Home/Products/Ansys Lumerical/FDTD
ANSYS LUMERICAL
Solves Maxwell’s equations numerically by the finite-difference time-domain method. Because no approximation restricts the direction or angle of propagation, three-dimensional behavior including back reflection and multiple reflections is handled directly.
What FDTD is
The analysis volume is divided into a grid, and the electric and magnetic fields are updated alternately at each time step. It applies to a wide range of geometries and can model material dispersion and anisotropy, which makes it the most generally applicable method for analyzing nanophotonic devices.
Because the calculation is in the time domain, launching a single broadband pulse and taking its Fourier transform gives the response across the whole wavelength range from one run. That matters when you are designing resonator Q factors or filter transmission spectra.
The cost, though, grows with the number of cells multiplied by the number of time steps. Where the analysis region is large compared with the wavelength, or a fine mesh, a long decay time or dispersive materials are needed, run time and memory use grow accordingly. Depending on the problem, it is worth choosing between FDTD and the RCWA and STACK solvers included in the same Lumerical FDTD product, or the FDE, EME and varFDTD solvers in the separate Lumerical MODE product.
What FDTD gives you
01
Solves Maxwell’s equations numerically in the time domain, with no approximation restricting the direction or angle of propagation. Accuracy depends on the mesh, the time step, the material model and the boundary conditions. Back scattering, multiple reflections and diffraction in every direction are included directly.
02
Using a broadband pulse with frequency-domain monitors, the response at many wavelengths can usually be obtained from a single simulation. Cases such as oblique incidence on a periodic structure may need BFAST or a wavelength and angle sweep.
03
From 2023 R2 onwards, 3D FDTD analysis on supported NVIDIA GPUs is available. Single-GPU, multi-GPU and distributed execution in supported environments are possible, but there are constraints on material models, boundary conditions, GPU specification, licensing and the number of SMs available. Setup changes between releases, so please check the requirements for the version you are using.
Where it fits
Its generality has a cost, and for some problems another method solves faster and more accurately. The following is a guide to choosing.
Use FDTD when
Consider another method
For periodic structures RCWA is generally faster, while FDTD suits non-periodic structures and more general analysis.
Typical applications
Characterizing grating couplers, directional couplers, ring resonators and MMIs
Verifying the response of a single meta-atom, and checking local focusing performance
Optical efficiency and color crosstalk of pixel structures including microlenses and color filters
Radiation patterns of antenna elements, and OPA element design
Light extraction efficiency of LEDs and OLEDs, and emission control in microcavities
Inputs and outputs
For the structure, materials, source and boundary conditions you set, it returns the electromagnetic field distribution, transmission, reflection and absorption spectra, S-parameters, far fields and more.
INPUT
| Geometry | Built-in primitives, GDSII import, STL and STEP import, parametric generation by script |
| Material | Fitting to measured n and k data, dispersion models, anisotropic materials, nonlinear materials |
| Light source | Plane wave, Gaussian, mode source, dipole, TFSF, imported field |
| Boundary conditions | PML, periodic, Bloch, symmetric and antisymmetric, metal |
| Mesh | Automatic generation, plus region-specific mesh refinement and conformal mesh |
| Monitors | Frequency domain, time domain, far-field projection, movie output |
FDTD
Finite-difference time-domain method
OUTPUT
| Spectrum | Wavelength dependence of transmittance, reflectance and absorptance |
| S-parameters | S-parameters including amplitude and phase between ports. These can be used for an INTERCONNECT circuit model or to create a CML (generating a compiled CML requires CML Compiler) |
| Electromagnetic field distribution | E and H distributions on a cross-section or through a volume, Poynting vector, intensity distribution |
| Far field | Radiation pattern, directivity, and focusing efficiency against numerical aperture |
| Resonant characteristics | Resonant wavelength, Q factor, mode volume |
| Data for handover | Beam-like fields from a frequency-domain monitor are passed to OpticStudio in ZBF format. A strongly confined near field needs a far-field transformation or similar first. Optical absorption distributions are passed to CHARGE and HEAT |
Analysis workflow
Skip the convergence check and you cannot judge whether a result is sound. Step 03 is where most of your time should go.
01
Import from GDSII or CAD, or define the dimensions as parameters. At the same time, check the fitting error of the material model against measured data over the wavelength range of interest. Get this wrong and everything after it is wrong.
02
Where symmetry can be used, symmetric and antisymmetric boundaries cut the computation substantially.
03
Vary the mesh, the simulation time, the number of PML layers and the size of the analysis region, and confirm that the result stops moving.
04
Vary the dimensions to map out the design space. Parallel execution on HPC and in the cloud is supported.
05
S-parameters to INTERCONNECT, absorption distributions to CHARGE and HEAT, beams to Zemax.
Frequently asked questions
Questions we are often asked before adoption.
Should I use FDTD or RCWA?
Where the structure is periodic and you are evaluating diffraction efficiency or phase by wavelength, angle and polarization, use RCWA. Where finite-size effects, local defects, or arbitrary geometry with no periodicity are involved, use FDTD. In metasurface design you can generate the meta-atom library with RCWA and, where needed, verify local and finite-size effects with FDTD.
How do I check convergence?
Vary the mesh accuracy, mesh overrides, the number of PML layers and the distance between the structure and the PML, and confirm that the result you care about has converged sufficiently. Also confirm that the simulation time and the auto shutoff setting are not affecting the result.
My simulation diverges. What should I look at?
Start with the material model and its fitting, the mesh, and the boundary conditions including the PML. If you cannot identify the cause, simplify the model and narrow down where the problem is by inspecting the field distribution.
Can I run it on a GPU?
Yes. FDTD can use a supported NVIDIA GPU to accelerate 3D simulations. There are version-specific requirements and limitations on which GPUs, licenses and features are available, so please talk to us before you buy.
See the requirements and limitations of the FDTD GPU solver →
What PC specification do I need to use FDTD?
It depends on the scale of the analysis, but RAM capacity and memory bandwidth matter most. For large analyses we recommend a CPU environment with ample RAM and high memory bandwidth, plus a supported GPU where appropriate. We will advise on a configuration to match the analysis you have in mind.
We can advise on analysis setup, convergence, and how to choose boundary conditions. Hands-on training working directly in FDTD and MODE is also available.
Lumerical support →
Training →
You do not need to share everything. Whatever you are able to tell us is enough.
The demo uses a structure close to your own. We can also give you an idea of the computational scale and the time it takes.