COMSOL Simulation Summit Panel Explores the Future of Multiphysics

At the recent COMSOL Simulation Summit (May 8, Santa Clara, California), I had a chance to moderate a panel titled "The Multiphysics Roadmap," focusing on how the latest advances in AI, GPU acceleration, and template building affect multiphysics simulation. The panelists were: 

• Bjorn Sjodin, SVP of Product Management, COMSOL
• Tony Lennon, Market Manager for Electrical and Multi-Domain Simulation, MathWorks
• Azi Riahi, Principal Product Manager, NVIDIA
• Professor Jennifer Dionne, Stanford University

To open the discussion, I asked Bjorn to confront the AI topic head-on. He revealed that, thanks to the natural language integration in AI chatbots, COMSOL users were already generating codes and simulation apps. "So a couple of years ago we added a tool called the Chatbot Window in COMSOL that can help you with programming. He also expected the tool would help COMSOL users interrogate complex models in everyday English. "You can upload a model, then ask questions: What do these parameters do? Can you explain this model to me?," he said.

COMSOL offers LiveLink for MATLAB, which functions as a bidirectional interface between COMSOL Multiphysics and MATLAB. Tony said, "[In the past] engineers had to become coders, and it took them away from engineering. The agentic workflow will allow engineers to return to doing what they're good at."

Similarly, Jennifer said her engineering students had to "learn to code -- essentially learning a different language. But now, that language is tied to the native language of any student or researcher, so that they can just focus on the building and creation of devices and systems." Her students use COMSOL software for nanophotonic device simulation. The use of COMSOL API and agentic tools has accelerated the pace of research, she noted.

Because of its origin in graphics hardware, NVIDIA is often considered a hardware maker, but Azi wanted people to know the company had evolved beyond it. Now, NVIDIA is "a company with an ecosystem for solving computationally intensive and difficult problems. Building the software layers on the hardware layers and accelerators is very important. That's why partnerships with companies like COMSOL that can benefit from accelerators is important. It's the software layer that enables these partners to solve the problems fast ... In the backend, NVIDIA provides complementary components for GPU acceleration," she said.

Part of NVIDIA's strategy is to develop CUDA-X libraries, covering the different disciplines, domains, and industries that can benefit from GPU acceleration. 

In 2014, COMSOL launched the Application Builder, aimed at letting users easily create a template or an app to tackle routinely executed simulation tasks. "The underlying physics model can be very complicated. It could have thousands of parameters. But in the app, maybe you only expose five of these parameters to the users. People are running these apps on their iPads or from the web ... And they work very nicely with GPU acceleration," said Bjorn.

Jennifer felt that, in the first year of college, it was important for the students to master the nuts and bolts of simulation. "With the electrical optical devices we are making, the results are only as good as the input, or the material properties. So it's important for them to understand the input they're using ... but once that foundation is laid, apps and AI can help you quickly iterate," she said.

Azi said, "As the simulation models grow, having the scalability and memory bandwidth to tackle them becomes a problem ... We will continue to address the unique needs of multiphysics. The evolution of our hardware components is synergistic to multiphysics." 

For the whole discussion, watch the video below: