Industrial Autonomy in the Era of Physical AI: Building the Autonomous Fab Stack

5:05 pm - 5:30 pm

Physical AI is pushing semiconductor manufacturing beyond smart factory toward true industrial autonomy, where perception, reasoning, and control are embedded across the entire fab stack. Fabs evolve from highly automated facilities into continuously learning systems that connect sensor-rich equipment, high-fidelity digital twins, and large-scale AI infrastructure into a single, coordinated intelligence fabric.

At the center is a Physical AI data flywheel that unifies equipment telemetry, process and recipe parameters, metrology and defect data, and logistics signals into a shared data and model layer. This flywheel powers self-improving AI agents that predict failures, adapt process windows, and optimize scheduling and material handling—turning every wafer, lot, and excursion into training signal for the next, more efficient run.

To realize this at scale, the autonomous fab stack spans three tightly coupled computing domains: edge intelligence near tools for real-time perception and control, GPU-accelerated simulation and digital twins for physics-accurate what-if analysis, and data center–class AI platforms for foundation models and global optimization across fab networks. This session presents a technical blueprint for how semiconductor manufacturers, equipment vendors, EDA partners, and software providers can act as a connected Physical AI ecosystem—aligning data models, APIs, and digital twin representations so that models, agents, and workflows can be shared and co-developed, rather than remaining in fragmented, siloed deployments.

Featured Speakers

Jonghwan Lee (invited)

Sr. Solutions Architect – Physical AI, NVIDIA

Jonghwan Lee is a Senior Solutions Architect for Physical AI at NVIDIA, with over eight years of experience at NVIDIA as a CUDA and AI development specialist. His work spans the full stack of accelerated computing and AI system architecture, from low-level GPU programming and performance optimization to large-scale AI deployment. As the industry enters the era of Physical AI, his current research and development efforts focus on Physical AI data flywheels, digital twin platforms, large-scale simulation, and robotics, enabling closed-loop learning between real-world physical systems and AI models. He works closely with customers to design end-to-end architectures that integrate simulation, perception, planning, and control for autonomous industrial systems.

Prior to joining NVIDIA, Lee spent eight years at the CTO SIC Center of LG Electronics, where he conducted in-depth research and development on mobile GPU architectures, neuromorphic processors, and DNN accelerator architectures. His work included architectural exploration, performance–power trade-off analysis, and hardware–software co-design for next-generation AI accelerators, contributing to energy-efficient and scalable computing platforms for embedded and mobile AI workloads.

Lee’s professional focus lies in building end-to-end Physical AI systems, leveraging deep expertise across computer vision, GPU and accelerated computing, AI accelerator architecture, autonomous systems, digital twins, simulation, and robotics. He continues to drive innovation at the intersection of data, simulation, and accelerated computing, enabling the realization of autonomous and intelligent factories.