NVIDIA has figured out how its GPUs will significantly speed

NVIDIA has figured out how its GPUs will significantly speed up the creation of photomasks for printing chips – this will make it easier to conquer new engineering processes

Nvidia reportedthat ASML, TSMC and Synopsys adopt the newly introduced cuLitho library. The tool is integrated into the software for designing photomasks used in the manufacture of chips and significantly speeds up the production of lithographic photomasks. Weeks of the heaviest workloads run in 8 hours on a cluster of NVIDIA GPUs. The company is certain that this is the way forward for semiconductor lithography.

    An example of a modern flotation template (photomask).  Image source: Intel

An example of a modern photomask (photomask). Image source: Intel

NVIDIA claims that running cuLitho on the GPU provides up to a 40x performance boost over current software. If we imagine many factories and design centers around the world designing photomasks, that’s tens of billions of hours of processing time per year. That’s megawatts and megawatts of energy that NVIDIA wants the semiconductor industry to keep from burning up.

The company estimates that 500 NVIDIA DGX H100 units do the same job as 40,000 CPUs. Not only does this reduce your carbon footprint, but it also helps you produce 3-5 times more photomasks per day using 9 times less energy compared to current configurations. A photomask that took two weeks to produce can now be processed in one night.

“The chip industry is the backbone of almost every other industry in the world, said Jensen Huang, founder and CEO of NVIDIA. “As lithography reaches its physical limits, the launch of NVIDIA cuLitho and the collaboration with our partners TSMC, ASML and Synopsys enables factories to increase throughput, reduce their carbon footprint and lay the foundation for 2nm and beyond place.”

Vivek Singh, Vice President of NVIDIA Advanced Technology Group, explained at the press conference that cuLitho will not only run on the latest NVIDIA Hopper architecture, but also on the company’s legacy GPUs, starting with the 2017 Volta architecture.

Photomask design is no longer a trivial task. Today, each layer of a chip can require several dozen photomasks, and this isn’t just a geometrically verified pattern. In order to make the device on silicon as small as possible and with sharp edges, different tricks are used with light and projection, when not only light comes into play, but also shadows and distortions and cross-shading and much more. The photomask has become an art that no longer rests on a person’s shoulder.

According to NVIDIA, ASML and Synopsys are busy integrating the cuLitho library into their software. TSMC will begin testing the software on production equipment in June this year, Jensen Huang explained.

As far as you can see from Huang’s talk, the cuLitho library doesn’t use AI algorithms yet. But there is no doubt that over time, full-length AI will be used in preparing chips for production. Machine learning is already being used in chip design, as Google mentioned a while ago. Offering the cuLitho library to industry is undoubtedly a good deed. At the same time, it is deeply embedded in the production and supply chain of the chips. And the further, the stronger this dependency becomes, because today the world of advanced chips depends on only one company – the Dutch ASML

About the author

Dylan Harris

Dylan Harris is fascinated by tests and reviews of computer hardware.

Add Comment

Click here to post a comment