Atomic-resolved structural and electric field analysis of the passivation interface of MIS-HEMTs Open Access

GaN-based metal–insulator–semiconductor high electron mobility transistors (MIS-HEMTs) have excellent physical and electrical properties. It partly depends on the characteristics of GaN, which has a wide bandgap (∼3.39 eV), a high critical breakdown electric field (∼3.5 MV/cm), a fast saturation drift speed (∼2.5 × 10⁷ cm²/s), good thermal conductivity (1.3 W/K cm), and high electron mobility (∼1500 cm²/V/s). These characteristics make GaN-based MIS-HEMTs have outstanding advantages in high-frequency, high-power, high-efficiency power devices, and radio frequency devices.^1,2 However, there are serious reliability problems existing in GaN-based MIS-HEMTs devices, such as gate degradation, surface charge trapping, and hot electron effect.^3,4 These reliability issues are related to the interface quality, which means that the structural disorder of the interface would cause the high interface density of states,^5,6 and the formation of a physical virtual gate, reverse breakdown, and threshold fluctuations.^7,8

The flatness and crystallinity of the passivation interface are dominated by the passivation layer preparation methods and passivation layer materials (Si₃N₄, ZrO₂, Al₂O₃, AlN, etc).^9,10 In order to minimize the damage to the surface during plasma enhanced chemical vapor deposition (PECVD), Zhang et al. inserted 3 nm Al before PECVD passivation; then, the gate leakage current was reduced by three orders of magnitude.¹¹ Atomic layer deposition (ALD) can be used to improve passivation interface crystallization. In particular, hollow cathode plasma enhanced atomic layer deposition Si₃N₄ can form 1.5 nm β-Si₃N₄ at the interface.¹² It was also found that before LPCVD, Si₃N₄ was also be pre-deposited by metal-organic chemical vapor deposition (MOCVD) to prevent LPCVD high temperature damage.¹³ Some passivation materials, such as AlN, ZrO_2, and SiO₂, are studied to improve the reliability issues. For example, the AlN layer was inserted under the Si₃N₄ passivation layer to improve response speed and reduce current collapse.^14,15 Cai et al. found that high dielectric constant (high-k) materials can increase the breakdown voltage by reducing the gate edge electric field.¹⁶ Deep level transient spectrum (DLTS), I–V, and C–V curves were used to study the performance of the GaN-HEMTs. It was found that reducing the interface state density can inhibit current collapse and increase the breakdown voltage. Furthermore, the key factor to reduce the interface state is improving the crystallinity and roughness of the interface and optimizing the passivation material.^17,18

However, the interface structure and its influence on the electrical properties of the interface are insufficiently studied. The current research studies only focus on the characterization of interface flatness and crystallinity, while the underlying physical mechanism is still unclear. In addition, it is difficult to directly correlate macroscopic electrical properties with microscopic interface structure. Differential phase contrast in a scanning transmission electron microscope (STEM DPC) is a widely employed technique for probing electric fields on the nanoscale or at the atomic resolution scale.^19,20

In this work, with the help of scanning transmission electron microscope (STEM)-high-angle annular dark-field (HAADF) and differential phase contrast (DPC) imaging technologies, we have explored the influence of different passivation materials (Si₃N₄, ZrO₂) and different crystal orientations on the atomic structure, electric field, and electric potential distribution at the passivation interface within the atomic resolution scale.

Figure 1 shows the schematic cross-sectional view of samples A and B. The investigated structure consisted of a 2 nm thick GaN cap layer, a 22 nm undoped Al_0.25Ga_0.75N barrier, a 330 nm GaN channel layer, and a 5.1 µm GaN buffer on a Si substrate.¹⁶ A 120 nm Si₃N₄ was deposited by PECVD at 350 °C as the passivation for sample A [Fig. 1(a)]. A ZrO₂/Si₃N₄ bilayer passivation with thicknesses of 22/120 nm was deposited for sample B [Fig. 1(b)]. The ZrO₂ layer was grown by ALD at 200 °C.

All TEM samples were prepared by a focused ion beam (FIB) microscope (Helios 5UX). The atomic structure and potential of the interface were characterized by a spherical aberration corrected scanning transmission electron microscope (Themis Z). Specifically, the structural information of the passivation/GaN interface is mainly obtained by the high angle annular dark-field (HAADF) imaging. The interface electric field information is obtained by the differential phase contrast (DPC) imaging in a scanning transmission electron microscope, and the interface electric potential characterization is obtained by integrated differential phase contrast (iDPC).

