Interlayer Investigations of GaN Heterostructures Integrated into Silicon Substrates by Surface Activated Bonding

Abstract

Thinning the buffer layer thickness between the GaN epilayer and Si substrate without introducing large residual stress is persistently desired for GaN-on-Si devices to promote their thermal budgets and low-cost, multifunctional applications. In this work, the GaN-on-Si heterostructures were directly bonded at room temperature by surface activated bonding (SAB) and the therein residual stress states were investigated by confocal micro-Raman. The effects of thermal annealing process on the residual stress and interfacial microstructure in SAB fabricated GaN-on-Si heterostructures were also systematically investigated by in situ micro-Raman and transmission electron microscopy. It was found that a significant relaxation and a more uniform stress distribution was obtained in SAB bonded GaN-on-Si heterostructure in comparison with that of MOCVD grown sample; however, with increasing annealing temperature, the residual stresses at the SAB bonded GaN layer and Si layer evolute monotonically in different trends. The main reason can be ascribed to the amorphous layer formed at the bonding interface, which played a critical stress relaxation role and transformed into a much thinner crystallized interlayer without any observable structural defects after 1000 °C annealing.

1. Introduction

Gallium nitride (GaN) is one of the most potential materials for next generation power and RF devices due to its advantageous unique physical properties such as wide bandgap, high electron saturation velocity, and high breakdown electric field [1,2,3,4,5,6]. GaN epitaxial layers have been integrated to various substrates such as sapphire, SiC, Si and diamond substrates through metal–organic chemical vapor deposition (MOCVD), molecular beam epitaxy, or microwave plasma chemical vapor deposition using various buffer layers [7,8,9,10,11,12]. Among them, thick buffer layers such as AlN, GaN, or AlGaN enable high-quality GaN layers grown on the Si substrates to alleviate the large mismatch in the lattice constant between the GaN epitaxial layer and the Si substrate (~17%) [9,12,13,14,15,16,17,18]. However, despite such lattice mismatch alleviation through thick buffer approaches, on one hand, the large residual stress generated in the epitaxial layer is inevitable due to the difference in the thermal expansion coefficients. On the other hand, using thick buffer layers would unavoidably result in a huge thermal boundary resistance at the interfaces, thus degrade the critical thermal performance of GaN-based devices [12].

For the sake of obtaining low residual stress and high quality ultrathin GaN/Si heterointerfaces, bonding GaN with Si using various thin buffer layers has attracted lots of interests and has achieved great progresses [19,20,21]; however, directly bonding GaN to Si substrate without using any buffer layer is still unrealized. Recently, the direct bonding of dissimilar materials has been achieved by surface activated bonding (SAB) technique at room temperature, in which the surfaces of the bonding materials are activated by Ar fast atom beam prior to bonding [22,23,24,25]. SAB can even enable bonding of dissimilar materials with a large mismatch in the lattice constant and the thermal expansion coefficients without using any buffer layer. It has been employed to form a variety of heterostructures such as GaAs/Si, 4H-SiC/Si, InP/Si, GaAs/GaN, and Diamond/Si [24,26,27,28,29]. A good thermal stability has been revealed in the 4H-SiC/Si and GaAs/Si heterostructures even at a high temperature near the melting point of Si and GaAs [24,26]. Since large residual stress could affect the electrical and optical properties of GaN-based devices in different manners [30,31], the residual stress behaviors in the SAB fabricated heterostructures need to be measured and fully understood. In addition, thermal annealing was found can modulate the defects, modify the interfacial structure and change the residual stress level of different semiconductor crystals and heterostructures [22,23,32,33,34], yet in situ monitoring and measuring the changes of such localized interlayer microstructure and residual stress during thermal annealing is still quite difficult. Among the present characterization methods, confocal micro-Raman spectroscopy can provide sub-micrometer resolution, showing great potential for in situ monitoring and visualization of the local sub-micrometer scale stress states and microstructure changes [22,23,35,36,37,38,39,40,41,42]. While the concreate atomic level interfacial microstructure of various heterointerfaces after different annealing conditions can be revealed through transmission electron microscopy (TEM) [22,23,33]. The knowledge on the stress states and microstructures within the bonding interface region is very important for better understanding the residual stress behaviors with temperature, as well as for improving the electrical and thermal properties of SAB fabricated heterostructures.

