Intel’s new chief executive officer Pat Gelsinger made a series of announcements at the Intel Unleashed event on March 23, outlining the company’s technology roadmap, the launch of a new business unit, and its plans to launch semiconductor manufacturing plants in Arizona.
Gelsinger started the presentation with an update on Intel’s 7nm transistor, announcing that the company has surmounted the technological barriers surrounding extreme ultraviolet lithography (EUV) and has architectured the 7nm process flow, increasing the use of EUV by more than 100 per cent.
“When Intel initially designed 7nm, EUV was still a nascent technology, so we developed our process to limit the use of EUV,” Gelsinger explained. “But this also increased the process complexity as EUV then matured and became more reliable, we experienced the domino effects of our 10nm delay, which pushed out 7nm and ultimately put us on the wrong side of the EUV maturity curve.”
Speaking of 10nm, Intel’s 10nm transistor process has seen numerous challenges in recent years. The technology, which was supposed to have launched around 2016, faced continuous setbacks due to over-ambition and timing, according to Intel’s previous chief executive officer Bob Swan. However, with the challenges finally surmounted, Gelsinger said that Intel’s 7nm process development is now on track.
As it worked on its transistors, Intel had also been developing its Foveros stacked chip package and embedded multi-die interconnect bridge (EMIB). These technologies allow Intel to mix and match different IPs for different markets and tune their performance and power consumption. As Gelsinger described it, Intel’s packaging can connect multiple process technologies and have them act as a single chip. He said that this hints at a larger design transition for Intel as it will move from system-on-chip (SoC) to system-on-package.
Intel showcased Ponte Vecchio during the event, its first high-end server GPU that uses both Foveros and EMIB, combining more than 40 chiplets and subsystems and housing up to 100 billion transistors. Gelsinger said that Ponte Vecchio can achieve petaflop performance and will be used in the Argonne National Laboratories Aurora exascale computer.
Intel IDM 2.0 – Intel devices manufacturing model
Intel is adjusting its manufacturing model to not only include internal and external production networks, but also to focus on bolstering production capabilities in the U.S and in Europe to establish a global supply chain.
Still, Gelsinger noted that Intel will deepen its business with TSMC, Samsung, Global Foundries and UMC.
Gelsinger promised that Intel will continue to build a majority of its product in its own fabs, but also mentioned that it will be using its third-party foundry capacity across its portfolio, including the modular chips used for Intel’s more advanced products, like Ponte Vecchio.
The semiconductor shortage continues to be driven by unanticipated demands during the pandemic. To fill that hole, Intel, like most other semiconductor manufacturers, is upscaling its foundries. Gelsinger announced a new push to increase US and Europe-based capacity to produce chips.
Gelsinger underscored that the semiconductor industry is expanding at a “torrid pace” and will reach a $100 billion addressable market by 2025.
Intel Foundry Services
To meet the elevated global demand for semiconductors, Intel has established the new standalone Intel Foundry Services led by Randy Thacker, who will be reporting directly to Gelsinger. Gelsinger said that a dedicated business division can better understand the customers’ needs and fulfil capacity commitments.
During the presentation, Gelsinger noted that Intel Foundry Services will provide a portfolio of IPs including x86 cores, x86 cores, graphics, media, display, AI, interconnect fabric, along with Arm and RISC-V ecosystem IPs. Moreover, the company will provide silicon design services to its customers.
He also co-announced a new Intel-IBM collaboration with IBM chief executive officer Arvind Krishna. The two companies will work together to “collaborate on critical leading-edge research that will advance silicon and packaging technology development.”
IBM designs its own processors for its Power series products. It isn’t clear what technologies the two companies will exchange.
New foundries planned for Arizona
To back its commitments, Intel also announced two new semiconductor manufacturing facilities (also known as fabrication plants or fabs for short) in Arizona. The two new fabs, costing around $20 billion in total, will begin construction this year. From construction to operation, the new investment will create over 21,000 jobs, including 3000 permanent high-tech jobs, 3000 construction jobs, and 15,000 long-term jobs in Arizona.
The facilities in Arizona marks the beginning of a series of global investments to bolster domestic and foreign expansions. Gelsinger highlighted Intel’s close relationship with the U.S. government and said the company is currently competing for a contract to build a domestic commercial foundry for the U.S. Department of Defense.
This isn’t Intel’s first foray into expanding production, but Gelsinger emphasized that Intel has learned from previous experiences and adapted to the current market.
“I’d also say that our first efforts were somewhat weak, and we learned a lot through them, but we didn’t really throw ourselves behind them,” said Gelsinger. “As you’ve heard today, with our major capacity investments, a separate organization, a separate P&L reporting directly to me, we’re going after this much more aggressively.”
Intel didn’t say what manufacturing the new fabs will focus on, but it did confirm that they will all be EUV capable, meaning they can scale to 7nm and beyond.
Intel said it welcomes partnerships from across the industry, competitors and otherwise.
“In some cases, there may be some cooperatition, but we see this as a great opportunity to create opportunities for differentiated products and technologies as never before,” said Gelsinger. “We’re also going to go to some–like you saw on our announcements today–people like Qualcomm, who might have been more [of a] competitor before. And we’re now going to say, ‘hey, let’s find ways to leverage our technologies in ways that weren’t possible before, can we become your foundry partner?’ We also will pursue customers like Apple and say: ‘Boy, [is it] possible to build and expand on your foundry capabilities as well?”
Intel expects to announce more fabs in the U.S. and Europe within the next year.
Other semiconductor players are also looking to grow their production. Recent news reports hinted that Samsung has filed applications to build a $17 billion semiconductor fab in Texas. TSMC is also planning on building a $12 billion fab in Arizona. The fab is scheduled to open in 2024 and will be able to churn out 2,000 semiconductor wafers a month.
Intel has already begun sampling its next-gen Alder Lake processor to customers. Alder Lake is expected to feature a hybrid design in which two core architectures–one for performance and one for efficiency–reside on the same chip.
Meteor Lake will begin tape-in this year. Source: Intel Unleashed.
Looking into the future, Meteor Lake, Intel’s high volume product for data centers and client CPUs, will become available in 2023. Intel expects Meteor Lake to begin tape-in in Q2 2021. The company is also now testing Sapphire Rapids, the successor to the Ice Lake data center processors that will be announced on March 31.
Gelsinger promised that “a majority” of its client’s wafer starts will be on Intel’s 10nm process by Q3 2021 including SuperFin transistors.