The typical STEM DPC setup is depicted in Fig. 2. Accordingly, the object is illuminated with a sharply focused electron beam, which is scanned over the object to record one DPC signal per scanning point.²¹ As illustrated by Fig. 2(a), the four detectors A, B, C, and D collect equal signal intensities with the electron beam passing through the vacuum. However, the electric field to the left causes the electron beam to be deflected to the right. As shown in Fig. 2(b), the signal amount of the A detector is larger than that of the C detector. The electric field distribution inside the sample can be calculated from the difference in signal intensity collected by different detectors. The following equation gives the relationship between the signal intensity collected by different detectors and the internal electric field of the sample:²² 

In this equation, $E⃗$ is the internal electric field of the sample, and I_A−C and I_D−B are the signal intensity differences between different detectors. $x̂1x̂2$ are the two mutually perpendicular coordinate axes of the detector surface. I_Sum and t, respectively, stand for the integrated intensity of four detectors to offset the influence of local absorption conditions and the thickness of the sample. α describes the calibration factor (function) between the signal gap and the electric field of different detectors. According to Eq. (1), in a uniform region of the sample thickness, the electric field can be obtained by the difference of the signal intensity of different detectors. The potential distribution is obtained by the integration of internal electric field $E⃗$⁠.

STEM has been carried out to characterize the atomic structure of the passivation interface as shown in Fig. 3. The highest contrast is the ZrO₂ layer and the darkest is the Si₃N₄ layer, due to HAADF image contrast is proportional to the square of the atomic number Z. To investigate the effect of two different passivation layers on the interface, the interface width and roughness are first discussed. Fig. 3(a) shows that the Si₃N₄/GaN interface has a width of only ∼2 atomic layers. However, the interface exhibits atomic disorder fluctuation, and the GaN surface is discontinuous at the depth of 1–2 atomic layers, which is caused by plasma damage of PECVD.¹¹ After the insertion of ZrO₂ by the ALD method, it has an atomically flat interface due to ALD with precise control. It is worth noting in Fig. 3(b) that the ZrO₂/GaN interface is wide (1–2 nm), but the ZrO₂/GaN interface has high flatness. Furthermore, it is observed that the ZrO₂ passivation layer has different orientations [(1) and (2) in Fig. 3(b)],²³ and the corresponding interface of the region (1) is found to have the best interface integrity. As mentioned in Ref. 16, the interface state density of the ZrO₂/GaN interface is lower than that of the Si₃N₄/GaN interface, and the density of states of ZrO₂/GaN varies by five times at different energy level depths. It is concluded that although the performance of the passivation interface has been greatly improved after the insertion of the ZrO₂ layer, the interface state density has great fluctuations. This fluctuation may be caused by the structural discontinuity between different materials at the interface and different crystal orientations of ZrO₂ [see Fig. 3(b)]. This is consistent with the U-shaped continuous distribution of the interface state density at the passivation interface.⁵ 

To investigate the reason for the reduction of interface states, the chemical composition change of the interface was first characterized by EDX mapping (see Fig. 4). The distributions of different elements are shown in Figs. 4(b)–4(e). In addition, the profile of the element distribution [along the red arrows in Fig. 4(a)] is shown in Fig. 4(f). It is indicated that the Ga element diffuses into the ZrO₂ region about 2 nm, as shown in Fig. 4(f). In addition, the Zr/O element ratios appear to gradually increase with distance from the interface and vary from 1/3 to 1/2 about 2.6 nm away from the interface. In other words, the O and Ga elements are enriched at the ZrO₂/GaN interface relative to the other elements.²⁴ 

To analyze the structure of the passivation interface in greater detail, the positions of A and B in Fig. 3 are enlarged, respectively, in Fig. 5. Figures 5(a) and 5(b) show the HAADF images of the Si₃N₄/GaN interface. It can be seen from Fig. 5(b) that the atoms on the surface are blurred, indicating that the Si₃N₄/GaN interface has poor order. It is known that the compressive stress created by the Si₃N₄ passivation layer enhances the interface disorder.²⁶ During the process of growing Si₃N₄ by PECVD, the surface is also damaged by high-energy ions.^11,25 Combined with the TEM results in Fig. 5(b), it is found that only the GaN surface atoms have a significant disorder. Therefore, it is speculated that the disorder of the interface structure comes from the combined effect of the damage of the PECVD growth process and the compressive stress of Si₃N₄, with the damage of PECVD playing a major role.¹⁶ Figure 5(c) is a magnification of the interface at position B (area 1) in Fig. 3(b). According to the result of the enrichment of Ga element and O element at the interface obtained in Fig. 4, it can be speculated that there is a compound composed of Ga and O at the interface. This is consistent with literature reports that the oxide passivation of GaN surface will form Ga₂O₃ at the interface. It is further found that ε-Ga₂O₃ exists regionally at the ZrO₂/GaN interface, which can improve the order of the ZrO₂/GaN interface.²⁴ As shown in Fig. 5(d), we constructed Ga₂O₃ models of different phases for comparison, and it is concluded that the interface structure in Fig. 5(d) is consistent with ε-Ga₂O₃. However, it is observed in Fig. 3(b) that the interface crystallinity at other areas is poor compared to the interface corresponding to the (1) orientation. It can be further analyzed that different orientations of ZrO₂ can influence the structure of the interface Ga₂O₃. This may cause the interface state density to change under different crystal orientations.

