Extended Defect Propagation in Highly Tensile-Strained Ge Waveguides

Abstract

Tensile-strained Ge is a possible laser material for Si integrated circuits, but reports of lasers using tensile Ge show high threshold current densities and short lifetimes. To study the origins of these shortcomings, Ge ridge waveguides with tensile strain in three dimensions were fabricated using compressive silicon nitride (SiN_x) films with up to 2 GPa stress as stress liners. A Raman peak shift of up to 11 cm⁻¹ was observed, corresponding to 3.6% hydrostatic tensile strain for waveguides with a triangular cross-section. Real time degradation in tensile-strained Ge was observed and studied under transmission electron microscopy (TEM). A network of defects, resembling dark line defects, was observed to form and propagate with a speed and density strongly correlated with the local strain extracted from both modeled and measured strain profiles. This degradation suggests highly tensile-strained Ge lasers are likely to have significantly shorter lifetime than similar GaAs or InGaAs quantum well lasers.

1. Introduction

Tensile-strained germanium has been studied as a possible laser material due to its nearly-direct bandgap [1,2] and its compatibility with conventional silicon integrated circuit fabrication [3,4,5,6]. Theoretically, a tensile strain of 0.7% (hydrostatic) or 1.4% (biaxial) could produce a direct bandgap in Ge [7,8,9,10]. The hydrostatic component of strain is responsible for the shift from an indirect to a direct bandgap. Biaxial strain has a weaker hydrostatic component due to Poisson contraction along the out-of-plane direction.

Many approaches to inducing tensile strain for Ge lasers and more efficient light emitters have been demonstrated [11,12,13,14,15,16,17,18,19]. The first optically pumped Ge laser used 0.15% biaxial strain plus degenerate n-type doping to partly populate the Γ valley with electrons despite the indirect bandgap [2]. Enhanced electroluminescence from Ge light-emitting diodes (LEDs) has been demonstrated based on tensile-strained Ge with different stressors, but power efficiency needs to be significantly improved for practical lasers [13,14,15,16]. Enhanced direct-bandgap photoluminescence has similarly been demonstrated [20,21,22]. Although electrically-pumped pulsed Ge lasers have been demonstrated, the reported threshold currents were very large and operating lifetimes short [23,24]. Efforts to increase efficiency and optical gain by inducing higher strain are often faced with higher dislocation density and nonradiative recombination, which increases the laser threshold. Also, the high circulating power in lasers can lead to the generation of extended defects in real time. This is observed as dark line defects (DLDs), which are dislocation networks and stacking faults generated in the laser due to high thermal, optical, and electrical power densities [25,26].

To study the stability of highly tensile-strained Ge in the presence of perturbations, this work exposed both strained and unstrained Ge to 300 kV electrons in transmission electron microscopy (TEM). Real-time defect generation observed in transmission electron microscopy (TEM) was then correlated with strain distribution analysis, Raman spectroscopy, and photoluminescence (PL). Tensile strain with a strong hydrostatic component was induced in Ge waveguides by compressive SiN_x stress liners [27,28], which are widely used for strained complementary metal oxide semiconductor (CMOS) transistors. In a wraparound geometry, as shown in Figure 1, compressive stress in the SiN_x induces significant tensile strain up to 3.6% in the Ge along two dimensions, just as planar biaxial strain techniques would do. In addition, the third dimension (longitudinal) is constrained, which enhances the total hydrostatic component of strain by preventing Poisson contraction that occurs in planar techniques.

2. Results

Figure 2 shows the top-incidence Raman spectrum of the fabricated Ge waveguides. Contributions from the substrate were ruled out by measurements at different locations and focal depths (supplemental Figure S1). The Raman pump laser had a wavelength of 488 nm, with a spot size of ~0.5 µm and a calculated penetration depth around 19 nm in Ge. The system was calibrated using a reference bulk Si crystal. A shift from the bulk Ge peak of 11 cm⁻¹ was observed for waveguides with a strongly (~2 GPa) compressive-strained SiN_x film. This shift in Raman corresponds to a sufficiently high strain, causing an expected direct bandgap and strong photoluminescence (PL) emission, with an effective hydrostatic strain up to 3.6% [29]. We attribute the difference from our COMSOL model to be due to the sharp peak geometry for the narrowest waveguide sample, which focuses strain on a narrow stripe along the top of the waveguide.

