TSMC has begun using NVIDIA’s cuLitho computational lithography platform to accelerate the production of next-generation semiconductor chips. The collaboration between TSMC and NVIDIA aims to push the boundaries of physics in semiconductor manufacturing, where computational lithography plays a critical role in transferring circuitry onto silicon wafers. This process, which involves complex computations from fields like photochemistry and electromagnetic physics, has traditionally been a bottleneck for chipmakers, requiring massive CPU-driven data centers.
NVIDIA’s cuLitho platform, powered by 350 H100 Tensor Core GPU systems, can replace up to 40,000 CPU systems in semiconductor foundries. This move significantly reduces time, space, and power requirements while speeding up production. With NVIDIA’s GPU-based systems, the computational workload that previously took tens of billions of hours annually can now be accomplished much faster. This breakthrough enables foundries to speed up the development of advanced technology nodes and high-performance chips that would otherwise be constrained by traditional CPU computing.
In addition to accelerated computing, NVIDIA is applying generative AI to further improve cuLitho’s efficiency. This AI-driven approach provides a 2x speedup in the optical proximity correction process, a critical step in semiconductor lithography that compensates for light diffraction during the printing of chip designs. The combination of accelerated computing and AI-driven algorithms promises to revolutionize the industry, allowing for new designs and calculations that were previously impractical.
Key Points:
• TSMC adopts NVIDIA’s cuLitho to accelerate semiconductor manufacturing.
• cuLitho with NVIDIA H100 GPUs replaces 40,000 CPU systems, reducing costs and time.
• Generative AI applied to cuLitho delivers an additional 2x speedup in chip design processes.
• New calculations, such as inverse lithography techniques, become practical at full chip scale with cuLitho.
• cuLitho addresses one of the most compute-intensive processes in semiconductor design.
“Our work with NVIDIA to integrate GPU-accelerated computing in the TSMC workflow has resulted in great leaps in performance, dramatic throughput improvement, shortened cycle time, and reduced power requirements,” said Dr. C.C. Wei, CEO of TSMC.
