1. Introduction

Controlling polarization state of light in photonic integrated circuits (PICs) is of great importance for high-speed optical communication networks. First, the single mode fiber used in communication networks does not preserve the polarization state and many components in PICs are polarization-sensitive. This polarization dependence along with the polarization mode dispersion degrades the performances of PICs at high modulation frequency. To solve the problem, polarization diversity systems were proposed [1–3]. Second, the polarization state can be utilized in polarization-division multiplexing (PDM) systems to double the spectral efficiency [4]. For the design of either the polarization diversity systems or the polarization multiplexing systems, a polarization splitter is the key component. A polarization splitter built on an Indium Phosphide (InP) substrate is especially of interest due to its capability of integration with active components including lasers, modulators, and photo-detectors [5,6].

A Multimode Interference (MMI) coupler is an excellent building block for polarization splitters due to its compactness, large bandwidth, and high fabrication tolerance [7]. A polarization splitter using two MMI 3-dB couplers as a Mach-Zehnder interferometer was proposed and demonstrated [8]. In this case, one of the arms is loaded with a dielectric layer and a metal layer on top, and the total device length was 3.3mm. An alternative method is to use an MMI itself as a polarization splitter, but it is very difficult to design an MMI-based polarization splitter on InP for two reasons: MMI couplers are inherently polarization-insensitive regardless of material systems, and strong polarization birefringence is difficult to achieve in InP material systems due to small index contrast between core and cladding layers compared to Silicon-On-Insulator (SOI) systems. As a result of the weak polarization birefringence, a long device length, e.g. a few millimeters long [9], is required for the MMI-based polarization splitters to separate TE and TM modes. Different methods such as the quasi-state MMI coupler [10] and the slot waveguide [11] have been proposed to reduce the total device length. However, none of these methods can reduce the device length without compromising the device performance or inducing extra fabrication difficulties.

In this article, we propose a novel design of a polarization splitter utilizing a 1-by-2 MMI coupler. A short phase shift section loaded with dielectric and patterned metal layers is incorporated into the MMI-based polarization splitter. This creates a strong polarization birefringence between TE and TM modes, resulting in very strong separation of TE and TM modes at the output waveguides. We also introduce a tilt in MMI, to adjust the phases both for TE and TM such that they are exactly at ± π/2 with opposite signs. The device was carefully optimized to satisfy good performance over the entire C-band (1528-1567nm) and also tolerate reasonable variations in the fabrication process.

2. Principle and design

The functional block diagram of the device is shown in Fig. 1 . The input signal (TE or TM), coupled into the 1 × 2 MMI, is split into two arms with equal phase and equal power. The phase shift section is designed to add an extra –π/2 and π/2 phase shift to the TE and TM modes, respectively, in the lower arm. When the electric fields from both arms are combined via the 2 × 2 MMI coupler, the electric field in one output coming from the cross arm has an extra –π/2 phase shift compared with that from the bar arm. The interference between electric fields with different phases causes the TE polarization mode to appear at the lower output whereas the TM polarization mode appears at the upper output.

 

Fig. 1 Functional block diagram of a polarization splitter based on MMI couplers. The arrows indicate the phases of the electric field.

Download Full Size | PDF

To create the birefringence needed for the phase shift, a waveguide loaded with a dielectric layer and metal layer is considered [12]. We conducted a 2D simulation on the effective refractive index n_eff of the structure consisting of 0.2 μm-thick gold layer, a SiN_x dielectric layer with varying thickness, a 0.6 μm-thick InGaAsP (λ = 1.3 μm) layer as waveguide core, and an InP substrate. Figure 2 shows Δn_eff (the difference between the n_eff with and without the 0.2 μm-thick gold layer) as a function of the SiN_x dielectric layer thickness, for TE and TM modes. The gold layer in the phase shift section tends to push the electric field of TE mode away from it, which reduces the propagation constant for TE mode, creating negative Δn_eff. On the other hand, the gold layer pulls the electric field of TM mode toward it, which increases the propagation constant for TM mode. As shown in Fig. 2, by reducing the SiN_x thickness, a stronger birefringence can be created. At the same time, a small change in the SiN_x thickness can cause a large change in the phase shift, so there is a trade-off. We chose 0.08 μm to balance the length requirement and sensitivity to the thickness variation, as will be discussed later. The propagation losses for TE and TM mode are 9.7 cm⁻¹ and 11.8 cm⁻¹, respectively, when the SiN_x thickness was 0.08 μm. In other words, a SiN_x dielectric layer plays an important role of balancing between birefringence and absorption loss. Without it, TM mode is strongly localized at the interface as a surface plasmon polariton and absorption coefficient becomes very high.

 

Fig. 2 2D simulation of Δn_eff , i.e, effective index difference with and without the gold layer, as a function of the SiN_x layer thickness

Download Full Size | PDF

The phase shift section described in Fig. 1 is achieved by using a tilted MMI coupler partially covered by a gold layer on a SiN_x layer as shown in Fig. 3(a) . In this design, the phase shift section (S₂ and S₃) is integrated into the 1 × 2 MMI and 2 × 2 MMI coupler, which helps to reduce the total device length. The partial gold layer in the phase shift section creates an extra negative phase shift to the TE mode and an extra positive phase shift to the TM mode in the lower arm. The length of the phase shift section is adjusted to give the TE and TM phase shift difference to be π. Furthermore, the tilt angle in the MMI joint [13], which gives the same amount of phase shift both for TE and TM modes, is adjusted, so that the phase shift for TE and TM modes become –π/2 and π/2, respectively.

 

Fig. 3 (a) The schematic top view of the proposed MMI-based polarization splitter and (b) the cross section view of the phase shift section (S2 and S3) with metal-dielectric cladding layer.

