NVIDIA has developed “cuLitho”, a software library designed to accelerate computational lithography, which will be implemented at TSMC starting June 2023. By combining cuLitho with the latest GPU hardware, computational lithography can be expedited, potentially improving yields and reducing chip costs. Additional benefits include a reduced carbon footprint, faster turnaround times, and advanced process nodes with smaller feature sizes.
Vivek Singh, NVIDIA’s vice president of accelerated computing, said that cuLitho not only accelerates mask manufacturing but also expedites the entire development cycle for foundries employing the technology. As computational lithography techniques have become more complex, the field has grown to consume tens of billions of CPU hours annually, representing the largest computational load in chip design and manufacturing.
Currently, computational lithography utilises two primary techniques: Optical Proximity Correction (OPC) and Inverse Lithography Technique (ILT). The more advanced ILT, supported by cuLitho, allows for larger wafer area focus, improved yields, and reduced chip costs. Singh highlighted that as miniaturisation advances, ILT is shifting from a “nice to have” to a “must-have” technology.
Using NVIDIA’s latest GPU hardware platform based on “H100” alongside cuLitho, computational lithography can be accelerated up to 42 times compared to previous speeds. Jensen Huang, CEO of NVIDIA, announced at the “GTC (GPU Technology Conference) 2023” that cuLitho would enable TSMC to reduce prototype cycle times, increase throughput, and lower the carbon footprint of manufacturing, preparing for the 2nm process node and beyond.
NVIDIA is also collaborating with EUV equipment manufacturer ASML and EDA tool provider Synopsys. ASML CEO Peter Wennink stated that the company plans to integrate GPU support into all of their computational lithography software, while Aart de Geus, CEO of Synopsys, highlighted the significant acceleration in performance brought about by the collaboration.