Advancing Physical AI with AI Physics and Accelerated Computing

Discover how innovators are combining physics-based simulation, accelerated computing, and artificial intelligence (AI) to enable physical AI innovations across industries.

Automotive lighting. Aerospace validation. Autonomous vehicles. Materials innovation. Complex lithography systems. Industrial AI workload support. These are just some examples of the many applications that already benefit from accelerated computing, AI physics, and physical artificial intelligence (AI), which connects digital AI to hardware that can sense and act in the physical world.

Despite physical AI’s rapid growth and broad range of potential applications, ensuring that it can successfully perform under real-world conditions is far from straightforward. Achieving this requires trusted simulation solutions, a highly accurate AI physics foundation, and accelerated computing resources capable of processing complex workloads.

NVIDIA accelerated computing and CUDA-X libraries accelerate Synopsys semiconductor engineering applications, helping teams simulate device behavior, explore materials, and evaluate manufacturing processes faster. For example, Ralf Siemieniec, senior principal engineer at Infineon Technologies, a world leader in semiconductor solutions, shares that “The use of graphics processing unit (GPU) acceleration can save a lot of time in case of large problems in 3D device simulation, due to the strong reduction of the solve time.”

Using four NVIDIA H200 GPUs, Infineon reduced simulation time from 24 days to less than 48 hours, reporting a 12x speedup. While this is a semiconductor simulation example rather than a physical AI application, it illustrates how accelerated computing can dramatically expand the scale and speed of physics-based engineering workflows.

To realize these benefits across industries, leading companies are using solutions from Synopsys to train, test, and validate physical AI systems and the underlying technologies powering them. Synopsys solutions offer a variety of benefits, including generating physics-accurate synthetic data at scale, accelerating robotics development, and validating autonomous systems before deployment. They are further strengthened by Synopsys and NVIDIA’s collaboration, which merges accurate physics simulations with immersive virtual environments and scaled compute.

What does this look like in physical AI applications? The following examples show how organizations are using simulation, synthetic data, digital twins, and accelerated computing to develop and validate intelligent systems that ultimately operate in the physical world.

Explore Three Real-World Examples of Synopsys-Enabled Physical AI

1: How FORVIA HELLA Accelerates Automotive Lighting and Uses T-Systems’ GPU-Accelerated Cloud Computing Solutions

The future of automotive lighting is only getting brighter. According to Dr. Andre Steffen, head of computer-aided lighting at FORVIA HELLA, automotive lighting is transforming into an intelligent, software-defined platform that supports safety, communication, personalization, and even brand identity. This evolution also creates challenges, including growing product complexity and pressure to reduce development time and costs.

FORVIA HELLA is addressing these challenges by moving from physical testing to exploring, optimizing, and validating entire lighting concepts in a virtual environment. To support this transition, FORVIA HELLA uses solutions from Ansys, part of Synopsys, including Ansys Speos computer-aided design (CAD)-integrated optical and lighting simulation software to create photorealistic renderings that virtually validate the optical performance of its lighting systems.

This is further accelerated by using T-Systems’ Industrial AI Cloud based GPU simulation service instead of an on-premise CPU environment, reducing the simulation time by 86% comparing 512 CPU cores to 32 NVIDIA RTX Pro 6000 GPUs.

By moving a significant portion of validation activities into a simulation-driven environment, FORVIA HELLA has reduced development iterations, shortened decision-making cycles, minimized costs, increased confidence in product performance and compliance, and shifted physical testing toward final confirmation rather than early-stage design discovery. This virtual engineering foundation becomes increasingly important as lighting systems evolve from individual components into intelligent, software-defined systems.

“At FORVIA HELLA, we believe that physical AI represents the next major leap in product development. Physical AI allows us to accelerate the development of next-generation lighting systems while reducing costs and supporting more sustainable engineering practices through fewer physical prototypes. The result is better products reaching the market faster, with greater performance and reliability from day one.”

— Dr. Andre Steffen, head of computer-aided lighting at FORVIA HELLA

2: How Leonardo Accelerates Mission-Critical Aerospace Validation

Helicopter perception and flight functions must operate safely across complex and degraded conditions. These can include severe weather, limited visual features, unpredictable turbulence, moving landing platforms, and unavailable Global Positioning System (GPS) signals. However, reproducing this broad range of conditions efficiently and with a high level of engineering confidence through physical testing alone can be costly and difficult.

Leonardo is exploring simulation-based development to virtually recreate diverse environments and sensor behavior, including challenging mission scenarios, all before physical testing begins. This approach is essential to physical AI development because it enables teams to develop, train, and validate AI-based perception and autonomous functions in realistic virtual environments before conducting costly or difficult physical tests.