Download Full Size | PDF

The geometrical parameters of the design are described as follows: the input waveguide has a width of W_input = 4.5 µm. The MMI section is composed of four sections, S₁, S₂, S₃, and S₄. The S₁ and S₄ sections do not contain the deposited gold layer, whereas the lower parts of the S₂ and S₃ sections are covered by deposited gold. The S₂ and S₃ sections are joined by an angled tilt of 0.45 degree. The MMI section has a width of W_MMI = 10 µm and a total length of L = 538 µm. The metal (gold) layer has a width of W_m = 4.5 µm and a length of L_m = 72 µm in total. The lengths of the S₁, S₂, S₃, and S₄ sections are 1/5L-1/2L_m, 1/2L_m, 1/2L_m, and 4/5L-1/2L_m, respectively. The upper output has a width of W_TM = 4.5 µm and is placed 3 µm from the center of the MMI coupler. The first section of the lower output has a width of W_con = 4.9 µm and a length of L_con = 47 µm; the second section of the lower output has a width of W_TE = 4.45 µm. The proposed device has an InP substrate, a 0.6 μm-thick InGaAsP (λ = 1.3 μm) layer as waveguide core, and 0.08μm-thick Silicon Nitride (SiN_x) layer as buffer layer. The gold layer is 0.2 μm thick. The cross section view of the phase shift section is shown in Fig. 3(b). A 10 μm-long polarizer, where a gold layer is directly deposited on InGaAsP is also included in the lower output waveguide, to absorb residual TM mode, while TE mode is passed with minimal loss. This gives about 7dB PER improvement for a TM Input. The total device length of less than 600 μm is much shorter than 1050 μm with InP MMI-based polarization splitter reported in [10].

3. Simulated performance and fabrication tolerance

The performance of the polarization splitter is simulated using commercial software FIMMWAVE employing the eigenmode expansion method [14]. The finite element mode solver is used for solving modes in all sections. Figure 4 shows the interference patterns of the MMI-based polarization splitter excited by (a) fundamental TE mode (TE₀) and (b) fundamental TM mode (TM₀) inputs. In this figure, the tilted MMI coupler looks like a straight waveguide, but it is because the tilt angle is small to be visible. As predicted, the single self-images for TE and TM modes build constructively near the upper and lower outputs, respectively. However, the self-image plane for TM polarization, Z_TM, does not coincide with the self-image plane for TE polarization, Z_TE. When the upper and lower outputs are placed at the TM self-image plane, an accurate TE self-image cannot be obtained at the lower output, which reduces the transmission. To solve this problem, a two-section TE output waveguide is designed. As shown in Fig. 4(b), the first section of the lower output waveguide functions as a mode converter, refocusing the divergent electric field of the TE mode back into a self-image, which was realized by adjusting the phase difference of the different MMI waveguide propagation modes.

 

Fig. 4 The electric field intensity in the MMI-based polarization splitter showing the wave propagation for (a) TM mode and (b) TE mode inputs.

Download Full Size | PDF

Figure 5 shows the simulation result of polarization extinction ratio (PER) and insertion loss for TE and TM modes as functions of wavelength from 1528 to 1567 nm, covering the entire C band. For both TE and TM polarizations, the polarization splitter of this design exhibits an insertion loss smaller than 0.7 dB and PER larger than 23 dB over the wavelength range of 39 nm. This shows that one of the advantages of MMI, i.e., relative insensitivity to wavelength, is still intact.

 

Fig. 5 Simulation result on (a) PER and (b) Insertion loss for TE and TM modes as a function of wavelength.

Download Full Size | PDF

The fabrication tolerance is also studied in this work. In the actual device fabrication, the most sensitive parameter is the SiN_x buffer layer thickness, which controls the phase shift, as shown in Fig. 2. Figure 6 shows the PER and insertion loss of TE and TM modes at the wavelength of 1528 nm and 1567nm, as functions of thickness error of SiN_x buffer layer. For both TE and TM polarizations, the polarization splitter of this design exhibits an insertion loss smaller than 1 dB and PER larger than 20 dB with the thickness variation of ± 6.2%, over the entire C-band. One of the method of achieving this requirement is ALE (Atomic Layer Deposition), which is capable of depositing an SiN_x layer with monolayer accuracy (<0.2nm) [15].

 

Fig. 6 Simulation result on (a) PER and (b) Insertion loss for TE and TM modes as a function of SiN_x layer thickness error.

Download Full Size | PDF

Improving the tolerance to manufacturing variations is important, and it is included in the device optimization process. All the possible combinations of tilt angle, MMI length (L_MMI), gold pattern length (L_m) and width (W_m) are used for the PER and insertion loss calculations, to generate the types of curves shown in Fig. 6(a) and 6(b). Then the best combination is chosen, such that it maximizes the SiN_x thickness range under which the PER is larger than 20dB and the insertion loss is less than 1dB both for TE and TM input modes at 1528 nm and 1567nm.

4. Conclusion

A novel MMI-based polarization splitter on an InP substrate is proposed. A patterned metal on a dielectric cladding layer creates birefringence, and a tilt joint of the MMI adjusts the phase such that phase shift for TE and TM mode are exactly -π/2 and π/2, resulting in separation of TE and TM modes within a short propagation distance. The total device length is less than 600 μm. Simulation shows that the device has a PER greater than 23 dB and an insertion loss below 0.7 dB over the wavelength range from 1528 to 1567 nm for both TE and TM polarizations. The fabrication tolerance is also studied, showing that for a thickness variation of ± 6% the insertion loss remains below 1 dB and the PER remains over 20 dB over the same wavelength range. The total device length is less than 600 μm. Although this device is proposed for InGaAsP/InP material systems, the device concept could also be readily applied to other material systems, such as SOI.