TSMC 3nm 'risk production' in 2021 paves the way to 2022 mass production
TSMC is getting closer to finalizing a 3-nanometer process that could be used on Apple's A-series or Apple Silicon chips that will be produced in 2022, with a report doubling down on rumors the chip foundry reportedly gearing up to start risk production of processors using the process in 2021.
![Chip wafers [via Taiwan Semiconductor Manufacturing Co., Ltd.]](https://photos5.appleinsider.com/gallery/36740-68650-tsmc-wafers-l.jpg)
Chip wafers [via Taiwan Semiconductor Manufacturing Co., Ltd.]
As a long-term supply chain member working with Apple to produce chips like the A12Z Bionic, TSMC has pushed to develop and use ever smaller scale fabrication processes. As it moves from the current 7-nanometer down to lower levels such as 5-nanometer, it is always looking to get down to even lower scales, including making chips using a 3-nanometer process.
Rumors from June put forward the idea that TSMC was on schedule to create chips using a 3nm process in 2022, with risk production anticipated to take place in 2021. In a report from MyDrivers citing a company official, the timing of risk production in 2021 will be officially announced by the chip foundry before the end of the year.
According to the company, the 3-nanometer process will provide quite a few benefits over earlier processes, even the relatively recent 5-nanometer. Compared to 5nm, 3nm will have a 15% greater transistor density, boast a performance increase of between 10% and 15%, and offer energy efficiency savings of between 20% and 25%.
Apple does tend to use the latest stable process TSMC develops and perfects in its chip designs. While it is almost certain that the 3nm fabrication process will be employed in an A-series chip destined for iPhone and iPad use, it remains to be seen whether it will also be used for another Apple-designed chip.
It is currently believed TSMC will be tapped to produce Apple Silicon chips, which will be used in new models of Mac as Apple transitions away from Intel processors over the next two years. As Apple Silicon and the A-series chips are ARM-based, as well as Apple's relationship with the supply chain producer, it seems extremely likely that TSMC is producing the chips for Apple, though it is unknown what processes would be used for their creation.
Chip wafers [via Taiwan Semiconductor Manufacturing Co., Ltd.]
As a long-term supply chain member working with Apple to produce chips like the A12Z Bionic, TSMC has pushed to develop and use ever smaller scale fabrication processes. As it moves from the current 7-nanometer down to lower levels such as 5-nanometer, it is always looking to get down to even lower scales, including making chips using a 3-nanometer process.
Rumors from June put forward the idea that TSMC was on schedule to create chips using a 3nm process in 2022, with risk production anticipated to take place in 2021. In a report from MyDrivers citing a company official, the timing of risk production in 2021 will be officially announced by the chip foundry before the end of the year.
According to the company, the 3-nanometer process will provide quite a few benefits over earlier processes, even the relatively recent 5-nanometer. Compared to 5nm, 3nm will have a 15% greater transistor density, boast a performance increase of between 10% and 15%, and offer energy efficiency savings of between 20% and 25%.
Apple does tend to use the latest stable process TSMC develops and perfects in its chip designs. While it is almost certain that the 3nm fabrication process will be employed in an A-series chip destined for iPhone and iPad use, it remains to be seen whether it will also be used for another Apple-designed chip.
It is currently believed TSMC will be tapped to produce Apple Silicon chips, which will be used in new models of Mac as Apple transitions away from Intel processors over the next two years. As Apple Silicon and the A-series chips are ARM-based, as well as Apple's relationship with the supply chain producer, it seems extremely likely that TSMC is producing the chips for Apple, though it is unknown what processes would be used for their creation.
Comments
It seems that should be 2022 in the headline instead of 2020...
I believed that even when they got to 100nm and I take everything as magic since then...
i no longer believe there is a limit!
We are certainly approaching silicon limits.
Even with other than silicon being used, theoretical limits say we can get this small, but how quickly and at what defect rate can that be done?
There is also the design time to most effectively use all the (theoretically available) transistors, as it's not quite as simple as copy/paste another CPU core with identical connecting logic in between.
Finally, there is the cost of developing and building new manufacturing tech and facilities: the smaller the process node, the more expensive it gets, by quite a bit.
After all, suppose Windows does switch to ARM causing Intel to fall flat on its face. Who benefits? Qualcomm. (You folks weren't thinking that Apple is going to sell its ARM chips to the likes of Dell or Lenovo to make Windows PCs to compete with Apple ones and do so at generally lower prices did you?) Did anyone think of that? Qualcomm will become the primary manufacturer of both mobile devices that have 85% market share against AND and servers/PCs that have 93% market share against macOS. And Qualcomm - unlike Intel - HAS gotten into plenty of conflicts with Apple over IP and lawsuits over the years. Since Qualcomm buys chips in much bigger volume than Apple ever will - again see market share - what on earth will happen if Qualcomm decides to buy up as much of TSMC's capacity as possible in order to delay their #1 competitor's pipeline? (This isn't theory ... Qualcomm HAS done this very thing before AND MORE THAN ONCE.)
