Demonstration of biaxially tensile-strained Ge/SiGe multiple quantum well (MQW) electroabsorption modulators with low polarization dependence

1 Introduction

Silicon-based optical interconnects are considered to be the future replacement of electrical interconnects using copper wires. Due to the indirect-bandgap nature of silicon, the silicon-based optical interconnects are bottlenecked by lack of realizable active devices. Hybrid integration techniques such as optical bench [1], [2] and flip-chip bonding [3], [4], [5] have been developed to achieve interchip and intrachip communications. Hybrid integration using III–V materials results in high optical loss and manufacturing cost. None of them can fulfill the need of monolithic and fully complementary metal oxide semiconductor (CMOS) compatible integration. Group IV direct bandgap semiconductor materials have drawn much attention recently as they are promising candidates to achieve the monolithic CMOS compatible integration of all active devices including lasers [6], [7], [8], modulators [9], [10] and photodetectors [11], [12], [13]. For modulators, quantum-confined Stark effect (QCSE) of Ge/SiGe multiple quantum wells (MQWs) has been demonstrated to achieve small footprint and low-energy consumption electroabsorption modulation [14], [15].

The polarization dependence of quantum well material comes from the asymmetrical spatial confinement [16]. Polarization dependence of Ge/SiGe MQWs has been theoretically investigated [17] and experimentally observed [18]. The polarization dependence of Ge/SiGe quantum well will greatly limit the design flexibility and application range. Absorption spectra of Ge/SiGe MQWs are polarization dependent and make the refractive index polarization dependent through K–K relationship. As a result, both electroabsorption and electrorefractive modulators based on Ge/SiGe MQW material are not suitable to be used in circumstance as polarization multiplexing. Besides, the incident light needs to be polarized to get predicted contrast ratio and insertion loss. In our previous paper, we proposed that biaxial tensile strain should be adopted to tune the polarization dependence of Ge/SiGe MQW electroabsorption modulators [19]. In this work, we report a low polarization dependence Ge/SiGe MQW electroabsorption modulators with suspended microbridge structure. The polarization dependence of unstrained Ge/SiGe MQW electroabsorption edge is caused by an asymmetrical spatial quantum confinement. By introducing biaxial tensile strain, the differences between the ground state heavy hole and electron (Γe1-HH1) transitions under transverse electric (TE) and transverse magnetic (TM) polarizations are eliminated. Low polarization dependence electroabsorption is achieved under 0.72% biaxial tensile strain. The fabricated device has a length of 900 μm with deep etching coupling facets. Under 0 V/2 V operation, the device has an extinction ratio of 4.3 dB for both TE and TM polarizations. The experiments confirm the feasibility of tuning the polarization dependence of Ge/SiGe MQW electroabsorption modulators. This new approach will promote applications of Ge/SiGe MQW electroabsorption modulators and photodetectors in low polarization dependence circumstances.

2 Design and fabrication

The device with suspended microbridge structure is schematically shown by Figure 1(a). The epitaxy design of the MQW wafer is shown in Figure 1(b). The Ge/SiGe MQWs are grown by low-energy plasma-enhanced chemical vapor deposition on a p-type [001] Si substrate. Four hundred nanometer boron-doped (1 × 10¹⁹ cm⁻³) Si_0.15Ge_0.85 buffer layer is firstly deposited to reduce dislocations and surface roughness. After the buffer layer is grown, an in situ 800 °C anneal process is proceeded and causes 0.19% biaxial tensile strain in the buffer layer [14]. The MQWs consist of 10 × 10/12 nm Ge/Si_0.19Ge_0.81 quantum wells surrounded by 50 nm intrinsic Si_0.15Ge_0.85 spacers. The whole structure is capped by a phosphorus-doped (5 × 10¹⁸ cm⁻³) n-type Si_0.15Ge_0.85 layer as the top contact. The strained lattice constant of the MQW region is designed to be equal to that of the lattice constant of the SiGe buffer layer after the in situ anneal process to achieve strain compensation (see Supplementary material). According to the TEM images of the material after epitaxy, there are dislocations at the interface of Si and SiGe materials caused by the crystal lattice mismatch between Si and SiGe. The quantum well and barrier interface is clearly showing a good crystal quality, as shown by Figure S1. The defect density is below 2 × 10⁷ cm⁻² according to the etching pit microscope test.

