Last week we had the news that Intel managed to get a client for its most modern 1.8nm process (Intel 18A) on which Faraday server CPUs are to be manufactured. Paradoxically, it seems that Intel will look elsewhere for its own chips. There has been information that the processors it plans for 2026 are to use TSMC technology. This raises questions about the quality of Intel’s silicon technology and what the company claims about it.
Nova Lake by TSMC 2nm process
The Taiwanese media (Taiwan Economic Daily) brought information that Intel is among those interested in manufacturing on TSMC’s 2nm process, which will be the company’s first technology using GAAFET type transistors. According to these reports, Intel has reportedly even signed contracts and has reserved capacity. This technology is supposed to start production in 2025, so it will probably lead to products sold only in 2026 (if not later).
According to insiders, it could therefore possibly be Nova Lake processors, which are still relatively far away. Now, after Meteor Lake, Arrow Lake is waiting for us this year, after that (2025) Panther Lake, which is supposed to use Intel’s 1.8nm process, and only then Nova Lake – possibly on TSMC’s 2nm process.
It must be said that this is not yet a given, the equation between the Nova Lake generation and TSMC’s 2nm process is only conjectured from the timing. It’s quite possible that Intel wants the 2nm process instead for discrete GPUs, ASIC AI accelerators, FPGAs, or perhaps a processor other than Nova Lake. It can also be just one of the tiles in the processor or GPU.
However, this is not as important here as the fact that Intel is interested in TSMC’s 2nm process commercially available from 2025 to 2026. At the same time, it has its own 2nm, and especially 1.8nm process (Intel 20A and 18A – their names mean 20 and 18 angstroms , which is the same as 2 nm and 1.8 nm). And these processes of his are supposed to be ready for production this year, so the products planned for 2025-2026 at TSMC could alternatively be made at 20A, or 18A at Intel. According to the numbering, the 18A process (1.8 nm) should be even better, or at least that is the intention to indicate. So there is a big doubt here, whether we can really take Intel’s new numbering seriously.
Intel processor roadmap shown at Intel InnovatiOn 2023 event. Nova Lake follows the Panther Lake generation
Author: Intel
Rather, it seems that TSMC’s 2nm process will be qualitatively more advanced than Intel’s 18A, which is why the former semiconductor leader is interested in it. After all, it would also make sense considering that TSMC’s technology will be available later. It would be remarkable, especially if the 1.8nm Panther Lake processors were released in 2025, and a year after that, the new generation Nova Lake would switch to a new technology, but it would be marked with a “worse number”. It would be interesting to see how Intel explains this.
Is TSMC’s 2nm process better than Intel’s 1.8nm process?
We have hinted at these doubts before. We usually defend the fact that the improved 10nm process (10nm Enhanced SuperFin) has been renamed to the 7nm process (Intel 7). It can be said that this new numbering will probably hold up with its parameters. Until then, it had to be explained everywhere that Intel’s processes have “behind” numbering and, for example, Intel’s 14nm process could be more competitive with TSMC’s 10nm technology. For Intel 7 and TSMC N7, it simply equaled.
More: Renumbering nanometers? Intel is said to want to rebrand production processes to keep up with TSMC
But the question is whether Intel’s new numbering will remain as “honest” and agreeable in the following generations. Before the renumbering, Intel had a 7nm process (with an improved version 7+) and a 5nm process (probably also with an improved version or versions) in the roadmap. We know that Intel rebranded the “old” 7nm process as Intel 4, so it jumped ahead a bit against the principle that its processes are a generation better, which would give the designation “Intel 5” (and “Intel 4” may then have been previously planned an improved “7nm+” process, similar to how TSMC has an improved N4 variant derived from the N5 technology). But what Intel did was to make the improved “7nm+” variant straight into “Intel 3”, which implies parity even with TSMC’s 3nm process. However, this is its own new generation, so here Intel is already hinting that its old 7nm technology should be almost two, and not just one, generation better. And there is reason to doubt that.
And it can probably be similar for later Intel processes, of which the Intel 20A and Intel 18A technologies have been announced so far. It can be assumed that these are new names for technologies that should probably have been called 5nm and 5nm+ in the old Intel markings. So, if the rule were to apply that Intel is one generation better, then we would expect that after rebranding they would be called Intel 30A and perhaps Intel 28A (at TSMC, 2 nm is already another big new generation, so we cannot label the former 5nm+ 2nm / 20A).
Procesy Intel 20A a Intel 18A
Author: Intel
If you look at it with this skepticism, it suddenly makes sense why Intel would then use TSMC’s 2nm process for the successor to the Panther Lake processor, which was already using its “1.8nm” 18A technology, even though it is supposed to be older and worse. In fact, it could probably be a technology half to a whole generation newer than Intel 18A (formerly “5nm+”).
It should be added that Intel’s development roadmap must have changed over the years, so we cannot guarantee that the equivalence presented in this way between what was once planned as a 5nm and 5nm+ process and the 20A and 18A technologies that are now underway is really valid. Of course, there is a certain chance that Intel really managed to get up to such a high pace that it caught up with the delay caused by the delays of the 10nm process, and still managed to improve the parameters in the next generations so that the 20A technology (previously 5nm) will really be comparable with TSMC’s 2nm process.
But after the experience of Intel’s failures over the past ten years, we would not bet entirely on this optimistic scenario. At the very least, we should not take this interpretation as somehow automatically plausible. It is probably appropriate to take Intel’s suggested equivalence of 2nm from TSMC with the Intel 20A process with a large margin and wait to see how it will look in practice.
Source: TechPowerUp, Taiwan Economic Daily
2024-02-11 01:20:45
#Intel #apparently #admitted #1.8nm #silicon #technology #inferior #TSMCs #2nm #process #Cnews.cz