TSMC deepens NVIDIA AI integration across chip manufacturing

Taiwan Semiconductor Manufacturing Co (TSMC) has expanded its deployment of artificial intelligence and accelerated computing technologies across its semiconductor development and manufacturing operations, according to an announcement made during NVIDIA’s GTC Taipei event. The partnership, detailed by NVIDIA, spans lithography, materials research, factory optimisation, and defect detection, marking a significant deepening of ties between the world’s largest chipmaker and the leading graphics processor manufacturer.

Central to this initiative is the implementation of NVIDIA’s cuLitho platform for computational lithography. TSMC reports that the technology has improved cost efficiency or processing cycle times by 20% to 50% compared with traditional CPU-based approaches. This acceleration targets one of the most computationally intensive stages of chip production, allowing for more efficient optimisation of chip patterning and production workflows.

In materials research, TSMC is utilising NVIDIA’s cuEST software for electronic structure simulation. The companies state that the platform delivers chemistry simulations up to 50 times faster than conventional methods. By shortening these simulation times, engineers can evaluate a broader range of material candidates, thereby accelerating research and development cycles for advanced semiconductor materials.

For manufacturing process optimisation, TSMC has integrated NVIDIA’s cuML machine learning library into its advanced process control systems. This integration enables the analysis of hundreds of thousands of manufacturing parameters across thousands of production stages, helping to identify inefficiencies and reduce process variation. TSMC notes that this has contributed to meaningful improvements in process consistency and operational performance.

The semiconductor manufacturer is also deploying NVIDIA H200 GPUs to support production scheduling and factory management. By leveraging GPU-accelerated computing for scheduling calculations, TSMC aims to better manage complex manufacturing constraints and optimise production flows within its fabrication facilities, resulting in measurable productivity improvements.

Quality control and inspection have also been enhanced through the use of NVIDIA’s Metropolis platform and the TAO Toolkit. These tools facilitate the development of advanced vision AI systems capable of identifying defects at the nanometer scale. The technology improves defect classification accuracy while reducing the manual labour required for model retraining, helping to streamline inspection processes and improve manufacturing yields.

Additionally, TSMC is evaluating NVIDIA Omniverse libraries for its FabTwin initiative, a virtual manufacturing environment designed to simulate and optimise fabrication facilities. This digital platform allows engineers to test equipment layouts and production scenarios in a virtual setting before implementing changes in physical facilities, a move intended to reduce deployment risks and improve planning efficiency.

NVIDIA founder and Chief Executive Officer Jensen Huang highlighted the strategic importance of this collaboration, noting that TSMC is bringing AI and accelerated computing directly into the fab to tackle complex design and manufacturing challenges. The announcement underscores the semiconductor industry’s broader shift toward adopting AI-driven tools to enhance efficiency and performance in chip production.