TSMC will use Nvidia’s cuLitho platform for chip design. This platform uses Nvidia GPUs, instead of the CPUs currently used, and artificial intelligence. Nvidia says cuLitho is faster and more efficient than existing methods.
CuLitho is one Numerical lithographyplatform used in making the so-called pictures, several of which are used during the chip production process. The masks act as patterns for the exposure steps in the chip manufacturing process. The chip designs are imaged on the wafer with light. These photographs must be ‘distorted’ to compensate for optical problems such as diffraction and this requires a lot of calculations.
Nvidia wants that the cuLitho platform can make that calculation more efficient by using GPUs, instead of the current data center CPUs. Nvidia claims that 350 H100 systems can take over the work of 40,000 CPU systems, although Nvidia does not reveal exactly what type of CPUs this applies to.
In addition, Nvidia says it has created algorithms that use generative AI for use with the cuLitho platform. This algorithm would speed up the cuLitho platform up to two times. With this generation AI, chip designers can also achieve ‘almost perfect’ indirect masks from indirect solutions to use According to Nvidia, this technology has been discussed among scientists for two decades, but until now it could not be used in practice because there was not enough computing power. According to Nvidia, this could be possible with cuLitho, allowing new chips and nodes to be designed faster. In addition to TSMC and Nvidia, ASML and Synopsys contributed to cuLitho development.
2024-10-09 12:03:00
#TSMC #Nvidia #GPUs #faster #chip #design
