Intel rewarded the patient this week as Pat Gelsinger made some very big announcements about the formation of a standalone foundry business, increased partnerships with TSMC for manufacturing of Intel processor products, and a huge promised investment in new manufacturing plants in Arizona. EETimes editor Brian Santo provided an excellent synposis and analysis of Gelsinger’s web presentation.
Up front, let us not ignore the elephant. Rumors have swirled for a very long time that Intel needed to get out of manufacturing because it was “no longer competitive” with TSMC and Samsung. Gelsinger definitely put an exclamation point on his response to the doubters with a commitment of $20 billion to build two new fabs in Arizona.
Intel’s contract manufacturing (Intel Foundry Services) will be a standalone business with its president, Randhir Thakur, reporting directly to the Intel CEO.
No fabs, no way
Looking back, comments made by Bolaji Ojo in EETimes were prescient: “An Intel with the right vision should be aiming for TSMC’s wafer foundry crown. An Intel with the right backbone should not be whining about having fallen one or two process technology behind TSMC. It should get back on the horse, mobilize and rearm its engineering workforce and assure investors it has what it takes to restore its technology leadership.”
It seems that a freshly minted Intel CEO has embraced this vision. More to the point, this challenge is very likely the reason Intel lured him back and Gelsinger jumped at the opportunity.
Ojo’s article was from July 2020, but he was forceful on the need for a radical change (that eschewed an approach promoted by activist investors that entailed dumping manufacturing) as far back as 2007.
Intel made a big bet. The pot will likely be sweetened with government cash. Hot on the heels of the CEO’s announcements, Jeff Rittener, chief government affairs officer for Intel, wrote a blog pointing out the need for more government investment and incentives.
Momentum is increasing. How much will it accelerate? Much of the conversation about US semiconductor production has naturally focused on Intel as the largest domestic semiconductor manufacturer. More than a few voices suggested Intel had fallen behind the semiconductor process technology curve and should just give it up (just one example, “Investors roast Intel over manufacturing failures”).
But in terms of turning manufacturing around in the US, it was really the courting of TSMC to build a plant in Arizona that may have started this ball rolling. That was arguably the tipping point when everyone realized the US government’s awakening to the strategic importance of domestic semiconductor manufacturing.
Considering the risks with manufacturing in Taiwan, TSMC may have been looking to get a toehold in a more secure location. TSMC was a little coy in its approach. It worked out well for the company in terms of incentives, but the broader community will benefit from another large semiconductor plant in Arizona.
There is a connection between these first movers for expanding semiconductor production in the United States. Pat Gelsinger indicated that Intel would “leverage our relationship with TSMC to deliver leadership CPU products for our client and data center customers.” This targets Intel’s 2023 product roadmap, but the relationship with TSMC appears set to flourish.
Intel is too low on the customer priority list to drive capital investment decisions at TSMC, but they might have some future influence. Apple definitely is in that position. With an existing Samsung plant and announced expansion in Texas, how much in the way of natural or man-made disruption to factories in Taiwan would have Apple pushing TSMC to ramp up production in the US? And TSMC rushing to oblige?
But if Gelsinger gets his way, Intel will be fabbing Apple chips in Arizona. Santo reported that “He stated flatly the company intends to try to win Apple’s business.”
We can clearly see momentum building for the continued flow of capital from Intel, Samsung, and TSMC into US-based IC manufacturing. But let’s refocus on Intel.
EETimes reporting of the Pat Gelsinger presentation noted that Intel’s packaging expertise was stressed and that the company would be shifting from a system-on-chip (SoC) approach to a system-in-package (SiP) model. According to Santo’s report, Pat Gelsinger repeated his point that advanced packaging technology from Intel would be an advantage for its foundry business.
We have certainly seen strong packaging moves by the other big foundries, especially TSMC, so it makes sense that Intel will need to develop its foundry business into a “one stop shop” in the way that TSMC has and Samsung continues to develop.
Intel has long touted its Foveros brand of 2.5D interposer technology and embedded multi-die interconnect bridge (EMIB). These technologies will both be important to the foundry business.
Beyond the big investment and announcement that Intel would contract with TSMC for 3nm production on its microprocessor products in 2023, we can look to this as more early stage, yet significant, momentum in the development of an ecosystem for chiplets.
This is most obvious considering that Intel is embracing the chiplet approach for its Meteor Lake design. There will be chiplets using Intel 7nm as well as TSMC 5nm. The various circuit tiles will be supported by the Foveros SiP platform. That sounds like mix-and-match to me.
There’s no better proof of the cook’s pudding than seeing him eat it.
Beyond manufacturing the chiplets and collecting them in a package for foundry customers, Intel is active in developing the interface protocols that allow chiplets to communicate. Intel has developed (and released royalty-free) its plainly labeled, Advanced Interface Bus (AIB).
If we think about some of the companies that stood with Intel for Gelsinger’s presentation, we can think of Intel’s strong foundry play as another sign of changing times. Amazon, Cisco, Ericsson, Google, and Microsoft (among others) could be viewed as system customers who once upon a time bought microprocessor chips to populate their own motherboard designs but now look to create that full system in a single packaged component.
Microsoft was a significant presence at the event with CEO Satya Nadella appearing in person. Microsoft will be changing with the times and would willingly use an Intel foundry as a new-age systems integrator, but future products will be leaving the conventional motherboard populated by a microprocessor, chipset, and memory for computer history society meetings.
Intel sees the path forward. From the outside looking in, it appears the giant woke up and rubbed its bleary eyes. Several pundits predicted grand new visions and subsequent success based on the return of Pat Gelsinger to the company. It is not clear how much of the new direction is attributable to Gelsinger, but surely a significant portion is if the accolades of former colleagues are any indication.
At the very least, he commanded the requisite respect and provided the necessary conviction to deliver a message that appears to have been well-received, inside Intel and out.