1. Introduction

Silicon photonics is an emerging technology that targets fabrication of photonic and electronic components on a complementary-metal-oxide-semiconductor (CMOS) compatible platform to take full advantage of the mature technologies available in the silicon semiconductor industry [1–3]. Ever increasing super-computing requirements have stimulated the development of silicon photonics in recent years [4], and one of the key components for the realization of silicon photonic links for these applications is the silicon-based modulator. Although significant progress has been achieved [5–12], silicon-based modulators remain challenging compared to other components on this platform. Owing to the weak electro-optical effect of silicon [13], most silicon modulator work utilizes the free carrier dispersion effect [5–9]. A practical device is usually a few millimeters long and requires 6-10V reverse bias. The power consumption of such a device is in the order of a few hundred milliwalts. Recently, Liu et al demonstrated that the electro-absorption (EA) effect in Ge/Si material is strong enough to provide a good modulation depth with much less power consumption [10-11]. However, all of the demonstrated modulator work so far has been focused on submicron waveguides, which still face some challenges that require substantial research effort to conquer, such as high fiber coupling loss and high polarization dependent loss (PDL). On the other hand, high performance optical devices based on 3µm SOI waveguides have been successfully demonstrated recently [14–17], but a high-speed modulator is still a missing element in this platform until our recent work on Ge EA modulators [18]. The free-carrier effect is too weak to use in the 3μm platform, but utilization of EA effect is one potential solution. The EA effect in Ge/Si, also known as the Franz-Keldysh (FK) effect for bulk material, is a strong effect that is intrinsically ultrafast and so is suitable for very high-speed modulation. Due to the indirect bandgap nature of the Ge (or GeSi), the strong EA effect in this material comes with high background absorption loss. The main target of the device design is to fully utilize material properties to achieve optimal performance with a feasible fabrication process.

Recently, we successfully demonstrated a Ge EA modulator capable of operating at 30GHz speed [18]. In this paper, we present the detailed design and fabrication of such 3µm SOI waveguide-integrated Ge EA modulator. The modulator uses a horizontally-oriented p-i-n Ge structure butt-coupled with a deep-etched silicon waveguide [18]. Both optical and electrical performances were optimized to ensure the best overall device performance. The design focused on achieving low transition loss from Ge to Si waveguides and balancing the trade-off between the insertion loss and the modulation depth of the device. Process simulations were performed to provide a feasible process flow for device fabrication, and a device was fabricated and measured to verify the design. Under a −4V_pp reverse bias, the device had a total insertion loss (IL) of 2.7-5.2dB and an ER of 4.9-8.2dB over the wavelength range of 1610-1640nm. The Δα/α value, which provides a measure of overall modulator performance, was estimated as being between 2.2 and 3.2, after taking removing the contribution of the transition loss to the total IL. The measurement results have confirmed that an EA modulator based on the FK effect is a feasible solution to realize high performance modulators using a CMOS compatible fabrication process.

2. Device structure and optical design of the EA modulator

A schematic view of the waveguide-integrated Ge EA modulator (half structure) is illustrated in Fig. 1 (a) . The device was fabricated on a six inch SOI wafer with 0.375µm thick buried oxide (BOX) and 3µm thick epitaxial-Si layer. The 2.7μm high Ge waveguide was grown in a deep-etched recess trench with 0.3μm thick Si remaining. The Ge waveguide width is designed to be 1.0μm to have better fabrication tolerance to a low-loss transition between the Ge and Si waveguides. A 2.4μm deep Ge ridge waveguide was fabricated with a 0.3μm remaining Ge slab sitting on top of the Si slab as shown in Fig. 1(a). The cross-sectional views of the deep-etched Si and Ge waveguides are sketched in Fig. 1(b)-(d). The figures also illustrate the optical modes of the Si and the Ge waveguides (with two Ge slab thicknesses, 0.3 and 0.6μm). From the optical mode profiles shown in Fig. 1(b) and (c), it can be observed that although the Ge waveguide has a 0.3μm additional slab on top of the Si slab, the optical mode centers of the Si and Ge waveguides have very little offset. This indicates that the transition between these waveguides will be smooth because the excitation of higher order modes is highly suppressed. However, for the thicker Ge slab case, the mode center shows a significant shift downwards and the mode is pushed into the Ge slab as shown in Fig. 1(d), where a 0.6μm Ge slab case is shown. Significant transition loss is expected from devices with thicker Ge slabs.