In this work, the direct wafer bonding of GaN-on-Si heterostructures without any buffer layers were realized by SAB at room temperature. The residual stress including the interlayer stress and microstructure in directly bonded GaN-on-Si heterostructures by SAB was systematically investigated and compared with that of GaN-on-Si heterostructures grown by MOCVD, using confocal micro-Raman spectroscopy and TEM. The interlayer microstructure and residual stress of SAB bonded GaN-on-Si heterostructures were found can be effectively modulated by appropriate thermal annealing, and their characters as well as the effects of annealing temperature were further investigated in detail.

2. Materials and Methods

The investigated GaN(0001) epitaxial layers were originally grown on 4-inch n-Si(111) substrates by MOCVD. Before growing GaN epitaxial layer, an AlN buffer layer with a thickness of ~170 nm was grown on the hydrogen cleaned Si substrate at 1060 °C. Then, a 600 nm-thick GaN layer was epitaxially grown on the AlN layer buffered Si substrates at 1000 °C, using trimethylgallium (TMG), trimethyl-aluminum (TMA), and ammonia (NH₃) as the Ga, Al, and N sources, respectively. In the following, acetone and ethanol were sequentially employed by means of ultrasonic bath to clean the above grown GaN epitaxial layers and the additional n-Si(111) substrates for 600 s. After being dried under N₂, both were set surface-to-surface in the sample chamber of the SAB facility, respectively. Their surfaces were then activated by the irradiation of Ar atom beam, of which the accelerate voltage is 1.5 kV, and the current is 15.6 mA. After surface activation, the GaN epitaxial layers were directly bonded to the target n-Si(111) substrates without using any intermediate layer or buffer layer at room temperature by the SAB technique [43,44], so that Si/GaN/AlN/Si heterostructures were fabricated. Finally, mechanical polishing and chemical wet etching were employed to remove the original epitaxial Si substrates and part of the AlN layer. The detailed process of fabricating a GaN-on-Si heterostructure through the SAB technique is illustrated in Figure 1.

Confocal micro-Raman spectroscopy was employed to systematically investigate the residual stress and interlayer microstructure changes in SAB fabricated GaN-on-Si heterostructures. Raman mapping measurements were performed on the surface and near the interface of GaN layers as well as on the GaN/Si cross-section for an area of 30 × 30 μm² with a step of 1 μm. A Renishaw InVia micro-Raman spectroscopy with a pinhole of 10 μm and a laser sport radius of ~1.5 μm was adjusted to obtain a confocal measurement status in the 180° backscattering configuration, equipped with a 488 nm excitation laser, a 2400 lines/mm diffraction grating, and a 50 × 0.6 N.A. objective lens. The effects of thermal annealing on the residual stress and interlayer structure in GaN-on-Si heterostructure were investigated by in situ confocal micro-Raman via integrating a high-temperature heating/cooling stage (Linkam TS-1500, Linkam Scientific Instruments, Salfords, UK) with a controlled temperature stability of ±1 °C. The SAB bonded samples were annealed in a N₂ gas atmosphere at a sequential temperature of 400, 700, and 1000 °C separately for 300 s. A Lorentz fitting to the GaN and Si Raman peaks was performed to determine their mode frequencies. The E₂^high mode of GaN was used for monitoring the residual stress of the GaN layers due to its higher sensitivity to the residual stress [45]. The high spectral stability enables the shift accuracy of the Raman frequency can be as low as ±0.02 cm⁻¹ within the mapping time, and these were also monitored by measuring the Raman peak change of the Si calibration wafer before and after each measurement to inspect the stability and accuracy of the spectroscopy. The interfacial microstructures of SAB as-bonded GaN-on-Si heterostructures and those after annealing were investigated using TEM (JEM-2200FS, JEOL Ltd., Tokyo, Japan).

