If 2023 still hasn’t had enough topics surrounding AI, NVIDIA is happy to announce that they are putting them into the manufacturing process of the digital brains behind every system in the future.
Called the cuLitho, it is a software library made for computational lithography – the process of carving every single bit of a processor onto a large silicon wafer through the use of a photomask layer. We’ll skip through the technical parts but there are some facts to know that the entire process takes quite an amount of time even at today’s production rate. The power of AI being far greater than humans when it comes to technicality and precision is where cuLitho is being applied to help break the physical limitation for 2nm processes and beyond.
TSMC and Synopsys are currently integrating cuLitho into their workflow while world-leading equipment maker ASML is in the discussion and preparation phase with NVIDIA in terms of adding it into all of its computational lithography software products in the coming days.
Well, let’s talk about how many benefits are there to adopting cuLitho. Citing internal estimation, it can speed up to 40X faster than the current lithography process by rivaling the work of 40,000 CPU systems with just 500 NVIDIA GDX H100 systems, further cutting down power consumption and raising efficiency. Meanwhile, the usual timing for a creation of a photomask is about 2 weeks but with cuLitho, it can be slashed down to less than 24 hours and I’m not even mentioning up to 9x less power needed. Crazy.
Let’s not forget AI’s ability to provide better decisions when it comes to design rules and circuit planning which can lead to higher density and yields in the long run. Less wastage = lower production cost = lower retail price.
So I guess the future where AI designing and powering their own hardware and software upgrades is just a matter of when.