GlobalFoundries adds a 12-nm node to its FD-SOI roadmap

If you think of GlobalFoundries these days, you probably think about the company's 14-nm FinFET process and the AMD Polaris graphics cards fabricated on it. 14-nm FinFET is just one of several processes that GlobalFoundries offers, though. As just one example, the company also makes 22-nm planar transistors on a fully-depleted silicon-on-insulator technology—FD-SOI, or 22FDX in GloFo parlance.

The company claims 22FDX is well-suited for devices that need to balance high performance and low power consumption with low cost, especially products that need to combine radio-frequency, analog, memory, and logic devices onto the same chip. 22FDX has apparently been successful enough that GloFo is adding a new node to its FD-SOI roadmap today: 12-nm FD-SOI, or 12FDX.

GlobalFoundries says 12FDX can deliver the performance of 10-nm FinFET transistors with better power consumption characteristics and a lower cost than 16-nm FinFET devices. It also emphasizes that 12FDX is a full node shrink that delivers a "15% performance boost over today's FinFET technologies" and "as much as 50 percent lower power consumption." The company is preparing to produce 12-nm FD-SOI transistors at its Fab 1 facility in Dresden, Germany, and it expects that customers will begin taping out products on the new process node in the first half of 2019.

Comments closed
    • Jigar
    • 3 years ago

    When they say 15% performance boost over today’s FinFET transistors are they comparing it with 22nm tech or 14nm tech ?

      • NeelyCam
      • 3 years ago

      And are they comparing to their FinFETs or someone else’s?

    • ronch
    • 3 years ago

    If you think of GlobalFoundries these days, you probably think about the company’s 14-nm FinFET process that they just licensed from Samsung because after they were spun off they totally became clueless about how to come up with new process nodes that have good yields and customers are glad they chose over competing processes from competing foundries.

    FTFY.

    /just being honest about how I regard them

    • just brew it!
    • 3 years ago

    Interesting. I’d assumed that SOI was just kind of fading away. Didn’t realize that they were taking it down to bleeding edge process sizes.

      • ronch
      • 3 years ago

      Well, GF must be desperate to get a new fab process out.

    • flip-mode
    • 3 years ago

    Given GloFo’s history of execution, they may as well add warp drive to their road map.

      • shank15217
      • 3 years ago

      Warp .0000000001 engage!

        • Chrispy_
        • 3 years ago

        she cannae take much more o’ this cap’n!

        </James Doohan>

    • Mat3
    • 3 years ago

    My understanding is that the 22FDX was a 14nm FD-SOI front end with a 28nm back end. Does this one have a shrink down to 20nm for the back end?

      • ronch
      • 3 years ago

      (14 + 28) / 2 = 21.

      So yeah, roughly 22nm.

      /s

    • codedivine
    • 3 years ago

    Wondering which customers are using the 22nm FD-SOI. Anyone knows?

      • crystall
      • 3 years ago

      Among those who designed the process STMicroelectronics and IBM first of all; then Freescale/NXP, Panasonic, Renesas and a number of smaller players as well as. The 28nm version proved very popular for low-to-medium volume ASICs which is unusual for a quasi-leading edge process. Cisco produced its switch ASICs using it for example.

    • DragonDaddyBear
    • 3 years ago

    Process technology would be a great article to go into depth with. There is a lot of information in the internet but it’s not distilled or written to TR’s standard.

      • Duct Tape Dude
      • 3 years ago

      Definitely, especially for the “10/12/14nm” crop. It’d be nice to have an updated article on the differences among them, what specs are mostly marketing, and what is meaningful with regards to consumer electronics.

    • bittermann
    • 3 years ago

    [b<]TAPING OUT[/b<] products on the new node in the [b<]FIRST HALF OF 2019[/b<] ladies and gentlemen! So basically they have the new name picked out so far, which they just changed two characters.

      • Jeff Kampman
      • 3 years ago

      This is really not so unusual. Semiconductor products take years to develop and companies who want to take advantage of this new process likely need that much time to build whatever device it is they want to fabricate.

        • faramir
        • 3 years ago

        Yes but what is unusual is how they brag (?) about whopping “15% performance boost over today’s FinFET technologies” and “as much as 50 percent lower power consumption.”

