Simulate Mode-Evolution-Based Polarization Rotator with High-Index Contrast

High-index contrast dielectric waveguides exhibit strong confinement, making the TE and TM modes very different. As a result, device characteristics, such as propagation constant and coupling strength become strongly polarization dependent.  Rotation of one of the polarization outputs allows a single polarization to be realized on-chip and the two paths to be operated on in parallel with identical structures. Mode-evolution-based polarization splitters and rotators have been proposed [1]. One approach is to use the Finite-Difference Time Domain (FDTD) algorithm to model these devices, but FDTD is very computationally intensive and requires a lot of computer memory. In this article, we will demonstrate the simulation of a mode-evolution-based polarization rotator based with high-index contrast using the RSoft Beam Propagation Method (BPM) tool, BeamPROP. The advantage of a BPM simulation when compared to an FDTD simulation is that it requires much less simulation time and memory (RAM). The design files can be accessed in RSoft Products section of the Customer Support Portal.

Figure 1 shows the proposed rotator structure. The polarization rotation can be created using only a pair of asymmetrically and oppositely tapered waveguide core layers. The core index material is silicon nitride and the cladding is silica. The principal axis of the structure and polarization state of the fundamental mode rotates in unison along the transition. Since we are going to investigate polarization state changes, a 3D full vector BPM calculation will be needed.  

Fig. 1: The schematic picture of the polarization rotator | Synopsys

Fig. 1: The schematic picture of the polarization rotator

Figure 2 shows the structure setup in the RSoft CAD Environment and the refractive index cross-sectional profiles at three different positions in the waveguide. We first calculate the TM fundamental mode at the input position, and then calculate both the TE and TM modes at the output position, as shown in Figure 3. The TM mode is launched into the waveguide structure, and a full vector BPM propagation is performed. The input TM mode and output TE/TM modes are monitored along the propagation. Figure 4(a) shows the propagation field and monitor values along the waveguide. It clearly shows how the input TM mode gradually evolves into the output TE and TM output modes, with ≈50% power in the TE mode and ≈50% in the TM mode at 25um structure length.

Structure set up in RSoft CAD and index cross section profiles at several waveguide positions | Synopsys

Fig. 2: Structure set up in RSoft CAD and index cross section profiles at several waveguide positions

Modes calculated with BeamPROP mode solver (a) Four field components of TM fundamental mode at input.  (b) Major E fields of TE and TM modes at output | Synopsys

               (a)                                                                                      (b)

Fig. 3: Modes calculated with BeamPROP mode solver
(a) Four field components of TM fundamental mode at input. 
(b) Major E fields of TE 
and TM modes at output. 

For comparison, a FDTD simulation was performed with RSoft’s FDTD tool, FullWAVE, for the same set up, as shown in Figure 4(b). The results from the FDTD and BPM simulations have close agreement.

BPM propagation and monitor results along the waveguide | Synopsys

(a)

FDTD propagation and time monitor results | Synopsys

(b)

Fig. 4: (a) BPM propagation and monitor results along the waveguide; 
(b) FDTD propagation and time monitor results

Finally, RSoft’s MOST scanning and optimization utility was used to simulate the propagation for different waveguide length with BeamPROP and FullWAVE. The results are shown in Figure 5. From Figure 5, we know the polarization will completely convert to the opposite polarization state for a waveguide approximately 200um. Again, the BPM and FDTD results show excellent agreement.  

Polarization state at output at different waveguide length for BPM and FDTD simulations | Synopsys

Fig. 5: Polarization state at output at different waveguide length for BPM and FDTD simulations

We also recorded the simulation time and memory cost for the above simulations, as shown in the table below. The simulations were performed on a single workstation with 2 x 10 core CPUs (E5-2650v3 @ 2.3GHz). To take advantage of the 20 cores, FullWAVE used multi-process clustering (20 processes) and BeamPROP used multi-threading (20 threads). As can be seen, BPM runs many times faster with much less memory compared to FDTD.

Simulation time and simulation memory | Synopsys

Note that the simulation time and RAM scale differently for BPM and FDTD. This is because the BPM equation is solved along the propagation axis, whereas the FDTD algorithm solves for all grid points at each time step during the simulation. Therefore, the simulation time for BPM is linear with the structure length, while for FDTD a 2x increase in length results in a 4x time increase. The required RAM for BPM is constant with structure length because the algorithm only saves field data over the cross-section; for FDTD, the required RAM increases linearly with the structure length since field data is saved over the entire structure.

Please contact the Photonics Technical Support Team at photonics_support@synopsys.com for more information.


References:

1) Watts et al. "Integrated mode-evolution-based polarization rotators," Optics Letters, 30 (2005).