 

Fig. 1 (a) Schematic view of the Ge EA modulator integrated with large core single mode SOI waveguide. (b)-(d) Cross-section views and optical mode profiles of the Si and the Ge waveguides with (c) 0.3μm and (d) 0.6μm Ge slab thicknesses. The amplitude difference between two adjacent contours is 3dB.

Download Full Size | PDF

A series of simulations were carried out to study the impact of the Ge slab thickness on the transition loss of the device with various Ge waveguide widths and slab thicknesses. The results are shown in Fig. 2(a) . The observation is that there is a cut-off Ge width for each different Ge slab thickness. Devices can have significant transition loss when their widths are narrower than the cut-off widths. The simulation shows that the cut-off width is smaller for the devices with thinner Ge slab remaining. For example, the device with 0.3μm Ge slab has a cut-off width of 0.7μm. Hence, devices with width> 0.7μm have negligible transition loss. However, with a 0.6μm thick Ge slab, a device with 1.0μm width can have about 1.3dB transition loss according to the simulation. A physical insight into the transition mechanism can be gained from the beam propagation simulation shown in Fig. 2(b) and (c). The optical mode evolution was simulated by the beam propagation method (BPM) for devices with 0.3μm and 0.6μm thick Ge slab, respectively. For the thin Ge slab case, due to good mode matching, negligible higher order modes are excited and this results in a very smooth transition with less than 0.15dB transition loss. For devices with thicker Ge slabs, significant power can be coupled to higher order modes which will be lost. As a result, the transition loss (two interfaces) increases to 1.3dB.

 

Fig. 2 (a) Simulated transition loss between Ge and Si waveguides excluding Ge absorption. (b) Beam propagation simulation of the devices with 0.3μm and (c) 0.6μm thick Ge slab, respectively.

Download Full Size | PDF

3. Device performance and electrical design of the EA modulator

The device design began with process-based simulations to ensure a feasible fabrication process could be created to achieve the design targets. In the insert of Fig. 3(a) , the simulated net doping concentration of boron (p+ type) and phosphorus (n+ type) after the ion-implantation and rapid thermal anneal (RTA) steps is illustrated for a device with a 1.0μm wide Ge ridge. An observation is that the doping concentration along the vertical direction is very uniform. A cutline plot (A-A’) in the middle of the Ge waveguide center is shown in Fig. 3(a) to illustrate the horizontal doping profiles. Due to a fast diffusion rate in Ge material, the phosphorus doping concentration attains a more box-like profile after the RTA step, which is desirable for achieving a more uniform electric field when applying a voltage to the p-i-n structure. Boron, however, does not have this unique property and the final doping profile retains a Gaussian shape with a doping concentration tail extending into the Ge intrinsic region. In spite of this, the final doping profile shown in Fig. 3(a) indicates a 0.85μm wide intrinsic Ge region within the 1.0μm wide Ge ridge. A smaller heavily-doped region is very desirable because it minimizes the free-carrier loss and more importantly minimizes the reduction of modulation efficiency caused by the unused electric-optical overlap region.

 

Fig. 3 (a) Simulated doping profiles for side-wall boron and phosphorus doping in Ge. (b) Simulation results of the electric field profiles for the device under reverse biases.

