Learning Forum

[Eva021]

Hi, how to get the bend loss has bothered me for a long time. I have try many solvers in Lumerical. For example, when I used the FDTD solver to simulate a bent waveguide, I set up power monitors at the input and output ports of a 90° bent waveguide and used a formula to calculate the bending loss. However, as the bending radius gradually increased to approximately 50 μm, the FDTD simulation of the bent waveguide became very slow and always terminated prematurely. The issue of large-bend waveguides requiring more operating memory cannot be resolved on my device.

So after reviewing the official documentation, I switched to use the FDE solver. However, the FDE solver has an issue: as mentioned in responses to other posts, the bending loss calculated by FDE does not include scattering loss. Additionally, based on my simulation results, the bending loss obtained from FDE simulations is consistently at the 10⁻⁷ order of magnitude or even smaller. What’s the problem?

I had to switch to the current FEEM. When running the official case, the bending loss was 278 dB/cm, consistent with the results on the official website. However, when I modified the structural parameters and material parameters of the waveguide while keeping other settings unchanged, and only adjusted the range of the bending simulation region and the size of the bending radius, with the bending radius set to 10 μm, the resulting loss is always at the 10^-10 order of magnitude, and the results of each simulation always change, not being a constant value, even I don’t adjust something other, just run again.

I have posted several questions about bending loss simulation, but I still haven't figured out how to address these issues to obtain accurate loss simulation results. Could you please help me resolve these issues?

[abhijith.dileepkumar]

 

Hi, thank you for the question.

Generally, FDTD is not recommended for simulating bend losses for large waveguide bends due to meshing constraints. It is inherently considered a time-consuming and computationally intensive method, when compared to other solvers like FDE and FEEM for the same waveguide design. With that being said, I can suggest some solutions that could potentially improve the accuracy and consistency of the bend loss results.

First and foremost, I would ensure that the correct mode is being selected during the simulation, for which you are trying to calculate the bend loss for. You can verify this by checking the effective index of the mode.

Since the size of the simulation region plays a critical role in determining loss accuracy, I would encourage you to perform a convergence test on how both the effective index and bend loss can change as you expand the simulation span. Specifically, you should increase the simulation span in the direction opposite to the bend to avoid truncating the mode profile and ensure that the loss values converge. However, care should be taken when you are changing the simulation span with respect to the waveguide. Please refer to this tutorial for more detailed information: Convergence Testing — Lesson 3

In terms of meshing, a good practice in both FDE and FEEM solvers, is to apply a fine mesh in the core region where the optical mode is concentrated, and a coarser mesh in the surrounding cladding and air regions. This not only improves simulation accuracy but also helps manage computational load effectively. Also, for bent waveguides, it is essential to make sure that the center of curvature lies outside the simulation region. This is a common issue that users tend to overlook in these simulations.

I hope these suggestions help. If not, feel free to come back with further questions and thoughts.

Best,

Abhijith