1. Introduction

Recent demand for low power consumption as well as high capacity transmission in photonic network systems necessitates the application of optical cross-connect systems; large-scale space-division optical matrix switches are key components in these optical cross-connect systems. Large-scale matrix switches composed of 2 × 2 Mach–Zehnder interferometer (MZI) optical switches with silica-based planar lightwave circuits (PLCs) and employing a thermo-optic effect have been investigated in order to exploit their advantages of low insertion loss, long-term stability, and high reliability [1–4].

The increase in the number of channels in a matrix switch necessitates the achievement of low crosstalk in a 2 × 2 element switch. Goh et al. proposed and demonstrated a double-MZI configuration to ensure low crosstalk levels of around –60 dB for the ‘cross’ state of the switch with silica-based PLCs (where an element switch operates as a cross/single-bar switch) [1].

As the thermo-optic coefficient of silicon (Δn/ΔT = 1.86 × 10⁻⁴ /K) is greater than that of silica, silicon photonics is a promising platform that can be employed to design optical switches with high integration and low power consumption [5–10]. However, for silicon waveguides, even a small dimensional error on the order of 10 nm in the fabrication immediately causes nonzero crosstalk of the MZI. Therefore, the approach in Ref. [1]. is important, because it entails the construction of a switch ensuring a good crosstalk performance from such imperfect MZIs. The objective of the present study is to implement the cascade connection in silicon MZIs for improving the crosstalk characteristics, which, to the best of our knowledge, has not yet been reported.

2. Device structure and fabrication

We consider a 2 × 2 array of MZIs shown schematically in Fig. 1 . Figure 1(a) and 1(b) shows its low-crosstalk operation for the ‘bar’ and ‘cross’ switching states, respectively. The element switch is an MZI with 3-dB directional couplers (DCs) and thermo-optic phase shifters in the arms, as shown in Fig. 1(c). Output ports of the first element switches (SW1, SW2) are connected to input ports of the second element switches (SW3, SW4) through an intersection or straight path. This switch connection can be regarded as a generalization of that implemented in Ref. [1]. In the present study, we assume that the top and bottom ports are for the input/output ports and that all the four element switches are set to either the ‘bar’ state or the ‘cross’ state. It should be noted that the optical signals from the two input ports never pass a common element switch. This 2 × 2 switch gives a bar or cross connection when the element switches are all set to the ‘bar’ state or the ‘cross’ state, as shown in Fig. 1(a) and 1(b), respectively. If the element switch has crosstalk, then leakage occurs at the other exit, where the presence of the signal is unintended. Let the leakage level in an element switch be –x ₁ dB and –x ₂ dB below the level on the main path for the ‘bar’ and ‘cross’ states, respectively, as shown in Fig. 1. For the ‘bar’ state, the leaked component in SW1 is further attenuated by –x ₁ dB in SW4, and it reaches Output-2 with a level of –2x ₁ dB. Similarly, for the ‘cross’ state, the leakage from Input-2 to Output-2 should have the level of –2x ₂ dB. It should be noted that the crosstalk performance of the intersection is also important, particularly for the cross-connection, as will be discussed later. In general, it is difficult to achieve an intersection of Si-wire waveguide with low crosstalk, because high-index contrast waveguides have strong diffraction [11–14]. In this study, we employ a vertically aligned DC as shown in Fig. 1(d). The optical coupling from In-A to Out-A and from In-B to Out-B should be inherently weak. High isolation should be maintained between In-A and Out-A and between In-B and Out-B irrespective of the deviation of the coupling ratio from 0 dB and of the operation wavelength.

 

Fig. 1 Schematic configurations of low-crosstalk 2 × 2 optical switch with MZI array as (a) ‘bar’ and (b) ‘cross’ states. Solid arrows show the ideal path of optical signals and dashed arrows show incident crosstalk in each component with a level of –x ₁ (dB), –x ₂ (dB), or –x ₃ (dB) below the power on the main path. (c) Element switch of MZI with thermo-optic phase shifter. (d) Intersection of vertically aligned directional coupler (DC).

Download Full Size | PDF

Figure 2(a) shows the microscopic image of the fabricated 2 × 2 thermo-optical switch. Si-wire waveguides having a cross-section of 450 nm × 220 nm were fabricated by electron-beam lithography and reactive ion etching on a Unibond silicon-on-insulator wafer with a 3-μm-thick oxide layer. Then, 40-μm-long thin-metal heaters with Pt were formed on a 2.5-μm-thick SiO₂ cover layer at the MZI arms, as shown in Fig. 2(b). The footprint of the MZI is 200 × 200 μm², which is much more compact than that with silica-based PLCs. Each end of the waveguide had a spot-size converter with a narrowed waveguide for field enlargement [15].

