Intel's new CEO Pat Gelsinger made several big announcements about its 7nm tech at today's Intel Unleashed: Engineering the Future event, revealing that the company has solved the primary issues that have resulted in an untenable delay to its 7nm products. As a result, Intel will tape in its first 7nm compute chip for desktop PCs, Meteor Lake, in the second quarter of this year, with the first chips shipping to customers in 2023. Intel also divulged that it will also ship its 7nm Granite Rapids data center CPUs in 2023.
Intel says that 7nm's issues stemmed from difficulties with series of steps in its manufacturing process and that it has leaned more heavily on EUV manufacturing to rearchitect that series of steps and simplify the design flow.
However, even though Intel has addressed the issues with its 7nm process and says that the majority of its products in 2023 will be produced in house, the delayed 7nm production schedule will leave the company in direct competition with chips built on more advanced nodes from competing foundries. As a result, Intel will also outsource the production of the CPU cores to TSMC for some of its key CPU models that will land in 2023. You can read more about that here.
Intel hasn't shared the details about its CPUs that will feature outsourced cores, but the company did share that its Meteor Lake processors are built on the 7nm process and feature the company's Foveros design. This technology allows for die-on-die logic stacking to fabricate 3D processors, as we see with Intel's Lakefield chips.
Meteor Lake chips are thought to come with a combination of Intel's Ocean Cove and Gracemont cores, meaning they'll follow the same hybrid arrangement found in Alder Lake, but in a 3D-stacked fashion. Early Linux hardware enablement code has already shown up for Meteor Lake, so it's clear that Intel is deep in the design process.
The 3D-stacked design could allow Intel to either use its own 7nm cores or swap in cores based on a process node sourced from a third-party foundry, like TSMC or Samsung, but Intel hasn't shared any details about its outsourcing strategy yet.
Intel's first 7nm server CPUs (Granite Rapids) will arrive in 2023, which is later than listed in earlier roadmaps that projected a launch in 2022, but on track with Intel's initial revised timeline.
In either case, that timeline is concerning in the face of AMD's continued execution with its EPYC data center chips – AMD's roadmaps outline its 5nm Genoa processors coming to market before the end of 2022. That's the obvious rationale behind Intel also using an outsourced TSMC node for some of its data center products that will launch in 2023. Again, it's hard to tell if those outsourced chips will come as chiplets/tiles that merely snap into the same Granite Rapids package, or if they will come as entirely different new models.
Intel hopes to achieve ~237 MTr/mm^2 with their 7nm process (https://www.anandtech.com/show/13405/intel-10nm-cannon-lake-and-core-i3-8121u-deep-dive-review/3), which if they do, would blow away TSMC's current 5nm process at ~173MTr/mm^2. Of course two years will give TSMC to catch up, but still.
So I wonder if Intel was simply biting off more than they could chew.
I'm a little confused...the slides from the AMD roadmap article link show Zen 4 / 5nm cpus coming by the end of 2021, not 2022.
there multiple roadmap. some zen3, some zen4
Timelines are from left to right. A "2022" at the end doesn't mean the end of 2022, it means the end of 2021/beginning of 2022. Just look at the second slide down titled "CPU Roadmap." The last date is 2021, but Zen 3 came out it 2020.
and the road map is from AMD. might be an old one?
You forgot to calculate the multiple years from Intel's first announcement date of 10nm to actual production. You gotta add at minimum a few more years to make it better! :p
There is NOTHING stopping Intel from using TSMC too!! This means any advantage AMD has will be gone once Intel uses TSMC....
AMD, Apple and most other TSMC clients secured their wafer starts years ago and likely have priority dibs on most capacity that gets released by other clients. Intel as one of the newer clients should be near the end of queue for everything.