1. Introduction

Photonic switch is a fundamental device in integrated optoelectronics and a key component for wide bandwidth high-speed fiber-optic networks. Photonic switches can be implemented using X or Y-shaped optical waveguides. The Y-shaped optical waveguide switch consists of one input waveguide and two output waveguides. The input waveguide is connected to the output waveguides at a branching point with a small branching angle. The 1×2 Y-branching digital photonic splitting switch (DPS) has an advantage in that the output optical power of the switch is insensitive to voltage variations or drift in the applied controlling voltage. When an appropriate controlling voltage is applied to one of the two output waveguides, propagation of the light is allowed in one of the output optical waveguides and blocked in the other output optical waveguide. The optical output power in either of the output waveguides can also be controlled and modulated by an appropriate voltage.

Desirable properties of the 1×2 DPS are low insertion loss, low crosstalk, low switching voltage, high switching speed, good reliability, and capability for highly integrated implementation. With the use of modal analysis tools, the Y-shaped DPS exhibits low wavelength sensitivity, polarization insensitivity, and highly desirable digital response. Previous developments are based on the advanced materials such as lithium niobate (LiNbO₃) [1], polymer [2], and III–V compound semiconductors [3–5], etc. Much attention has also been paid to compact [6], low crosstalk [7], and voltage-length reduced [8] DPSs. However, demonstrated LiNbO₃ based DPSs require a relatively high operation voltage, whereas polymer-based switches exhibit very slow speed because of the use of thermo-optic effect and environmental instability. On the other hand, these devices using InP- and GaAs-based materials suffer from poor fabricating process and poor integration with Si-based chips.

To achieve good reliability and highly monolithic integration with Si-based chips, SiGe material was used to fabricate multifunctional and highly integrated photonic devices [9,10,15,16] because of low propagation loss (<0.5 dB/cm) in the wavelength region of λ=1.3 ~1.55 µm. SiGe epitaxy has the advantage that its fabrication techniques are compatible with large scale Si integration. Due to greater electron mobility and smaller bandgap in germanium, remarkable improvements in the performance of devices can be achieved, with virtually no changes in existing all-silicon designs. SiGe is also relatively easy to fabricate using existing silicon facilities.

DPS devices with a large number of inputs-outputs such as 1×N, N×1, N×M, and N×N block DPS arrays can be constructed. In this work, we report the first DPS fabricated in Si-based SiGe material. We have used an epitaxial SiGe layer as a core optical waveguide. The symmetrical 1×2 DPS was designed for free carrier injection for operation at 1.55 µm. It can be useful for dense wavelength division multiplexed (DWDM) networks and associated applications. To reduce the switching length, and at the same time to reduce the switching voltage and power consumption, two widened carrier injection regions are employed in the two output arms of the switch. The measured device performances are satisfactory.

2. Device description

Figure 1(a) shows a schematic diagram of the proposed symmetrical Y-branch DPS with two widened carrier injection regions and (b) its cross-section perpendicular to A-A. In the ridge waveguide, an impurity-induced index step at the p/p+ interface is used for vertical containment of the light. Lateral confinement is provided by the SiGe/SiO₂ ridge wall, and an effective index is induced by its geometry. Below the top surface of each branching waveguide, an abrupt n+p junction is made to inject the carriers into the waveguide when it is forward biased as shown in Fig. 1(b). The principle of the DPS is based on the plasma dispersion effect, in which the variation of the refractive index of SiGe is related with its carrier concentration. In the operation, the incident light with a wavelength of λ=1.55 µm will be split into the branches 1 and 2 of the Y-branch. When the n+p junction of the Y-branch is forward biased, a large number of electrons will be injected to the p-region of the waveguide causing an increase of carrier concentration. This will lead to a decrease of the refractive index of the SiGe layer at 1.55 µm which will be cut-off at branch 1 and/or branch 2.

 

Fig.1. Schematic diagram of a symmetrical waveguide Y-branch digital photonic splitting switch (a) top view and (b) A-A cross-section view. (not in scales)

Download Full Size | PDF

3. Simulation

The device parameters were determined by the following considerations. The thickness of the rib waveguide was chosen for the single-mode rib waveguide operation to be 2.5 µm, the widths of input rib waveguide, branch 1 and branch 2 were chosen to be 8 µm. To enhance the switch performance and therefore to reduce the switching power consumption, the widths of the two carrier injection regions were chosen to be 10 µm, but still it will support single mode operation. The separation of the two outputs waveguide at the ends is only 16 µm, permitting the use of only a single output fiber at one time. The etching depth of all ridge waveguides was 1.2 µm for Ge content of 4%. The effective index method was employed to analyze the behavior of the mode propagation and determine the branch angle of the switch. In our optimum design, the Y-branching angle of the switch is optimized to be 1.273 degree. The total length of the switch is 900 µm and the length of the active region is L _active=150 µm. The distance of the electrode to the facet of the output waveguide is 50 µm. The device performance is simulated by the Beam Propagation Method (BPM) and the results are shown in Fig. 2. From Fig. 2(a), we can see that for a Y-branch, the output optical powers at the two arms are 47.6% when a switch is ON for both the branches. This means the insertion loss, which is defined as -10log10(P_total-output/P_total-input), is 0.2 dB. For a DPS response as shown in Fig. 2(b), the output optical powers in the two arms are 58.7% and 3%, whenever a switch is ON for branch 2 and OFF for branch 1. Exactly similar results are obtained whenever the branch 2 is OFF and branch 1 is ON, as shown in Fig. 2(c). The calculated insertion loss is 2.1 dB and the crosstalk is -12.9 dB in both the cases.

