Overview
- The companies announced at NVIDIA GTC Taipei on June 1, 2026 that TSMC is adopting multiple NVIDIA GPU libraries and AI toolkits inside its fabs to accelerate design‑to‑production workflows.
- TSMC is using NVIDIA cuLitho for computational lithography, which NVIDIA says can improve cost effectiveness or cycle time by 20 to 50 percent compared with CPU‑based lithography.
- For materials and transistor modeling, TSMC deployed NVIDIA cuEST for electronic‑structure simulations that NVIDIA reports run about 50 times faster on GPUs than traditional methods.
- TSMC is also applying NVIDIA cuML for large‑scale process analytics, Metropolis and the TAO Toolkit for vision AI defect classification at nanometer scale, and CUDA on H200 GPUs to speed production scheduling.
- The move deepens a nearly 30‑year partnership and, if company claims hold, could lower wafer waste and shorten product ramps for customers, though the performance figures come from company announcements and have not been independently verified.