On the IEDM convention, Intel shared its course of expertise roadmap and its imaginative and prescient for chip designs that might be out there within the subsequent three to 4 years. As anticipated, Intel’s next-generation fabrication processes — Intel 4 and Intel 3 — are on monitor for use for high-volume manufacturing (HVM) in 2023 and 2024, respectively. Moreover, the corporate’s 20A and 18A manufacturing nodes might be prepared for HVM in 2024, which signifies that 18A might be made out there forward of schedule, a slide printed by IEEE Spectrum (opens in new tab) suggests.
Intel’s Applied sciences Between Now and 2025
|Node:||Intel 7||Intel 4||Intel 3||Intel 20A||Intel 18A|
|Standing:||HVM||Prepared Now||Prepared in H2 2023||Prepared in H1 2024||Prepared in H2 2024|
|Notable Merchandise:||Raptor Lake, Sapphire Rapids||Meteor Lake||Granite Rapids, Sierra Forest||Arrow Lake||Future Lake, Future Rapids, IFS|
NOTE: Course of expertise readiness doesn’t imply HVM begin.
Intel 4 Prepared In the present day, Intel 3 Due in H2 2023
Subsequent 12 months Intel will launch its 14th Technology Core codenamed Meteor Lake CPU, its first mass-market shopper processor that includes a multi-chiplet (or multi-tile) design with every chiplet set to be made utilizing a special course of expertise. Intel’s Meteor Lake merchandise will comprise 4 tiles: the compute tile (CPU cores) made utilizing Intel 4 course of expertise (aka 7nm EUV), the graphics tile produced by TSMC presumably utilizing its N3 or N5 node, the SoC tile, and the I/O tile. As well as, the tiles might be interconnected utilizing Intel’s Foveros 3D expertise.
Meteor Lake’s compute tile is arguably probably the most thrilling a part of the package deal as a result of it is going to be made on Intel 4 (beforehand generally known as 7nm), the corporate’s first manufacturing node that may use excessive ultraviolet (EUV) lithography. This fabrication course of is prepared for mass manufacturing, in keeping with Intel, although it is going to be deployed for the HVM of Meteor Lake’s compute chiplet solely a number of months from now. Protecting in thoughts that Intel powered on this compute tile in October 2021, it’s not stunning that the node is prepared for manufacturing by now. What’s a bit surprising is that Intel doesn’t verify that this course of expertise is used to make Ponte Vecchio’s Xe-HPC compute GPU tiles, as planted two years in the past.
Intel will begin utilizing EUV practically 4 years after TSMC, which started to provide chips on its N7+ node in Q2 2019. Intel wants to make sure that its 4nm-class node performs as much as expectations and delivers good yields, as it is going to be the primary node to reach after the corporate’s relatively unfortunate 10nm household of processes that didn’t carry out as anticipated early in its lifecycle and which prices are increased than the corporate hoped a number of years in the past.
Since Intel has to meet up with its rivals Samsung Foundry and TSMC, its Intel 4 course of expertise will already be joined by its Intel 3 fabrication node (3nm-class) in 2023 ~ 2024. This course of might be manufacturing-ready within the second half of 2023, primarily based on knowledge shared by Intel. Will probably be used to make Intel’s codenamed Granite Rapids and Sierra Forest processors, that are high-profile merchandise for the corporate. Sierra Forest is anticipated to be the corporate’s first knowledge heart CPU to make use of energy-efficient cores and can compete in opposition to varied Arm-based choices with excessive core counts.
Intel already has to work on Xeon ‘Granite Rapids’ samples, so it appears to be like just like the design of the CPU is prepared, and the node itself is on monitor for HVM 2024.
“The primary stepping of Granite Rapids is out of the fab, yielding nicely, with Intel 3 persevering with to progress on schedule,” stated Pat Gelsinger, chief government of Intel, at the latest earnings name. “Emerald Rapids is exhibiting good progress and is on monitor for the whole 12 months 2023, Granite Rapids could be very wholesome working a number of OSs throughout many configurations, and with Sierra Forest, our first E-core product offering world-class efficiency per watt, are each solidly on monitor for 2024.”
