Intel Investing $20B in Building Foundry Business

Article By : Brian Santo

Intel is spending $20 billion to build two foundries in Arizona, with plans to serve customers in North America and Europe.

Intel is going to dive headlong into the foundry business, starting with a $20 billion investment in not one but two fabs in Arizona. New CEO Pat Gelsinger said the company is establishing the contract manufacturing endeavor as a standalone business called Intel Foundry Services (IFS).

IFS will be run by Randhir Thakur, a longtime executive at Applied Materials who joined Intel in 2017 to manage the company’s global supply chain; he most recently held the title of chief supply chain officer. Thakur will report directly to Gelsinger.

“We are ready to engage with customers today,” Gelsinger said.

That Intel is getting deeper into foundry services is exactly the opposite of what some Intel critics wanted to hear for a lot of reasons. Once considered the most advanced chip maker in the world, Intel’s manufacturing operations have stumbled; the company has been eclipsed by Taiwan Semiconductor Manufacturing Company (TSMC) and Samsung, and it will be exceedingly difficult to catch up. Partly to compensate for those stumbles, Intel itself is committed to using other foundries (those plans have not changed at all). Building and operating fabs is fabulously expensive and risky. Even though Intel’s doing some limited foundry work now, being a foundry is both a business model and a technological approach, neither of which Intel has ever excelled at. At a macro level, manufacturing moves to countries with trailing economies for reasons.

Some were hoping Intel would bundle all of that and use it as an opportunity to start getting out of manufacturing.

When Intel’s board hired Gelsinger, that was a signal that restoring its manufacturing was every bit as valid an option as getting out. That Intel might vastly expand its manufacturing operations and commit to a globally competitive commercial foundry business was an option that coalesced with only with extraordinary circumstances.

First there was the trade war the Trump Administration started with China, which severed an incalculable number of links in the global supply chain. Then the COVID pandemic hit, disrupting global commerce even more. Global warming finally got chaotic enough to create an exceedingly rare freeze in Texas that knocked out Renesas and Infineon plants for weeks, while a highly unusual drought in Taiwan is stressing TSMC and UMC. Throw in an entirely random fire that recently gutted the cleanroom of another Renesas facility, this one in Japan, and let’s not forget the occasional earthquakes and tidal waves that always seem to take out a fab here and there.

All of that has served to overwhelm what had been the prevailing economic view: the free-market insistence that The Lowest Cost Must Win. Now there’s growing sympathy for the notions that supply chain integrity, national economic health, and national security should all be considered along with cost.

Various defense agencies around the world have always been keenly aware that relying on a potential adversary for critical materials and goods is problematic. After the supply chain disruptions ensuing from the trade war and the pandemic, just about everybody has come to the conclusion that manufacturing capacity has to be more widely and carefully distributed. And with recent parts shortages, many companies around the world — including those making vehicles, smartphones and game consoles — are simply desperate for access to IC manufacturing capacity.

The answer to most of these problems is to create more leading-edge manufacturing capacity. The answer to some of those problems is to put that new trusted and secure capacity somewhere in a western country that already has the wherewithal to host more.

In short, the western world wants more chip manufacturing capacity, and Gelsinger’s Intel is jumping on the opportunity. Intel might not be the perfect company to provide that solution, but it has advantages few if any other companies can match. Despite recent troubles it is still one of the most advanced IC manufacturers on the planet, and it can afford to build fabs on its own.

Gelsinger said foundry services will be a $100 billion market opportunity by 2025, and that Intel’s new IFS fabs are going to address market requirements in the US and Europe, as well as defense needs in the US and the European Union.

Again, Intel already does some foundry work, and he said the company is ready to engage “today.” He said the IP available to foundry customers will include X86 cores, graphics, media, display, interconnect, fabric, along with Arm and RISC-V IPs.

Again, the first new fabs will be in Arizona (on the Ocatillo campus in Chandler), but Gelsinger several times referred to capacity in Europe. He was no more specific than that, but the implication was that Intel might build subsequent facilities there. The company currently has manufacturing facilities in Ireland and in Israel.

He said the $20 billion commitment is all Intel money, and Intel is ready to proceed completely on its own. The strategy does not depend on a penny of state or federal support, Gelsinger said. He added, “Of course, we want incentives…, but we announced this without getting incentives first. We’re going first; we’re putting our chips on the table, and we’re confident we’ll see positive response from around the world.”

Who might contract with IFS?

Intel lined up an all-star cast of companies that endorsed at least the concept of Intel radically expanding its foundry activities. Amazon, Cisco, Ericsson, Google, IBM, IMEC, Microsoft, and Qualcomm all lent their names.

Beyond them, Gelsinger has no shortage of ambition for IFS. He stated flatly the company intends to try to win Apple’s business.

