So… bad news good news department. Intel appears to be skipping over the 20a process to put more focus on delivering 18a. Another story today revealed that Broadcom decided 18a wasn’t satisfactory yet and decided not to use it at this time either. May need an extra quarter (at least) to come ready. A third story here suggests that it will be 2027 before Intel sees “meaningful” revenue from the fab division. (I’m not sure whether that means “meaningful sales” or “sales that can offset the vast expenditures needed to scale up.”) It’s ugly weather for Intel Foundry, that’s what I’ll say.
Will TSMC be a beneficiary? Almost certainly – what other choices are there?
Will AMD be a beneficiary? Hard to say. Creates extra pressure at TSMC that might compete for capacity AMD could use. Unless it’s all bleeding-edge capacity and AMD is still aiming for one node behind Apple et alia. Could also create a delay in Arrow Lake that would improve AMD’s competitive position on the desktop. On the whole I’d say it’s a net win but others may dig deeper and draw a different conclusion.
In a surprise move, Intel announced today that it no longer plans to use its own ‘Intel 20A’ process node with its upcoming Arrow Lake processors for the consumer market. Instead, it will use external nodes, likely from partner TSMC, for all of Arrow Lake’s chip components. Intel’s only manufacturing responsibilities for the Arrow Lake processors will be packaging the externally manufactured chiplets into the final processor.
…
The node change comes after Intel initially demoed a wafer of Arrow Lake processors fabbed on the 20A node at its Innovation 2023 event, which indicated the chips were already far along in the development cycle. At the time, Intel said Arrow Lake would come to market in 2024. Since then, industry rumors pointed to the 20A node only being used for a subset of the Arrow Lake family, while the remainder would use a TSMC node.
Intel says its crucial next-gen ‘Intel 18A’ node remains on track for launch in 2025. It has now shifted its engineering resources from 20A to the newer 18A node, which the company says was spurred by the strength of the yield metrics for 18A. Intel again noted that it had reached a sub-0.40 D0 defect density (def/cm^2) for 18A, a critical measurement of the yield rate for a process node. A process node is usually considered production-worthy and healthy once D0 reaches 0.5 or below.
It appears that Intel will now leapfrog over its 20A process entirely and avoid the capital expenditures required to bring the node to full production. Eliminating the always-eye-watering ramp costs of a new node, particularly one as advanced as 20A, will surely contribute to the company meeting its cost-cutting goals.
The Intel 20A node was never planned for many products due to the company’s fast-track move to the more advanced 18A node as it races to meet its goal of delivering five nodes in four years, so building out an extensive 20A production would have limited returns. However, Intel’s 20A served as a vehicle for several new advances, like RibbonFet Gate-All-Around (GAA) technology, which is Intel’s first new transistor design since FinFET arrived in 2011. It also marked the debut of the company’s PowerVia backside power delivery tech, which routes power for the transistors through the backside of the processor die.