To study the electrostatic potential distribution near the passivation/GaN interface, Fig. 6 shows the integral differential phase difference (iDPC) image near the interface after passivation with different materials. The contrast in the iDPC image corresponds to the potential level.^19,27 Figure 6(a) shows the iDPC image of the Si₃N₄/GaN interface, where the Ga columns show the highest brightness. Figure 6(b) shows contrast profiles of areas A and B, and the mean intensity of Ga columns (area A or B) is decreasing significantly at the interface. According to Eq. (1), the potential profiles of areas A and B are basically the same, and the potential peak value of the surface atomic layer at the interface decreases significantly.

The results of the potential study near the ZrO₂/GaN interface are shown in Fig. 6(c). The C and D regions, respectively, represent different orientations of ZrO₂. It is interesting that both Ga black contrast areas of 1–2 nm, which means that there is a low potential region near the interface. A more detailed potential distribution was obtained from Fig. 6(d). The GaN potential decreases continuously along the interface and reaches the lowest potential at 1.1–1.4 nm from the interface. As a result, there is such a low-potential area near the interface, which can restrain the movement of electrons to the interface. Additionally, this reduces the probability of electrons being trapped by the interface. In addition, this interfacial potential redistribution is partly due to the formation of 1–2 nm Ga₂O₃.²⁸ 

The potential distribution around the ZrO₂/GaN interface in the C and D regions after ZrO₂ passivation is also compared in Fig. 6(d). Under different ZrO₂ orientations, the potential distribution differs within 1 nm from the surface in the GaN region, but the potential distribution inside the GaN is basically the same. The results show that when the crystal orientation of the ZrO₂ passivation layer changes, the potential distribution near the GaN interface will also be different, and it may be one of the reasons for the fluctuation of the interface density of states.

To complement the discussion on the electric field distribution of the interface under different orientations of the ZrO₂ passivation layer, the DPC technology and Avizo software were used, and the results are shown in Fig. 7. Positions 1 and 2 in Fig. 7(a) represent two different orientations of ZrO₂. Figure 7(b) shows the iDPC image of the same area. Figure 7(c) is a differential phase contrast (DPC) image, in which the color wheel in the upper right corner represents different directions, the saturation of the color represents the intensity of the electric field, and the direction of the electric field points from the center to the outside of the circle. In Fig. 7(c), it is interesting that there is a strong and periodic electric field in the passivation layer in position 1, while the electric field in position 2 is weaker and not periodic. It is noted that the interface of position 2 and GaN has significantly stronger electric fields than that of position 1.

In addition, the Avizo software in Fig. 7(e) is used to calculate the electric field information obtained at this position. The arrow and color in Fig. 7(e), respectively, stand for direction and the intensity of the electric field. Figure 7(f) is the enlarged view in the red box in Fig. 7(e). There is no obvious difference between the intensity and the direction of the electric field in the GaN body, but there is a strong electric field at one or two atomic layers of the interface in the position 2 region, as shown in the white dashed box in Fig. 7(f). It agrees with the results in Fig. 7(c). This fully illustrates that the different crystal orientations of ZrO₂ will not only influence the Ga₂O₃ structure at the interface but also affect the electric field distribution at the interface.

In summary, the structure and electric field information of the Si₃N₄/GaN and ZrO₂/GaN interfaces were investigated by STEM. We found that the flatness and crystallinity of the passivation/GaN interface were improved after inserting a layer of ZrO₂ before Si₃N₄ passivation; in particular, forming ε-Ga₂O₃ at the interface can improve the interface quality. It was also found that a low potential region of 1–2 nm was formed near the inner interface of GaN, which may be able to prevent electrons from moving to the interface. Further research shows that the different orientations of the ZrO₂ passivation layer will influence the Ga₂O₃ structure at the interface and the electric field of 1–2 atomic layers at the GaN interface. Both of them will affect the interface density of states. Therefore, it is necessary to pay attention to the crystal orientation of the passivation layer when considering the interface state density during passivation.

This work was supported by the National Key R&D Program of China (Grant No. 2017YFB0404100), the National Natural Science Foundation of China (Grant Nos. 62104247 and 62104245), and the 2020 Key R&D Program—Industry Foresight and Key Core Technologies of Jiangsu Province (Grant No. BE2020004-1).

The authors have no conflicts to disclose.

The data that support the findings of this study are available from the corresponding author upon reasonable request.