3. Results and Discussions

To evaluate the overall residual stress level, confocal micro-Raman spectra of near interfacial GaN layer and Si layer in both SAB and MOCVD fabricated GaN-on-Si heterostructures were measured and analyzed, and their typical GaN E₂^high and Si F_2g modes were separately shown in Figure 2a,b. We note that the Raman peak of unstressed GaN E₂^high mode is relatively dispersive in the literature [10,46,47]. Therefore, we measured a high quality supportless GaN substrate with a thickness of ~0.35 mm as a referenced stress-free bulk value, in which the Raman peak of GaN E₂^high mode was observed at 567 cm⁻¹. Similarly, Si F_2g mode was referenced to the stress-free bulk value of an n-Si(111) substrate whose Raman peak was measured at 520 cm⁻¹ under the same Raman configuration. For MOCVD grown samples, in comparison with the above stress-free bulk value, the E₂^high peak of the GaN layer shifts to higher wave numbers which indicates that the residual stress was compressive. In contrast, the residual stress in SAB bonded GaN layers was mainly tensile stress (Figure 2a). The compressive stress here obtained in the MOCVD grown GaN layer near the interface is normally originated from the synergistic effects of lattice mismatch induced defects (such as dislocations, vacancies) in the GaN and the stress transfer by the strain relief buffer layer [10,46,47,48]. For the small tensile stress in the SAB bonded GaN layer near the interface, it is likely ascribed to the significant stress relaxation by the amorphous layer formed at the interface. While in the Si layer, no matter for the MOCVD grown samples or the SAB bonded samples, compressive stress was normally resulted (Figure 2b). Similarly, the synergistic effects of interfacial lattice mismatch and stress transfer also resulted in different compressive stress states in the Si near the interface for both samples. The difference is that the MOCVD grown samples have relatively larger compressive stresses but narrower peak width which indicates better crystal quality of Si near the MOCVD grown AlN/Si interface. The smaller stress but relatively poor Si quality resulted in the SAB bonded samples is mainly ascribed to the surface amorphization during the SAB process, which leads to a more disordered local structure but owns a better relaxation effect.

In further, the Raman peak shift mappings of GaN E₂^high mode and Si F_2g mode extracted from their Raman spectra by referencing to their stress-free bulk values in SAB fabricated GaN-on-Si heterostructures were shown in Figure 2c,d, respectively. The Raman peak shift of GaN E₂^high mode and Si F_2g mode in GaN-on-Si heterostructures fabricated by SAB was found to range from −0.43 to 0.46 cm⁻¹ and 0.22 to 0.72 cm⁻¹, respectively; in addition, their averaged Raman peak shifts in corresponding layers of SAB fabricated samples were near 0 and 0.46 cm⁻¹, respectively. These results imply that tensile stresses were mainly resulted in the GaN layer while compressive stresses were mainly existed in the Si layer of the SAB as-bonded GaN-on-Si heterostructures.

According to the known relationship between biaxial stress and Raman shift, the optical phonon frequencies of both GaN E₂^high mode and Si F_2g mode shift linearly with stress in the following formulas, respectively [49,50],

$${\sigma }_{\mathrm{xx}}\left(\mathrm{GaN}\right)= {\Delta }_{\omega }\left(\mathrm{GaN}\right)/\left(2.9{ \mathrm{cm}}^{-1}{\mathrm{GPa}}^{-1}\right),$$

(1)

$${\sigma }_{\mathrm{xx}}\left(\mathrm{Si}\right)= {\Delta }_{\omega }\left(\mathrm{Si}\right)/\left(2.3{ \mathrm{cm}}^{-1}{\mathrm{GPa}}^{-1}\right),$$

(2)

where ${\Delta }_{\omega }\left(\mathrm{GaN}\right)$ and ${\Delta }_{\omega }\left(\mathrm{Si}\right)$ is the residual stress in the GaN layer and Si layer, while ${\Delta }_{\omega }\left(\mathrm{GaN}\right)$ and ${\Delta }_{\omega }\left(\mathrm{Si}\right)$ is the measured Raman peak shifts (in the units of cm⁻¹) of the GaN layer and Si layer with respect to their stress-free bulk values. We used these formulas to calculate the residual stress values in the GaN layer and Si layer of GaN-on-Si heterostructures. In average, a significantly relaxed residual stress of 0.0004 GPa (tension) was resulted in the GaN layer of SAB fabricated GaN-on-Si heterostructures, while a relatively large residual stress of −0.0462 GPa (compression) was formed in the GaN layer of MOCVD grown GaN-on-Si heterostructures, which is in consistent with those reported values of GaN grown on Si(111) substrates using the same MOCVD methods [46,51]. Additionally, as can be seen in Figure 3a,b, there was a larger stress variation in the GaN layer of MOCVD grown GaN-on-Si heterostructure, indicating a more inhomogeneous stress distribution was formed.

