1. Introduction

Silicon photonics has become a very attractive research area in the past decade due to the potential of monolithically integrating photonic devices with complementary-metal-oxide-semiconductor (CMOS) microelectronic circuits on this platform [1,2]. Such an integration approach is crucial for the successful realization of next generation low cost optical links for dataCOM and TeleCOM applications, and has further application potential in areas such as chemical and biochemical sensing. Recently interest is increasing in the integration of optical links into microprocessors to facilitate high performance and low-cost super computing [3]. Significant research effort in this area, has led to the demonstration of essential building-block components, including silicon based lasers [4–6], photodetectors [7–11], and modulators [12–17]. Among these, high speed silicon based modulators are critical components that have proved difficult to realize in practical devices. Owing to the weak electro-optical effect of silicon [20], most demonstrated waveguide based silicon modulators utilize the free carrier dispersion effect [12–16]. The fastest silicon modulator demonstrated uses this effect and operates at 40Gb/s [16] but has a limited extinction ratio of 1dB. To achieve an acceptable extinction ratio, the device is usually a few millimeters long and works at 6-10V reverse bias because of the weak free carrier effect. The power consumption of this type of device is in the order of a few hundred miliwalts. Recently, Liu et al demonstrated a waveguide integrated GeSi modulator based on the electro-absorption (EA) effect with 1.2GHz modulation speed [17,18]. The EA effect is known as the Franz-Keldysh (FK) effect in bulk semiconductors and the quantum-confined Stark effect (QCSE) in quantum-well (QW) structures [19].

To date all modulator work has been focused on submicron SOI waveguides. However, submicron SOI waveguides suffer from high fiber coupling loss, high polarization dependent loss, large waveguide birefringence and phase noise [10,11,21,22]. Some of the shortcomings of the existing submicron devices may become major obstacles for realizing next generation on-chip optical interconnects. Overcoming all of these challenges with submicron waveguide-integrated devices still requires substantial research effort. On the other hand, high performance optical devices based on 3µm SOI waveguides have been demonstrated recently and showed potential of providing more practical and easier solutions [10,21,22]. A high-speed modulator based on 3µm SOI waveguides is a key missing building block and can potentially provide a solution to practical future optical interconnect and communication systems. Since the free carrier effect is too weak to use owing to the small optical and carrier overlap, large core SOI waveguide-based modulators utilizing other electro-optical effects need to be considered. The EA effect in Ge/Si is a strong effect and is intrinsically an ultrafast process suitable for very high-speed modulation.

In this paper, we demonstrate a Ge EA modulator that can operate at 30GHz modulation speed. The device is integrated with single mode 3μm thick large core SOI waveguides [21,22], and uses a horizontally-oriented p-i-n structure fabricated in a deep-etched silicon trench. The Ge EA modulator section is butt-coupled to a deep-etched silicon waveguide, which transitions adiabatically to a shallow-etched 3μm single mode SOI waveguide [21,22] (see Fig. 1(a) ). The demonstrated device has a very compact active region of 1.0 × 45µm² and exhibits a strong EA effect in the wavelength range of 1570-1640nm. The total insertion loss of the device (including the transition loss) is measured as being between 2.5 and 5dB in the wavelength range of 1610-1640nm. Around 4-7.5dB extinction ratio has been achieved over this wavelength range. It is estimated that the transition loss between Si and Ge waveguides is around 1.3dB for a 1.0µm wide Ge ridge. The insertion loss induced by the Ge absorption from the modulator section is 1.2-3.8dB. Therefore, the Δα/α value of the reported device is estimated in the range of 2-3.3 with around 55kV/cm electric field. The 3dB bandwidth measurement shows that the device can operate at more than 30GHz under −4V_pp bias. A clear eye-opening at a transmission rate of 12.5Gbps demonstrates the capability of high-bit-rate large signal modulation.

 

Fig. 1 (a) Schematic view of the Ge EA modulator integrated with large core single mode SOI waveguide. (b) Cross-section views and optical modes of the deeply-etched Si waveguide region and (c) the Ge modulator region. The amplitude difference between two adjacent contours is 3dB.

