While Intel is struggling to push out Intel Core processors fabricated on 10nm processes into production, the company dropped a bombshell on Wednesday: Executives began talking about manufacturing 7nm chips in 2021.
During the company’s investor meeting on Wednesday, newly appointed chief executive Bob Swan said that the company has begun planning for the 7nm manufacturing generation as part of a strategy to regain manufacturing process leadership. At the end of the day, though, Swan said that “product leadership is going to trump all,” not process leadership.
That’s an unusual statement to make from an executive leading a chipmaker that has depended on manufacturing leadership for literally decades. In fact, Swan seemed prepared to set aside precedent for a more practical approach. Instead of an “Intel Inside” predicated on CPUs, Swan said that Intel’s market opportunity was now far larger, and predicated on an “XPU” strategy that could include CPUs, GPUs, programmable logic like FPGAs, and even 5G chips.
AMD, of course, has already announced 7nm products, such as its next-generation “Navi” GPU, that it will ship this year. (AMD and Intel use different manufacturing processes, so the two numbers don’t directly correlate.)
However, Swan said he’s not committed to furthering investment into chips like memory. Swan told investors that the company is halting plans to add manufacturing capacity for NAND flash chips, and is considering a manufacturing partnership similar to the one it struck with Micron Technology for the production of 3D XPoint chips, which Intel branded as Optane. (That partnership has since dissolved.)
Swan reiterated that Intel has decided to discontinue plans to compete in the smartphone 5G modem market. Intel has yet to decide what to do with its 5G technology, and whether Intel will continue trying to develop 5G modems for devices like PCs.
Still, the 7nm plans are the most eyebrow-raising news. Leaked roadmaps suggest that Intel will lean heavily on 14nm technology throughout 2019. Swan acknowledged that Intel’s fabs were supply-constrained leading into the first half of the year. By the second half of 2019, Swan said, the pressure on Intel’s fabs will lessen, and the company will again be able to satisfy customer demand.
As Intel has said before, it plans to have 10nm parts on shelves by this holiday season, based upon the Ice Lake chip it’s qualifying this quarter. Server parts based on 10nm will debut in the first half of 2020. By 2021, Intel said, it hopes to have 7nm parts in production—without saying for which products, where they’ll be manufactured, and how many of them there will be.
Intel’s use of 7nm will also incorporate Extreme Ultraviolet (EUV) lithography for the first time, a new lithography technology that uses shorter wavelengths of light. EUV technology will be used in subsequent 7nm generations, which Intel is calling 7nm+ and 7nm++, said Venkata (Murthy) M. Renduchintala, the group president of the Technology, Systems Architecture & Client Group. Renduchintala said that Intel would deliver one “Moore’s Law” scaling at the beginning of a node and another further along—which Intel later clarified as saying that Intel would deliver performance and scaling at the beginning of a manufacturing node, plus another performance midway through, plus various optimizations near the end.
Renduchintala also indicated that Intel will adopt new packaging techniques that will allow multiple cores, each with its own optimized process technology, to be implemented within the same package. This heterogenous integration, he said, would allow smaller chiplets to be interconnected, allowing for more complex, optimzied designs.
What this means to you: Can Intel suddenly reverse years of manufacturing problems and return to a tick-tock pace of yesteryear? It’s an aggressive commitment, and we’ll see if Intel can pull it off.
This story was updated at 9:49 PM with additional detail.