Taiwan Semiconductor Manufacturing Company (TSMC) is incorporating Nvidia’s cuLitho platform into its chip production process, aiming to transform computational lithography—a crucial and highly complex stage in semiconductor manufacturing. Lithography, the most compute-intensive task in chip production, involves intricate computations in electromagnetic physics, photochemistry, and computational geometry.
Traditionally, semiconductor foundries rely on extensive data centers to perform the necessary computations, consuming billions of computational hours annually. However, the physics of these processes has begun to reach its limits. Nvidia’s cuLitho platform, powered by 350 Nvidia H100 Tensor Core GPU systems, offers a breakthrough solution. The company claims cuLitho can replace around 40,000 CPU systems, drastically reducing production time, power consumption, and the space required for data centers.
TSMC CEO C.C. Wei highlighted the significant benefits of integrating cuLitho into the company’s workflow, stating that it has led to dramatic improvements in performance, throughput, and power efficiency.
Nvidia has further enhanced cuLitho by incorporating generative AI capabilities, which promise an additional 2x speedup in processes like optical proximity correction (OPC). OPC is critical in ensuring the accurate transfer of circuit patterns from photomasks onto silicon wafers.
The introduction of accelerated computing and AI into lithography could also bring previously impractical techniques into mainstream production. One such technique is inverse lithography, which involves designing a circuit by first defining the final pattern on the wafer and then computing the optimal photomask pattern. This approach, though long studied, has remained largely in research labs due to the overwhelming computing demands.
By leveraging cuLitho’s accelerated computing power, inverse lithography techniques could become viable for next-generation semiconductor manufacturing, enabling faster development of new chip capabilities and paving the way for more complex and efficient chips.
In short, TSMC’s partnership with Nvidia marks a new chapter in semiconductor innovation, promising significant gains in both efficiency and technological advancement.
