TSMC is using Nvidia’s CUDA-X and AI tools to speed up advanced chip design and manufacturing in its fabs.
The world’s top chipmaker already builds Nvidia’s most advanced AI platforms, but now it is swallowing more of its supplier’s own dog food.
Nvidia said TSMC is using accelerated computing and AI across semiconductor design, manufacturing, lithography, inspection and fab operations. As chips move to more advanced nodes, getting them from design to mass production has become a computational monster.
Computational lithography, transistor simulation, process control and wafer inspection need huge simulation workloads and real-time optimisation. TSMC is applying Nvidia technologies to improve turnaround time, energy efficiency, yield and operational productivity in advanced fabs.
Nvidia founder and chief executive Jensen Huang said: “Nvidia and TSMC have worked together for nearly three decades to push the limits of computing. TSMC is bringing Nvidia AI and accelerated computing into the fab itself, tackling some of the world’s most complex design and manufacturing challenges with simulation, optimisation, and AI to improve speed, efficiency, and yield for the next generation of chips.”
TSMC chairman and chief executive C.C. Wei said the pair had built a long-standing partnership around next-generation computing.
“TSMC and Nvidia have built a long-standing partnership rooted in advancing the technologies that make the next generation of computing possible. By using NVIDIA accelerated computing and AI across fab operations optimisation, lithography, process control, and inspection, TSMC is strengthening our technology leadership and manufacturing excellence to support our customers’ future products and success,” he said.
TSMC is using Nvidia CUDA-X libraries and AI models to accelerate computational lithography, transistor modelling, process control and fab scheduling. For computational lithography, TSMC is using Nvidia cuLitho, a GPU-accelerated library for chip mask design.
Nvidia claims cuLitho delivers a 20 to 50 per cent improvement in cost effectiveness or cycle time compared with CPU-based computational lithography. For transistor, equipment, and process simulation, TSMC uses Nvidia cuEST, a GPU-accelerated electronic-structure simulation library.
TSMC is using Nvidia cuML to speed large-scale analytics on Nvidia GPUs for advanced process control. That lets TSMC chew through hundreds of thousands of process parameters across thousands of steps as inputs for machine learning models.
For fab operations, CUDA-powered scheduling on Nvidia H200 GPUs has brought claimed productivity improvements by managing constraints and streamlining production paths.