Figure 1:

(a) Schematic of the device. (b) Epitaxy design of the multiple quantum wells (MQWs) structure. (c) The fundamental mode of the suspended waveguide.

The fundamental mode of the suspended waveguide is shown in Figure 1(c). The fundamental mode shown in Figure 1(c) is simulated by using the three-dimensional (3D) finite-difference time-domain software. The total waveguide consists of several segments with different widths and the 3 μm wide central active waveguide is chosen after detailed calculation. The waveguide offers multimode performance. The overlap factor between the optical field and the MQWs region is 0.42. The active area length of the device is 20 μm, and the total waveguide length is 900 μm. The active waveguide has a width of 3 μm. At the butt coupling end, a length of 40 μm waveguide is used to couple light with focusing lensed fiber. The coupling waveguide is 8 μm in width. Between the active waveguide and the coupling waveguide, there is taper-shaped waveguide with a linearly decreasing width. The biaxial tensile strain is introduced through under-etched suspended microbridge structure [20], [21], [22]. The strain accumulation effects of the suspended microbridge structures rely on the residual tensile strain of the suspended material. The strain distribution simulation result is obtained by using a 3-D finite element method (3-D FEM). At the central area of the microbridge, the strain is uniformly distributed (see Supplementary material).

The fabrication process is shown in Figure 2. To ensure the alignment accuracy, the electron beam lithography (EBL) is used to form the patterns of the metal marks, waveguide, contact window and metal contacts. Figure 2(a) shows the fabrication of metal marks used for alignment. The metal marks are used in steps (b), (d) and (e).

Figure 2:

The schematic diagram of the fabrication process.

(a) Metal marks fabrication; (b) waveguide lithography and etching; (c) deposition of passivation layer; (d) contact window lithography and etching; (e) contact evaporation; (f) deep etching; (g) wafer dicing and protecting layer deposition; (h) microbridge window etching; (i) wafer splitting and wet etching.

Figure 2(b) shows the waveguide lithography and etching. The waveguide is patterned by EBL and etched by inductively coupled plasma etcher. The etching depth of the waveguide layer is 630 nm and stops at the p-type buffer layer. A set of cross-shaped marks are also fabricated in this step. The cross-shaped marks are used in the following ultraviolet (UV) exposures in steps (f) and (h). The waveguide is along [100] direction.

Figure 2(c) shows the deposition of passivation layer by using plasma-enhanced chemical vapor deposition. Hundred nanometer SiO₂ is firstly grown, which serves as the main insulation material. Two hundred nanometer Si₃N₄ is then grown, which is the protecting and mask layer.

Figure 2(d) shows the contact window lithography and etching process. The etching depth of this step is 220 nm and stops at the SiO₂ layer. Wet etching using buffered hydrofluoric solution with 40 s is carried out. The buffered hydrofluoric solution has a high selectivity for SiO₂ and the Si₃N₄ layer that serves as a mask layer [23], [24], [25].

Figure 2(e) shows the metal contact lithography, evaporation and lift-off. The material of the metal contact is 20/80 nm Cr/Au. A low power ultrasonic machine is used during the lift-off process.

Figure 2(f) shows the deep etching process of the coupling facet. Bosch process is carried out to form deep etching trench [26], [27], [28]. The Bosch recipe uses C₄F₈ as passivation gas and SF₆ as etching gas. The width of the deep etching trench is 500 μm. The total etching depth is measured to be 101.5 μm and the period is 0.84 μm. The etching facet is smooth within each cycle and corrugation like on the whole. Corrugation causes slight diffusion, and the period is too small to cause significant interference or form surface grating. On the other side, such scalloping shape of coupling facet avoids strong Fresnel reflection.

