Radio frequency simulation methodology

The majority of Lumerical's customers do simulations in the UV-Vis-IR wavelength range. For this reason, many default settings in the software and suggested simulation setup guidelines are appropriate for these wavelengths. This page explains some common issues and the methodology that must be considered when running simulations at RF frequencies, such as in the MHz, GHz,and sub-THz ranges.

Simulation time

At lower frequencies, the simulation time property (found in the general tab of the simulation region) must be increased. The default value of 1000fs corresponds to hundreds of optical cycles at visible frequencies. When operating at lower frequencies, you must increase the simulation time accordingly. It is suggested to use nanosecond units for MHz and GHz (1-100ns) and picosend units for sub-THz (100-500ps).

Material data (refractive index)

The materials defined in the Material Database generally have refractive index data around the UV-Vis-IR wavelength range. In most cases, the default material data does not extend to the GHz, MHz ranges. In such cases, you must add your own material data into the database. See the creating sampled data materials page for instructions. A non-dispersive dielectric or (n-k) model is often sufficient for most substrates.

Metals

At UV-Vis-IR wavelengths, it is often necessary to correctly model the dispersive nature of metals. At lower frequencies, the material dispersion is less important, and metals can often be treated as ideal. Making this approximation means that it is not necessary to find the actual refractive index values. It can also allow a much larger simulation mesh to be used. To treat metals as ideal, use the PEC (Perfect Electrical Conductor) material model or a non-dispersive 2d or 3d conductivity model. PEC is an ideal metal with zero absorption and 100% reflection.

Very small features

In some cases, extremely sub-wavelength features may exist in your real device. For example, a 1um thick metal layer in a device designed to operate at 300MHz. The layer is only one millionth of a wavelength thick! Making the simulation mesh small enough to resolve this feature will result in extremely long simulation times. Such situations deserve careful consideration. It is often possible to make some type of reasonable approximation to make the simulation run faster. For thin metal layers, it is often reasonable to make them much thicker in the simulation than they are in reality. It is also possible to use a 2D sheet to represent an object with infinitesimal thickness. See the metamaterial tips page for details

Mesh size

At RF frequencies, often the feature size is extremely subwavelength and the default mesh settings are not enough. It is often more efficient (in terms of simulation time), to place mesh override regions in the simulation domain instead of increasing the mesh accuracy in the mesh settings. These mesh override regions should be placed in regions the fields are concentrated, such as the the interfaces between dielectric, vacuum, and metal (particularly near metal edges). In most cases it is sufficient to only override the mesh over the cross-section of waveguides and transmission lines, but testing should be done on the extent of the mesh override regions and the mesh size. See the Using the non-uniform mesh  and Mesh refinement pages for details.

PML Thickness

As just mentioned, RF simulations often require mesh override regions near metallic surfaces. As as result, a graded mesh will be present in the simulation. This grading will impact the PML's performance because the PML layer's thickness is dependent on the length (normal to the PML boundary) of the adjacent mesh cell. As this length decreases, the PML becomes thinner and its performance gets worse. Effort should be made to allow the grading of the mesh cells to be a large enough size such that the PML is at least λ/4 thick. Using a time monitor near the PML, it is possible to gauge the PML's performance by observing the magnitude of the reflections and adjust the number of layers to accordingly. If significant reflections occurs, the number of PML layers should be increased.

In cases where the mesh size is very fine in the directions transverse to the PML's surface but coarsely meshed normal to the PML's surface (i.e., a high aspect ratio mesh grid of 10:1 or more), the PML's performance will be impacted and reflections may occur at its boundary. In these cases, the number of PML layers should be increased. While the exact PML thickness will be specific to each case, a total thickness of λ or more may be required.

Source Pulse

As the center frequency of the source pulse tends to zero frequency (i.e, DC), the pulse will contain a small DC component (and very low frequencies) that can limit the noise floor of the simulation. In bandstructure simulations, this DC component is necessary to obtain the entire resonant response of a device. However, in transmission/reflection studies (such as involving Ports on networks), this DC component will decay slowly or not decay at all causing long simulation times and/or the autoshutoff to never be triggered. Enabling the 'multifrequency mode injection' flag in the port helps in suppressing this DC component in the pulse but at the expense of longer simulation times. Another option is to manually change the 'time domain' settings of the pulse, delaying and spreading out the pulse to lower the DC component or turning off the 'optimize for short pulse' setting in the port or mode source.

Scattering parameters

In the analysis of many RF devices, we are typically interested in frequency-domain information such as the input impedance, standing-wave ratio, and the scattering parameters. In FDTD, Ports are a simple way to obtain a device's frequency network response in which the scattering parameters can be found directly. Other network parameters such as ABCD, impedance, and admittance parameters can be derived from the scattering parameters.

It is recommended to place the ports at least λ/4 away from any discontinuity in the transmission-line or waveguide that the port is exciting. This ensures that any evanescent modes excited at the discontinuity will decay and not impact the phase of the scattering parameters.

When the fields of a waveguide mode are entirely contained with the physical boundaries of the waveguide, the size of the port should equal with the size of the boundary. If the fields of a waveguide are not confined, the size of the port should be set to capture a majority of the mode's power.

Normalization of Power

At UV-VIs-IR wavelength, the sourcepower provides an accurate measure of the total amount of power the source injects into the simulation. However, the accuracy of the sourcepower function is impacted by the highly varying fields seen at RF frequencies and any injection errors from mode mismatch. In these cases, the amount of input power in a simulation can be found from the port's or mode expansion monitor's expansion results (see Quarter-wave monopole for details).

Advanced Option - Snap PEC to yee cell boundary

When the snap pec to yee cell boundary option is enabled in the FDTD object's advanced option tab, the interface of any PEC is forced to align with the Yee cell boundary (more details can be found in FDTD Solver). In general, this can force the geometry of the PEC to either grow or shrink, which can adversely impact the simulation results. Therefore, it is recommended that in most RF applications with PEC's, this option be turned off. However, in antenna applications, turning this option on can lead to improved accuracy in the calculated radiation efficiency. To gauge its impact, the user can run two simulations with and without the option turned on. This option is not used for non-PEC metals assigned finite conductivity or permittivity models.

Parameter definitions for antennas

Due to different definitions about the power for antennas, users need to know the definitions we used in the script.

Power

Efficiency and Gain

See also

Ports

Metamaterials methodology

Creating sampled data materials