        Taping out in 2019, coming to market in 2020 at the earliest.

    • WhatMeWorry
    • 3 years ago

    “The company claims 22FDX is well-suited for devices that need to balance high performance and low power consumption with low cost”

    Can’t that be said for any device?

      • chuckula
      • 3 years ago

      No, there are plenty of products where you can sacrifice one of those goals for one of the other goals.

        • DeadOfKnight
        • 3 years ago

        Pretty sure this is the case for almost anything. I guess all this is saying is that Moore’s law, or at least a diminished version of Moore’s law, is alive for one more process node. It’s a sign of progress, but not a sign of hope. We likely won’t see much further progress like this until they get EUV lithography and 450mm wafers sorted out, at least for the price. All this talk of alternative materials like carbon tubes and nanowires reeks of lower perf/W/$. Hopefully stronger competition can ease the pain (I’m looking at you, AMD).

      • Neutronbeam
      • 3 years ago

      Not if they’re made by Apple.

    • blastdoor
    • 3 years ago

    So is this something they bought from IBM?

      • chuckula
      • 3 years ago

      “Bought” implies that they paid something for the technology.
      IBM paid GloFo (not the other way around) to take its fabs and technology off of its hands.

        • jihadjoe
        • 3 years ago

        The guys at GloFo must be the kings at deal making. They get paid by AMD to not make their chips, and now they get paid by IBM to take their [s<]fans[/s<] fabs and technology... Edit: autocorrect strikes again!

          • the
          • 3 years ago

          Sounds great until you realize the costs in maintaining a fab.

          • ronch
          • 3 years ago

          Oops, seems you already beat me to it.

        • ronch
        • 3 years ago

        I talked to IBM too. Told them I’d be happy to grab their fabs if they pay me.

        They never called back. I tried my best to sound nice and professional. Did I say something that turned them off??

        Anyway, it’s totally bonkerz how GF seems to be good at striking deals where they actually GET PAID NOT TO MAKE CHIPS or GET PAID TO ACQUIRE FABS. Seems like a totally legit way to MAKE MONEY. /s

    • tipoo
    • 3 years ago

    AMD after paying a few hundred million to poke a small hole in the WSA: DOH!

      • NTMBK
      • 3 years ago

      SOI has been promising the world and failing to deliver for years now. AMD dropped it after 32nm for a reason!

    • tipoo
    • 3 years ago

    Is the fab actually sized differently, or just a naming marketing change? I.e TSMCs 20nm and 16nm didn’t actually change size, just added finfet.

    [url<]http://www.extremetech.com/wp-content/uploads/2014/12/Cell-SizeComparison.png[/url<] Like TFS vegeta said about power levels, Fab naming is bullcrap.

      • Jeff Kampman
      • 3 years ago

      They claim it’s a real shrink. I’m sure there’s a certain element of marketing to it, but I can’t exactly do much but take them at their word.

      • the
      • 3 years ago

      This is a full node change from their 22 nm FD-SOI process, so 22 nm SOI -> 12 nm SOI.

      This will likely be used for the POWER9+ refresh as IBM migrated to 14 nm FinFET for the vanilla POWER9. eDRAM reportedly works better on SOI than bulk processes..

        • tipoo
        • 3 years ago

        Gochya. So “14nm” (and 16 which is 20 for TSMC) was the stopgap, this is the real deal, sorta thing.

          • the
          • 3 years ago

          14 nm was bulk FinFET, something different. They’re both full node jumps from their predecessors: 20 nm planar bulk -> 14 nm FinFET and 22 nm SOI -> 12 nm SOI.

          [url=http://www.eetimes.com/author.asp?section_id=36&doc_id=1329887<]Supporters of SOI are claiming some big benefits as process nodes continue to shrink.[/url<] So while SOI is used sparingly today due to costs or specific use-cases (IBM's massive eDRAM on POWER chips is due in part to the properties of SOI), it could become increasingly popular for general usage as bulk process costs continue to increase. Offering a SOI process alongside a bulk FinFET option is a good strategy. Conceptually, [url=http://www.eetimes.com/document.asp?doc_id=1327035<]SOI and FinFET can be used on the same process too.[/url<]

Pin It on Pinterest

Share This