Figure 2(g) shows the wafer dicing and deposition of SiO₂ protecting layer. The wafer is diced along the deep etching trench so that it can be separated into bars before wet etching. The gap between the cutter and the bottom tray is set to be 150 μm so that the chip does not break after dicing. The protecting layer can avoid electrochemical corrosion during the wet etching process [29].

Figure 2(h) shows the fabrication of the microbridge window. The pattern is formed by an UV exposure machine. The etching depth of the microbridge window is 1.8 μm.

Figure 2(i) shows the wet etching process. The separated bar has a width of 2.8 mm and is wet etched by tetramethylammonium hydroxide (TMAH) solution. The TMAH solution eats off the Si material through the microbridge window and forms suspended microbridge structure as the solution has a high selection ratio for Si material [30], [31], [32]. The wafer we use is <001> orientated and the waveguide is along <100> direction. Wet etching of silicon in TMAH solution is anisotropic. Although the maximum etching rate can be realized at <111> direction, the bridge does not have to be along or perpendicular to <110> direction (for in-plane 2D condition) to form a suspended microbridge. During the formation of suspended microbridge, the wet etching happens when the reactants contact with each other, the effect of anisotropy can be concluded as shaping the outline of the etched under cut. The microbridge is surrounded by the etching solution with a large contact area. The shape of the etching pit is pyramid like because of etching anisotropy.

The SEM images of the fabricated device are shown in Figure 3. There are three different areas according to the difference of strain. The bridge area has a tensile strain due to the strain accumulation effect of suspended microbridge structure [20]. Strain of the arm area is relaxed, and the blank area has a 0.19% tensile strain caused by in situ anneal process, which is tested by microarea X-ray diffraction (XRD). The XRD curves are shown in Figure 3(d). The Si(004) peak is utilized to deduce the residual strain value of the SiGe buffer layer. The X-ray source is Cu-kα1 with a wavelength of 0.15406 nm. The experiment result shows several clear satellite-peaks overlapped by the Si_0.15Ge_0.85 buffer peak. As the buffer layer is thicker than quantum wells, the Si_0.15Ge_0.85 buffer peak is higher than the satellite-peaks. The simulation result shows that there is 0.19% biaxial tensile strain in the buffer layer after epitaxy. Figure 3(b) shows the suspended microbridge area. Figure 3(c) shows the deep etching facet. The coupling end is deeply etched so that the lensed fiber can reach the coupling waveguide.

Figure 3:

(a) SEM image of the fabricated device with suspended microbridge and deep etching facet after splitting. (b) Detailed image of the active area. (c) Deep etching facet of the coupling end. (d) X-ray diffraction (XRD) experiment and simulations of the wafer after epitaxy.

3 Measurements and discussions

3.1 Raman characterization

According to the study of Gassenq [33], there is a corresponding relationship between the Raman peak location and the strain value for SiGe material. We take Raman tests to analyze the strain value in the buffer layer. As the Ge contents of buffer layer and cap layer are equal, the strain in the buffer layer and cap layer are equal when the pseudomorphic crystal is formed during the wafer epitaxy. The Raman peaks of the cap layer are shown in Figure 4. The Raman peak for the 0.19% biaxial residual tensile strained blank area is at 297.31 cm⁻¹. For the central biaxially tensile-strained bridge areas, the corresponding Raman peaks are at 296.64 and 296.09 cm⁻¹ for 0.72 and 1.04% biaxially tensile-strained microbridges. The strains of the arm areas are relaxed, and the corresponding Raman peak locations are 297.48 and 297.50 cm⁻¹.

Figure 4:

(a) Raman peaks of the device with 0.72% biaxial tensile strain in the cap layer. (b) Raman peaks of the device with 1.04% biaxial strain in the cap layer.

3.2 Extinction ratio and absorption spectra

We use a polarization maintaining focusing lensed fiber and a single mode focusing lensed fiber to couple light into and out of the device waveguide. The images of the device under test are shown in Figure 5. Light from the source is polarized by a direction adjustable linear polarizer. The transmitted light is fed into an optical spectrum analyzer. An Erbium-doped fiber amplifier (EDFA) is used as a wide spectra light source to adjust the coupling angle of the butt coupling focusing lensed fibers.

