NVIDIA’s cuLitho library has been adopted by ASML, TSMC, and Synopsys to expedite the processing of lithographic photomasks, reducing heavy computing workloads to only eight hours. The library will help to reduce energy consumption and improve throughput in semiconductor lithography. By using the NVIDIA DGX H100 CPUs, it is expected that photomask production will increase three to five times per day while utilizing nine times less energy. CuLitho runs on NVIDIA’s Hopper architecture and enhances the speed of performance by up to 40 times. AI algorithms are not yet used in the cuLitho library, but NVIDIA confirmed that full-length AI would be used in chip design. Advanced chips are currently only supplied by ASML, which deeply embeds itself in the production and supply chain of chips.
NVIDIA’s cuLitho Library Speeds Up Chip Design for ASML and TSMC
The introduction of NVIDIA’s cuLitho library has been adopted by ASML, TSMC, and Synopsys to enable fast and efficient processing of lithographic photomasks. The library reduces the workload of heavy computing tasks that typically run for weeks to just eight hours on a cluster of NVIDIA GPUs. The use of the cuLitho library is considered the future of semiconductor lithography.
According to NVIDIA, the performance of cuLitho on the GPU enhances its speed by up to 40 times compared to current software, resulting in tens of billions of hours of processor load per year. This enhancement aligns with the energy-saving goals of the semiconductor industry.
By utilizing the NVIDIA DGX H100 CPUs, it is expected that 500 CPUs will perform the same job as 40,000 CPUs, reducing carbon footprint and increasing photomask production by three to five times per day, utilizing nine times less energy than current configurations.
NVIDIA’s founder and CEO, Jensen Huang, stated that the semiconductor industry is the backbone of nearly every industry worldwide. Huang also stated that through NVIDIA cuLitho, in collaboration with TSMC, ASML, and Synopsys, fabs can increase throughput, reduce carbon footprint, and lay the foundation for 2nm and beyond.
The cuLitho library is expected to run on the latest NVIDIA Hopper architecture, including the 2017 Volta architecture. ASML and Synopsys are currently integrating the cuLitho library into their software. In June of this year, TSMC will start testing the software on production equipment.
Constructing photomasks has become a tedious task, with each layer of a chip requiring several hundred photomasks. This process involves different techniques, such as shadow, distortion, cross-shading, and more, with light and projection.
NVIDIA confirms that AI algorithms are not yet being used in the cuLitho library, but full-length AI will be used in chip design, as previously discussed by Google. Currently, the world of advanced chips depends on the Dutch company, ASML, as it deeply embeds itself in the production and supply chain of chips.
Don’t miss interesting posts on Famousbio
