TSMC’s 3nm process is wanted by AMD, Apple, Broadcom, Intel, Mediatek, Nvidia and Qualcomm

This year, products built on the 5nm generation of TSMC processes (i.e. 5nm and 4nm process) will enter the PC segment, and the next big step, which will take place in about two years, will be the 3nm generation. However, this process is already attracting unusually strong attention not only from potential customers, but also from the media and analysts. You could even say that he stirred up emotions.


For example, the Commercial Times reported that Apple will start production of the SoC M2 Pro/Max on the 3nm process as early as next month. In turn, analyst Ming-Chi Kuo responded by stating that TSMC states in its financial outlook that the 3nm process will only start generating profit in the first half of 2023, so it is possible that the M2 Pro/Max will still be produced on the 5nm process. In other words, the 3nm process would be delayed and production wouldn’t start until after the New Year. Other sources, on the other hand, confirm the start of production as planned, in September.

Where can the truth be? I would have one hypothesis that would be able to combine the start of 3nm production next month with the first revenues of 3nm production in the first half of 2023. We know that TSMC is getting high upfront payments with the 3nm generation. However, these are really deposits, i.e. funds that the customer subsequently chooses back in the form of removed chips, which are no longer (again) valid. So TSMC can have the advance from Apple accounted for long ago, and this year TSMC can supply Apple with 3nm chips without Apple paying anything that would show up on TSMC’s books as revenue from the 3nm process.


Another important customer that has reserved 3nm capacities is AMD. It can be expected that for the year 2023 it will mostly only be about the production of test samples, while serial production will be used at the end of the mentioned year at the earliest.

AMD is expected to use this technology for APUs Strix Pointwhich will carry the kernels Zen 5 (and probably some kernel derivative Zen 4). Information about large / chiplet processors (Ryzen, Epyc) from the generation Zen 5 are not clear yet. According to some sources, it may be a 4nm process, according to others, 3nm.


Although Intel is now rebelling against this fact and trying to give the impression that it never intended to use the 3nm process for Meteor Lake processors, in fact, the company’s CEO visited TSMC directly to acquire capacities, and that Intel was counting on the 3nm process was confirmed by a number of independent sources. From the available reports, I dare to infer that the 3nm process should have been used at least for the Meteor Lake-U GPU tile, the dimensions of which are simply so small that if (this small tile) were manufactured with a 5nm process, its graphics performance will be reduced intergenerationally.

It was the rumors that Intel would eventually not produce tiles on TSMC’s 3nm process that gave rise to hypotheses about the insufficient yield of the 3nm process.

But as the slide above shows, at least on the N3E process variant, TSMC achieves significantly higher yields than it achieved with the 5nm process at the same preparatory stage.


Mediatek has not yet made an official statement, but according to various rumors, it seems that it has been developing the Dimensity 10000 3nm SoC since last year, which is supposed to become a competitor to the Qualcomm Snapdragon 8 Gen 2. However, no further information is available yet.


At Nvidia, the 3nm process will almost certainly be used for the next generation Blackwell, the successor to this year’s novelties. Rumor has it Blackwell will continue to be a monolith, which theoretically could complicate Nvidia’s position against RDNA 4 (which will probably also be produced by the 3nm process, so Nvidia will lose even the half-generation lead it gained this year by using 4nm lines compared to 5nm used by AMD), however, it is too early for catastrophic scenarios.

So far, the sources cannot even agree unanimously on whether it is Blackwell the successor of the game Lovelace or computing Hopper and the mere fact that there are rumors of a monolithic solution does not necessarily mean that it actually is. For example, Nvidia has in the past downplayed the benefits of unified shaders so heavily that it convinced practically everyone that the G80 GPU would not support them, and the result was a big surprise when the opposite was announced at release.

More information is not available on Broadcom and Qualcomm’s 3nm products. According to some sources, the latter is supposed to be ready to use Samsung’s 3nm lines if it is not satisfied with the situation at TSMC, but for now this seems more like a lever to negotiate more favorable terms with TSMC.

Finally, a journalistic gem. News of TSMC’s seven clients came from a Chinese-speaking environment. It’s no big secret that the Western world receives such news through Google Translate. That there is a need to edit machine translations, also not. But it doesn’t always happen, so we could read on the Guru 3D website, for example, that the 3nm process will be used by companies such as (I quote): “Super Micro, Huida, Qualcomm, MediaTek, Broadcom.”

I assume you don’t know Huidu. However, Guru 3D is lucky that it uses Google Translate and not DeepL, which translates the same text as: “Supermicro, Pfizer, Qualcomm, MediaTek, Broadcom.” That would be a discussion full of conspiracy theorists cheering that the chips in vaccines finally confirmed. Leaving aside that Supermicro or Super Micro is the Chinese name for AMD, then Huida or Pfizer is actually 輝達 = Nvidia.

Source: Diit.cz by diit.cz.

*The article has been translated based on the content of Diit.cz by diit.cz. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!

*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.

*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!