Figure 5:

(a) Photograph of the stage. (b) Microscope image of the fabricated device with two lensed fibers, top view. (c) Oblique view.

The experiment results and corresponding simulations are shown in Figure 6. When the focusing lensed fibers are aligned to the device waveguide, a Fabry–Perot interference filtering spectrum is observed with 1.136 nm free spectral range around 1547 nm wavelength, as shown in Figure 6(b). Otherwise, the spontaneous emission spectrum of the EDFA as Figure 6(a) is received.

Figure 6:

(a) Transmission spectrum when the focusing lensed fiber is aligned to the Si substrate, (b) transmission spectrum when the focusing lensed fiber is aligned to the device waveguide, (c) finite-difference time-domain (FDTD) simulation model, (d) FDTD simulation result of the transmission spectrum around 1547 nm wavelength. (e) The transmitted light power under TE polarization, (f) the transmitted light power under TM polarization. (g) Extinction ratios under 0 V/2 V operation, (h) extinction ratios under 0 V/4 V operation.

After the focusing lensed fibers are aligned to the device waveguide, the EDFA is replaced by a wavelength tunable laser with a wavelength tuning range of 1450–1520 nm. The transmission light is detected by an optical power meter. Reverse bias voltage is applied to the device using a Keithley 2401 source meter. The optical powers received by the power meter under 0, 2 and 4 V reverse bias voltages are shown in Figure 6(e) and (f). The total insertion loss from the light source to the power meter is −37.5 and −39.6 dB for TE and TM polarizations under 1478 nm wavelength. The polarization dependent insertion loss is mainly caused by the difference between TE and TM polarization coupling losses. The insertion losses caused by the linear polarizer are tested to be −8.2 and −8.5 dB for TE and TM polarizations correspondingly. The insertion loss caused by butt coupling is calculated to be −23 and −26.5 dB for TE and TM polarizations, respectively. The transmission loss and material absorption loss is estimated to be −6.3 dB for TE polarization and −4.6 dB for TM polarization. As the absorption edge of the biaxially tensile-strained MQWs material under 0 V bias voltage is around 1475 nm wavelength, the transmitted light is absorbed by the material and the received optical power increases with wavelength under 0 V bias voltage for both TE and TM polarizations. Under 2 and 4 V reverse bias voltages, the absorption peaks are red shifted under QCSE, and the minimum optical locations are red shifted correspondingly.

Figure 6(g) and (h) shows the extinction ratios of TE and TM polarizations. Under 0 V/2 V operation, the device has 4.3 dB extinction ratios for both TE and TM polarizations at 1478 nm wavelength. Under 0 V/4 V operation, the device has 6.5 dB extinction ratio for both TE and TM polarizations at 1480 nm wavelength. The experimental results indicate that low polarization dependence modulations are achieved around 1480 nm wavelength under 0 V/4 V operation. Eliminations of absorption contrast differences between TE and TM polarizations agree well with our theoretical predictions [19]. Our simulations showed that proper biaxial tensile strain broke the transition selection rules caused by spatial asymmetry of quantum well structure. The phenomenon can be summarized as the admixture effect between heavy hole and light hole states at the zone center, under certain crystal distortion. As a result, the Γe1-HH1 transition is significantly enhanced by introducing biaxial tensile strain. The maximum extinctions under 0 V/4 V operation are 9.3 and 7.9 dB for TE and TM polarizations correspondingly. Under 2 V reverse bias voltage, the dark current is 7.2 mA/cm², also indicating good crystal quality and a reverse break down voltage higher than 12 V. The energy per bit is estimated to be 131 fJ/bit with a leakage current of 131 pA under 2 V reverse bias voltage.

