NVIDIA's Computational Lithography Technology to Be Introduced to Advanced Semiconductor Processes

▲NVIDIA introduces accelerated computing to computational lithography (Photo: NVIDIA)

As more transistors are integrated into advanced nodes and accuracy requirements become more stringent, the computational time required for the most demanding workloads in semiconductor manufacturing has outpaced Moore's Law in recent years. Future nodes will require more detailed calculations, but the available computational bandwidth of current platforms cannot accommodate all of them, slowing the pace of semiconductor innovation.

Changing foundry processes often requires modifying OPC, creating bottlenecks. Libraries that help eliminate these bottlenecks and support new solutions required for new technology nodes, such as curved masks, high-NA EUV lithography, and subatomic photoresist modeling, are gaining attention.

Nvidia recently announced a groundbreaking technology that brings accelerated computing to the field of computational lithography. This, semiconductor leaders such as ASML, TSMC, and Synopsys expect, will accelerate the design and manufacturing of next-generation chips, as current production processes approach the limits of what physics can achieve.

Global foundry TSMC and electronic design automation leader Synopsys are integrating the new NVIDIA cuLitho software library for computational lithography into their software, manufacturing processes, and systems. Equipment manufacturer ASML has also announced a close collaboration with NVIDIA on GPUs and cuLitho, with plans to integrate GPU support into all of its computational lithography software products.

This is said to enable chips with smaller transistors and wires than currently available, while also shortening time to market. It could also improve the energy efficiency of large-scale data centers, which operate 24/7, to power manufacturing processes.

“The chip industry is the foundation of nearly every industry on earth,” said Jensen Huang, CEO of NVIDIA. “With lithography reaching the limits of physics, NVIDIA is introducing cuLitho and working with our partners TSMC, ASML, and Synopsys to help foundries increase throughput, reduce their carbon footprint, and build the foundation for 2nm and beyond.”

Running on GPUs, cuLitho delivers up to a 40x performance leap over traditional lithography, the process of creating patterns on silicon wafers, accelerating massive computational workloads that currently consume hundreds of billions of CPU hours each year. This allows 500 NVIDIA DGX H100 systems to do the work of 40,000 CPU systems, running all parts of the computational lithography process in parallel, reducing power demands and potential environmental impact.

In the short term, foundries using cuLitho will be able to produce three to five times more photomasks—the templates for chip designs—per day, using nine times less power than current configurations. Photomasks that used to take two weeks to produce can now be processed overnight.

“The cuLitho team has made remarkable progress in accelerating computational lithography by moving expensive operations to the GPU,” said CC Wei, CEO of TSMC. “This will make a significant contribution to TSMC’s continued semiconductor expansion by opening up new possibilities for lithography solutions, such as reverse lithography and deep learning, to be applied more broadly in chip manufacturing.”

“We plan to integrate support for GPUs into all of our computational lithography software products,” said Peter Wennink, CEO of ASML. “Our collaboration with NVIDIA on GPUs and cuLitho will provide tremendous benefits to computational lithography and semiconductor scaling, especially as we move into the era of high-nanometer EUV lithography.”

“Computational lithography, and optical proximity correction (OPC) in particular, is pushing the limits of compute workloads on advanced chips,” said Aart de Geus, CEO of Synopsys. “By running Synopsys OPC software on the cuLitho platform, we have dramatically accelerated performance from weeks to days.”