More simulations is a different way to work.
Faster is useful. Cheap enough to run thousands is what changes the design process. Each run costs little, so you can ask questions you used to skip.
Design sweeps
Vary every dimension and see the whole landscape, not a few hand-picked points.
Optimization
Let an optimizer call the solver as many times as it needs. Thousands of calls stop being a budget problem.
Tolerance and yield
Simulate manufacturing variation across thousands of cases and find out which designs survive the fab.
Today's pricing was built for a handful of runs.
You pay per seat, per year. The licence count, not your hardware, decides how many jobs can run at once.
You pay for every simulation. A thousand-run study costs a thousand times as much, so the meter decides how far you explore.
Running more simulations should be the obvious choice, not a budget request. LiminusFDTD is built, and will be priced, around that.
Fast, and accurate enough to trust.
We ran the 16-million-cell Mie-sphere example that Tidy3D publishes, end to end on a cloud A100, and finished 11% sooner than Tidy3D's published time.
Our run is from October 2026. Tidy3D does not disclose the hardware behind its time, so this compares the same job, not the same chip.
Validated on 6 benchmark problems and 7 NVIDIA GPU types. The full report is available on request.
Teams that design, and teams that manufacture, light-shaping optics.
Metalenses
Sweep pillar geometry across the lens, then test the design against fab variation before production.
AR / VR optics
Tune waveguide gratings across angle, wavelength and polarization in one study.
Image sensors
Model pixel-scale structures, where a few nanometres change the result.
Integrated photonics
Sweep couplers, splitters and resonators across the whole process window.
See it run on your own problem.
- 1
Send us a problem
Pick one you already simulate elsewhere, so there is a result to compare against. We can work under NDA.
- 2
We run it
We set it up and run it on LiminusFDTD, on NVIDIA GPUs.
- 3
You get the comparison
We send back the results, the accuracy against your reference, and the cost per run.
Watch light move through a device before anyone builds it.
FDTD splits space into a grid of tiny boxes and steps time forward a tiny amount at a time. A computer can then follow how light or radio waves spread, bounce and pass through a structure.
It is the trusted way to check designs too small or intricate to solve by hand. It is also slow, so most teams run a few simulations and leave the rest of the design space unexplored.
The grid update is one simple step repeated billions of times, which is exactly the work a GPU is built for. LiminusFDTD runs the whole solver on the GPU.
Built by a physicist who got tired of waiting weeks for a result.

Jim wrote the first version of LiminusFDTD at university, frustrated by how slow and expensive the industry-standard simulation tools were. A single run could take weeks. His own solver let him run dozens of simulations in hours.
He has now written FDTD codes for thirteen years. He holds a PhD in Physics and was Senior Professional Staff at Johns Hopkins APL before founding Liminus Labs to put that speed in other engineers' hands.
Jim is advised by industry veterans in semiconductor design software and enterprise SaaS.
A multi-billion-dollar market that GPUs are reshaping.
Every optical and radio device is simulated before it is built. We start with metalens makers, then photonics more broadly: AR/VR optics, image sensors, LiDAR and integrated photonics.
We are raising a seed round. Contact us and we will send the deck.
Have a design you want to sweep?
Send us a problem you already simulate somewhere else. We will run it on LiminusFDTD and share the results, the accuracy and the cost per run.
info@liminuslabs.com