GlobalFoundries fires up its 7-nm Leading Performance forges

A few days ago Globalfoundries announced that its "7LP" 7-nm FinFET process is ready for prime-time, meaning that customer companies can start building actual products on the process. That's a pretty quick turnaround, given that the company announced its plans to skip the 10-nm node and move directly to 7 nm in September of last year. GloFo says that compared to its 14-nm products, its 7LP process can offer a 40% performance improvement in the same power envelope or a whopping 60% power reduction within the same frequency range.

Normally we wouldn't be as excited about an "LP" process, but in this case the initials actually stand for "Leading Performance." Unlike TSMC, who usually brings out low-power-optimized version of its new processes first, GlobalFoundries' new tech is intended for high-performance products. In fact, the company specifically mentions that its 7-nm chips could operate at 5 GHz, and suggests "server, data center, [and] ASICs" as potential targets for the process.

We are obviously not privy to the fine details of GlobalFoundries' business, but it doesn't take great clairvoyance to recognize that AMD is probably one of the foundry's largest customers. Suddenly that year-old roadmap that we saw back in May—which listed 7nm CPUs launching in the latter part of 2018—doesn't look so far-fetched anymore. GlobalFoundries says that it expects the first 7LP products to launch about a year from now, and that volume production should pick up in the second half of next year.

Comments closed
    • Gadoran
    • 2 years ago

    Too bad GloFo and Samsung have lost Qualcomm, Mediatek and Apple on 7nm.
    So this upcoming process is not good as expected, marketing is nice but real customers are better IMO.
    The worst thing is the Samsung huge loss of Qualcomm, this is why Samsung is thinking about a spin off of all foundry activities.

      • freebird
      • 2 years ago

      Their loss is AMD’s gain…I’m sure AMD will love being able to get as much 7nm production as they want from GloFo.

    • synthtel2
    • 2 years ago

    A real HP process from someone other than Intel, at the same time as Intel’s new stuff is increasingly LP? This should be interesting. *grabs popcorn*

    • Platedslicer
    • 2 years ago

    Is there any reason to believe that GF is going to deliver on this one when it underdelivered/was late with every major promise in recent memory?

      • Gadoran
      • 2 years ago

      Any reason

      • freebird
      • 2 years ago

      Yes.

      Global Foundries bought IBM chip business back in 2014 and now they (IBM staff) are fully integrated/merged into Global Foundries chip production.

      “According to IBM, compared to existing leading-edge 10nm chips, nanosheet-based 5nm technologies can deliver as much as 40 percent performance enhancements at fixed power, or 75 percent power savings at matched performance.”
      [url<]http://www.pund-it.com/blog/ibm-research-alliance-details-new-nanosheet-transistors-for-5nm-chips/[/url<] And Global Foundries continues to extend collaboration and investments in technology for future nodes with the IBM Research Alliance which I expect to be extended. Along with: “GLOBALFOUNDRIES is committed to an aggressive research roadmap that continually pushes the limits of semiconductor technology. With the recent acquisition of IBM Microelectronics, GLOBALFOUNDRIES has gained direct access to IBM’s continued investment in world-class semiconductor research and has significantly enhanced its ability to develop leading-edge technologies,” said Dr. Gary Patton, CTO and Senior Vice President of R&D at GLOBALFOUNDRIES. “Together with SUNY Poly, the new center will improve our capabilities and position us to advance our process geometries at 7nm and beyond.” [url<]http://electroiq.com/blog/2016/02/suny-poly-and-globalfoundries-announce-new-500m-rd-program-in-albany/[/url<]

        • NoOne ButMe
        • 2 years ago

        They did not “buy” all of IBM’s process R&D, or get exclusive access to this.

        I think either 7nm or 5nm Samsung incorporates some IBM stuff.

          • freebird
          • 2 years ago

          I didn’t say they bought ALL of IBM’s process R&D, DID I? I said they bought their chip business in 2014.
          [url<]https://www.extremetech.com/computing/192430-ibm-dumps-chip-unit-pays-globalfoundries-1-5-billion-to-take-the-business-off-its-hands[/url<] And AS I DID state, Global Foundries is PART of the IBM RESEARCH Alliance, in which IBM R&D develops the 7nm & 5nm processes and then it is up to Samsung & Global Foundries to determine when and how the will bring is online in their manufacturing plants. But yes IBM is developing and doing the research. Anyone that reads any articles about the IBM Research Alliance should already KNOW that... But I guess the uniformed are entitled to make stupid posts in comment threads.

    • Bumper
    • 2 years ago

    A 5ghz Ryzen 2 will be a nice upgrade for any Ryzen 1 buyers.

      • Spunjji
      • 2 years ago

      I find myself even more excited about the prospects for an APU. Especially if they finally stick a nice EDRAM cache on it…

    • ronch
    • 2 years ago

    7LP kinda looks like 7UP if you’re as blind as a bat. 7nm Ultra Performance. Why lead when you can be ultra?

      • CuttinHobo
      • 2 years ago

      7UP yours!!

      [url<]https://youtu.be/WsMFZBDIcFs[/url<]

        • derFunkenstein
        • 2 years ago

        I have that shirt somewhere still. I got it when that commercial was new.

    • ronch
    • 2 years ago

    Can’t wait to see what Zen 2 or 3 at ‘7nm’ can do.

      • ImSpartacus
      • 2 years ago

      Yeah, we often take Intel’s clock advantage for granted, but we forget that Intel has a healthy process advantage.

    • blastdoor
    • 2 years ago

    Has anyone compared the hypothetical 7nm process from GF to Intel’s 10 nm?