Third, Samsung entered 3nm risk testing this year and will produce the world's first 3nm chips next year. It will be their own Exynos chips, likely the ones used in the international version of the Galaxy Note 30. It will not be ready in time for the Galaxy S30, as their new 5nm chip will be used instead, as well as in international versions of the Galaxy Note 20. So Samsung is going to beat Apple getting to 3 nm by 1 year.
Qualcomm will introduce their own 5nm chips with integrated 5G modems next year, the 875, a few months after Apple launches their own 5nm A14. Rumor has it that they are going to charge $100 more than the 865 and people are not pleased. While previously Qualcomm's chips were made by TSMC, their 5nm chips will be made by Samsung. Who knows when Qualcomm will reach 3nm as - unlike Samsung - they likely don't have a design ready yet.
Finally Intel states that they will reach 3nm by 2025. AMD will reach 3nm - using TSMC's foundries - around 2023. However, Intel's transistor design is denser, so a 3nm AMD chip is roughly equivalent to a 5nm Intel chip, a 5 nm AMD chip equal to a 7nm Intel chip and so forth. Thanks to the work of Bill Keller, who got Intel past their road block, Intel will release their first 7nm chip later this year, and that is when all the people who are crowing "AMD has surpassed Intel!" will basically be silenced. And yes, Intel's 7nm chips will add performance and efficiency to their already considered "best available" desktop, workstation and server i7, i9 and Xeon chips. So, Intel is going to be releasing new chip generations on a smaller process every 2 years just like Apple and everyone else.
Bottom line: Intel isn't going anywhere. Even if Apple Silicon beats them - which I still sincerely doubt but I will concede that point to those who feel otherwise - the performance and efficiency improvements are going to make the gap between Macs and Windows machines more than close enough for Lenovo, Dell, HP, Acer and Asus to continue to use them rather than try to find a better performing ARM replacement than Qualcomm (which currently does not exist and there are none on the horizon) and in the process of being stuck with Qualcomm chips that already are no better than half as powerful as Apple Silicon but will have to run most applications in emulation.
Now what WILL happen is a shift from Intel to ARM for CHROMEBOOKS. The only reason why it happened yet is Oracle's copyright lawsuit against Google. That FINALLY ends in October. Shortly after Google will certify ChromeOS Linux for Qualcomm and Exynos - and likely take it out of beta also - which means we will see Exynos-based Chromebooks from Samsung and a raft of Qualcomm-based Chromebooks from Lenovo, HP, Acer, Asus and possibly even one from LG (who doesn't make many Chromebooks) though likely not Dell (who is an x86-64 loyalist and also doesn't make many Chromebooks) in 2021. But even there, the only reason why this will happen is because ChromeOS on ARM performs similarly to ChromeOS on Intel and there are no app compatibility issues (if anything Android apps run better on the ARM-based Chromebooks than the x86-64 ones, and most of the main Linux applications were ported to ARM ages ago, which is why Apple is encouraging Windows bootcamp and virtualization users to switch to Linux instead for Apple Silicon). But because ChromeOS is more of a tablet OS akin to iPadOS than a desktop/workstation/server OS akin to macOS, Windows and Ubuntu, the Chromebooks will only cost Intel sales of devices in the Celeron, i3 and i5 range. It won't affect Intel's i7, i9 or Xeon business at all.
Intel missed the 7nm mark and still is marketing chips with security flaws. 7nm won't hit the market till sometime in 2021. So is Apple to wait?
https://wccftech.com/first-apple-silicon-featuring-12-cores/
The A13: 6 cores. The A14 is widely rumored to have 8. The MacBook chips won't be mobile chips like the A12/A13/A14. They are going to be desktop and server class chips with 12 cores.
We don't know the core breakdown ... but I think that having 8 efficiency cores would be ridiculous for a device that is primarily going to be plugged in (not a smartphone or tablet that is going to rely primarily on a charged battery). I think that it will be 6 performance/6 efficiency at minimum. Possibly even - my own personal hope - 8 performance and 4 efficiency, although that configuration would not be possible in a fanless device. (You COULD use the A14 in a fanless MacBook Air however).
OK, I was wrong. How wrong? Who knows. Ampere ARM server chips have 64-128 cores. But Apple is just getting started. Who knows how many cores their 3nm chips are going to have in 2022.
That being said ... the last time Qualcomm - and MediaTek - went past 8 cores (5 years ago) things went badly and they haven't tried since. (Samsung hasn't even tried to surpass 8 cores.) So Lenovo, Dell and the rest are going to be stuck with Intel and AMD no matter how badly Apple is going to boatrace them.
And anyone who buys an Intel-based Mac instead of a 12 core Apple Silicon Mac is out of their mind. If they want to run with clearly inferior tech they should just go ahead and buy a Lenovo or Dell Windows machine. Sorry but facts are facts.
I still hope they will make a new Mac Pro that allows plugging in multiple Apple Silicon and/or Intel cores on discrete cards. Let me have 256 threads of ARM64 and appropriate GPU wave frontage (~128,000 CU) and I don’t need to think about Intel ever again.