Moreover, the effects of thermal annealing on the stress and interfacial microstructure of GaN-on-Si heterostructures fabricated by SAB were further explored. The averaged Raman peak shifts and the calculated residual stresses in GaN and Si layers of SAB fabricated GaN-on-Si heterostructures under different temperatures annealing, as well as those in GaN and Si layers of MOCVD grown GaN-on-Si heterostructures at room temperature, are displayed in Figure 3c,d, respectively. The magnitude of error bars reflected the homogeneity of Raman peak shifts and stress distribution in GaN epilayer and Si substrate over the mapped area of 40 × 40 μm². Note that the averaged Raman peak shifts and residual stress as well as their error bars in the GaN layer and Si layer of SAB fabricated GaN-on-Si heterostructures were smaller, indicating the stress was significantly relaxed and more homogeneously distributed than that of MOCVD grown samples. It was also found that the averaged Raman peaks of GaN and Si in SAB fabricated GaN-on-Si heterostructure shift to lower wavenumbers with increasing annealing temperature, which decrease to about −0.47 and 0.13 cm⁻¹, respectively, after 1000 °C annealing; these results reveal that the tensile stress in GaN increases while the compressive stress in Si decreases, respectively, with the increase in annealing temperature. These different annealing temperature dependent changes of residual stresses were later found being closely correlated to the temperature dependent changes in the interfacial structure of GaN-on-Si heterostructures during the annealing process.

For the sake of better understanding the relaxation nature of residual stresses and their annealing temperature dependent behaviors in SAB fabricated GaN-on-Si heterostructure, the cross-sectional microstructure and stress were further investigated by micro-Raman spectra. As displayed in Figure 4a, the GaN/Si interface was vertically aligned to the moving direction of the laser beam, which is adopted to linearly scan along the white-dotted line that rightly across the bonding interface from the GaN side to the Si side. Similar to the above annealing method, we monitored the effects of annealing temperature on the interfaces by mapping the cross-sectional Raman spectra. From Figure 4b, significant changes were observed at the as-bonded and 1000 °C annealed GaN/Si interfaces. When just bonded by SAB without annealing, the interlayer region of GaN-on-Si heterostructure was wide and the Si elements diffuse much deeper into the GaN side due to the SAB process. While after 1000 °C annealing, narrower interlayer region and shorter penetration depth of the Si elements can be found diffusing into the GaN side. Figure 4c,d show the detailed Raman spectra evolution along the white-dotted line before and after 1000 °C annealing. As can be clearly seen, when the GaN/Si interface was just bonded without annealing, the Si F_2g peak was broad, but after 1000 °C annealing the Si F_2g peak became much sharper, this implies better Si crystallinity was formed. At some positions within the interlayer region, we can also observe weak GaN peaks, from which the stress was found to be as large as 0.49 GPa, meanwhile, the splitting of Si F_2g mode and GaN mode was accompanied. Such splitting behaviors were probably originated from the seriously distorted lattice induced by the large mismatch (17%), as can be inferred from the weakened splitting intensity when the stress in Si was relaxed after high temperature annealing.

Raman spectra can conveniently and sensitively detect the above interfacial microstructure changes or even the evolution under different annealing temperature, yet it is still an indirect observation. To directly examine the local nanoscale and atomic structure, TEM was performed at the SAB bonded GaN/Si interfaces. As shown in Figure 5a,b, high-resolution cross-sectional TEM images under high magnification condition for the bonding interfaces of both as-bonded and 1000 °C annealed GaN-on-Si heterostructures were performed, respectively. In general, sharp bonding regions and interfaces without any observable structural defects such as micro voids and cracks were evident for both as-bonded and annealed samples. This reflects the highly reliable stability and high quality of this kind of SAB bonded GaN/Si heterointerfaces, even after the near-melting high temperature annealing at 1000 °C. It is also noted that a graded transition region of about 10 nm thick amorphous-like interlayer was formed at the as-bonded interface. However, after 1000 °C annealing, this amorphous layer vanished at the bonding interface; instead, a well-crystallized interlayer with a thickness of only several atomic layers thin was formed (Figure 5b), which is very likely to be silicon nitride due to its easier formation at high temperatures [50].