Intel’s 18A Moved in to H2 2024
Taking part in meet up with TSMC and Samsung is essential, however to return its course of expertise management, Intel must leapfrog each of its rivals. That is set to occur someday in 2024 when the corporate unveils its 20A (20 angstroms, or 2nm) node that may use its gate-all-around transistors branded RibbonFET in addition to bottom energy supply referred to as PowerVia. Intel expects its 20A node to be manufacturing prepared within the first half of 2024; it is going to be used to make — amongst different issues — chiplets for the corporate’s codenamed Arrow Lake processors for shopper PCs in 2024.
Intel’s 20A would be the trade’s first 2nm-class node, and it’ll additionally extensively use EUV to maximise transistor density, present first rate efficiency enhancements, and decrease energy consumption. In 2024, it’s set to compete in opposition to TSMC’s third-generation 3nm-class (N3S, N3P) course of applied sciences designed for enhanced transistor density and efficiency. It stays to be seen how these three nodes stack in opposition to one another. Nonetheless, Intel is setting the bar very excessive for its 20A course of because it concurrently introduces two main improvements (GAA, BPD).
And but, 20A just isn’t probably the most superior course of expertise that Intel plans to begin utilizing by late 2025. The corporate can also be readying its 18A (18 angstroms, 1.8nm) manufacturing node that guarantees to additional enhance PPA (efficiency, energy, space) benefits for Intel and its Intel Foundry Companies prospects.
For 18A, Intel initially deliberate to make use of EUV instruments with 0.55 numerical aperture (NA) optics, which is about to offer an 8nm decision (down from 13nm within the case of at present used EUV instruments with a 0.33 NA). However ASML’s manufacturing of Excessive-NA EUV gear will solely be prepared in 2025, whereas Intel targets its 18A to be ready for manufacturing within the second half of 2025, forward of its rivals.
Since it’s attainable to get to an 8nm decision for post-3nm-nodes with multi-patterning utilizing current-generation EUV instruments (although it will lengthen manufacturing cycles and will doubtlessly have an effect on yields), Intel is prepared to take some further dangers with 18A and use ASML’s Twinscan NXE:3600D or NXE:3800E to make chips on this node because it believes that it’ll convey it undisputed market management.
Because it seems, the primary 20A and 18A take a look at chips have been taped out already.
“On Intel 20A and Intel 18A, the primary nodes to profit from RibbonFet and PowerVia, our first inside take a look at chips and people of a significant potential foundry buyer have taped out with silicon working within the fab,” stated the pinnacle of Intel. “We proceed to be on monitor to regain transistor efficiency and energy efficiency management by 2025.”
System Know-how Co-Optimization
Each 20A and 18A manufacturing nodes will extensively use EUV instruments (and doubtlessly even Excessive-NA EUV instruments), making chips produced on these applied sciences extraordinarily costly. Even right this moment’s giant monolithic 4nm and 5nm chips are pricey to develop, validate, and produce, which is why multi-tile designs like Intel’s Ponte Vecchio are gaining recognition. At 2nm and 1.8nm, it should make sense to disaggregate high-performance designs additional.
To take action, Intel believes that an all-new new ‘outside-in’ design strategy might be wanted. Intel envisions that a number of years down the highway, chip designers will be capable of disaggregate features of a single chip right into a multi-chiplet design after which produce chiplets utilizing probably the most optimum expertise to fulfill their efficiency, energy, and price targets. Intel calls such strategy system expertise co-optimization (STCO). For instance, since logic scales higher than SRAM, it is smart to provide logic and caches utilizing completely different nodes (for optimum prices and efficiency) after which sew them collectively utilizing applied sciences like Foveros or EMIB.
Given such an strategy, a profitable foundry must supply varied nodes for various chiplets and aggressive packaging applied sciences. Because of this Intel wants to offer the perfect logic expertise (i.e., 20A and 18A) forward of its rivals to make sure that it makes probably the most profitable components of these upcoming multi-tile designs.