Intel places 1nm course of (10A) on the roadmap for 2027 — additionally plans for totally AI-automated factories with ‘Cobots’
We have already lined the first major announcements from the Intel Foundry Direct Connect occasion, however it seems that Intel miscommunicated a few presentation that was regarded as beneath NDA — now Intel says the presentation is open to be reported to the general public, so we now have some extra information to share: Intel’s previously-unannounced Intel 10A (analogous to 1nm) will enter manufacturing/growth in late 2027, marking the arrival of the corporate’s first 1nm node, and its 14A (1.4nm) node will enter manufacturing in 2026. The corporate can also be working to create totally autonomous AI-powered fabs sooner or later.
Intel’s Keyvan Esfarjani, the corporate’s EVP and GM and Foundry Manufacturing and Provide, held a really insightful session that lined the corporate’s newest developments and confirmed how the roadmap unfolds over the approaching years.
Picture 1 of 2
Right here, we will see two charts, with the primary outlining the corporate’s Ok-WSPW (hundreds of wafer begins per week) capability for Intel’s numerous course of nodes. Notably, capability usually signifies what number of wafers might be began, however not the entire output — output varies based mostly on yields. You will discover there is not a label for the Y-axis, which might give us a direct learn on Intel’s manufacturing volumes. Nevertheless, this does give us a strong concept of the proportionality of Intel’s deliberate node manufacturing over the following a number of years.
Intel didn’t specify the arrival date of its coming 14A node in its earlier bulletins, however right here, the corporate signifies it is going to start manufacturing of the Intel 14A node in 2026.
Much more importantly, Intel will start manufacturing/growth of its as-yet-unannounced 10A node in late 2027, filling out its roster of nodes produced with EUV know-how. Intel’s ‘A’ suffix in its node naming conference represents Angstroms, and 10 Angstroms converts to 1nm, that means that is the corporate’s first 1nm-class node.
Intel hasn’t shared any particulars in regards to the 10A/1nm node however has instructed us that it classifies a brand new node as no less than having a double-digit energy/efficiency enchancment. Intel CEO Pat Gelsinger has instructed us the cutoff for a brand new node is round a 14% to fifteen% enchancment, so we will count on that 10A could have no less than that degree of enchancment over the 14A node. (For instance, the distinction between Intel 7 and Intel 4 was a 15% enchancment.)
Capability for the Intel 4 and Intel 3 processes would not construct as shortly as 20A/18A, however that is not stunning — nearly all of the corporate’s wins for its third-party foundry enterprise have been with the 18A node, which Intel says is according to plan. As proven, Intel will even steadily draw down its total manufacturing of its 14nm, 10nm, Intel 7, and 12nm nodes because it transitions to the EUV-enabled nodes.
Notably, Intel’s slide has a footnote that claims, “*roadmap: Ultimate Scale, Tempo and Processes depending on enterprise circumstances and incentives,” mirroring the corporate’s ongoing statements that funding from the CHIPS Act will impact its ability to scale production.
Intel’s 18A and 20A nodes have been in no less than some type of manufacturing since 2023, which is not too stunning — the corporate received a RAMP-C contract for 18A with the US government again in 2021, and firms like IBM, Microsoft, and Nvidia have been engaged on producing take a look at chips since (Understandably, Intel will not share details with the general public, however it has received a $1 billion award from the US Authorities for this system).
Moreover, Intel’s 20A finds the corporate integrating two new applied sciences without delay – bottom energy (PowerVIA) and GAA transistors (RibbonFET). To de-risk the method and keep away from stumbles prefer it noticed with 10nm, Intel introduced in April of 2022 that it had run a unique taste of 18A by its fabs, an inner take a look at node with solely bottom energy paired with normal FinFETs, to ensure backside power worked separately before being integrated into the final 18A node. As such, we will count on that 20A wafers have been flowing by an Intel fab for fairly a while.
As proven on the appropriate, Intel will even aggressively ramp up its superior packaging manufacturing capability for Foveros, EMIB, SIP (silicon photonics), and HBI (hybrid bond interconnect). Superior packaging capability has been a key choke level for the present shortages of AI accelerators. This elevated capability will guarantee a gentle provide of superior processors with complicated packaging, together with HBM.
Intel’s ramp of superior packaging capability is explosive — the corporate had little or no manufacturing capability for these interconnects in 2023. As an apart, Intel lately wrapped up all of its inner packaging efforts utilizing normal packaging; it’s now all-in on superior packaging and can use OSATs (outsourced meeting and take a look at firms) for traditional packaging duties.
The second slide within the above album visualizes how Intel’s transfer to working as an exterior foundry will permit it to extend each the quantity of manufacturing for every of its nodes and the size of time that every node is in manufacturing, thus maximizing revenue from its fab and gear expenditures because it serves its buyer orders over longer durations.
Picture 1 of 3
Esfarjani additionally shared particulars about Intel’s globe-spanning operations. Along with its current services, the corporate plans to speculate $100 billion over the following 5 years on expansions and new manufacturing websites.
The slides above define the varied areas of node manufacturing, with 18A occurring in Fab 52 and 62 in Arizona. In distinction, the superior packaging and 65nm foundry operations for Tower will happen in Fab 9 and 11X in New Mexico. Intel did not share the place it plans to provide its 10A node, and it additionally has ongoing expansions in Ohio, Israel, Germany, Malaysia, and Poland.
This geographically distributed manufacturing capability, spanning each chipmaking and packaging, permits Intel to have world redundancy in its operations whereas additionally providing its foundry prospects the choice of leveraging a provide chain solely positioned in America.
Picture 1 of 2
As we illustrated in our protection of our tour of Intel’s Penang, Malaysia facilities, the corporate leans closely on automation in its foundries. Intel now plans to make use of AI in all segments of its manufacturing flows, from capability planning and forecasting to yield enhancements and precise floor-level manufacturing operations, in a ’10X moonshot’ effort.
Esfarjani did not present a timeline for the corporate’s moonshot effort however stated it is going to impression each facet of its operations sooner or later. That features the introduction of AI “Cobots,” that are collaborative robots that may work together with people, and intensive robotic automation within the manufacturing course of.
Within the meantime, Intel will proceed to aggressively pursue any and all potential customers for its operations. You may learn extra about these efforts in our interview with Stu Pann, the SVP and GM of Intel Foundry Companies, who’s tasked with making Intel Foundry the world’s second-largest foundry by 2030.