NVIDIA cuLitho targets the calculations behind chipmaking

NVIDIA announced cuLitho, a GPU software library, on 21 March. It claimed up to 40 times faster computational lithography. TSMC and Synopsys were integrating it; ASML planned GPU support across its computational-lithography products.

Photo: © 2020 Will Buckner, NVIDIA sign, 31 October 2020. CC BY 2.0. Cropped for display. Historical identification, not announcement imagery.

The company projected three to five times more photomasks per day and described processing a mask overnight instead of two weeks. Those are supplier claims about a specific workload, not a measured fortyfold increase in finished chips.

ASML’s 2022 annual report, published in February, explains the distinction. Computation optimizes mask patterns and machine settings, while inspection and process-control systems manage manufacturing variation. Its chipmaking sequence also includes deposition, coating, baking, etching and other steps.

The same report placed customers’ High-NA EUV research in 2024–2025, targeting high-volume manufacturing in 2025–2026. That timetable was a plan, not an achieved production milestone.

Read together, the records identify where faster computing could help: preparing and refining the patterns a fab will print. They don’t establish how much an entire factory’s output would rise.