Home/Technical Information/Design and Analysis Guides
Design and Analysis Guides
A checklist of the Ansys Lumerical FDTD settings that are easiest to overlook: mesh size and mesh order, material fitting, simulation time and span, boundary conditions and source span, with what to check and why each setting is what it is.
This article covers the settings in an FDTD simulation that are easily overlooked or set badly (Ansys Lumerical FDTD product overview and main features). It first offers a checklist you can refer to when setting up a simulation, then explains each point on that checklist in more detail.
The key parameter that determines the accuracy of the FDTD algorithm is the number of mesh cells per wavelength. With the default “auto-nonuniform mesh” in Lumerical FDTD, the “mesh accuracy” setting automatically generates a mesh with a fixed number of cells per wavelength.
Early simulations do not need a very fine mesh, so a mesh accuracy of 2 or 3 is sufficient. Even in an early simulation, however, you may need to refine the mesh in particular areas of the simulation region using a mesh override object. Regions that need a mesh override include device geometries with fine features that the default mesh cannot resolve, such as thin layers, and metal-dielectric interfaces where the field changes abruptly. For rectilinear geometry it is best to use a mesh override so that mesh cells align with the geometry. The images below, for example, show a thin layer: first with the default mesh, then with a mesh override region whose “dy” setting guarantees four mesh cells aligned exactly with the thin layer in the Y direction.
Default mesh:
With a mesh override:
An index monitor can be used to display the mesh before running the simulation and confirm that every feature is resolved. You can also display the mesh grid by clicking the “View simulation mesh” button on the toolbar to the left of the viewport.
Running a convergence test after the first simulation lets you determine the mesh settings needed for accurate results. (How to run a convergence test, step by step)
When geometry objects overlap, the mesh order determines which object’s index is used, and the lower mesh order takes precedence. You should use an index monitor to check that the correct index is used in the mesh.
More detail here Mesh Order)
In the material explorer, check the fit of every material in the simulation. The fit must follow the data points closely, with no gain and no sharp peaks.
More detail here: Modifying the Material Fits
An FDTD simulation ends under three conditions: the simulation time expires, the auto shutoff threshold is reached (meaning the fields have decayed sufficiently), or the fields diverge. After running a simulation you can confirm how it ended by checking the “status” result of the FDTD region object, or by checking the log file.
status =
In general it is best for a simulation to end at the auto shutoff threshold. Ending a simulation because the simulation time has expired can introduce error into the frequency-domain results.If the simulation ends because the simulation time was reached, increase the simulation time.A longer simulation time may be needed when there is a resonant structure or a long propagation distance.
More
The span of the FDTD region should be set so that the PML boundary is half a wavelength away from the sides of the geometry objects being simulated. “Wavelength” here means the longest wavelength in the source spectrum, accounting for the refractive index of the material between the object and the boundary. Objects expected to extend beyond the simulation region are the exception, such as a substrate or cladding, or the ends of input and output waveguides. A simulation of a straight waveguide running simply along Z is one such case.
When using PML boundaries, use the “standard” PML profile by default. Use the “steep angle” profile for periodic simulations and for simulations where light propagates at large angles. Use the “stabilized” profile only when you have divergence problems.
When the structure and the source are periodic, use “Periodic” boundaries for a source at normal incidence, “Bloch” boundaries for a narrowband source at an angle, and “BFAST” boundaries for a broadband source at an angle (What to watch for in broadband, oblique-incidence analysis with BFAST). When using symmetry boundaries, confirm that both the geometry and the source are symmetric. Whether to choose symmetric or antisymmetric depends on the polarisation of the source. As a rule of thumb, the arrow showing the source polarisation should be parallel to boundaries of the same colour and perpendicular to boundaries of a different colour.
More detail here: PML boundaries), Symmetry boundaries), Periodic boundaries)
A common mistake is to set the source span smaller than the span of the injected field. This is seen particularly often with Gaussian and mode sources. Truncating the source this way causes scattering and other injection errors. You should confirm visually that the source field has decayed sufficiently at the edge of the source span (an amplitude of 10^-3 to 10^-4). A log scale is useful for this. Placing a DFT monitor or a movie monitor also lets you judge whether the source is being injected correctly. Of the DFT monitor results below, the upper image shows a source span too narrow for the injected field and the lower image shows a source span appropriate for it. Note the scattered light behind and beside the source in the image with the inappropriate span.
Narrow source span:
Appropriate source span:
Products and software
Analysis technologies
Our engineers can advise on building a design flow for your application, and on validating an analysis model, from practical experience.