To achieve this, Leonardo used Ansys AVxcelerate Sensors autonomous vehicle sensor simulation software to model camera, radar, and lidar performance and generate synthetic multisensor data. Through this, the team connected sensor simulation results with perception and autonomous-flight functions in synthetic mission environments. Open Universal Scene Description (OpenUSD)-based workflows and NVIDIA Omniverse libraries also help Leonardo teams assemble, share, and reuse critical assets, thereby increasing efficiency and managing complexity.

Leonardo’s broader digital twin approach, meanwhile, combines mission-environment simulation, physics-based models, synthetic data generation, and AI-enabled workflows to support the development and virtual testing of autonomous aircraft and other embodied systems.

3: How INYO Mobility GmbH Develops and Validates Autonomous Vehicles

The INYO-Cab is an autonomous four-seat vehicle that provides first- and last-mile public transport services. Before it can hit the road, it must meet stringent safety and regulatory standards while maximizing passenger comfort and energy efficiency. Achieving these goals requires collecting, analyzing, and processing a large amount of data.

Because physical testing and data collection alone cannot generate enough training and validation data, INYO turned to Ansys simulation and validation solutions. Since the INYO-Cab's autonomy is driven by physical AI, accurate and reliable simulations are crucial for comprehensive training, explains Marcus Zwick, INYO's founder and CEO.

To implement these solutions, INYO worked with CADFEM, an Ansys Elite Channel Partner. Through this collaboration, CADFEM used AVxcelerate Sensors software and the larger Ansys AVxcelerate Autonomy autonomous vehicle development and safety validation toolchain to analyze INYO-Cab’s autonomous behavior capabilities. The team also created a digital twin using the Ansys Twin Builder simulation-based digital twin platform and NVIDIA technologies.

The INYO-Cab

“The Ansys tools and CADFEM’s application expertise yielded outstanding results, helping the vehicle obtain approval from the authorities.”

— Marcus Zwick, founder and CEO of INYO

For startups such as INYO, these solutions can make advanced development and validation more affordable and accessible. INYO used Synopsys solutions to validate its vehicle against German urban traffic conditions, iterate in hours rather than weeks, and plan remote supervision before deployment.

Explore Three Companies Using Synopsys for Accelerated Computing and AI Physics to Enable Physical AI

While not all organizations are developing physical AI systems directly, many are providing the foundational technologies that make physical AI possible. The following examples highlight how accelerated computing and AI physics are accelerating innovation across engineering disciplines.

1: How EMD Electronics Transforms Materials Innovation

As the semiconductor industry enters the AI and physical AI era, the need for greater performance continues to grow. This shift is increasing demand for advanced materials, including those needed for 3D system integration. This is where EMD Electronics comes in. EMD Electronics, the North American electronics business of Merck KgaA, Darmstadt, Germany, develops materials and equipment for a wide range of semiconductor processes, helping customers improve semiconductor technology performance. “Our materials enable our customers to push the boundaries of semiconductor technology performance,” says Jacob Woodruff, head of technology scouting and partnerships at the company.

“With our collaboration with Synopsys, we have a first-of-its-kind opportunity to connect our deep materials expertise with trusted, physics-based digital engineering to model material behavior, explore complex interactions, and validate choices earlier."

— Surésh Rajaraman, executive vice president and chief technology officer at EMD Electronics

To advance materials innovation, EMD Electronics must establish predictive links among material composition, structure, process, and device outcomes. According to Woodruff, understanding and predicting structure-property relationships remains difficult for complex materials. This difficulty stems from the multiphysics phenomena involved in semiconductor manufacturing and the vast disparities in length and time scales.

To overcome these challenges, EMD Electronics uses machine-learned interatomic potentials and quantum-mechanical simulation in Synopsys QuantumATK atomistic simulation software to explore semiconductor materials and predict their behavior. QuantumATK software combines NVIDIA accelerated computing with CUDA-X libraries to accelerate atomistic simulations. This helps researchers evaluate larger materials design spaces faster and identify promising candidates for further development.

Looking ahead, EMD Electronics views its collaboration with Synopsys as a foundation for more connected, simulation-driven digital workflows spanning materials, processes, devices, packaging, and system-level performance. The company expects these capabilities to reduce development cycles, improve predictive accuracy, and enable earlier identification of promising materials, thereby accelerating the path from discovery to manufacturing adoption.

2: How imec Advances the Future of Semiconductor Innovation

As semiconductor scaling approaches fundamental physical limits, lithography systems and their associated technology challenges are also becoming more complex. In extreme ultraviolet (EUV) and high-numerical-aperture (High-NA) EUV lithography, for example, interconnected variables are making experimental optimization difficult. This is not the only challenge either. Advanced experimental tools are costly and limited, while process windows continue to shrink.