 

Fig. 2. Simulation results of the switch using beam propagation method, (a) switch-ON state on both branches, (b) switch-ON state on branch 2 and OFF state on branch 1, (c) switch-ON state on branch 1 and OFF state on branch 2.

Download Full Size | PDF

4. Experiment

The samples were grown in an UHV-CVD system at 900°C. The substrate used is a p⁺-type Si (100) wafer. A~5 nm p-type Si buffer layer with a concentration of about 2×10¹⁶ cm^-3 had been grown followed by a 2.5 µm p-type strained Si0.96Ge0.04 core waveguide layer (2×10¹⁶ cm^-3). On the top of the sample, an abrupt n⁺p junction was formed by growing a 5 nm n⁺-type Si with a concentration of 1×10¹⁸ cm^-3. The device was fabricated by a standard Si device technology. The ridge waveguides were shaped by lithography and reactive ion etching. Then, the n⁺ and p⁺ ohmic contact electrodes were deposited by evaporating 1.0 µm thick Al layer followed by alloying at 450°C. The whole device surface was passivated by a SiO₂ layer. The chips were diced and the input and output facets of the switch were polished. The incident light from a single-mode fiber was coupled to the waveguide.

5. Results and discussion

A GaInAsP laser diode with a wavelength of 1.55 µm was used as light source and was coupled into the single-mode input waveguide directly. To improve mode matching and coupling efficiency, the pigtailed single-mode fiber was specially designed with a tapered and lensed end-face with an antireflection coating. The tapered length is 37 mm, beam diameter is 2.5 µm, and the working distance is 6 µm. The magnified near-field radiation patterns at output branches 1 and 2 were imaged by an infrared vidicon. Without applying modulation voltage on both n⁺p junctions, two bright output patterns of 1.55 µm were observed from branch 1 and 2, respectively. With the branch 1 at forward bias (active state) while the branch 2 at zero bias (passive state), the output optical power of branch 1 decreases and the output optical power of branch 2 increases with the applied voltage at branch 1. Figure 3 shows the output optical power versus the applied forward bias. It illustrate that the switch operates as an on/off device at a threshold voltage of 1.0 V and a corresponding threshold current of 85 mA on one of the output waveguide arms. In this situation, the calculated insertion loss is 3.2 dB. In this work, the modal mismatch loss is ignored because the beam size of 2.5 µm from the tapered single-mode fiber matches the SiGe core layer thickness of 2.5 µm. From Fig. 3, we can further see that when the forward bias reaches 2 V, the active branch was totally cut-off. When the branch 1 is at zero bias while the branch 2 is at forward bias, the output optical power of branch 2 decreases and the output optical power of branch 1 increases with the applied voltage on branch 2. The totally cut-off voltage of the light at branch 2 is also 2 V. Under this condition, the injection current density is 6.3 kA/cm² and the power consumption is 190 mW. The measured insertion loss is 5.2 dB, and the crosstalk is -9.6 dB. It should be emphasized that during the experiment, the thermo-optical effects were negligible because the device was mounted on a copper heat sink and the temperature of the device during the measurement was continuously monitored using a temperature controller.

This device is also a splitter combined with a blocking attenuator on each arm. As a result there is a minimum 3 dB loss per switch. The injection current density of 5–7 kA/cm² in this work of the SiGe DPS is smaller than the reported of 9 kA/cm² by Liu et al. [11] and 22 kA/cm² in InAgAsP/InP carrier injection-based PDS by Abdalla et al. [5]. In comparison to other optoelectronic material DPSs reported in the literature, the digital response power of 85 mW and cut-off power of 190 mW in this work for SiGe material is higher than those previously reported by Vinchant et al. (60 mW in GaInAsP/InP by carrier injection) [3] and Keil et al. (45 mW in polymer by thermo-optic effect) [12]. Our results are similar to that reported by Yang et al. (165–180 mW in polymer using the thermo-optic effect) [7] and superior than those reported by Siebel et al. (200 mW in polymer by thermo-optic effect) [2], Abdalla et al. (800 mW in InGaAsP/InP by carrier injection) [5], Toyoda et al. (380 mW in silicone resin by thermo-optic effect) [13], and Lee et al. (250 mW in polymer by thermo-optic effect) [14]. The results show that the proposed DPS is a promising device to fulfill monolithic integration with other Si-based optoelectronic devices and for 1.31 to 1.55 µm fiber-optic communication and networking applications.

 

Fig. 3. Output optical power versus the applied forward bias.

Download Full Size | PDF

6. Conclusions

A symmetrical 1×2 digital photonic splitting switch has been fabricated in SiGe materials by UHV-CVD growth. The operation of the switch is found to be optimized at 2 V forward bias. The measured insertion loss is around 5.2 dB, and the crosstalk is -9.6 dB.