Photon Design, a global leader in photonic simulation CAD software, is the first company to simulate Thin Film Lithium Niobate (TFLN) bends natively, in just 10 seconds, with its FIMPROP simulation tool, which uses Eigen Mode Expansion (EME) processing. FIMPROP delivers results as rigorous as industry-standard FDTD tools, but hundreds of times faster.

Because TFLN is a birefringent optical material that’s refractive index changes with voltage, so it is commonly used for the high-speed, phase switching needed for Mach-Zehnder Modulators (MZMs). These are central to coherent data communications, AI and co-locatedoptics processors. Rigorous TFLN modelling is, therefore, essential for developing these
emerging, high-data-rate, photonics applications.

Dr Dominic Gallagher, CEO of Photon Design, said: “EME achieves its efficiency by modelling only the live material regions of the TFLN adiabatic bend, rather than the full bounding volume. TFLN bends can have large bounding volumes so we used a 500um² ‘S’ bend as a benchmark, demonstrating rigorous simulation results in just 10 seconds. Not only does FDTD process the entire bounding volume, but its runtime also increases with both volume and simulation duration. This means that every time the bend dimensions double, the FDTD simulation runtime increases eightfold.”

“Photon Design completed the benchmark TFLN ‘S’ bend simulation in just 10 seconds on a standard CPU-only laptop. By comparison, estimates of simulation run times for FDTD on computer CPU are more than one hour and even when using costly, cloud-based, GPU processing, it’s run times are still around 10 minutes. This performance difference reflects
the computational demands of FDTD, which requires more than 7.5 trillion calculations to produce the benchmark ‘S’ bend, versus only a handful of calculations using EME.

“Since TFLN design is highly iterative, with engineers often running more than 100 simulations for each design, processing speed is a critical factor when choosing a simulation tool.”

Authors from [1] compare simulation runtimes of two respected FDTD vendors. Taking the results of their ring resonator example, a value for the simulation’s giga Yee-cells per second can be found. This was selected at 15 cells/ wavelength, the resolution at which these results are shown to converge to a reasonable degree.

This simulation rate is then applied to the volume of our example 500um x 500 um S-bend to approximate runtime values of ~17,300 seconds (4 hours 48 minutes) for results on CPU and ~490 seconds (~8 minutes) for the same vendor run on their GPU service that is charged at an additional cost at time of writing.

[1] - https://arxiv.org/abs/2506.16665