At GTC Taipei, NVIDIA said TSMC is now using NVIDIA accelerated computing and AI across the heart of its advanced semiconductor design and manufacturing flow — from computational lithography and transistor simulation to fab-floor process control, defect inspection and digital-twin planning of new tools. The deepening alliance turns the world's largest foundry into a heavyweight reference customer for NVIDIA's CUDA-X, Metropolis and Omniverse stacks.
CUDA-X Libraries Speed Up The Hardest Steps
TSMC has wired NVIDIA's GPU-accelerated libraries into the steps that gate every advanced node. cuLitho, NVIDIA's lithography library, delivers a 20–50% cost or cycle-time improvement against CPU-based computational lithography. cuEST runs electronic structure simulations roughly 50x faster than CPUs for new semiconductor materials. cuML ingests hundreds of thousands of process parameters spanning thousands of steps to compress variation across TSMC's most complex fabs.
Scheduling 200,000 Wafer Steps With H200
CUDA-powered scheduling on NVIDIA H200 GPUs is doing the heavy combinatorial work behind production throughput, letting TSMC manage the constraint stack — tools, recipes, materials — at a higher resolution than before. The companies frame the result as "notable improvements in fab productivity" without quoting a specific yield delta. "TSMC is bringing NVIDIA AI and accelerated computing into the fab itself," said NVIDIA founder and CEO Jensen Huang.
Vision AI On Nanometer-Scale Defects
For inspection, TSMC is layering NVIDIA Metropolis and the NVIDIA TAO Toolkit on top of its in-house tooling to do automated defect classification at nanometer scale, cutting re-labelling and retraining cycles as recipes and inspection equipment change.
FabTwin: An Omniverse Digital Twin Of A New Fab
TSMC is also piloting NVIDIA Omniverse to build "FabTwin," a virtual fab environment for evaluating tool layouts and process flows before any concrete is poured. C.C. Wei, TSMC's chairman and CEO, framed the partnership in continuity terms: "TSMC and NVIDIA have built a long-standing partnership rooted in advancing the technologies that make the next generation of computing possible." For the broader market — already absorbing Europe's EU Chips Act 2.0 push, fresh rack-scale AI announcements at Computex and a robotaxi reset — it cements NVIDIA as both supplier and operational backbone to the chips that make AI run.
Reporting based on coverage from NVIDIA Newsroom and GlobeNewswire (NVIDIA GTC Taipei).