Intel Charts Manufacturing Course to 2025

Article By : Brian Santo

Intel is renaming its process nodes, and provided details about its next five. It also said AWS and Qualcomm are customers.

Intel CEO Pat Gelsinger has had enough. 10 nanometers will continue to be 10 nanometers — Intel is stuck with that. But as of today, the next version of 10 nm, aka SuperFin, will instead be Intel 7, and 7 nm will henceforth be Intel 4 (no “nanometers”). There will then be an Intel 3, after which Intel will then cease even thinking in nanometers; after 3, Intel will be alluding to angstroms, starting with nodes it will call 20A and 18A.

A graphic representation of the ribbonFET, Intel’s version of the gate all-around (GAA) transistor.

That’ll be five nodes taking Intel into 2025. The roadmap the company just announced includes a new transistor architecture — a gate all-around variant Intel calls the ribbonFET; a new interconnect technology called PowerVia that uses the back of the wafer; and the announcement that Intel is contributing to an evolution of EUV technology with lithography specialist ASML.

The renaming of nodes seems like it might be specious flapdoodlery from a company trying to get everyone to forget that it stumbled and lost the technology lead it once had, but the node designations used by its competitors to make Intel sound like it’s lagging are already almost meaningless marketing blah-blah.

So why shouldn’t Intel reset its nomenclature? Intel has always argued that the performance it can deliver at any given node is better than the performance enabled by the like-numbered nodes of its competitors, anyway. The company has made a good case, for example, that it’s 10nm process is equivalent to TSMC’s 7nm process.

The key issues for IC designers are performance, performance improvement from node to node, and the pace of the progression from node to node. Intel famously missed its own deadline for delivering the 10nm process node (now in full production swing).

Losing the technology lead (or being perceived as having lost the lead) was bad enough, but stumbling on the node progression was just as bad, if not worse. An IC manufacturer’s customers have their own roadmaps, and if their key chip supplier can’t get them where they need to go, then it’s time to consider finding a new key supplier.

Which explains why Gelsinger since taking the reins of Intel has repeatedly vowed that Intel would start progressing from one node to the next on a regular cadence, each coming with substantial performance gains. Delivering is essential. If Intel were to experience another failure to progress from one node to the next in a timely manner, that would not just jeopardize existing relationships, it would be fatal to Intel’s brand new foundry business, Intel Foundry Services (IFS).

Today Intel announced a fairly detailed roadmap for node progression:

  • Intel 7 will be introduced this year and will be in production in 2022
  • Intel 4 will be introduced late in 2022 and will be in production in 2023
  • Intel 3 will be introduced later in 2023, implying production in 2024
  • Intel 20A will be introduced early in 2024
  • Intel 18A is scheduled to be introduced in early 2025

Intel also mapped processors on its product roadmap to some of its upcoming process nodes.

For example, Intel 7 will be featured in products such as Alder Lake for client in 2021 and Sapphire Rapids for the data center, which Intel said is expected to be in production in the first quarter of 2022. Intel 4 will be used to create Meteor Lake ICs for client and Granite Rapids for the data center.

Nodes and manufacturing processes

Intel said it will “fully embrace” EUV lithography with Intel 4. The company will continue refining finFET technology through Intel 3; after that, starting with 20A, it will shift to a gate all-around (GAA) structure. Everyone working on advanced process nodes is anticipating moving from finFET to GAA. Intel has its own approach to GAA, however, that it’s calling the ribbonFET.

An image of the ribbonFET from Intel’s presentation.

Intel also talked about a new interconnect technology called PowerVia. Traditionally, interconnect is positioned on top of transistors. Historically that’s worked fine, but at smaller dimensions with new architectures, Intel said inefficiencies arise. PowerVia describes a process of putting interconnect on the underside of the chip. The company plans to experiment with it for Intel 3, so that it will be ready to commercialize fully in Intel 20A.

Intel said it is partnering with Qualcomm to develop a major smartphone platform at the 20A node. If the relationship holds, Qualcomm’s endorsement would be significant. Intel has been trying to achieve a foothold in the smartphone market for some time, with minimal success.

To hammer home its commitment to manufacturing, Intel said it is working with ASML to help define, build, and deploy a refinement on EUV technology, called high numerical aperture EUV. Intel said it will have the first production high-NA EUV system in the industry. Intel said high-NA EUV would contribute to ribbonFET improvements in the 18A process node.

For years Intel has been boasting that it leads in new packaging technologies, which will be critical for improving semiconductor performance moving forward. The company said IFS has signed up AWS as its first customer for packaging solutions.

Image of the PowerVia from Intel’s presentation.

Node designations

Historically, there used to be some relationship between the minimum size of a transistor gate and node designations. It’s arguable when that relationship fully broke down. Some say it was at the 32 nm node, which arrived around 2010. Intel today said it was back in 1997, which would have been right about the time the industry was shrinking down to 0.18 micron. Whenever the disconnect started, there’s no relationship now.

That’s why, according to Tirias Research analyst Keven Krewell, “The number reset was overdue. Intel had been doing 14nm+++ stuff that was meaningless to anyone outside of Intel. TSMC meanwhile would create intermediate nodes like 8nm. That said, it’s really hard to make a detailed transistor to transistor comparison,” he wrote.

Asked about Intel’s plan for the next five nodes, Krewell wrote, “The roadmap is extremely aggressive, but Intel feels it’s ready for this leap forward. Each new node is tied to a specific product, so it will be transparent how Intel is performing. The Qualcomm win for 20A is years in the future, but impressive that Qualcomm is willing to publicly commit to it.

“Packaging is important part of the story, with Intel getting AWS as a customer,” Krewell added.

Intel made its announcements in a conference call Monday afternoon. During the question-and-answer session following the company’s presentations, Gelsinger was asked about other customers beyond AWS and Qualcomm. He stopped to note that the work with Qualcomm demonstrates that IFS customers will get access to the most advanced manufacturing technology concurrent with Intel getting it to use for internal purposes.

Gelsinger said Intel was talking with other customers beyond Qualcomm, but begged off on naming any. He did say that some are in the industrial segment, some in automotive, and some are semiconductor companies — including some traditional rivals — who need foundry. “IFS is off to the races,” he enthused.

This article was originally published on EE Times.

Brian Santo is Editor-in-Chief of EE Times. He has been writing about technology for over 30 years, for a number of publications including Electronic News, IEEE Spectrum, and CED; this is his second stint with EE Times (the first was 1989-1997). A former holder of a Radio Telephone Third Class Operator license, he once worked as an engineer at WWWG-AM. He is based in Portland, OR.

 

Subscribe to Newsletter

Leave a comment