TSMC Implements Nvidia‘s Advanced Computing Solution to Streamline Chip Production
Taiwan Semiconductor Manufacturing Company (TSMC) is set to enhance its semiconductor production capabilities by adopting Nvidia’s cuLitho platform, aimed at accelerating the computational lithography process crucial for advanced chip manufacturing.
Computational lithography is a vital component of chip fabrication and represents the most compute-intensive task within the semiconductor production workflow. It involves complex calculations across various scientific disciplines, including electromagnetic physics, photochemistry, and computational geometry.
Historically, semiconductor foundries have relied on expansive data centers to perform these computations. However, traditional systems consume an astonishing tens of billions of hours annually and are approaching the physical limits of computing power.
Nvidia’s cuLitho platform leverages the power of 350 Nvidia H100 Tensor Core GPUs, effectively replacing the need for 40,000 CPU systems. This transition promises significant reductions in production time, energy consumption, and data center footprint.
“Collaborating with Nvidia to incorporate GPU-accelerated computing into TSMC’s processes has led to remarkable enhancements in performance, substantial throughput improvements, and shorter cycle times, all while decreasing power requirements,” stated TSMC CEO C.C. Wei.
In addition to enhancing computational efficiency, Nvidia has integrated generative AI capabilities into the cuLitho platform, potentially doubling the speed of operations by refining optical proximity correction (OPC). OPC is a critical step in semiconductor lithography that improves the precision of transferring circuit patterns from a photomask onto silicon wafers.
The acceleration of computational lithography will facilitate faster mask production, thereby expediting the development of new chip technologies. It also opens the door to previously impractical calculations, such as inverse lithography techniques, which researchers have explored for two decades. This innovative method allows designers to start with a desired final wafer pattern and compute the optimal photomask design while accounting for optical and physical distortions.
Until now, the high computing demands of these techniques have kept them confined to research labs. However, the implementation of Nvidia’s cuLitho platform could pave the way for their practical application in manufacturing next-generation semiconductors.
Related topics:
Xbox Eyes Android Expansion Following Court Ruling Against Google
Trek Unveils CarBack Bike Radar for Enhanced Cycling Safety
Asahi Linux Breaks Barriers, Enables Windows Gaming on Apple Silicon Macs