Intel’s newly appointed CEO, Pat Gelsinger, has unveiled his plan for the future of the company. The company made a number of announcements late on Tuesday that collectively sketch what the next few years have in store and how Intel will respond to the manufacturing and CPU design leadership challenges it’s facing.
Here are some of Gelsinger’s biggest announcements:
$20B For New Arizona Fabs
Intel wants to break ground on two new semiconductor fabs in Arizona, with both facilities coming online by 2024. Both would be built on unspecified leading-edge nodes (possibly a 7nm+ in Intel parlance). Intel specifically stated these fabs would deploy EUV hardware, which it is expected to deploy at that node. Total employment at the facilities could be roughly 3,000, with ~15,000 in long-term ecosystem support jobs.
Intel’s Ocotillo campus in Chandler, Arizona, is the company’s largest U.S. manufacturing site. Four factories are connected by a mile-long automated superhighway to create a mega-factory network. In March 2021, Intel announced it will invest about $20 billion to build two additional factories at the Ocotillo campus. (Credit: Intel Corporation)
This is Intel’s response to the capacity expansions both TSMC and Samsung announced in late 2020 and early 2021. All three leading-edge foundries have committed themselves to stay on the leading edge, and Intel will not be selling its foundries or spinning off its manufacturing division into a separate company.
Gelsinger affirmed 7nm development remains on track at Intel and the company plans to ship Meteor Lake, a 7nm tile-based CPU (Intel is referring to chiplets as “tiles,” but an Intel tile and an AMD chiplet aren’t exactly the same thing) in 2023. The company’s new manufacturing strategy will be known as “IDM 2.0.” This encompasses the building of new fabs (mentioned above) and the radical changes to Intel’s foundry services and manufacturing partnerships (discussed below).
Tick-Tock, ‘Unquestioned CPU Leadership Performance’
PAO (Process — Architecture — Optimization), the Intel improvement model meant to succeed Tick-Tock, is dead. Tick-Tock is back. Intel first adopted Tick-Tock in the wake of the Prescott disaster in 2004. Under Tick-Tock, Intel first introduces a new manufacturing process (a tick), followed by a new architecture built using that manufacturing process (a tock). In practice, this gave us Westmere (tick) and Sandy Bridge (tock), Ivy Bridge (tick) and Haswell (tock), Broadwell (tick), and finally Skylake (tock).
Post-Skylake, Intel’s roadmap was all tocked up.
Intel is promising it will iterate again in this stepwise fashion, with a return to cadenced improvement. Gelsinger is also promising that Intel will return to “unquestioned CPU leadership performance” by 2024 or 2025. Meteor Lake, which arrives on 7nm CPUs in 2023, is part of that process. Between then and now we’ve got Alder Lake for certain and possibly another family in-between — depends on when in 2023 Meteor Lake is expected, and we’ve seen rumors go both ways on the existence of an in-between part.
Intel Will Use Outside Foundries for Future Manufacturing
Which cores? Which products? We still don’t know. Intel has reiterated it will use outside foundries for products as needed. This time around, Gelsinger’s comments could be read to imply Intel CPUs could be built at outside foundries in client and data center markets, though no product announcements or new manufacturing agreements have been made.
Intel hasn’t technically committed to building high-performance CPUs at TSMC or Samsung, but the company is edging closer to that mark with references to “high performance compute products” as opposed to “GPUs and FPGAs.” It isn’t a big deal for Intel to build GPUs and FPGAs with other companies, but CPU manufacturing has always been the company’s bread and butter. Even hinting it could move these products elsewhere is a big deal. According to Intel:
“Gelsinger said he expects Intel’s engagement with third-party foundries to grow and to include manufacturing for a range of modular tiles on advanced process technologies, including products at the core of Intel’s computing offerings for both client and data center segments beginning in 2023.”
Intel Will Build ARM, x86, and RISC-V CPUs for Third Parties
In addition to the above, Intel is re-entering the client foundry market. Even more surprisingly, Intel claims to be willing to build x86 CPUs for third parties. Intel Foundry Services will exist as a separate business unit within Intel.
This time around, Intel is working with industry partners to create a standardized toolchain and to better support customers. The company wants to offer clients the option to integrate a variety of solutions in AI, FPGAs, GPUs, and CPUs. Critically, customers will be able to create x86 CPU designs.
We’ve seen Intel talk a big game about launching a foundry effort before, only to fail to follow through, but the company never put x86 on the table or implied it would be willing to license x86 CPUs. Intel doesn’t plan to offer a full architectural license like ARM — think more along the lines of a specific core, like being able to license a Core i7-8700K.
Conclusion
Pat Gelsinger has just committed Intel to the largest overhaul of its manufacturing and CPU partnerships in decades. Intel’s roadmap calls for the company to build out a new foundry business with cooperation from Cadence and Synopsis (and a new partnership with SiFive) while simultaneously building two new fabs, working to launch products built at third-party foundries, executing the repair of Intel’s 7nm tech, and conducting the long-term research required to regain its leadership position.
There are reasons to be skeptical. Intel’s past efforts to enter the foundry business didn’t go well. TSMC, Samsung, Apple, Qualcomm, AMD, Nvidia — none of them represent an easy target. Intel is betting that its interconnect technologies like EMIB and Foveros, combined with a renewed commitment to manufacturing excellence, can help it regain its footing in an industry it led for decades.
The breadth of these announcements speaks to an energy that has long been missing from the company. Intel spent too much of the 2010s fruitlessly chasing the mobile market and practically marching in place where x86 computing was concerned. The plans the company announced today may or may not work, but they reflect a willingness to reevaluate where the industry is in 2021. It may be down, but Intel is determined not to be out of this fight.
Now Read:
0 Commentaires