      • tsk
      • 2 years ago

      Since samsung and GF share technologies and samsungs 10nm is similar to intels 14nm, I assume GFs 7nm will be similar to intels 10nm.

        • Beahmont
        • 2 years ago

        Honestly? If that statement by GF is correct and not BS, then, to the best of my knowledge, GF will be ahead of Intel as of the last time they talked about their processes. Way back when Intel was pitching their [url=https://techreport.com/review/31660/intel-defends-its-process-technology-leadership-at-14nm-and-10nm<] way to measure process tech[/url<] because the nm system is pretty much broken at this point, they talked about how their 14nm+++ tech was better than their 10nm tech at that moment. From what I've seen and given some private conversations with someone at Intel, the problem wasn't yields at 10nm, but they were, and may still be, having problems scaling up the power-clock scale. They were doing better than their target for low power chips, but something was absolutely Not-Working-as-Inteneded(tm) when they tried to test build the higher clocked chips. The rumor was that until that was figured out they just were not going to put 10nm into large scale production.

          • tsk
          • 2 years ago

          I guess it’s really hard to tell, unless everyone used the metric recently suggested by intel.
          [url<]https://techreport.com/review/31660/intel-defends-its-process-technology-leadership-at-14nm-and-10nm[/url<] Intel is supposed to start manufacturing ARM on their 10nm node, so maybe then we can get some pretty direct comparisons of A75 chips on different nodes.

          • DavidC1
          • 2 years ago

          Beahmont: I rarely upvote people because I don’t like favoritism in any way but I did for your post.

          About the second paragraph in your post.

          Yea that makes so much sense. The absolute highest clock doesn’t reach 9GHz. That’s with a very highly pipelined architecture like Bulldozer or Prescott based Celeron. Funny the Celeron part, its like the regular Prescott is too complex or something.

          The air overclocks are maxed out at 5.5GHz or so. You are talking about just CPU-Z screenshots on a golden sample. There’s a barrier that just can’t be breached. Across microarchitecture, across process, across companies with vastly different R&Ds.

          What happens if 5GHz is really the limit?
          -Near future: 5GHz overclocks
          -Even further: 5GHz Turbo, 5GHz overclock
          -Even further down: 5GHz Base, 5GHz overclock

          Sub-9GHz on the best case golden sample tells me how even engineers with all those degrees and intellect can’t tell much about the future. Intel had 6GHz Prescott in plans. Just a few years must have passed when the dream totally crashed.

          • NoOne ButMe
          • 2 years ago

          the non-yielding issues are the same that 14nm had.
          Probably just magnified by the harder shrink.

          reformated easier read.

          • ImSpartacus
          • 2 years ago

          I don’t know if that’s better or worse than the rumors that 10nm yields simply sucked too much to use it on a big chip.

        • chuckula
        • 2 years ago

        Actually while GF licensed from Samsung for 14nm, that deal does not appear to apply to 7nm, where GloFo is more independent.

        You can read more about roadmaps and some of the differences between GloFo and Samsung’s plans here:
        [url<]http://www.anandtech.com/show/10704/globalfoundries-updates-roadmap-7-nm-in-2h-2018[/url<]

          • Gadoran
          • 2 years ago

          Obviously there is something in common between two processes.
          Anyway there isn’t any major customer at the orizon and TSMC lead is pretty clear.

        • NoOne ButMe
        • 2 years ago

        That was one node shared, technological knowledge may be shared, but after 14nm they each have different teams building the nodes.

    • Demetri
    • 2 years ago

    “In fact, the company specifically mentions that its 7-nm chips could operate at 5 GHz”

    Pretty nice to hear, especially considering they only claimed “3 GHz Operation” for 14LPP and Zen was able to hit 4.

    [url<]https://i.redd.it/oajvimprrw3z.png[/url<]

      • ImSpartacus
      • 2 years ago

      They said “above” 3 GHz.

      That’s not analogous to “up to” 5 GHz.

        • Spunjji
        • 2 years ago

        Sure, but it smells like an increase however you slice it. Bug-fixed Zen at 4.5Ghz+ without abysmal power characteristics should absolutely be competitive in 2018. Not revolutionary, but enough to keep things on track and close that gaming gap.

    • orik
    • 2 years ago

    I thought AMD’s roadmap to get Zen 2 on 7nm meant it wasn’t coming for 2+ years, but color me surprised.

    Hopefully there are good yields.

      • freebird
      • 2 years ago

      I’ve read too many conflicting articles/timelines on Zen2, Zen3 or Zen+… Supposedly, a refresh was scheduled for 2018, but only AMD knows if they were designing their 1st enhancements to Zen (whether Zen+ or Zen2) on 14nm+(enhanced 14nm) or 7nm.

      I lean towards an enhanced 14nm+ with some minor tweaks to get it out by mid-2018. Then maybe a full Zen 2 (significant arch improvements, cache/memory latency) on 7nm for early 2019 allowing to get good yields.

      Wonder if that means Navi (Professional version) could be here for X-mas 2018 or will they push it back so Vega 20 can get some revenue 1st… still all in all, good for AMD and good for the consumer…

        • NoOne ButMe
        • 2 years ago

        AMD claims tape out of Zen2 7nm products (so somewhere between 14nm and 10nm compared to Intel’s nodes) in 2H2017.

        If this is on GloFo, we may see a limited release in 1H2018, followed by a true volume ramp at 2H2018.

      • Waco
      • 2 years ago

      That’s a quick turnaround from 14nm to 7nm…it would be impressive if AMD was able to get a chip out in the next year running at much higher clocks (for consumers) or much lower power at similar clocks (servers).

        • smilingcrow
        • 2 years ago

        The claimed power reductions make it particularly interesting for laptop chips.

Pin It on Pinterest

Share This