Intel news has been coming out a lot lately:
Intel 3 is ready to go;
18A node picks up a big Microsoft order;
Back to global foundry technology leader in 3 years;
Becoming the world’s second-generation foundry by 2030;
……
Intel’s great transformation seems to have reached a new critical point. At the recently concluded Direct Meeting, “Blue Giant” Intel announced that IFS (Intel Foundry Services), Intel’s foundry services division, has officially changed its name to Intel Foundry, and at the same time has completely updated its roadmap, including the debut of Intel’s 14A (1.4nm) process.
As the most central part of IDM 2.0, “foundry” means that Intel no longer only makes chips for itself, and has begun to manufacture chips for other chip design companies and Intel’s product business units and other customers like TSMC, which has forced Intel to accelerate its efforts to catch up with TSMC, Samsung, and other technology leaders in the chip manufacturing process.
Before this, Intel’s original IDM (vertically integrated manufacturing) model had been unsustainable: chip manufacturing process technology route miscalculation, resulting in a comprehensive backwardness of the process, in the 14nm spent too much time and energy, but also affects its products are abandoned by Apple, was AMD beyond, and even by the consumer market is widely quoted as “toothpaste factory”.
But today, nearly three years after the IDM 2.0 transition, many things have changed, especially the new wave of generative artificial intelligence, and the trend of foundry from the manufacturing of the chip more and more extended to the packaging and other aspects. As current Intel CEO Pat Kissinger said:
This is a pivotal moment in the evolution of the Intel foundry. We see an opportunity to reposition the Intel Systems Foundry.
Expanding the ‘four-year plan’, Intel’s process roadmap refreshes again
According to Kissinger’s plan in 2021, Intel will update five process nodes in four years. After Intel 7, Intel 4 has already landed on the latest Core Ultra mobile processors, officially claiming that it represents the most significant architectural change for Intel in 40 years.
On the other side of the coin, Intel 3 (nominally 3nm) is ready to go, with Intel announcing last July that the Intel 3 process is on track in terms of capacity and performance for the two Xeon processors to be released in 2024 – Granite Rapids and Sierra Forest (in the P- and E-Core product lines). The two Xeon processors to be released in 2024 – Granite Rapids and Sierra Forest (in the P-Core and E-Core product lines) – will both be built using the Intel 3 process.
As Intel’s first high-volume EUV (Extreme Ultraviolet) node, Intel 3 will also be available in different versions over the next few years, including Intel 3-T with support for through-silicon vias (TSVs) this year, Intel 3-E with expanded capabilities in 2025, and Intel 3-PT with higher performance based on second-generation TSV technology even later.
After that, Intel will officially launch 20A (2nm) and 18A (1.8nm) processes with PowerVia backside power technology.
The 20A node was ill-fated. Originally Intel Arrow Lake and Qualcomm are planning to use the 20A process, but subsequent news has pointed out that Arrow Lake has shifted to TSMC 3nm. including Kissinger has just confirmed that the two upcoming processors will be manufactured using the TSMC 3nm process.
According to a report last year by analyst Ming-Chi Kuo of Tianfeng International, Qualcomm has long since stopped developing Intel 20A chips. It is also rumored that the 20A process will be used only by Intel’s product division and will not be actively offered to third-party customers. Intel’s latest roadmap doesn’t include any plans for a successor to the 20A process, confirming the previous rumors to some extent.
In contrast, 18A is getting more and more attention. Not only is it widely supported by semiconductor ecosystem vendors such as Synopsys (an EDA giant) and ARM, but it is also “favored” by more and more chip design companies, and Microsoft CEO Satya Nadella also made an official announcement at this direct meeting:
Microsoft CEO Satya Nadella also announced at the event that Microsoft will build chips based on the Intel 18A process.
Microsoft did not explicitly say the details of this chip, but at the end of last year, Microsoft just released two closely related to artificial intelligence self-developed chip – Maia 100 AI chip and Cobalt 100 CPU. if not unexpected, is that these two chips will use Intel’s latest 18A process.
In addition, Intel has just recently completed the flow of the 18A main product Clearwater Forest, and subsequently plans to launch the performance-enhancing 18A-P node.
In this version of the roadmap, in addition to a series of iterative versions of the increase, the most noteworthy part or the first announcement of the 14A (1.4nm) and its iterative version of the 14A-E, which will be the key to Intel’s real return to the global foundry technology leader.
From 14nm to 14A, Intel wants to overtake TSMC
If the launch of the 14nm process Skylake from 2015, to 2025 Intel officially put into mass production of the 18A process, it is exactly 10 years. In these 10 years, Intel has been stuck in the upgrade from 14nm to 10nm for 7 years, and the time left for Intel to cross over from 10nm to 18A (1.8nm) is only full counting:
3 years.
It would be a technical and engineering “miracle” if it could be realized on schedule. And by then, Intel will have completed catching up with the leading groups (TSMC and Samsung) in terms of process technology.
According to the current roadmap of Intel and TSMC, Intel will realize the mass production of 18A process in 2025, and TSMC 2nm will be officially put into mass production in 2025.
It should be noted here that while Intel’s 18A process is nominally a 1.8nm process, and Kissinger has repeatedly emphasized the process’ lead over TSMC’s 2nm process, the actual comparison of transistor densities shows that the two are not far apart, and both use the next-generation GAA (full wrap-around gate) architecture.
From Fin FET to GAAFET, Photo/Samsung
At present, Intel 18A has more direct advantages, or from the unique back power supply technology, of course, TSMC 2nm will also have some unique technical advantages. But overall, the Intel 18A process is the actual target of TSMC, Samsung’s 2nm process.
In other words, Intel will have to go to the next process node – 14A before it can really overtake TSMC.
According to Intel’s plan, they will develop the 14A process by 2027, and will use ASML’s High-NA EUV (High Numerical Aperture Extreme Ultraviolet) lithography for the first time at that node. Lithography giant ASML took to X (formerly Twitter) late last year to announce that the first High-NA EUV lithography machine was being loaded and shipped from its headquarters in Veldhoven, Netherlands, for delivery to Intel’s Oregon facility:
“We are very pleased and proud to be delivering our first high-NA EUV system to Intel,” said the company.
According to ASML CEO Peter Wennink, the price of a single High-NA EUV is between 300 million and 350 million euros (about 2.26 billion yuan to 2.64 billion yuan), and Intel’s first rush to buy it just shows the urgency and importance it attaches to reclaiming its position as a technology leader.
In conclusion
Will Intel complete this painful and lengthy transition?
It’s hard to say. But before Kissinger returned to Intel in 2021, the ‘blue giant’ had reached the point where it had to change. Three years later, Intel still has a lot of problems, IDM 2.0 transformation also continues to promote in the controversy, they did not choose to split the wafer foundry business, but the foundry and product business into two independent financial accounting parts.
The new wave of generative AI has brought more questions about Intel, relying on CPUs at the expense of GPUs has allowed Intel to miss a lot of opportunities, and has significantly increased the risk of “losing ground” in the data center market. But at the same time, generative AI also makes everyone, especially Intel, see the huge potential of reinventing the PC.
These are the challenges that Intel is most directly facing at the moment, and the only thing we can be sure of is that Intel’s “Great Transformation” has not yet reached a definitive conclusion.
