Intel splits in two, signs deal with Microsoft

Intel is splitting down the middle, separating its Product and Foundry lines into two separate businesses.

CEO Pat Gelsinger made the announced at Intel's Foundry Direct Connect event in California this week.

"We want to be the foundry for the world. If we're going to be the Western foundry at scale, we can't be discriminating in who's participating in that," he said.

The official launch of Intel's contract chip manufacturing business will see it compete with Asian foundry giants TSMC and Samsung. It will go by the name Intel Foundry rather than the original name it provided, Intel Foundry Services. Gelsinger shared that name when he declared his ambition to build a significant business from manufacturing chips designed by other companies, in addition to the products it develops, in early 2021.

Intel Foundry is a key part of Gelsinger's comeback plan and IDM 2.0 strategy for Intel. An evolution of its integrated device manufacturing model, IDM 2.0 calls for increased investments in manufacturing in addition to revitalizing its contract chip manufacturing efforts and expanding its use of external foundries in instances where it will benefit Intel's products.

Reiterating its goal of displacing Samsung to become the world's second-largest foundry by 2030, the semiconductor giant said it remains on track with Gelsinger's plan to introduce five advanced manufacturing nodes in four years to regain "process performance leadership." Each successive node offers a significant increase in performance-per-watt.

These nodes are offered to Intel's own chip design teams as well as external chip designers. Both groups will be treated as customers and, as a result, will have to compete for business with Intel Foundry according to Craig Orr, vice president and general manager of marketing and product for the chip manufacturing business.

This means an external chip designer could take away manufacturing capacity from Intel's own product groups by placing a large enough order with Intel Foundry.

In a briefing, Orr said this ensures there is "clarity in terms of financials" for Intel Foundry, which will become a reportable segment with the company's upcoming Q1 earnings report, so the business is "able to benchmark and get very competitive against" other foundries.

It also shows potential and existing Intel Foundry customers that it's independent from the rest of the semiconductor giant's businesses, he added.

"It helps our customers feel the trust because we're now independent, we can make decisions about capacity allocations, technology, preferences and so on independently based on what's best for that [business with a profit and loss statement]," Orr said.

However, Orr said Intel expects the development of the contract chip manufacturing business through Intel Foundry to benefit the company's own products. That's because with technology development becoming more expensive with each node transition, the addition of external customers will help "drive down the cost for everyone."

Orr also referenced Intel's strategy to provide chip designers with an expanded chip manufacturing capacity in the West to offset the heavy concentration of foundry plants in Asia. He also said Intel Foundry aims to become the industry's "most sustainable foundry," with which the company has made great progress by largely relying on renewable energy for its fabs.

"There's also a great benefit to the industry in the sense that you get this sustainability benefit and the resilience of your supply chain," he added.

Reaffirms plans for Intel 3, Intel 20A and Intel 18A

Having already introduced products on the first two nodes in Gelsinger's plan - Intel 7 and Intel 4 - the company said the next process, Intel 3, is ready for high-volume manufacturing. The first product to use Intel 3 is a new generation of Xeon server CPUs with a focus on high core density code-named Sierra Forest, set to launch in the first half of this year.

The final two nodes, Intel 20A and Intel 18A, are on track to become ready for manufacturing this year. Intel expects the latter to give it an edge over the most advanced processes from TSMC and Samsung by 2025.

Orr said Intel Foundry customers will be able to start designing products today with Intel 18A, with full product design expected to be enabled in Q2. One of the first products to use Intel 18A, a follow-up to Sierra Forest called Clearwater Forest, is set to launch in 2025 and has already taped out, a crucial step before manufacturing can begin, he added.

"The reason this is so significant for foundry customers is because it demonstrates the foundation of building blocks upon which we believe other future systems for AI will be made," Orr said.

Microsoft joins as Intel 18A customer

Microsoft has already been announced as a customer for a chip design it plans to manufacture on the Intel 18A process.

"We are in the midst of a very exciting platform shift that will fundamentally transform productivity for every individual organisation and the entire industry," Microsoft chair and CEO Satya Nadella said in a statement.

"To achieve this vision, we need a reliable supply of the most advanced, high-performance and high-quality semiconductors. That's why we are so excited to work with Intel Foundry, and why we have chosen a chip design that we plan to produce on the Intel 18A process," he added.

Intel reiterated that it has other foundry design wins with customers across Intel 18A, Intel 3 and the mature Intel 16 node.

Expanded process roadmap

For the first time since 2021, Intel provided an update to its process roadmap.

The company will follow Intel 18A with Intel 14A, set to begin commercialisation in late 2026. It is expected to be the first node to use an advanced form of lithography called High NA EUV (Extreme Ultraviolet).

The semiconductor giant also said it's developing modified versions of Intel 3, Intel 18 and Intel 14A that have been optimised to improve performance, introduce new features or implement its Foveros Direct 3D stacking technology for advanced chip designs.

For Intel 3, the company plans to extend the node this year with Intel 3-T. The "T" suffix indicates that it will use Foveros Direct to connect chips with a technology known as through-silicon vias.

Orr said Foveros Direct will "become the foundation for future AI technologies."

Beyond 2024, Intel will follow with Intel 3-E, with new features, and Intel 3-PT, a version using Foveros Direct and that boasts performance improvement, indicated by the "P" suffix. The company also plans to introduce Intel 18A-P and Intel 14A-E.

"What we're doing is we're going to have a road map for our customers to address new markets and new market segments and improve their offerings while we get to keep our fabs full from this new road map," Orr said.

Intel says ‘Systems Foundry' approach sets it apart

Intel said the systems foundry approach of its contract chip manufacturing business sets it apart from rivals by offering "full-stack optimization from the factory network to software."

This is especially important for the fast-growing AI computing space, which Orr said is "not scalable" today with the way current systems, technologies and supply chains are set up.

"We're not going to be able to bring AI everywhere the way the industry operates today," he said.

This is a problem, according to Orr, because "with the number of computations needed to train an AI model" doubling every 10 months, "the number of chips being shipped for AI are growing at a commensurate rate, although a little bit less because of supply challenges, year over year."

He added, "If you were to take those AI chips that are going to be shipped next year and turn them on for 60% of the time running at the rated power, it'll consume more power and more electricity than 61% of countries in the world."

Intel is working on new technologies to enable the continued growth rate for AI computing with the "scalability and efficiency that's needed" to enable it everywhere.

The foundry business is developing several layers of technologies in concert with each other, from the process technology, transistors and interconnects; to the design platform, core and accelerator technologies and advanced packaging technologies; to the system architecture, software and applications and workloads, said Orr.

"We think we're really well suited to create this next generation of systems that will help our customers and our industry commercialise AI and do so in a way that's sustainable," he said.

This article was first published on CRN