Simulation and GPU Acceleration Power Data Center Photonics

Artificial intelligence (AI) workloads continue to reshape data center infrastructure. Increasing processor and accelerator performance places growing demands on bandwidth, power efficiency, and connectivity across increasingly large-scale systems. Traditional electrical interconnects are increasingly challenged by these requirements. Photonics is emerging as a critical technology for high-speed, energy-efficient data movement and for scaling the next generation of AI infrastructure.

Companies like iPronics — which is advancing optical circuit switching technology for AI and data-center networks — and Alcyon Photonics — which focuses on photonic building blocks and subsystems for optical communications — are using Ansys Lumerical simulation software to accelerate the development of these optical innovations. By combining physics-based photonic design tools with graphic processing unit (GPU)-accelerated computing, both organizations are reducing development cycles, improving device performance, and helping bring next-generation optical networking solutions to market faster.

iPronics: Simulating Intelligent Optical Switching for AI Networks

For iPronics, simulation is a foundational part of developing its optical circuit switching (OCS) technology, which is designed to improve the efficiency and scalability of data center networks. Optical switching has emerged as a promising alternative to traditional electronic switching because it can reduce power consumption while enabling higher bandwidth connectivity between growing clusters of GPUs.

"Simulation is central to the development of OCS technology for data centers," says Irene Ayuso, PDK Lead at iPronics. The company uses Ansys Lumerical MODE optical waveguide design tool, Ansys Lumerical FDTD advanced 3D electromagnetic finite-difference time-domain (FDTD) simulation software, Ansys Icepak electronics cooling simulation software, and Ansys Lumerical Multiphysics opto-electronics component simulation software to design and optimize every photonic component within its switching platform.

"Hundreds of simulations are required to develop critical functions such as light coupling, splitting, polarization management, and phase shifting," says Ayuso. "Through simulation, we can evaluate optical, thermal, and electrical performance long before manufacturing begins, significantly reducing development risk and validation time."

Simulation provides iPronics with the ability to evaluate hundreds of design variations and identify the optimal balance of performance, manufacturability, and reliability. Instead of depending on multiple fabrication cycles to validate designs, engineering teams can predict device behavior digitally, reducing both development risk and time to market.

As AI infrastructure grows, these workflows become increasingly important. Modern optical switching architectures contain complex interactions between optics, electronics, and thermal effects. By using simulation to understand these interactions early in the design process, iPronics can improve first-pass design success and accelerate the deployment of technologies that connect large-scale GPU clusters.

"We run a very large number of simulations to deliver optimized, robust designs, so GPU acceleration has completely changed how quickly we can explore and validate components."

— Irene Ayuso, PDK Lead at iPronics

GPU Acceleration Drives Faster Design Cycles

The scale and complexity of photonic simulations are driving a growing need for high-performance computing. Many of the optical components developed by iPronics require full 3D FDTD simulations, which are among the most computationally intensive electromagnetic analyses used in engineering.

By running Ansys Lumerical FDTD with NVIDIA CUDA acceleration on NVIDIA H100 GPUs in an AWS cloud environment, iPronics has achieved remarkable performance gains. The FDTD solver was 11-13 times faster compared with a 96-core central processing unit (CPU) environment. For edge coupler optimization, a 20-point parameter sweep that previously required roughly nine days can now be completed in about 18 hours. Cell throughput increased from roughly 2.3-2.6 billion cells per second on CPUs to approximately 29-31 billion on GPUs. The result is dramatically higher simulation throughput, faster design iteration, and significantly lower computational costs.

"We run a very large number of simulations to deliver optimized, robust designs, so GPU acceleration has completely changed how quickly we can explore and validate components," says Ayuso. "This dramatically increases simulation throughput, enabling us to explore large design spaces through extensive parameter sweeps and then validate the most promising solutions using high-fidelity simulations. As a result, we gain a much clearer understanding of design sensitivities and performance trade-offs."

These gains allow engineers to perform larger parameter sweeps, evaluate more design alternatives, and optimize complex photonic structures much faster than before. Rather than waiting days for simulation results, teams can iterate in near real time, accelerating innovation while reducing development costs. For AI data centers, this means faster delivery of the optical technologies needed to support increasingly dense GPU deployments.

Alcyon: Building the Optical Foundation for Data Center Connectivity

Alcyon specializes in designing the photonic building blocks that make advanced optical communications possible. The company develops components such as multimode interferometers (MMIs), directional couplers, multiplexers, polarization management devices, and other integrated photonic subsystems used in optical transceivers and networking equipment.

Simulation plays a central role in Alcyon's design methodology. Engineers use advanced electromagnetic modeling and foundry-specific process simulations to evaluate how devices will perform under real manufacturing conditions. This enables the company to analyze fabrication tolerances, alignment sensitivities, optical losses, and wavelength performance before any physical prototypes are produced.

"Our use of Ansys Lumerical has been foundational to the design and development of photonic building blocks and subsystems that form the core of Alcyon Photonics’ business," says Panaiot Zotev, tech lead, building blocks at Alcyon.

One key advantage of this approach is the ability to create highly accurate compact models that can be integrated into larger photonic circuits and systems. These models help customers develop increasingly complex optical architectures while maintaining confidence that designs will perform as expected when fabricated.

For data center applications, this simulation-driven process is especially important. AI clusters demand optical interconnects that are highly efficient, reliable, and scalable. By validating designs virtually, Alcyon can optimize performance and improve manufacturability while reducing development timelines.

"Our use of Ansys Lumerical has been foundational to the design and development of photonic building blocks and subsystems that form the core of Alcyon Photonics’ business."

—Panaiot Zotev, Tech Lead, building blocks at Alcyon.

Accelerating Photonics Innovation with GPUs

Alcyon also benefits from the rise of GPU-accelerated simulation. Traditional photonic simulations often involve extensive computational resources and long runtimes, limiting the number of design iterations engineers can realistically perform.

GPU acceleration significantly changes this equation. Faster simulation times enable Alcyon engineers to evaluate larger design spaces, test additional operating conditions, and automate optimization workflows. This capability is particularly valuable when developing advanced photonic integrated circuits for data center communications, where even small performance improvements can have a meaningful impact on system efficiency.

"The GPU acceleration feature in Lumerical, combined with our NVIDIA RTX A5000 GPU, has reduced simulation runtimes by more than a factor of 8 compared with a 24-core CPU. This acceleration has also shortened overall development cycles by minimizing the number of tape-outs required to achieve validated lot-to-lot device performance, reducing it to just two," says Zotev. "This has not only enabled us to accelerate the development and optimization of individual photonic building blocks, but has also made previously prohibitively time-consuming simulation and analysis techniques feasible."

By combining simulation, automation, and GPU computing, Alcyon can move from concept to validated design more quickly while maintaining high levels of accuracy and reliability.

Accelerating Photonics for the AI Infrastructure Era

As AI infrastructure continues to scale, the challenge is no longer simply adding more compute. Moving growing volumes of data efficiently between processors, accelerators, memory, storage, and networks is becoming equally important. Photonics will play an increasingly important role in addressing that challenge. At the same time, advanced simulation, GPU acceleration, and access to scalable compute help photonics companies explore larger design spaces and bring these technologies to market faster. iPronics and Alcyon Photonics demonstrate how these two trends are coming together. Modern compute infrastructure is accelerating the development of the photonic technologies that next-generation AI infrastructure increasingly depends on.

Learn more about Ansys Lumerical by visiting us at NVIDIA GTC Berlin at booth #7011.