The high intensity of the laser used in Raman spectroscopy can itself lead to an apparent shift in the peak position. To determine the upper limit on laser power, the power dependence of the Raman spectra was measured both in the far field (Figure 3b) and centered on a strained waveguide (Figure 3c), as shown in Figure 3. Waveguides showed a stronger power dependence than the far field, with a thermal shift of up to −3.5 cm⁻¹, indicating that the tall waveguide geometry was insufficient to dissipate the heat from laser powers >2 mW. Therefore, unless otherwise noted, the Raman measurements in this paper used lower laser powers.

Figure 4 shows a comparison of normal-incidence PL for different strains. The excitation wavelength was 1310 nm, which has a penetration depth in bulk Ge of 1.5 µm. In the highly strained waveguides, PL intensity was very weak. Since both Raman and TEM (below) confirm significant strain, the weak PL indicates a large number of non-radiative recombination centers. The sample with a lower (~1 GPa) compressive stress SiN_x film, i.e., weakly strained Ge, had a PL intensity higher than the highly strained Ge but lower than the unstrained control sample; furthermore, a much smaller PL peak wavelength shift was observed.

To understand the reasons for weak PL emission, we used transmission electron microscopy (TEM) to investigate the strained-Ge/SiN_x interface. The highly strained Ge sample is shown in Figure 5. Even using the shortest possible imaging time to acquire focus (~15 s), dark spots were seen near the Ge/SiN_x interface. Smaller but similar spots were observed in all samples with compressive SiN_x, even in the far field where Ge strain was minimal, but more pronounced for higher stress SiN_x, consistent with more intense ion bombardment of the surface during nitride deposition. With increasing electron-beam exposure time, the dark spots grew larger and darker, indicating a more severe structural degradation. To assure the same electron flux for direct comparisons, beam current and magnification were kept as close as possible between samples.

Highly strained Ge was found to degrade rapidly under TEM imaging conditions. To show the time-dependence of the TEM image contrast, a series of TEM images were taken as a function of beam exposure at a constant dose rate, as shown in Figure 6. The damage is compared across different regions by imaging after certain time intervals. There is no degradation for the region where Ge has relaxed or is not intentionally strained, while the degradation is apparently severe near the strained Ge/SiN_x interface. For the highly strained sample, the degraded area quickly increases in size, propagating into the bulk Ge from the Ge/SiN_x interface. In order to obtain a quantitative comparison, the brightness of all images was normalized to that of SiN_x. With dark spots denoting defective areas, the portion of the defective area in the same region (delineated by a dashed box in Figure 6 to account for sample drift) for both low strain and high strain samples is shown as a function of electron exposure time in Figure 7. Defects were observed to propagate faster and deeper for highly strained samples, but progressive degradation was observed for all samples with tensile strain, and no degradation was observed for samples without strain.

3. Discussion

We attribute the observed degradation to instability in the highly strained Ge. Figure 8 shows a cross-section HRTEM strain analysis for the same sample as Figure 2. Extracted values of tensile strain were comparable with the values from Raman, averaging +3.0% (tensile) along [022] and −1.2% along [0$\overline{2}$2], although quantitative accuracy is limited in this technique due to the large number of defects. Qualitatively, the strain was also found to be non-uniform, with tensile strain highest at the Ge-SiN_x interface. Figure 9 shows that the pattern of degradation follows the distribution of strain calculated using an elastic strain model in COMSOL. No degradation with time (darkening in TEM) was visible in the bulk Ge below the waveguide, where SiN_x would exert little or no strain. Rapid and deep degradation was observed near the tip of the triangular waveguide where the SiN_x stress liner geometry exerts the most strain due to geometry. Intermediate values of strain may only lead to slight darkening near the Ge-SiN_x interface. The figure shows a snapshot of darkening after just 15 s at low magnification, corresponding to a lower dose than previous figures.

Although the theoretical tensile strength of Ge is predicted to be 14–20 GPa [30,31] while undergoing an elastic deformation, experimental measurements range from 15 to 18 GPa [32,33] for Ge nanowires to just 40–95 MPa for bulk Ge [34]. This study indicates that tensile-strained Ge is prone to premature relaxation nucleating at defects, interfaces, or asperities. Achieving sufficient tensile strain to produce a direct bandgap therefore requires extremely gentle fabrication and passivation techniques with no nucleation points for relaxation. Figure 8 indicates that the highly strained Ge degraded in a distinct pattern. The strain analysis in Figure 8 also confirms that the defective areas are networks of dislocations and stacking faults propagating in real time, and that there is a correlation between defective area and strain distribution.

