NVIDIA has announced something interesting because they are incorporating AI into the production process of the digital brains of modern electronics through the reveal of the NVIDIA cuLitho.
The NVIDIA cuLitho is a software library designed for computational lithography that utilizes photomask layers to carve out each nit and bit of a processor onto a sizable silicon wafer. We’ll skip the technical details about lithography but it’s important to realize that even at the current manufacturing rate, the entire process takes a long time. As such, CuLitho is being used to help overcome the physical restriction for 2nm processes and beyond because AI is significantly more powerful than humans when it comes to accuracy and precision.
The great software stack is already in the process of being integrated into workflows by TSMC, Synopsys, and other industry leaders. ASML is currently in talks with NVIDIA about incorporating it into all of its computational lithography software solutions in the near future as well.
There are several advantages of adopting cuLitho which includes an official estimate of rivaling 40,000 CPU systems’ worth of work with just 500 NVIDIA GDX H100 systems, resulting in up to 40X faster production across the entire lithography process. Or with a tangible example, it normally takes 2 weeks to create a photomask but cuLitho can complete it just overnight, and that’s before even mentioning the up to 9x lower power requirements.
Some general benefits can also be found such as better decisions when it comes to circuit planning and design principles, which could eventually result in higher yields and densities leading to lower production costs and subsequently retail prices resulting from less waste.
So when will the time AI starts creating and fueling its own hardware and software upgrades comes?
