Nvidia NB-Cell transforms chip design with AI

The Nvidia NB-Cell tool

Nvidia is a pivotal company in the growth of the AI market, so it is no surprise that it also leans heavily on these technologies itself. As the company’s employees have explained, artificial intelligence is applied at several points throughout Nvidia’s in-house chip design process.

How Nvidia uses AI in chip development

Further detail was shared by Nvidia’s Chief Scientist, Bill Dally. As an illustration, he pointed to a tool called Nvidia NB-Cell. With its help, moving a standard cell library onto a new semiconductor process - a task that previously took roughly 10 months for a team of eight - can now be completed in a single night using just one GPU. What is more, the output can sometimes surpass what is achieved using the traditional approach.

Bill Dally’s comments on NB-Cell

"So we try to use AI everywhere we can in the design process, and there are many examples of that. For example, every time we get a new semiconductor process, we have to port our standard cell library to that process. We’re talking about roughly 2500-3000 cells, and it used to take a team of eight people about 10 months to do that - 80 person-months.

Then we developed a reinforcement learning-based programme called NB-Cell. I think we’re now on NB-Cell 2 or 3. And it happens overnight on one GPU. The results are actually better than human-designed implementations, on metrics like cell size, dissipated power and delay. They match or exceed human-designed implementations.

That’s a huge productivity increase, and it removes a barrier to moving to new processes, because now we can port cell libraries very quickly"