2nm process node will not enter commercial production, there's no Arrow Lake variants manufactured in-house by Intel
Last week, Intel officially launched the first of their next-gen Core Ultra 200 CPUs, which brings new architectures as well as more advanced silicon technology. That raised optimistic expectations , but those are getting cast in doubt now. Intel has cancelled its 2nm manufacturing process node (Intel 20A), which was part of its plan to catch up with TSMC. However, the follow-up process nodes are still being developed.
Intel sort of tried to bury this news in the announcement of “continued momentum of the 1.8nm node”, which the company says looks promising (according to earlier plans, it was supposed to be ready for mass production by the end of 2024). And this, according to Intel, allows it to redirect resources and engineers from working on the 20A node (2nm technology) to Intel’s 18A node (1.8nm technology).
In reality, this is probably not good news, as this seems to mean the cancellation of the entire 2nm technology. It has been essentially fully developed, but it will not be commercially deployed in production. And this has not escaped media attention despite the attempt to hide the bad news.
This measure appears to be part of the cost-cutting Intel has resorted to after recent financial results that did not look good at all. Intel says directly that this measure will allow it to save investment spending. Intel CFO Dave Zinsser has meanwhile said outright that the company will “skip over productizing” of the 20A node in order to save capital expenditures and overall costs. Reportedly, this is expected to save up to half a billion dollars. This was stated during the Citi Global TMT Conference.
Arrow Lake processors will only be made outside of Intel
It’s been known for some time that much of the Arrow Lake processors that represent Intel’s next generation (Core Ultra 200) in laptops and desktops would be manufactured on TSMC’s 3nm process node – the largest competitor in silicon technology. This was confirmed for the highest-performance version of the CPUs, which was to include a CPU chiplet with 24 cores (8 P-Cores, 16 E-Cores), the so-called stepping B0. In addition to this, SoC chiplets and GPU chiplets are also produced at TSMC.
But Intel was still supposed to produce chiplets with 6+8 cores for cheaper models on its own 20A node. This silicon was to be used in the Core Ultra 5 and Core Ultra 3 desktop processors (called stepping C0). We discussed their models and specifications in this article:
Read more: Intel’s first 2nm chips: 65W Arrow Lake, Core Ultra 3 and 5 parameters
But the cancellation of the 2nm process node probably means the end of these plans. Intel now states that Arrow Lake will be “manufactured primarily at TSMC”, which doesn’t sound like a statement that would categorically rule out 20A being used for Arrow lake, but if Intel cancels spending on “productization” of the 2nm process node, that probably means that nothing will be produced on it at all.
So the C0 stepping processors will either all be made from the larger and more expensive 3nm B0 chips, or Intel will transfer the C0 stepping to production at TSMC (it’s possible the company may have already done so, quietly). All versions of Arrow Lake for laptops will probably also be made on the TSMC 3nm process node. These were also supposed to have CPU chiplets with 6+8 cores.
Intel will retain a certain share of the overall production of Arrow Lake processors – the company will do the packaging of the processors (i.e. assembling individual pieces of silicon together and fitting them into a BGA or LGA package) in its own production lines. Arrow Lake does use advanced 3D packaging, so Intel’s part actually deserves mentioning in this case, but the company’s technology will not contribute to the actual production of “high-tech” silicon dies (or chiplets or tiles) themselves.
Doubts over the future of Intel’s silicon technology
There has been speculation that what Intel may be doing is essentially bringing to market what it previously referred to as the 20A node, but rebranding it as the Intel 18A process, while quietly abandoning the originally planned version of the 18A node. However, this is probably an oversimplified idea based on the fact that Intel has cancelled the deployment of High-NA EUV technology on the Intel 18A node, which was originally supposed to be one of the differentiating features of 18A. So from a layman’s perspective, 20A and 18A may not appear to be that different, as both technologies were supposed to have RibbonFET transistors and PowerVIA technology. But this is probably a mistaken view, there are certainly going to be differences between the technologies.
Intel still plans to deploy High-NA EUV technology. But it’s supposed to be introduced on the following 1.4nm node (“Intel 14A”). We wrote about it here:
Read more: Intel announces 1.4nm process, first node with high-NA technology
Perhaps what could be closer to truth is that the Intel 20A node can be seen as an “early” version of Intel 18A, similarly to the strategy Samsung is pursuing, where for each new generation, the Korean company is separately introducing an “early” variant first, followed by a “Plus” node, where the parameters (for example, in the achieved clock speeds) are typically much better. Intel also states in its press release that the experience gained on Intel 20A will be part of the rollout of 18A. The implication Intel wants to suggest is probably that it 20A testing worked so well that the 18A process will be available soon enough that there is no point in completing the rollout of the 20A technology.
However, this can also be understood as meaning that the 20A node was completed too late. What is probably also important is that there was no use for it, even with the potential to produce some of the Arrow Lake processors. It was generally expected that producing a portion of the desktop models on Intel’s 20A node would have an economic benefit for Intel (as it would be cheaper than outsourcing to TSMC), but from the current news it seems that even that didn’t pan out. Although it is possible that the problem with 20A’s rentability was fixed costs, where individual processors produced on the 20A node would be cheaper but this would not offset the overall cost of equipment and line ramp-up at the volumes that Intel would be making.
This is not good news either. Not so long ago, Intel’s own fabs and large economic scale were thought to be a decisive advantage against all competitors. But current developments suggest the exact opposite – that the company cannot afford its own fabs anymore because what it would produce in them is not able to make enough money to offset the fab buildup cost.
If the 20A’s process node sacrifice is only a temporary strategy and everything really improves again with the advent of the 1.8nm process node (on which the Panther Lake processors are to be produced, which will come out in a year, but only for laptops), maybe the decision will pay off for Intel. However, there is probably a risk that the more and longer Intel keeps outsourcing CPU production to TSMC, the harder it will be for the company to shift the entire business back to its own fabs and regain a position where Intel is back on the technological cutting edge and able to stay there. There is a danger that manufacturing processors using the services of an arch rival could become one of those proverbial stopgap solutions that ultimately turn out to be permanent.
Sources: Tom’s Hardware, Intel
English translation and edit by Jozef Dudáš
⠀
