The semiconductor giant is utilising Nvidia's cuLitho platform, a suite of tools to enhance computational lithography processes — which comes as TSMC is ramping up production of Nvidia's latest GPUs, including the H200 series and the highly anticipated Blackwell architecture, which has faced setbacks due to hardware complications.
Subscribe today for free
Computational lithography is a crucial process in chip manufacturing where companies like TSMC use algorithms and simulations to optimise the photomasks used in creating chips.
The process has become increasingly important and time-consuming as chip features have shrunk to nanometer scales but it ensures that the tiny structures on a chip can be accurately printed.
Nvidia has developed a library called cuLitho, which the company claims can speed up certain computational lithography calculations by up to 40x using GPU acceleration, potentially helping chip manufacturers like TSMC to reduce design and production times.
According to Nvidia, a typical mask set for a chip can take 30 million or more hours of CPU compute time, requiring sizable data centres within semiconductor foundries to power workloads.
With the cuLitho library, Nvidia suggests just 350 of its H100 GPUs running cuLitho could replace servers, reducing costs, space, and power for manufacturers like TSMC.
Where chip photomasks previously took two weeks to process, Nvidia suggests its cuLitho library can perform the task overnight, with 3x to 5x more masks able to be generated in a day.
“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,” Dr C.C. Wei, CEO of TSMC, said at Nvidia’s GTC conference earlier this year.
TSMC has now moved into production leveraging the cuLitho library, at a time where the company is hard at work building Nvidia’s next generation of chips.
TSMC was due to start Blackwell production earlier in the year, but a pushback was enforced after it uncovered a flaw in the B200 chips’ processor die.
Nvidia was forced to rework the design, with some shipments expected at the end of the year pushed back to 2025.
The company confirmed the issue has since been resolved, with demand for the powerful GPU “well above supply.”
RELATED STORIES