 

Fig. 2 (a) Microscopic image of fabricated 2 × 2 optical switch. (b) Schematic image of cross section of Si-wire waveguide with thin-metal heater.

Download Full Size | PDF

3. Characterization

The transmittance of the fabricated device was measured with a tunable laser diode having an operation wavelength range of 1520–1630 nm. The output of the laser diode was coupled to a focusing lens module. The module has a polarizer with a purity of over 40 dB, and it sets the launched polarization to a transverse electric-like (TE) mode. The output light through the device was coupled to a lensed fiber in order to measure the transmitted power with a photodiode. A propagation loss of 8 dB/cm and a coupling loss of 4.5 dB/facet between the lens fibers were measured for the single waveguide. Each MZI was set to either the ‘cross’ or the ‘bar’ state by applying electrical current to the heater.

Figure 3(a) shows measured results of the element switch of a single MZI, which was fabricated in the same sample as a reference. The power consumption and response time of the element switch were 40 mW and 30 μs, respectively. The insertion loss was 16 dB. Theexcess loss of the element switch itself was then estimated to be 4 dB. On the same sample, we also laid a stand-alone intersection pattern for evaluation. Figure 3(b) shows the measured transmittances of the fabricated intersection from In-A to Out-A and Out-B. Figure 3(c) shows measured results of the 2 × 2 optical switch with an MZI array from Input-1 and Input-2 to Output-2 for each state. The insertion loss, including excess losses through the two element switches, was 20 dB.

 

Fig. 3 Measured results of (a) element switch of single MZI, (b) intersection, and (c) 2 × 2 optical switch with MZI array. Input and output ports are defined in Fig. 1. Solid and dashed lines show transmittances to diagonal and straightforward ports, respectively. The color of lines corresponds to the signal paths in Fig. 1.

Download Full Size | PDF

We define the crosstalk level as the difference between the transmitted power level of the main input signal and the leakage level of the other input signal. Figure 4 shows the crosstalk levels calculated from the measured results with smooth polynomial curves that fit the data. The lowest crosstalk levels for the single MZI switch were –25 dB (at wavelength of 1550 nm) and –20 dB (1570 nm) for the ‘bar’ and ‘cross’ states, respectively. The crosstalk of an MZI is associated with the deviation in the behavior of the two splitters from that of an ideal 3-dB coupler. For Si-wire waveguides, an error of 10 nm in the waveguide width causes the splitting ratio shift 3 to 4%. Then, the MZI can have crosstalk no better than –25 dB. When the deviations in two couplers are identical, power imbalance due to the first coupler is nullified by the second coupler for the transmission to the diagonal port. This is the reason why we observe better crosstalk in the bar state. The lowest crosstalk level of the intersection was –35 dB (1560 nm), presumably resulting from the back-scatterings due to the sidewall roughness in the waveguides [16]. The lowest crosstalk levels of the switch with an MZI array were –50 dB (1550 nm) and –30 dB (1570 nm) for the ‘bar’ and ‘cross’ states, respectively. For the bar state, the figure of –50 dB is two times that for a single MZI (–25 dB). This improvement is in good agreement with the crosstalk levels expected from the present design. On the other hand, for the ‘cross’ state, the figure of –30 dB falls short of target (–40 dB). This degradation is due to the intersection. If some leakage occurs at the intersection (–x ₃ dB), it is transmitted through the signal path, as shown in Fig. 1(b), and cannot be eliminated in SW4. The observed result indicates that the leakage from the intersection limits the crosstalk for the ‘cross’ state.

 

Fig. 4 Crosstalk performance calculated from the measured results for (a) ‘bar’ state, (b) ‘cross’ state, and intersection. The gray lines show polynomial curves that fit the data.

Download Full Size | PDF

We emphasize that the crosstalk characteristics were certainly improved by using the cascade connection. Our study also validated the importance of the intersection. In addition, there is a fair prospect that the latest fabrication processes of silicon-wire waveguides, which reduce the sidewall roughness, would make the vertically aligned DC a more ideal intersection.

4. Conclusion

We have demonstrated a low-crosstalk 2 × 2 thermo-optic switch with silicon wire waveguides using a 2 × 2 array of MZIs. A vertically aligned DC was employed as an intersection; this was beneficial in improving the crosstalk characteristics by means of a cascade connection. Lowest crosstalk levels of –50 dB and –30 dB were obtained for the ‘bar’ and ‘cross’ states, respectively; these values were noticeably lower than those in the case of a single MZI. The power consumption of one MZI element switch was 40 mW, and the total power consumption of the device was at most 160 mW. This is still much lower than that of single-MZI switch with silica-based PLCs. This can be further reduced by forming heat-insulating grooves beside the heaters [10].