High NA EUV was supposed to debut on Intel’s 14A node towards the end of the decade. It has turned up early, on Panther Lake, patterning layers that are also qualified to run on the old tools.
A machine that costs about $400mn is now printing a handful of the layers inside a laptop processor. ASML said on Wednesday that Intel Foundry has entered high-volume manufacturing on a subset of its Core Ultra Series 3 chips, code-named Panther Lake, using High NA extreme ultraviolet lithography. It is the first time any chipmaker has shipped a volume logic product patterned with the tool.
Panther Lake runs on Intel 18A, the node behind the chips that launched at CES in January and now ship in more than 200 laptop designs. Nobody expected High NA to show up there.
The plan, for years, was to hold the tool back for 14A, the node after next. It has arrived early instead, patterning what ASML carefully calls “specific Intel 18A layers” at Hillsboro, Oregon.
What High NA buys is resolution. Its optics gather light at a numerical aperture of 0.55 rather than the 0.33 of a standard EUV scanner, which cuts the smallest printable feature from 13nm to 8nm in a single exposure.
That is worth money because the alternative is multiple patterning: exposing the same layer two or three times, with etch and deposition steps in between, to assemble a pattern the optics cannot resolve in one go.
Every extra pass adds cost, adds cycle time, and gives the layers another opportunity to drift out of alignment. Single-exposure demonstrations at imec, on logic and DRAM patterns, were the early evidence that one pass could do the work of several.
The hardware history is short. Intel and ASML integrated the industry’s first commercial High NA system at Hillsboro in 2024, and Intel passed acceptance testing on the second-generation TWINSCAN EXE:5200B in December, at 175 wafers an hour and 0.7nm overlay accuracy.
There is a catch, and ASML put it in its own release. The 18A layers in question are “dual-qualified”, certified to run on High NA and on the older NXE machines at yields the company says are matched.
Panther Lake, in other words, does not need High NA. Intel is running the tool because it owns one, the wafers have to come out of somewhere, and every wafer through it is data.
Naga Chandrasekaran, who runs Intel Foundry, framed it as output rather than ambition. Qualifying the High NA option on select 18A layers means “our existing fleet of tools are providing customers with increased output,” he said, “while we develop future options” for later nodes.
The economics are why the industry has spent three years arguing about when to switch. At roughly $400mn a scanner costs twice what a standard EUV machine does, and it images half the field, so a full-size pattern has to be stitched from two exposures.
TSMC has been the conspicuous holdout. Chairman C.C. Wei told shareholders in June that the company owns High NA tools and is running research on them, and that cost is the only thing keeping them out of mass production. Its A13 and A12 nodes, both aimed at 2029, are not expected to need it. That position has softened only in tone since TSMC’s first public doubts in 2024.
Intel’s next step remains 14A, where High NA is designed in rather than bolted on. The node is on track for risk production in 2028, with the 0.9 process design kit due in October, the point at which prospective foundry customers commit to volumes.
The spending carries on regardless. Intel put €5bn into its Irish fab last week, one of the few EUV sites in Europe. Intel itself declined to comment on the announcement.
ASML had the better day of the two. It raised 2026 revenue guidance to €43-45bn from €36-40bn on AI demand, and said it would expand EUV and DUV capacity by 30% in each of the next two years. Chief executive Christophe Fouquet told reporters in May that the first logic and memory products off High NA machines were months away. He was right, which is not the usual outcome for a lithography roadmap.
Get the TNW newsletter #
Get the most important tech news in your inbox each week.