Intel to be TSMC’s 1st customer for 3nm CPUs and GPUs,

Intel to be TSMC’s 1st customer for 3nm CPUs and GPUs

According to the new information, Intel would become TSMC’s first 3nm customer by 2023 or 2024. Apple would be the other customer interested in this lithography, but Intel will also want to invest in this technology for its Arc Celestial and Battlemage graphics cards, as well as for its Meteor Lake processors.

Pat Gelsinger is clear that the future lies in pushing the process forward, and so he has spearheaded the idea of using TSMC for this purpose. It has already been revealed at the investor day that Arrow Lake and Meteor Lake will come with 4nm and 2nm respectively in conjunction with 3nm chipsets from TSMC.

Intel Meteor Lake and Arrow Lake with TSMC behind them, 3nm!

There is talk of the IDM 2.0 strategy and Intel Foundry’s billion-dollar investment to try to upgrade chip foundries in the US. However, the Taiwanese are way ahead of 10nm and 7nm and offer a very reliable 5nm node. So if you can’t beat your enemy, join him: a partnership between Intel and TSMC. Specifically, this collaboration will cover Arc graphics cards as well as Meteor Lake and Arrow Lake processors. These chips will have the Intel 4 process and will launch between 2023 and 2024.

The Alchemist GPUs will be manufactured on a 6nm process, the Arc Celestial on a 5nm process, and the Battlemage on 3nm. In the case of Arrow Lake and Meteor Lake, they will come with three different manufacturing processes: 4nm and 2nm on their dies and 3nm chipsets. So we are talking about a long-term alliance between Intel and TSMC: from 2023 to 2026, 3 years in chip manufacturing. The GPU and CPU chips will be manufactured using Intel’s own Foveros and EMIB technologies.

Intel’s roadmap is no small thing, but we’re talking 5 different nodes in 4 years. This idea was born on the premise of regaining an undisputed leadership position, which relates to AMD Ryzen’s slump in the sector. Another reason to turn to TSMC is that they plan to go for 3D designs, i.e. stacking (like 3D V-Cache). This is because chip area can be increased to extend Moore’s Law: 1 billion transistors on a chip by 2030, that’s the goal! Analysts expect TSMC to start mass-producing 3nm chips by the end of 2022. Samsung, on the other hand, is venturing into GAA production earlier this year, with the upcoming Exynos SoC being the first chip to feature the technology.

The production sprint should happen in early 2023, the source said. However, we doubt TSMC’s production capacity given the number of orders the company has to take. We recommend reading the TSMC process. What do you think of this strategy?