The two companies are using Nvidia's accelerated computing tools across lithography, defect inspection, and factory scheduling
Bloomberg / Getty Images
TSMC is using several Nvidia software libraries on Nvidia GPUs to address some of the most computationally intensive stages of chip production. Benchmarks cited by both companies put cuLitho's performance advantage at 20% to 50% over CPU-based workflows, measured in either cost savings or cycle time reduction. Chemistry simulations that once relied on conventional methods can now be completed up to 50 times faster through cuEST, accelerating TSMC's evaluation of new semiconductor materials.
On the process control side, TSMC is using Nvidia's cuML machine learning library to analyze hundreds of thousands of manufacturing parameters across thousands of production steps, allowing engineers to reduce process variation. Production scheduling across TSMC's fabs is handled in part by Nvidia H200 GPUs, whose computing capacity gives engineers greater flexibility in navigating manufacturing constraints and keeping output moving efficiently.
TSMC is also using the Nvidia Metropolis platform and the Nvidia TAO Toolkit to improve defect detection. Using computer vision AI, the company said it has improved its ability to identify defects at nanometer scale while reducing the need for repeated data labeling and model retraining as production conditions change.
In addition, TSMC is exploring Nvidia's Omniverse libraries to build what it calls FabTwin, a virtual environment for simulating and evaluating chip factory layouts before committing to physical construction or capital spending. The virtual-first approach is intended to improve planning efficiency and accelerate decisions about tool placement and workflow design, the company said.
"TSMC is bringing Nvidia AI and accelerated computing into the fab itself, tackling some of the world's most complex design and manufacturing challenges with simulation, optimization and AI to improve speed, efficiency and yield for the next generation of chips," Nvidia CEO Jensen Huang said in a statement.
TSMC CEO C.C. Wei said the expanded use of Nvidia tools is intended to strengthen TSMC's manufacturing capabilities in support of its customers' future products.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.