To face these challenges, imec has seen the benefits of GPU-accelerated Synopsys S-Litho simulation software and the S-Litho EUV feature, accelerated by NVIDIA GPUs, to evaluate advanced lithography processes. These solutions help researchers model imaging, masks, resists, stochastic effects, and process-window performance for deep ultraviolet (DUV), EUV, and High-NA EUV applications. The Synopsys solutions also support cross-department collaboration and provide access to GPU-accelerated computing. In a benchmark using imec’s model, Synopsys demonstrated up to a 40× speedup with GPU-accelerated S-Litho.

“(Synopsys solutions) allow us to investigate process sensitivity, optimize mask designs, evaluate future lithography concepts, and gain insight into failure mechanisms that would be very difficult or expensive to study experimentally.”

— Eric Hendrickx, R&D manager at Imec

For imec, "the most important benefit has been confidence,” explains Eric Hendrickx, R&D manager at Imec. Rigorous simulation enables imec researchers to reduce experimental costs, make technology decisions earlier, and evaluate future options — sometimes years before production hardware exists. As a result, teams can concentrate resources on the most promising options. This will only become more important as physical AI and GPU-accelerated simulation empower the evaluation of millions of possibilities while remaining grounded in reality. “Together, these technologies will shorten development cycles, improve engineering confidence, reduce cost, and accelerate the introduction of future semiconductor technologies that would otherwise be too lengthy to develop,” Hendrickx adds.

3. How T-Systems Is Providing a Foundation for Physical AI

Physical AI and advanced simulation ultimately depend on infrastructures capable of running increasingly demanding workloads at scale. The T-Systems Industrial AI Cloud (IAIC) provides accelerated computing capacity for AI, simulation, digital twins, and other industrial workloads in a secure and compliant environment.

As product complexity grows, engineers can use the IAIC’s GPU-accelerated infrastructure to “run large-scale simulations, explore deeper design spaces, and make better-informed decisions as products move from concept to deployment,” says Björn Weidenmüller, senior vice president, AI Factory Go-To-Market. This infrastructure provides organizations with a trusted accelerated computing foundation for physical AI, simulation, digital twins, and virtual testing at scale.

For example, designs approved by FORVIA HELLA were also tested using the accelerated computing capabilities provided by T-Systems’ data centers. Using an automotive exterior lighting workload approved by FORVIA HELLA, Synopsys performed tests in the T-Systems sovereign cloud environment that demonstrated a 14.4 times speedup on 32 NVIDIA RTX 6000 PRO Blackwell Server Edition GPUs compared with a 256-core CPU configuration, reducing simulation time from more than 4.5 hours to less than 19 minutes. The near-linear scaling up to 32 GPUs supported rapid turnaround of large-scale optical simulations, enabling engineering teams to accelerate design exploration and decision-making.

Looking ahead, continued integration of T-Systems infrastructure and Synopsys solutions could further help engineering teams scale simulation, explore more design options, and make informed decisions earlier in development.

Re-Engineering Engineering With Synopsys Solutions

Across industries, a common theme is emerging: companies are relying on simulation, AI physics, digital twins, and accelerated computing to solve increasingly complex engineering challenges before they reach the physical world.

Whether validating autonomous aircraft, developing intelligent lighting systems, or advancing semiconductor manufacturing, organizations are using these technologies to reduce development cycles, improve validation, and increase confidence in product performance.

These technologies are also unlocking new potential collaborations that can advance entire industries. For instance, “as semiconductor architectures become more complex, materials innovation is increasingly central to advancing performance, power efficiency, reliability, and scale,” says Surésh Rajaraman, executive vice president and chief technology officer at EMD Electronics. “With our collaboration with Synopsys, we have a first-of-its-kind opportunity to connect our deep materials expertise with trusted, physics-based digital engineering to model material behavior, explore complex interactions, and validate choices earlier. This is how we see the next phase of materials innovation: an integrated, data-driven ecosystem in which AI and physics work together to help our customers bring the next generation of semiconductor technologies to market faster.”

As AI continues its expansion into the physical world, the ability to accurately simulate, test, and validate real-world behavior will become a competitive advantage. These real-world examples demonstrate how Synopsys solutions are helping companies bridge the gap between AI and reality, enabling the next generation of innovation through physical AI.

Learn more about Synopsys physical AI solutions. You can also chat with us about physical AI, accelerated computing, and more at NVIDIA GTC Berlin, booth #7011 from October 20-22, 2026.