Korea Accelerates Physical AI Integration in Manufacturing and Mobility with NVIDIA Technologies

South Korea is significantly advancing its physical AI ecosystem by collaborating with NVIDIA to integrate accelerated computing and AI techniques across several industrial sectors. This initiative capitalizes on Korea's manufacturing and engineering strengths to enhance semiconductor production, autonomous vehicle development, and advanced robotics.

In semiconductor engineering, Samsung Electronics is expanding its partnership with NVIDIA to accelerate materials research, process development, and device engineering. Samsung employs NVIDIA’s CUDA-based libraries, such as cuISS and cuDSS, to speed up computations involving sparse linear systems critical for technology computer-aided design (TCAD). Their sMOL platform utilizes NVIDIA cuEST to execute quantum chemical calculations up to five times faster, enhancing material property simulations.

Samsung is also applying NVIDIA cuEquivariance to optimize tensor operations in training atomistic AI models used for semiconductor research. A notable achievement includes using NVIDIA PhysicsNeMo to develop thermal and mechanical analysis workflows, where AI homogenization combined with GPU acceleration delivered approximately a 1,600-fold reduction in simulation time, from 18 hours on CPUs to just 41 seconds on GPUs.

In the autonomous vehicle sphere, Kakao Mobility is advancing level 4 self-driving capabilities by integrating its proprietary driving software with NVIDIA’s vehicle computing platform. Utilizing NVIDIA Hyperion, powered by DRIVE AGX Thor, and NVIDIA Halos for safety validation, the collaboration focuses on deploying autonomous systems in real urban environments like Seoul’s Gangnam district, initially on Kia’s purpose-built vehicles.

LG Electronics, in partnership with NVIDIA, has demonstrated practical robotic manipulation and manufacturing tasks using LG’s CLOiD robot powered by the NVIDIA Isaac GR00T platform. The collaboration established a physical AI data flywheel involving real-world data collection, simulation using NVIDIA Cosmos, AI model training, and onboard computing through NVIDIA Jetson AGX Thor, creating a reusable framework to enhance robot capabilities in manufacturing environments.

These projects emphasize a shift from experimental AI models to operational physical AI systems, enabling more efficient manufacturing processes, improved autonomous navigation, and smarter robotic automation in real-world settings.

While these advancements showcase substantial performance improvements and integration efforts, challenges remain in scaling physical AI deployments widely, ensuring robust safety validation, and maintaining sovereignty over AI technologies through local collaboration and ecosystem development.

Looking ahead, tools like the NVIDIA Agent Toolkit are anticipated to further streamline engineering simulations and AI model integration, enhancing automated analyses and design optimization across industries in Korea and beyond.