Download Full Size | PDF

After the process simulation, device modeling was carried out. A typical simulated electric field distribution for the device under −4V reverse bias is shown in the insert of Fig. 3(b). Along the cutline (B-B’), the electric field profiles along the horizontal direction of the Ge p-i-n structure under various reverse biases are plotted in Fig. 3(b). At 0V bias, a built-in electric field around 5kV/cm is observed. From this, we can calculate the forward turn-on voltage of the p-i-n junction as 0.35V. This is very close to the experimental result. When increasing the reverse bias, the electric field inside the intrinsic region increases. Under −4V bias, the electric field in the intrinsic region reaches about 55kV/cm with a very uniform distribution along both vertical and horizontal directions. Simultaneously, the depletion region extends farther to the side walls and leads to an increased intrinsic Ge region. This effect can further improve the device performance: 1) it increases the effective electrical-optical overlap and reduces the free-carrier loss from the doped areas, hence improving the modulation efficiency; 2) it decreases the junction capacitance and the 3dB bandwidth of the device. The junction capacitance is estimated to be about 25fF for a 45μm long device. With such a small capacitance, greater than 50GHz 3dB bandwidth is achievable due to the extremely small RC time constant.

Using the data shown in Fig. 3 and the material model provided in [11], we can calculate the device absorption loss induced by the electric field. The simulated transmission of the device as function of Ge length under 0V and −4V bias conditions is shown in Fig. 4(a) . The simulation excludes the transition loss to avoid complicating the situation. The simulation is carried out at 1610nm wavelength. From the model, the IL and ER of the device can be calculated and the results are plotted in Fig. 4(b). The figure also demonstrates the approach we used for the device design. To illustrate this, we set the design targets as: IL<4dB, and ER>6dB, with −4V bias. The blue and yellow areas in the figure represent the targeted working regimes satisfying either the IL or ER requirements, respectively. It is straightforward to see that the overlap of these regions gives the working regime that satisfies both design targets. For this example, we found out devices with length ranges between 35 and 55μm can provide IL 2.8-4dB and ER 6-9dB at a wavelength of 1610nm.

 

Fig. 4 (a) Simulated transmission of the device versus device length under two bias conditions. (b) Simulated insertion loss and extinction ratio of the device under −4V bias.

Download Full Size | PDF

4. Measurement results of the fabricated EA modulator

A Ge EA modulator with active area of 1.0×45μm² was fabricated using the optimized process steps described above. The active performance of the device was measured to verify our design. Figure 5 shows the measured IL and ER of the device under different biases. The results reveal a strong EA effect induced by the applied bias. The device has an IL of 2.7-5.2dB and an ER of 5-8.2dB under −4V bias in the wavelength range of 1610-1640nm. From a scanning electron microscopy (SEM) image, we measured the Ge slab thickness and found it to be approximately 0.6μm. According to the simulation, the transition loss for this device is 1.3dB. Excluding the transition loss from the total IL, we can estimate the Δα/α value of the fabricated device, where Δα and α represent the bias induced absorption change and the background absorption without bias. The results are shown in Fig. 5 (b). Δα/α value ranging between 2.2 and 3.2 have been achieved with an electric field of around 55kV/cm. To verify our design, the IL of the device without transition loss is estimat ed as being 3.9dB and the ER is 8dB. Both figures are close to the simulation results of 3.5dB and 7.5dB, respectively.

 

Fig. 5 (a) Measured extinction ratio and insertion loss of a Ge EA modulator device with 1.0μm width in the spectral range of 1570-1640nm. (b) Estimation of Δα/α value for the device. The double arrow indicates the best working regime. The insert shows the SEM image of the fabricated device.

Download Full Size | PDF

5. Conclusions

We presented design and fabrication of a waveguide-integrated Ge EA modulator on the 3µm SOI platform. The device is optimized both optically and electrically for operating over a wavelength range of 1610-1640nm. The design takes the fabrication process into account to target a feasible solution to this type of modulator device. Measurements were made on a fabricated device to verify the design. Under a −4V_pp reverse bias, the device demonstrated a total IL of 2.7-5.2dB and an ER of 4.9-8.2dB over the operating wavelength range. The Δα/α value was estimated to be between 2.2 and 3.2 with an electric field around 55kV/cm. We have proved that high performance silicon waveguide modulators can be realized by using the strong FK effect in Ge/Si material.