Download Full Size | PDF

2. Device structure and measurement results

Figure 1(a) shows the schematic view (half structure at input end) of the demonstrated Ge EA modulator integrated with a large core single mode SOI waveguide. The device was fabricated on a six inch SOI wafer with 0.375µm thick buried oxide (BOX) and 3µm think epitaxial-Si layer. The process began with the formation of a Si deep-etched recess area with 0.3µm thick remaining Si slab for Ge selective-area epitaxial growth. A 100nm thin buffer Ge layer was selectively grown in the recess area at low temperature (400°C) followed by a 3.0µm thick Ge layer selectively grown at high temperature (670°C). The Ge was intentionally grown thicker than the silicon epitaxial layer to compensate for thickness reduction in a later chemical-mechanical polishing (CMP) step. The target Ge thickness is 2.7µm after CMP. The Ge waveguide was formed by etching a 2.4µm Ge ridge. The resulting Ge section was protected by photoresist with the Si sections exposed and the wafer was further etched to form the single mode Si waveguides and the deeply-etched adiabatic transition trench between the shallow ridge waveguide and the deep ridge waveguide (see Fig. 1(a)). Cross-sectional views of the deep-etched Si and Ge waveguides are shown in Figs. 1(b) and 1(c). The figures also show the optical mode profiles of the waveguides. From the mode distributions, it is seen that the mode centers of the two waveguides are matched well. With an optimized design, the light input from the left side Si waveguide can be butt-coupled to the Ge waveguide with a transition loss less than 0.15dB. After waveguide etching, the Ge waveguide was implanted with boron and phosphorous on the sidewalls and the Ge slab as depicted in Fig. 1(a) to form a horizontally-oriented p-i-n structure and p- and n-type ohmic contact areas. The metal contacts for both p and n were formed by depositing and patterning a Ti/Al metal stack on top of the heavily doped areas. Figure 2 shows the top and cross-sectional scanning electron microscopy (SEM) images of a fabricated device with 1.0μm Ge width. From the images, the Si taper and the deeply-etched trench can be clearly observed. We also measured the remaining Si and Ge slab thicknesses to be 0.23μm and 0.6μm, respectively. The beam propagation simulation predicts that the thicker Ge slab (compared to the 0.3μm target thickness) can impose 1.3dB additional transition loss to a device with a 1.0μm wide Ge ridge due to the mode mismatch. This is a correctable process and the insertion loss of the EA modulator is expected to show 1.2dB improvement after the correction.

 

Fig. 2 Top view and cross-sectional view SEM images of the fabricated Ge EA modulator.

Download Full Size | PDF

The EA effect of the device was studied by measuring the transmission spectrum for various bias voltages. Typical transmission spectra of the demonstrated waveguide based EA modulator device are shown in Fig. 3(a) . The Ge waveguide length is 45μm. The spectrum has been normalized to the response of a straight waveguide without the Ge section. The plot clearly illustrates the EA effect of the Ge waveguide under different reverse biases. As the reverse bias increases, the absorption edge of the device is tilted towards the longer wavelength range. A strong EA effect has been observed from the measurement results. Figure 3(b) shows the measured extinction ratio and insertion loss of the demonstrated device. In the wavelength range of 1600-1640nm, the device can achieve more than 4-7.5dB extinction ratio with a 2.5-5dB total insertion loss. The efficiency of the EA effect of an EA modulator can be measured by the figure of merit Δα/α, which is the ratio between extinction ratio and insertion loss of the device. The measured insertion loss includes the transition loss between Si and Ge waveguides, which is not related to the EA performance. Therefore, an estimation of the transition loss is necessary for a clear understanding of the real device performance.

 

Fig. 3 (a) Typical measured transmission spectra of the Ge EA modulator device. (b) Measured extinction ratio and insertion loss of the EA modulator device in the spectrum range of 1570-1640nm. The double arrow indicates the best working regime.

Download Full Size | PDF

The transmission loss can be estimated by measuring the device loss at longer wavelengths, for example, at 1640nm, where the absorption loss is small. From experimental measurement [18], the absorption coefficient of Ge at 1640nm is about 60cm⁻¹, which corresponds to 1.16dB absorption loss for a 45μm long device. The transition loss is hence estimated to be 1.34dB given that the total insertion loss at this wavelength is 2.5dB. The numerical simulation also confirms this estimate. For the device with a 1.0µm wide Ge ridge, the transition loss is calculated as being around 1.3dB due to the unintentionally fabricated thicker Ge slab (0.6µm versus the targeted value of <0.3µm). This calculation agrees well with the experimental data. Using this value for the transition loss, the excess loss induced by the Ge absorption is 1.2-3.8dB. Δα/α can be easily calculated and is in the range of 2-3.3 with around a 55kV/cm electric field, which is within the range of the experimental results from bulk material [18]. This figure of merit can be improved further (Δα/α >4) by using narrower devices with the cost of higher transition loss and lower modulation speed. Further, the working wavelength range of the device can be tailored by using GeSi material. For example, with a Si composition at 0.75%, the working wavelength can shift to around 1550nm [17].