Figure 7 shows the simulated and experimental absorption spectra of the 0.72% biaxially tensile-strained 10/12 nm Ge/Si_0.19Ge_0.81 MQWs. The experimental absorption spectra are deduced from the photocurrents under 0 V voltage. We use an eight band k·p method to simulate the polarized momentum matrix elements and absorption spectra of the biaxially tensile-strained MQWs [19]. The simulation result agrees well with the experiments in terms of predicting the wavelength locations of the absorption edges. However, the experiments observe higher absorption slopes. Lock-in amplifier is used with 1 kHz chopping frequency during the measurement of the photocurrent.

Figure 7:

Absorption spectra of the 0.72% biaxially tensile-strained 10/12 nm Ge/Si_0.19Ge_0.81 multiple quantum wells (MQWs).

To achieve a 1550 nm wavelength polarization-insensitive electroabsorption modulation, the reverse bias voltage should be increased properly, and the Ge/SiGe MQWs epitaxy and structure design should be revised correspondingly to achieve polarization independent absorption under the desired bias voltage. It is achievable theoretically according to our previous simulation result and still requires more experimental results [19].

3.3 High frequency response

We use a 30 GHz vector network analyzer (VNA) to test the high frequency response of the device (see Supplementary material). Radio frequency signal from port 1 of VNA is applied to the device contacts through a bias-Tee and a GSG probe. N-contact area is 10 μm in length and 1 μm in width. With such small contact area, the voltage is uniform for the central active area. According to our simulation, the capacitance of the device is 172 μF which limits the bandwidth of the device at high modulation speed. The polarized light is modulated by the device and amplified by an optical amplifier. The output light of the OA is fed to a high frequency photodetector. The electrical signal of optical amplifier is fed to port 2 of VNA.

The frequency response of the system including the fabricated device, OA and PD is Rf(IM + OA + PD). In the calibration test, we replace the fabricated device by a 40 GHz lithium niobate intensity modulator and test the total frequency response Rf(IM₀ + OA + PD). The calibrated frequency response of the fabricated modulator is Rf(IM) = Rf(IM + OA + PD)/Rf(OA + PD) = Rf(IM + OA + PD)· Rf(IM₀)/Rf(IM₀ + OA + PD). The frequency response of the modulator used in the calibration test is from the user manual of the product.

Figure 8 shows the high frequency response of the fabricated device. The 3 dB bandwidth of the device under 0, 2 and 4 V reverse bias voltages are 7.4, 8.8 and 12.1 GHz for TE polarization. For TM polarization, the corresponding 3 dB bandwidths are 7.5, 8.7 and 11.4 GHz under 0, 2 and 4 V reverse bias. The experiments show that the fabricated Ge/SiGe MQW electroabsorption modulator has a good uniformity between TE and TM polarizations. The operating bandwidth is limited by the junction capacitance and resistance. The waveguide length is 900 μm with a capacitance of 172 μF according to our calculation. On the other side, compact isolation design is excluded to maintain the biaxial tensile strain of the suspended microbridge structure, resulting in parasitic capacitance. To increase the operating speed of the suspended microbridge device, the modulator structure, such as the taper waveguide coupler, active waveguide length, should be adapted to achieve a compact design and better isolation [34].

Figure 8:

The high frequency response of the fabricated device (normalized S₂₁ response): (a) TE polarization, (b) TM polarization.

4 Conclusions

We demonstrate a biaxially tensile-strained Ge/SiGe MQW electroabsorption modulator with low polarization dependence in aspects of extinction ratio, insertion loss and 3 dB bandwidth. Suspended microbridge structure is utilized to introduce biaxial tensile strain into the Ge/SiGe MQWs. The suspended microbridge structure allows us to precisely control the biaxial tensile strain value of the active region. The Raman tests show that 0.72–1.02% biaxial tensile strain is achievable in the cap layer of the Ge/SiGe MQWs material in this way. The fabricated device has 4.3 dB extinction ratio for 1478 nm for both TE and TM polarizations. Under 2 V reverse bias voltage, the measured 3 dB bandwidths for TE and TM polarizations are 8.8 and 8.7 GHz correspondingly. The proposed method enables precise control of the Ge/SiGe MQW electroabsorption devices and broadens the applications of Ge/SiGe MQW electroabsorption modulators for waveguide integrated polarization-independent circumstances.