Intel CFO confirms that 14A will be more expensive to use than 18A — Intel expects 14A fabrication process to offer 15-20% better performance-per-watt or 25-35% lower power consumption compared to 18A

(Image credit: Intel)

Intel's 14A manufacturing technology (1.4nm-class) technology — the company's first fabrication process that was designed both for Intel and its foundry customers from the ground up — will be more expensive to use than the company's upcoming 18A production node according to Intel. The reason for that will be use of ASML's next-generation Twinscan EXE:5200B High-NA lithography machine with a 0.55 numerical aperture optics.

"14A is more expensive than 18A," said David Zinsner, chief financial officer of Intel, at the Citi’s 2025 Global TMT Conference. "It is not significantly [more expensive] in terms of investment. [But] it is a higher wafer cost, for sure and partly that is because we, we are expecting to use High-NA EUV tools in 14A, which was not the case in 18A."

Intel expects its 14A fabrication process to offer 15% – 20% better performance-per-watt or 25% – 35% lower power consumption compared to 18A. The new manufacturing technology features RibbonFET 2, an upgraded gate-all-around transistor structure, and PowerDirect, a backside power delivery network that connects power wires directly to source and drain of transistors. A key innovation of 14A is Turbo Cells, which enhance CPU and GPU frequency by optimizing critical timing paths using high-drive, double-height cells within dense standard cell libraries, which boosts speed without major area or power compromises. However, to enable 18A production, Intel needs to use next-generation lithography systems that offer better resolution and therefore does not need to rely on multi-patterning.

ASML's Twinscan EXE High-NA EUV systems achieve a 8nm resolution in a single exposure, a notable leap from the 13.5nm resolution of current Low-NA EUV tools. Although Low-NA systems can also reach 8nm using double patterning, this approach increases process complexity and can impact yield. However, High-NA EUV reduces the exposure field by 50%, requiring chipmakers to adjust their design strategies. Moreover, the new systems are more expensive than existing litho machines.

ASML's current-generation Low-NA EUV Twinscan NXE:3800E tool costs around $235 million per unit, whereas its upcoming High-NA EUV Twinscan EXE:5200B (or more advanced) tools are expected to cost $380 million per unit. Due to the high cost and unique challenges posed by the technology, adoption of High-NA EUV vary: Intel is pushing for early deployment with 14A several years down the road from now, while TSMC is taking a more cautious approach.

Modern fabs designed to produce chips on leading-edge process technologies currently cost from $20 billion to $30 billion depending on capacity, so adding four High-NA tools at $1.52 billion will hardly increase fab cost significantly. Meanwhile, 14A R&D costs billions of dollars. As a result, once all the extra costs are added, they really make 14A process technology more expensive than 14A, which is why Intel Foundry needs external customers to justify extra costs.

Earlier this year Intel announced that if it does not land a single significant customer for 14A, it would either slowdown its development, or will abandon it completely as developing and ramping up of a leading-edge process technology is a too expensive endeavor for a single company these days. However, Intel may no longer have a choice, but to finish development of 14A and then start production of chips on this technology as Intel Foundry is an indispensable part of Intel that cannot be spun off. Under the terms of the deal with the U.S. government, under which Intel converted its grants into equity, Intel should control at least 51% of Intel Foundry over the next five years as otherwise it will breach its contract with the U.S. government.

TOPICS

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

6 Comments Comment from the forums

acadia11

18A should have been the process where Intel fixed its tooling to be a proper client manufacturer as a foundry as opposed to the focus they had on their own products. The Intel second mindset is tamtamount if they want to rival TSMC customer centric monopoly. Beating TSMC and SAmsung to High NA is their Zen, if they execute they are properly positioned to turnaround foundry. They learned their lesson with EUV being delayed in their 7n processes. So this time around they know to be first with the power consumption demands and core counts AI will drive its about time Intel got in thr game proper like. The more expensive part they can address with thr proper volume and yields ….
Reply
dalek1234

"they really make 14A process technology more expensive than 14A"
So something that cost a specific amount also cost more than that amount. This is some quantum mechanics stuff.
Reply
nogaard777

dalek1234 said:
"they really make 14A process technology more expensive than 14A"
So something that cost a specific amount also cost more than that amount. This is some quantum mechanics stuff.
Yeah that headline is definitely some 🤦 stuff. Of course a newer more refined node costs more than it's predecessor.
Reply
nogaard777

acadia11 said:
18A should have been the process where Intel fixed its tooling to be a proper client manufacturer as a foundry as opposed to the focus they had on their own products.
So you're saying they should have bet the farm on customers that don't exist yet instead of their own designs that are a guaranteed user?

While it's mostly agreed that Intel foundries has caught back up to TSMC why would any TSMC customer shift to Intel when they're perfectly happy at TSMC? Intel needs to be a better choice before anyone will take a gamble on IF.

Much like processors why do the OEMs keep using Intel when Zen4 and 5 have caught up with Intel? Because there's nothing wrong with Intel and they have proven time and again that they have the capacity to deliver the millions of chips they need at a good price. As much as gamers in the diy space love to point at X3D to claim AMD is in the lead X3D is an overpriced 1 trick pony that OEMs can't rely on. Why would an OEM use an X3D in a laptop that doesn't use a dedicated GPU? That would be stupid.

When it comes to the mainstream chips Arrow Lake competes rather well against non-X3D Zen5 and their laptop chips never had the raptor lake failures that have been fueling clickbait articles for a year. There's no reason an OEM would gamble on AMD when they're perfectly happy with Intel with more capacity at better prices.

AMD can't break into the GPU business because they need to be notably faster and cheaper than Nvidia. Which they aren't. AMD can't break into the laptop business because they need to be notably cheaper and faster than Intel. Which they aren't. Intel can't break into the 3rd party wafer business unless they're faster and cheaper than TSMC. Which they aren't.

Customers need to become unhappy with TSMC before they'll shift to Intel Foundries. For now that's not the case, but if TSMC keeps raising prices that may change soon in the future.
Reply
bit_user

nogaard777 said:
Of course a newer more refined node costs more than it's predecessor.
Well, not if you're talking about per-transistor, which is usually how cost-comparisons are done between foundry nodes. It's only somewhat recent that MTr/$ (Million Transistors per dollar) of new nodes has stopped being cheaper than prior nodes.

Yes, I think wafer cost has pretty much always gone up. But, you'd traditionally gain enough density to more than compensate for that, which is how chips have generally been able to get ever more complex without getting massively more expensive.
Reply
TerryLaze

nogaard777 said:
why would any TSMC customer shift to Intel when they're perfectly happy at TSMC? Intel needs to be a better choice before anyone will take a gamble on IF.
...
....
There's no reason an OEM would gamble on AMD when they're perfectly happy with Intel with more capacity at better prices.
For both cases, there is no gamble, tsmc is over booked for years to come, everybody wants to make more products than what tsmc alone can handle, every customer of tsmc can also use intel either to make the same thing at two places at half the time (or at least reduced time) or for completely different products or support chips.

Same for AMD and OEMs, they can use intel all they want and also make some AMD models if they want to make something intel doesn't have a good fit for (or for whatever other reason)

In both cases they don't have to switch all of their production from one to the other.

The only gamble is the one that they make every time no matter what, will we be able to sell everything we produce or not.
Reply

Show more comments