Apple and Intel Reportedly First in Line for TMSC’s 3nm Chipset Process

Posted on July 2, 2021 by Paul Thurrott in Apple, Hardware, Mobile with 28 Comments

Intel Starts Shipping "Kaby Lake" Chipsets

A new report claims that Apple and Intel will be the first to use TMSC’s upcoming 3-nanometer chipset production technology next year.

Citing multiple sources, Nikkei Asia reports that both firms are already testing their chip designs against TMSC’s 3-nm manufacturing process technology, and that both expect to begin producing chips in the second half of 2022.

As the publication notes, the most advanced chipset manufacturing process today, which is from TMSC, can produce 5-nm chips. This process is used to create the Apple A-series chips in the iPhone 12 family. But most chipset processes result in larger chipsets—or, more accurately, chipsets with more space between their transistors. In the PC space, 10-nm and even 14-nm are still very common.

The advantages of moving to smaller manufacturing processes are clear: The smaller the size between transistors, the more energy-efficient and powerful the resulting chipsets. TMSC says that its 3-nm processor technology will result in chips that are 25 to 30 percent more efficient and offer 10 to 15 percent more performance.

Apple intends to use the 3-nm chipsets in its iPad Pro product line first, while Intel is aiming at a bigger market: Notebook PCs and other portable computers. “Currently the chip volume planned for Intel is more than that for Apple’s iPad using the 3-nanometer process,” one Nikkei Asia source told the publication.

And the stakes are arguably higher for Intel, which has lost market share to AMD while struggling to improve its own manufacturing processes. Intel once controlled over 80 percent of the PC chipset market, but its share has fallen to just north of 70 percent. AMD, meanwhile, has surged to 20 percent market share, from 11 percent in 2019.

And from a technical perspective, Intel, which has historically built its own chips, still hasn’t made the transition to 7-nm designs internally, and it isn’t expected to do so until 2023. It plans to improve its own processes while it takes advantage of third-party fabricators like TMSC in the meantime.

Tagged with

Join the discussion!

BECOME A THURROTT MEMBER:

Don't have a login but want to join the conversation? Become a Thurrott Premium or Basic User to participate

Register
Comments (28)

28 responses to “Apple and Intel Reportedly First in Line for TMSC’s 3nm Chipset Process”

  1. crunchyfrog

    What happens after they get much lower than 3 nanometers? Can they really go much smaller?

    • Paul Thurrott

      Yeah, was wondering the same. Do they just disappear?

      • scottmichaud

        Yeah we're getting down to the "countable atoms" level... although (as other comments mention) the nm isn't exactly literal.


        I am wondering what will happen when we completely max-out radio signals over silicon transistors. There's been the occasional research paper talking about things like optical integrated circuits. Much more frequency to work with (although the wave is a photon at that point). Not sure how much work will be required to actually create switchable logic gates that are faster than electrical circuits, or even how many people are actually working on it (versus continuing to polish our current techniques until they disappear).


        I'd be curious if anyone has a genuine prediction (that's not just regurgitated quantum hype -- although, who knows).

      • rosyna

        There a minimum limit of about 1nm before quantum tunneling allows electrons to jump the border between transistors, making the chip unusable.

    • bkkcanuck

      There is a limit, but the next advancement of significance is the replacement of the EUV (extreme ultraviolet lithography) with x-ray based lithography (so more potential for shrinkage), 3D layering (assuming you can make your bottom layer be a frying pan). There is ongoing research on new materials (silicon is cheap, easier and mature -- but not efficient in comparison) such as GaN (Gallium Nitride) and SiC (Silicon Carbide). I believe GaN can switch at an order of magnitude faster (1000 times faster or efficiency - not sure exact numbers), but the cost and density of the transistors (and number of defects in a wafer) is still much higher. If they were able to perfect the technology the efficiency gain would lead to a lower power loss (much cooler chips)... which could mean more density by having a multilayered chip etc without it turning into molten lava. I think we will likely hit a wall, and slow down considerably at some point ... before the new materials become a better option... but there is lots of research going on.

    • bkkcanuck

      I thought that 2-3nm would be hitting the wall, but then I heard an interview of Jim Kellar (who 'might' be more knowledgeable than me on the issue ,,, )... and I gather from him he expects there is more headroom for progress (of course if they had a materials breakthrough that could give a big big boost).


      www.youtube.com/watch?v=AFVDZeg4RVY

    • ben lee

      They've already moved beyond this. The nano meter nomenclature no longer actually refers to the distance between transistors. Dr Cutress (he writes for Anandtech) has a good explanation of how it actually works at the start of this video on IBM's 2 nano meter process.

    • wright_is

      Researchers released details of a 2 atom thick switch today... So there is still some room left.

  2. bkkcanuck

    It is my understanding that Intel is using TSMC for TWO chips (leading edge process has a much more restricted ability to churn out new chips in volume). One a server chip and one a laptop chip... I am guessing both of those are going to be for more niche products, usually the early chips released are the smaller side (though they could be used in a chiplet design). Most Intel chips in the mainstream laptop market are of a larger size chip. So my guess is that both of these chips will be more niche in comparison to the current Intel line (maybe an ultralight laptop - lower power; and the server may be a newer design)... just a guess though.