Similar to the amorphous layers observed at the SAB fabricated Si/SiC, Si/GaAs, Si/Diamond interfaces [22,24,26,29], the amorphous layer formed at the GaN/Si bonding interface should be originated from the amorphized activation of sample surfaces by fast Ar atom beam during the SAB process, followed by tight bonding under appropriate pressure, so that a robustly connected heterointerface can be formed. The amorphous layer normally has graded composition and disordered structure with enough variation space to alleviate the large lattice mismatch between GaN and Si (~17%), therefore, the significant relaxation behavior of residual stress in both GaN layer and Si layer of SAB fabricated GaN-on-Si heterostructure could only be correlated to the amorphous layer formed at the bonding interface. In different to previously bonded heterostructures by SAB [22,24,26,29], this amorphous layer between GaN and Si not only owns a stably welding function to work as an interfacial adhesion layer, but also plays an air-spring effect across the bonding interface to serve as a stress relief layer, similar to the examples of GaN layers transferred onto Si substrate through an Au-Si bonding process, in which the Au intermediate layer acted as a relaxation layer [47]. However, the GaN/Si heterostructure with an Au intermediate layer is not suitable for fabricating high-frequency power devices, because large parasitic capacitance would be generated between the device channel and the Au intermediate layer.

As to the strong dependence of residual stress on the post-annealing temperature in SAB fabricated GaN-on-Si heterostructure, it is in fact correlated to the thickness change of the amorphous layer under annealing, which continuously decrease with increasing annealing temperature until disappear after annealing at 1000 °C. This continuously thinning amorphous layer makes it more and more difficult to ease up the residual stress within GaN and Si caused by their large lattice and thermal expansion mismatches after high temperature annealing. Moreover, because of the larger thermal stress than the shocking residual stress remained in GaN after the bonding as well as the larger thermal expansion coefficient of GaN than that of amorphous and crystalized silicon nitride, the tensile stress in the GaN layers became larger at high annealing temperature, while the compressive stress in Si became smaller, thus resulting in the stress cross-change with increased annealing temperature (see schematic for the stress evolution in Figure 6). We also noted that the amorphous layer recrystallized into a several atomic layers thin silicon nitride layer after annealing at high temperature. The recrystallization mechanism is not difficult to understand as it is similar to the examples of previously observed direct reaction of nitride with silicon at the grown GaN/Si interface [52]. Meanwhile, thermal annealing can promote the recrystallization of those amorphized regions of GaN and Si near the bonding interface as introduced by the initial surface activation, which also contribute to the vanish of the amorphous layer as well as the more and more significant interlayer thinning after high temperature annealing. However, this ultrathin silicon nitride layer is very difficult to observe in our present Raman spectra to determine its stress state due to the large laser beam radius and diffraction limited spatial resolution.

4. Conclusions

In summary, the residual stress, the interlayer stress, and microstructure in GaN-on-Si heterostructures directly bonded by SAB at room temperature were systematically investigated through confocal micro-Raman spectroscopy and TEM techniques. Unlike a larger compressive stress usually generated in the GaN epilayer and Si of MOCVD grown GaN-on-Si heterostructures, a significantly relaxed small tensile residual stress and a more uniformly distributed smaller compressive stress were separately obtained in the GaN layer and Si layer of SAB as-bonded GaN-on-Si heterostructures. The main reason was due to a surface activation resulted amorphous layer forming at the bonding interface that acted as a role of not only interfacial adhesion but also stress relieve. In addition, thermal annealing was found can significantly modulate the residual stress and interlayer microstructure in SAB bonded GaN-on-Si heterostructures. The residual stresses monitored via in situ micro-Raman at both GaN layer and Si layer were observed monotonically change with increasing temperature. Additionally, a several atomic layers thin crystallized silicon nitride layer was observed by TEM at the GaN/Si interface after 1000 ˚C annealing, while a relative thick amorphous interlayer was observed at the as-bonded interface. The stress cross-change and interlayer thinning with increasing annealing temperature are in fact originated from the thickness evolution of the amorphous layer. This work provides a new strategy to fabricate low-cost high quality GaN-on-Si heterostructures with potentially high thermal management performances.