti layer may depress the fierce reaction between Ni and SiC to improve the interfacial roughness and greatly reduce the number of voids. The microstructure characteristics reveal the reliability of contact^[20], showing that the Ni/Ti/Ni/SiC structure might be more stable than Ni/SiC.

During the process of the formation of Ni/SiC ohmic contact, Si diffuses into the Ni layer and reacts with Ni to form Ni$_{2}$Si, leaving elemental C atoms at the interface. The C atoms diffuse through the Ni-silicide layer to the surface, leaving a number of C vacancies at the interface. On the one hand, the C vacancies act as donors, leading to an increased electron concentration under the contact, decreasing the width of the depletion layer and the barrier height simultaneously, which facilitate electron transport along the vertical direction^[21]. On the other hand, the carbon diffuses through Ni-silicide layer and becomes graphitic carbon and accumulates at the reaction layer and on the surface, which causes poor adhesion for packaging^{[10, 11]}. The graphitic carbon, which is the byproduct of Ni/SiC during high temperature annealing, does not affect the specific contact resistivity^[4]. Low specific contact resistivity could still be achieved with no graphitic carbon by using NiSi/SiC and Ni$_{2}$Si/SiC structure^{[11, 22]}. In our presented Ni/Ti/Ni/SiC system, the formation of ohmic contact is similar to that of Ni/SiC, and the specific contact resistivity of Ni/Ti/Ni/SiC is of the same order as the Ni/SiC specific contact resistivity. Besides, the intermediate titanium layer efficiently prevents the carbon from accumulating on the surface by forming TiC compounds. With the carbon-free surface of the Ni/Ti/Ni/SiC structure, the Ti-carbide and Ni-silicide at the surface can efficiently improve the adhesion strength of the Ti/Au overlayer^[9]. Therefore, for good mechanical reliability during the device packaging process, it is important to eliminate the carbon clusters by forming carbide. The graphitic carbon can be effectively consumed by the Ti buffer layer. Although introducing a Ti buffer layer may bring a little additional cost, it is still very cost-effective because of the great improvement of the interface and adhesion. So the Ni/Ti/Ni structure is suitable for forming electrodes without sacrificing contact resistivity.

4. Conclusion

We investigated the ohmic contact of Ni/Ti/Ni to n-type 4H-SiC using XRD, SEM and Raman spectroscopy. There is no carbon observed for all the Ni/Ti/Ni samples, and the bonding strength to the overlayer is greatly improved. The specific contact resistivity is 3.16 $\times$ 10$^{-5}$ $\Omega$·cm$^{2}$ for Ni/Ti/Ni/SiC annealed at 1050 ℃, which is of the same order as Ni/SiC (3.7 $\times$ 10$^{-5}$ $\Omega$·cm$^{2})$. These imply that the bottom Ni layer reacts with SiC and effectively forms an ohmic contact, and the introduction of an intermediate Ti layer is helpful to improve the interface and eliminate carbon on the surface. The Ni/Ti/Ni/SiC structure is preferential to Ni/SiC for SiC power device applications.