  3. codymesh

    how the hell did Intel get in front of AMD in the line? AMD has far larger customer for TSMC than Intel, unless Intel put down a gigantic (and I mean gargantuan) order, there's no way TSMC gives Intel first dibs before AMD

    • bkkcanuck

      I don't remember AMD ever being in the leading edge (i.e. they have been unwilling to pay the substantial premium to do so). The leading edge volume production is a fraction of that that is available after all the issues in the leading edge have been worked out (2nd wave of orders on a node).

  4. waethorn

    Is this why China has been so keen on declaring a takeover of the country?

    • Truffles

      No, it wouldn't be a factor at all as Taiwan is a manufacturer of leading edge chips using machines it buys elsewhere. In other words, it's the development of the machines that is the strategically important technology, not the factories where those machines are actually used to make stuff. I picked up that distinction after listening to an interesting podcast interview with a fab guy who I think worked in Taiwan.


      Interestingly, the Taiwanese guy said that the current leading edge machines are made by a European company, and the devices are so delicate that their operation is more art than science and they have frequent breakdowns due to weird factors like micro-variations in local gravity. As a consequence, the companies that make the machines don't just sell the machines but also supply their own engineers to fix them.


      The way the market works is that fabs like TSMC make a billion dollar bet on the technology being proposed by the companies that make the machines, and sometimes the bet pays-off really well, and sometimes the bet turns bad. He said that Intel made a series of bad tech bets, while TSMC made some good bets.


      He said that the Chinese government hadn't taken any notice of chip manufacturing as a strategic industry until a Chinese company contracted to buy one of the European machines and the US quickly slapped on sanctions preventing any company or person from anywhere in the world from aiding chip fabrication in China. Obviously the Chinese government took notice and being China decided to build a chip manufacturing technology stack from the ground up that will be entirely domestic rather than using vulnerable western supply chains.


      To that end, China set up multiple large scale development projects employing the local fab engineers who, due to the sanctions, were now twiddling their thumbs. It's also offering enormous salaries to attract foreign talent who will essentially be making a decision to live in China for their entire career for fear of being wanted US criminals if they leave. The fab guy predicted that China will catch up to the current (ie the podcast was in 2020) state of the art in a few years (say, 2025) and will be on an equal footing by about 2030. The European companies are pretty much screwed because they're not allowed to sell to China (or even Russia) and the US government is now subsidising US manufacture of the machines and requiring US-based fabs to use those products. That leaves Europe with limited customers - mainly in Taiwan.


      So, in short, invading Taiwan doesn't help China get in to advanced chip manufacturing technology because it's the development of the machines that are the strategic technology, not the factories themselves.

  5. bkkcanuck

    A better comparison than 7nm, 5nm, 3nm would be transistor density.


    The machines used to manufacture and EUV Lithography are made by a third party.


    Peak Quoted Transistor Densities (MTr/mm2):

    TSMC 10nm - 52.51

    TSMC 7nm - 91.20

    TSMC 5nm - 171.21

    TSMC 3nm - 292.21* Estimated


    Intel 14nm - 44.67

    Intel 10nm - 100.76

    Intel 7nm - 237.18* Estimated


    IBM 2nm - 333.33 (will be licensed)

    • bluvg

      Thank you for correcting this continued misconception between different manufacturers' process technologies. Many look at the manufacturer-provided "x nm" process number and assume Intel is FAR behind. They are behind, but the numbers are not equivalent.


      Furthermore, it's also often assumed that a single process technology is used for entire chip, which is not at all the case.

    • Oreo

      Transistor density is indeed an important metric, but not the only one. Two more that come to mind are yields and operating frequencies. Intel’s 10 nm process hasn’t been financially viable for the longest time, which is why it hasn’t moved all of its products over. And Intel had to reduce frequencies when it went from 14+^n nm to 10 nm.

      • bkkcanuck

        Metrics of yield come to profitability -- irrelevant to the end user. If Intel wants to subsidize to remain in the market (i.e. not losing too much), that is a business decision. Operating frequencies for the most part are irrelevant to the vast majority of users, it is more about how much performance the chip can provide for the tasks and at a certain point become a case of diminishing returns... as the user hits the peak performance less and less. Enterprise don't pick the chips with the highest performance, it is more about price / performance and overall performance (more cores) -- and the density and operating cost. For end users, the most important market is mobile devices (6 billion ARM ISA chips manufactured per year, and the laptop market outstrips the desktop market by more than 2 to 1 margin.

    • neilorourke

      IBM and a 2nm process? <cough>josephson junctions, anyone?<cough>

  6. arrowd

    Isn't the acronym TSMC?

  7. angusmatheson

    Why wouodnTMSC do this with Intel first? Intel is actively completing with them with its own fans and using their process is a stop gap. AMD is all in using TMSC. My only thought is Intel spent crazy money on this. I have wondered how much of AMD amazing comeback and Apple brilliant chip design is simply using TMSC more advanced manufacturing process. If Intel is in on it they all may end up the same.

    • bkkcanuck

      TSMC is going to offer this to customers that are committed to long term manufacturing with them that are willing to pay the premium to be an early adopter. It is a major investment to build out capacity and those that are willing to commit and fund part of that will be given priority.

  8. jchampeau

    It's TSMC, which stands for Taiwan Semiconductor Manufacturing Company.

Leave a Reply