High speed performance of the EA modulator was demonstrated by measuring the 3dB bandwidth and eye-diagrams at high transmission rates. An Agilent vector network analyzer was used to measure the 3dB bandwidth of the device. The high-speed signal and DC bias voltage were applied to the modulator device through a bias-tee and a high-speed probe. The output modulated light signal was coupled to a fiber, amplified using an L-band (1570-1620nm) erbium doped fiber amplifier (EDFA), then coupled into a commercial high-speed photodetector with 50GHz response speed. The converted RF signal was fed into the network analyzer and measured. The system was calibrated in advance to factor out the effect of the RF system, including the cable, the bias-tee, and the photodetector response. The measured frequency responses of the modulator device for various reverse biases are shown in Fig. 4(a) . With a −4V bias, the device can achieve 30GHz 3dB bandwidth. From a previous study [17], it is known that the 3dB bandwidth of EA modulators is only limited by the RC time constant of the device. Reducing the device capacitance is an effective way to achieve higher modulation speed. When increasing the reverse bias, the intrinsic region of the device increases due to depletion, therefore, the capacitance of the p-i-n junction is reduced. This makes the RC time constant smaller and the modulation speed higher. Such behavior can be observed from the measurement results shown in Fig. 4(a). Devices with shorter length and wider Ge p-i-n structure width can operate at higher speed, however they suffer from lower modulation efficiency (extinction ratio). We have clearly demonstrated that EA modulators using the FK effect can indeed operate at >30GHz speed with good (4-7.5dB) extinction ratio. This is comparable to the speed achieved by a MZ modulator which is much larger and has much higher power consumption [14,15].

 

Fig. 4 (a) Measured frequency response of a Ge EA modulator with dimension of 1.0 × 45µm. (b) Measured eye-diagrams for various wavelengths at 12.5Gbps transmission rate with −4V_pp bias. The dashed lines are the fitted curves.

Download Full Size | PDF

The eye-diagram measurement used a similar experimental setup. A pseudorandom binary sequence (PRBS) signal with (2²³-1) pattern length at a 12.5Gbps transmission rate was used for the measurement although the device can achieve much higher transmission rates. The PRBS signal was amplified by a commercial modulator driver with ~4V_pp. The signal was combined with −2V DC bias using a bias tee and applied to the modulator. The modulated light signal was amplified by an L-band EDFA and coupled into a commercial photodetector attached to a digital communication analyzer. The eye-diagrams at three different wavelengths, 1590, 1600 and 1610nm were measured and are shown in Fig. 4(b). Eye openings with >5dB extinction ratio were achieved with only 4V reverse bias over greater than 20nm wavelength range. With a 30GHz 3dB bandwidth, the device is capable of achieving a 40Gbps data transmission rate [15].

Note that the demonstrated device with an active area of 1.0μm x 45μm is estimated to have a low capacitance of only 25fF. A simple formula to calculate the average energy consumption per bit for the dynamic modulation is given by energy/bit = 1/4CV², where C is the junction capacitance, and V is the reverse bias voltage. It is calculated that the energy/bit for the demonstrated EA modulator at −4V bias is only 100fJ/bit. The energy consumption level of the reported device is much lower than for Mach-Zehnder type modulators [14,15]. With a 25Gbps data transmission rate, the dynamic power consumption is only 2.5mW under −4V bias. Additionally, the photo-generated currents at “on” and “off” states also contribute to the total power consumption. However, they are not considered in the estimation due to the dependence on the input optical power.

3. Conclusions

In conclusion, we have demonstrated a compact, high-speed, low driving voltage Ge EA modulator integrated with large core silicon-on-isolator (SOI) waveguides. The Ge EA modulator employed a horizontally-oriented p-i-n structure. The demonstrated device has a very compact active area of 1.0 × 45µm², shows a strong EA effect in the wavelength range of 1570-1640nm and has achieved 4-7.5dB ER with 2.5-5dB total IL, including the transition loss between Si and Ge waveguides, in the wavelength range of 1610-1640nm. A Δα/α value of 2-3.3 was achieved under −4V _pp bias. The 3dB bandwidth measurement shows that the device is capable of operating at more than 30GHz speed. A clear eye-opening at transmission rate of 12.5Gbps is demonstrated with −4V_pp bias. The device has a low dynamic energy consumption of 100fJ/bit and low power consumption only 2.5mW at 25Gbps transmission speed under −4V_pp bias. The device can be made to operate in the C-band wavelength range by using GeSi material with suitable Si composition. This type of Ge/Si EA modulator will be particularly useful for use in inter-chip optical interconnects where compact size and low energy consumption are highly desirable.