Pat Gelsinger’s Intel will evolve from lone wolf to touting modular systems-on-packages with third-party foundry collaboration
Teams with IBM, creates foundry biz to build x86, Arm, RISC-V cores, schedules in-person conf for October
Intel’s new CEO Pat Gelsinger on Tuesday outlined his strategy: getting manufacturing processes right so that Chipzilla can create modular products, and either build them itself or work with third-party foundries. And do that all even as Intel makes a serious tilt at the foundry business itself.
Intel calls this plan “Integrated Device Manufacturing 2.0” (IDM 2.0). The gist of it is that over the years Intel built all the capabilities it needs to design and make semiconductors by itself. Intel now wants its processes to be more flexible so that work can happen in more locations. If that means Intel collaborating with third-party foundries to build some products, or some elements of a product, Chipzilla is fine with that.
Gelsinger said this plan is both a big change and business as usual. “Complementary and strategic use of third-party foundries is an underappreciated fact” at Intel, he said.
Another announcement was a new standalone Intel foundry business. Gelsinger said the new outfit, which will take years to build and have responsibility for its own P&L, will aim to win business from all comers. And it'll be more comprehensive, and less half-arsed and narrow, than Intel's previous attempts at a foundry, we're told.
He mentioned Apple and Qualcomm as the kind of silicon customer he craves, while hyperscale operators will be lured with the promise of the chance to create heavily customised kit using all of Intel’s tech or designs preferred by customers.
Intel will happily make silicon based on Arm or RISC-V designs as well as x86. SiFive, for one, today said its configurable RISC-V core portfolio will be available via Intel's foundry services.
- Pat Gelsinger promises Intel can go back to the future – in memo to staff shared with world+dog
- Intel is over GPUs and CPUs – it's all about 'XPUs' now that OneAPI code-abstraction tool is golden
- Chip fab Intel said to be using better chip fab TSMC to make 5nm Core i3 processors, 20% of its non-CPU parts
- Intel couldn't shrink to 7nm on time – but it was able to reduce one thing: Its chief engineer's employment
- Intel's 7nm is busted, chips delayed, may have to use rival foundries to get GPUs out for US govt exascale super
Gelsinger also wants Intel’s own products to become modular and take on more roles. Systems-on-a-chip, he said, will give way to systems-on-packages. And those new devices will be customizable because they’ll comprise integrated tiles of silicon, each with their own function.
Intel already has packaging tech to make this possible, and one product in particular using this approach: the Ponte Vecchio GPU. And if some of those tiles are made elsewhere – such as by TSMC, cough, cough – either due to manufacturing capacity constraints or to take advantage of specialist capabilities, Intel won’t mind a bit.
To make this all happen, Intel has teamed with IBM to create joint R&D teams “focused on creating next–generation logic and packaging technologies.” Gelsinger also named Samsung, TSMC, UMC, and Global Foundries as third-party foundries that Intel will use “strategically” when required for its own products, and as participants in the wider IDM 2.0 effort.
The foundry business is also about securing better ROI from Intel’s fabs: Gelsinger said that as factories age, their tech remains relevant for products beyond Chipzilla’s core offerings. The new CEO said he expects operations as foundries-for-hire will prove lucrative for years.
The new CEO also ring-fenced $20bn to build a pair of new fabs in Arizona, USA, which he said will help the foundry business and generally ensure Intel has the capacity it needs. Future investments in the European Union and America were mentioned, including an Intel bid for a fab dedicated to military applications.
All the new builds were framed as a response to US and EU desires for supply chains that start and end on their own soil, for security reasons. Gelsinger also said that the world currently suffers from an imbalance of supply, with too much of it in Asia.
However, when discussing third-generation scalable Xeon CPUs due to be launched in the coming weeks, Gelsinger mentioned Chinese giants Alibaba and Baidu as happy testers of the new silicon.
Complementary and strategic use of third-party foundries is an underappreciated fact
Gelsinger also addressed Intel’s slow rollout of 7nm manufacturing process, attributing it to a desire to avoid reliance on extreme ultraviolet lithography (EUV) tech that was not proven when the silicon giant started work on designs. Intel’s plans resulted in more complex manufacturing processes that took longer to develop than hoped. Meanwhile, EUV matured, and rivals put it to work leaving Chipzilla eating their dust.
Intel now has now nailed EUV and 7nm products, some of them using the new system-on-package approach, will flow from 2023. In the here and now, Gelsinger said “the majority of client CPUs” will be built on a 10nm process by the end of 2021.
Gelsinger’s first day in what he called his “dream job” as Intel CEO was on February 15. It seems highly unlikely that much of what he described today was invented, or proposed, after his arrival. But he took responsibility for the strategy and tactics articulated in his talk.
Gelsinger also announced one more contentious item in his talk: a new in-person event, called Intel ON, in San Francisco during October 2021. VMware, Gelsinger’s last employer, has already said its August 2021 VMworld event will be virtual.
What a difference a few weeks makes. You can catch the chief exec's announcements in the video below. ®