TSMC and NVIDIA transform chip manufacturing with cuLitho and AI

TSMC, a leader in semiconductor manufacturing, announced it is entering production using NVIDIA’s cuLitho computational lithography platform. According to the NVIDIA blog, this strategic collaboration aims to accelerate the manufacturing of advanced semiconductor chips.

The role of computational lithography

Computational lithography is an important process for transferring circuits to silicon, including complex calculations involving electromagnetic physics, photochemistry, and distributed computing. Historically, this step has been a bottleneck due to its compute-intensive nature, requiring large data centers and consuming billions of CPU hours each year. A typical set of chip masks can require more than 30 million CPU hours, making it a costly and time-consuming process.

Advancing with NVIDIA’s cuLitho

NVIDIA’s cuLitho platform introduces accelerated computing to this process with a system based on 350 NVIDIA H100 Tensor Core GPUs, which can replace a 40,000 CPU system. These advancements dramatically reduce production time, cost and resource consumption, allowing TSMC to push the limits of its current semiconductor manufacturing capabilities.

Dr. CC Wei, CEO of TSMC, highlighted that the integration of GPU-accelerated computing provides a dramatic leap in performance, improved throughput, and reduced cycle time and power requirements. This development was discussed at the GTC conference earlier this year.

Enhancing Generative AI

Beyond accelerated computing, NVIDIA has integrated generative AI into its cuLitho platform. This integration doubles the speed of the optical proximity correction process and improves mask creation. Generative AI helps generate near-perfect inverse masks, takes light diffraction into account, and accelerates processes over traditional methods.

The combination of accelerated computing and AI is transforming semiconductor lithography, a field that has seen little rapid change over the past 30 years. These technologies enable more accurate simulation and realization of complex mathematical techniques that were previously hindered by resource limitations.

Implications for the semiconductor industry

Significant speedups in computational lithography accelerate the development of each mask in the manufacturing process, reducing the overall cycle time for new technology nodes. With cuLitho, techniques such as reverse lithography, once impractical due to time constraints, now become feasible, paving the way for the next generation of powerful semiconductors.

This collaboration between TSMC and NVIDIA marks a pivotal moment in semiconductor manufacturing and demonstrates the potential to advance technology by combining cutting-edge computing and AI.

Image source: Shutterstock