NVIDIA's CEO Jensen Huang is optimistic about the use of GAA-focused processes in the future, claiming that they could bring a massive boost to performance.
NVIDIA Is Set To Employ GAA Technologies With TSMC's N2 Process, Likely Being Featured In Feynman AI Lineup
We all know that Jensen isn't a firm believer in Moore's Law or lets it dictate the trend of performance brought in by newer architecture. NVIDIA's key focus, not just with AI but also other products, is to frame their lineups so that they defy the traditional performance scaling norms, and this is what the company has done over the course of many years, which we'll discuss ahead. This is why when Jensen was asked how he sees GAA (Gate-All-Around) technology (via EETimes), he claimed it would make a performance difference but wouldn't "change the world."
NVIDIA's CEO claims that GAA transistors could bring in a 20% performance boost with a particular generation, although he didn't specify in what area we are looking to see changes, suggesting that the statement was vague. However, in the case of large-scale AI clusters, the key factor to consider is how individual components are being managed and connected with each other since, by focusing on this, NVIDIA has managed to achieve far more significance than what Moore's Law intends for.
The chip industry has seen uncertainty around GAA adoption up until now, mainly since Samsung was the first company to announce a fully developed process based on the technology, the 3nm GAA. Despite the Korean giant's optimism towards their node, the company has failed to achieve industry-standard yield rates, and the last time, they were said to be at 20 percent, far lower than the required figure. As far as TSMC is concerned, the Taiwan giant is said to have introduced GAA with their N2 (2nm) lineup of processes.
For NVIDIA, we are projecting it to utilize GAA technology with their recently-announced Feynman AI architecture, which is set to release in 2028 and will likely employ TSMC's N2 process. Team Green doesn't adopt the latest process in their product lineup; rather, it waits a few years, likely to get the node to become more mature in the market. However, for Team Green, the "Huang law" is what matters the most, which is to drive performance through architectural advancements, and this is how the company has scaled up chip efficiency by over 1,000x in a decade.
News Source: Tom's Hardware