Microsoft CEO Satya Nadella appeared in person for the announcement and spoke with great enthusiasm about what Microsoft and Intel once accomplished together and about what they might yet accomplish together in the future — but he offered no specifics.

Still, Nadella’s appearance is significant. Intel designs chips for Microsoft’s Surface laptops, but Intel’s production problems tested the patience of many of its partners, reportedly including Microsoft. Nadella’s appearance alone is a positive sign the relationship will likely continue.

The question of who might contract with IFS is compounded by the question of who would actually rely on a competitor to produce their leading chips? Intel obviously tried to address some of that concern with the endorsements mentioned above, and also by setting up IFS as a separate, standalone vertical operation with its own profit & loss (P&L) responsibility, reporting directly to Gelsinger.

Gelsinger highlighted “Ponte Vecchio,” Intel’s first exascale graphics processing unit. It combines 40 tiles in an EMIB package. He said Argonne National Labs will use it with its Aurora supercomputer.  (Credit: Walden Kirsch/Intel Corporation)

How can a company that so famously has had manufacturing problems start offering to build chips for the most demanding customers in the world?

Intel’s manufacturing troubles were the first thing Gelsinger addressed in his announcement Tuesday (well, the first thing after listing his new guiding principles for Intel: market leadership, execution, innovation, hire the best talent). He said Intel erred by being too cautious about EUV lithography equipment when began designing circuitry at the 10nm node. To compensate, the company’s designs got excessively complicated, and that led to production problems. The reluctance to trust EUV persisted through the 7nm node, and so did the production problems.

The company is confidently using EUV now, Gelsinger said, and it believes it is leaving its production problems behind it.

Whither 5nm? (Not to mention 3nm?)

But that doesn’t address the fact that Intel is now behind TSMC and Samsung. Both are producing chips at 5nm, and both are developing 3nm process technologies.

This is where the company’s new research collaboration with IBM presumably comes into play. Intel invited IBM CEO Arvind Krishna to talk about how the two companies would collaborate on semiconductor and packaging technology.

And that’s pretty much all the detail either Gelsinger or Krishna provided.

EE Times asked Gelsinger specifically about competing at 5nm and 3nm. Gelsinger responded that the company has a roadmap, it plans on accelerating its cadence of process innovations to yearly, and he referred again to the collaboration with IBM. He then turned to Ann Kelleher, who last summer was given responsibility for technology development of Intel’s 7nm and 5nm nodes. Kelleher spoke confidently about the 7nm node and about EUV, but neither Intel exec’s response included the words “five nanometer” or “three nanometer.”

Throughout Tuesday’s announcement, Gelsinger several times stressed the company’s packaging expertise, specifically its Foveros and EMIB technologies. Foveros enables multiple chips to be stacked in a single package. EMIB, meanwhile, is Intel’s distinct variation of chiplets technology (the company refers to “tiles” rather than “chiplets”). The idea there is to break up complex ICs into subsystems — chiplets (or tiles) — so that they can be mixed-and-matched and connected inside packages designed to host them. Gelsinger several times noted that Intel is shifting from the system on chip (SoC) approach to a system in package (SiP) model.

While Gelsinger didn’t provide much illumination about Foveros or EMIB, these and similar packaging technologies will be increasingly significant when moving into the next era of semiconductor technology, which some refer to as the more than Moore (MTM) era. As silicon is nearing its limits, there will soon be little if any practical gain in pushing to subsequent process nodes (assuming that will still be possible). At that point, performance improvements are going to have be achieved in other ways, and one of those ways will be clever packaging of chiplets.

Randhir Thakur

What Gelsinger did say, repeatedly, was that Intel’s expertise with advanced packaging technology will be an advantage for its foundry business.

As for Intel’s own business, the company’s first demonstration that it is getting past its manufacturing flubs will be the 7nm Meteor Lake processor, which Gelsinger said is being built with EUV technology, and is set to tape out by July, with a commercial introduction expected in 2023. Meteor Lake, he said, will also rely on Intel’s Foveros technology. (Elsewhere it’s been reported that at least one of the chips in the stack will be built by TSMC in a 5nm process.)

He said the company’s first 7nm Granite Rapids server CPUs are also due in 2023. The company will work with TSMC to produce those.

Intel is almost ready with its third-generation Ice Lake version of its Xeon processors; those are due in early April.

Gelsinger said customers have samples of the Xeon Sapphire Rapids processor. The production ramp for that product will begin in the first half of next year.

Intel is the biggest integrated device manufacturer (IDM) — the biggest company that builds chips primarily of its own design. Intel intends to remain competitive at the leading edge of process technology for its own designs, will also rely on other leading foundries when necessary, and will expand its business with IFS. Gelsinger called the three-pronged approach “IDM 2.0.”

Subscribe to Newsletter

Leave a comment