Nucleation of these defects could be driven by energetic fabrication techniques such as ion bombardment during either the reactive ion etching (RIE) or the dual-frequency plasma enhanced chemical vapor deposition (DF-PECVD) of compressive SiN_x. Such damage could then propagate into the strained bulk area during TEM observation. However, the degradation effect was more severe on the top of the waveguide, which received the least exposure to ion bombardment (protected by the mask) during RIE etching but comparable bombardment during SiN_x deposition. Also, dark spots existed at the interface for all samples with strained SiN_x, even in the far field where the Ge was effectively unstrained. Therefore, we attribute this surface damage due to ion bombardment during the nitride deposition. Therefore, we do not believe RIE etching contributed significantly to the sensitivity of the Ge under TEM e-beam exposure.

As mentioned above, the Ge waveguides that had lower strains either due to wider features or lower stress (1 GPa) SiN_x also demonstrated the formation of defects in real time, but at a slower rate. We were unable to establish a threshold of maximum strain for which the defects would not form; they formed in all strained Ge waveguides, but much slower for lower strains. We conclude that the strain necessary to achieve a fully direct bandgap is well beyond the threshold for structural stability in the presence of perturbations such as TEM or, by analogy with DLDs in GaAs/AlGaAs lasers, intense optical illumination [35] and free carrier injection [36,37].

4. Materials and Methods

Ge ridge waveguides with widths from 0.5 µm to 80 µm were patterned by contact lithography on (001) Ge wafers and a 2 µm deep etch by RIE. Due to mask erosion during RIE, the sidewalls of the waveguide were sloped by approximately a 4:1 ratio. The narrowest (0.5 µm) waveguide thus had a triangular cross-section, which enhanced the local strain near the tip of the triangle. Next, 1 µm thick compressive-strained SiN_x stress liners with a stress of either 1 GPa (low strain) or 2 GPa (high strain) were deposited by DF-PECVD. As discussed above, the mechanism for inducing compressive strain in SiNx by DF-PECVD is by ion bombardment, mostly H⁺. Figure 1a shows a schematic of the waveguide structure, indicating the components of stress from the SiN_x film applied along the in-plane and out-of-plane directions. Figure 1b shows a scanning electron microscopy (SEM) image of a cleaved fabricated waveguide structure with rectangular cross-section. The strain distribution in the waveguide structure was simulated using COMSOL Multiphysics. Figure 1c shows one of the simulated waveguide structures, predicting that the upper region is tensile-strained while the lower regions are under compression. Figure 1d shows the optical mode profile predicted from the strain profile [38]. Cross-section high-resolution TEM (HRTEM) images were taken in a FEI Titan 80–300 TEM system. The strain in fabricated waveguides was characterized by a WITec Alpha 300R confocal Raman microscope system. HRTEM was also used for strain analysis using the Digital Analysis of Lattice Images (DALI) software package [39]. The specimens were tilted along the zone axis for TEM imaging. All the TEM specimens were prepared by focused ion beam (FIB) milling.

5. Conclusions

In summary, highly tensile-strained Ge waveguides were observed to rapidly degrade under the perturbation of an electron beam in TEM, while lower-strained Ge degraded more slowly, and no degradation was observed in unstrained or compressively strained Ge, even on the same wafer. Although the use of SiN_x stress liners induced tensile strains in the Ge of up to 3.0–3.6%, as measured by Raman and HRTEM strain analysis, and PL peak emission shifted to lower energies, PL emission was very weak. Strain distribution analysis showed a non-uniform strain profile with an average strain of +3.0% (tensile) along [022] and −1.2% (compressive) along [0$\overline{2}$2]. Degradation closely followed the pattern of strain according to TEM strain distribution analysis as well as numerical elastic strain models. Since low-strained Ge and comparably strained InGaAs do not show such rapid degradation, it appears that tensile Ge is more fragile than comparable InGaAs quantum wells for lasers. Specifically, we believe local variations in strain near Ge surfaces or interfaces can lead to the rapid formation of dislocations and stacking faults. The results indicate that tensile Ge is better suited for low-energy devices such as detectors and modulators than it is for lasers, where high optical powers and free carrier recombination will accelerate the formation and propagation of such defects.