Skip to content
← All posts
6 min read

What is FDTD, and why it suits PCB RF simulation

Short answer
FDTD, the finite-difference time-domain method, solves electromagnetics by stepping the electric and magnetic fields forward in time on a regular grid. You launch a broadband pulse and watch it propagate, so a single run gives results across a whole frequency band. Each cell update depends only on its neighbors, which is why FDTD parallelizes so well on a GPU.

FDTD is one of the workhorse methods in electromagnetic simulation. The idea is older than most commercial tools and the math is approachable. This is a plain explanation aimed at RF engineers who want to know what the method actually does and when to reach for it.

The core idea

Maxwell's equations relate how electric and magnetic fields change in space and in time. FDTD turns those continuous relationships into a grid of small cells and a clock. At each tick, it updates the electric field from the surrounding magnetic field, then updates the magnetic field from the surrounding electric field. This leapfrog in time, on the staggered grid known as the Yee cell, is the whole engine.

You excite the structure with a short pulse. Because a short pulse contains a wide band of frequencies, you do not run one frequency at a time. You let the pulse propagate, record the response over time, and take a Fourier transform to get the behavior across the entire band from a single run.

Strengths and trade-offs

StrengthTrade-off
One run covers a broad bandHigh-Q resonators need many timesteps to settle
Transient and time-domain behavior is naturalCell size is tied to the smallest feature and shortest wavelength
Each cell update is local, so it parallelizesCurved or angled geometry needs fine cells to resolve
Maps cleanly onto GPU hardwareOpen boundaries need an absorbing layer (PML)

Why GPUs and FDTD fit together

An FDTD cell only needs its immediate neighbors from the previous timestep. There is no global system to solve at each step, just a huge number of small, identical, independent updates. That is the exact shape of work a GPU is built for. Updating millions of cells in parallel is why a GPU FDTD solver can run tens of times faster than a single-threaded CPU version. We measured about 68.5x against openEMS on a 46.8M-cell patch run in the benchmark post.

FDTD vs FEM for PCB RF

  • FDTD is time domain. One run gives the whole band, and it scales on a GPU. It suits broadband PCB RF such as antennas, filters, and transmission lines.
  • FEM is frequency domain. It is strong for arbitrary 3D geometry and resonant structures, but it solves one frequency at a time.

Neither is universally better. For the planar PCB RF that most engineers do day to day, FDTD on a GPU is a strong fit, which is the bet RayRF is built on. If you want to see it on a real design, the patch antenna walkthrough runs the method end to end.

Frequently asked questions

What is FDTD in simple terms?
FDTD, the finite-difference time-domain method, simulates electromagnetics by stepping the electric and magnetic fields forward in time on a grid. You launch a pulse, let it propagate cell by cell, and record the response. Because the pulse is broadband, one run gives results across a whole frequency range.
What is FDTD good for and not good for?
FDTD is good for broadband problems, transient behavior, and structures that fit a regular grid, and it parallelizes well on GPUs. It is less convenient for very high-Q resonators that take many timesteps to settle, and for geometry that does not align with a rectangular grid without fine meshing.
Why is FDTD a good fit for GPUs?
Each FDTD cell update depends only on its immediate neighbors from the previous timestep, so the work is local and massively parallel. A GPU updates millions of cells at once, which is why a GPU FDTD solver can run tens of times faster than a single-threaded CPU implementation.
FDTD vs FEM: which should I use for PCB RF?
For broadband PCB RF such as antennas and filters, time-domain FDTD gives the whole band from one run and scales well on a GPU. FEM is frequency-domain and strong for resonant and arbitrary 3D geometry but solves one frequency at a time. Many PCB RF tasks are well served by FDTD.
See FDTD run on your own board

RayRF is a GPU FDTD simulator with a GUI. Draw a structure, run it, and watch the fields. 30-day free trial, no card required.

Start the free trial
FDTDexplainerelectromagnetics