Samsung takes EUV lithography into production for its 7-nm LPP node

Samsung announced yesterday evening that its 7-nm Low Power Plus (LPP) process technology is ready to roll. The company says it has completed development of that next-generation process technology and is now turning out production wafers with it. That milestone is remarkable because 7-nm LPP is the only process technology in the world so far to use extreme ultraviolet lithography, or EUV, tools to inscribe features on silicon. Samsung's adoption of the technology is especially ambitious because the company is using it for critical layers in the stack of elements that go into a modern chip, not just for low-risk features like contacts and vias (as GlobalFoundries had planned to do before it abandoned 7-nm development entirely).

Samsung's S3 EUV production line in Hwaseong, South Korea

Even with industry-wide challenges such as source power and its consequent constraints on wafer throughput, as well as machine uptime and yields, Samsung is rosy on the prospects of EUV technology in its own fabs. The company says using EUV tools allows it to reduce the number of masks needed to produce features at critical levels of the chip, thereby improving productivity—one of the largest expected benefits to foundries of the move to this exotic lithography method.

A cutaway of an ASML EUV scanner similar to the ones Samsung is employing for 7-nm LPP

As one example of its own innovations in dealing with the challenges of EUV, the company says it has developed its own defect-detection capabilities for EUV masks to prevent flaws from making their way to wafers—another difficulty that had bedeviled would-be adopters of this tech. The net result of using EUV scanners for its 7-nm LPP chips is that Samsung can claim a 40% increase in areal density over its 10-nm process with up to 20% higher performance or 50% lower power. To assist potential customers in implementing their designs on this process, Samsung is providing a wide range of standard IP and supporting design tools to partners like Ansys, ARM, Cadence, Mentor, Semco, Synopsys, and VeriSilicon.

Along with TSMC, whose 7-nm process is already producing customer silicon without the use of EUV, Samsung's transition to production-readiness for 7-nm LPP marks the second domino to fall in the race to next-generation process nodes. Only Intel's 10-nm process remains in the works at the leading edge, and the blue team hasn't promised production silicon from that process until late next year. Intel now finds itself firmly in the unenviable position of playing catch-up to pure-play foundries in silicon technology (at least on paper), and that is a remarkable turn of events for a company that used to claim it was a generation ahead of those foundries in process tech. We'll be eager to see what customers Samsung lands as it ramps 7-nm LPP.

Comments closed
    • spanky1off
    • 11 months ago

    I have no idea about this technology but just my claim to fame is I work in the 3 towers that are just visible in the top right of the top picture. These are also Samsung buildings where I happen to teach English to several department managers there. These are some of Koreas brightest minds, mostly PHDs grads from the SKY universities (Seoul National, Korea, Yonsei universities or Korea’s equivalent of the Ivy League). The huge factory buildings you see behind the construction area are where they make the ideas of these guys. Basically SSD, NAND flash or device solutions (integrating products with other companies like intel,nvidia etc) or something lol. I can share this because its not any kind of secret, though personally ive no idea exactly what, which is probably for the best. Great bunch of guys by the way. My favorite English classes by far.

      • tipoo
      • 11 months ago

      *waves at picture*

    • Shobai
    • 11 months ago

    Is that ‘50% lower power ‘ figure also ‘up to’?

      • Zizy
      • 11 months ago

      Like usual. Up to: A) 20% higher performance OR B) 50% lower power.
      So, iso-performance, you can reach 50% lower power. Iso-power, 20% higher performance. Or say 30% lower power, 10% higher performance.

      Only area is additive to the performance or power gains, but even there I would expect the very same chip simply shrunk down wouldn’t reach neither 20% higher performance nor 50% lower power.

    • DeadOfKnight
    • 11 months ago

    When are they getting those 450mm wafers? That was supposed to be the big thing for reducing prices.

      • Stonebender
      • 11 months ago

      Will only ever happen if another brand new factory is built. The cost to retrofit an existing Fab to use 450mm wafers is astronomical

        • freebird
        • 11 months ago

        Which comes first? The chicken or the egg; No factories or equipment to process 450nm wafers and nobody producing silicon wafers that large.

          • Sahrin
          • 11 months ago

          Tooling comes first; making ingots is cheap/easy by comparison.

            • JustAnEngineer
            • 11 months ago

            Funding comes first. If you can’t get the short-term bean counters to invest in the future, you’re dead in the water.

            • Sahrin
            • 11 months ago

            lol, no, demand comes first. You can’t build something unless you have someone to sell it to.

            • LASR
            • 11 months ago

            Lol, no. Are you stupid or just trolling?

            It’s an innovation driven space. Demand is zero until you productize something and market it well.

            > You can’t build something unless you have someone to sell it to.

            Sure you can. Nobody wanted an iPhone until it was actually shown to the world. Of course by that time, the money was there, the innovation happened, and the factories were cranking these out in volume all before any demand for it existed.

            Every tech business works the same way. Innovation drives demand. Not the other way around.

            • Wonders
            • 11 months ago

            Tooling is first.
            > Funding is first.
            >> Demand is first.
            >>> Innovation is first.
            Actually, I’m pretty sure PrincipalSkinner is first.

      • Sahrin
      • 11 months ago

      450 mm isn’t really an advantage for microprocessors, because most of the process steps are per die not per wafer. Doesn’t matter if the wage is 100 mm or 450 mm, the number of steps is the same per wafer.

      • Zizy
      • 11 months ago

      Never, most likely. It turns out 450mm doesn’t solve many problems, especially as the defects are pushing industry towards smaller dies. The expensive steps are the litho ones anyway, so you don’t save a lot by having bigger wafers, yet you need to change all the tools, which comes with all sorts of problems – the main being cost, but space is an issue as well. Plus during research phase, you would require less 450mm wafers in a box to carry them around, these wafers get heavy.

      It was thought solar cells would benefit from moving to 450mm wafers, but stuff there is moving away from mono Si wafers, plus the wafers there are moving from the high quality ingot stuff (or even the experimental FZ wafers) towards cheaper processes. Perf/$ is the most important parameter there.

    • Acidicheartburn
    • 11 months ago

    It boggles my mind how far Samsung has come so quickly. Growing up they used to just be one of those second rate electronics companies you wouldn’t normally want to choose over Sony, Toshiba, GE, Nokia, etc. Now they’re the giant of multiple industries, leading not only in volume of product but innovation. It’s crazy to me.

      • DeadOfKnight
      • 11 months ago

      Same thing with Hyundai. I guess South Korean companies just hadn’t expanded their reach until the turn of the century.

        • Acidicheartburn
        • 11 months ago

        Hyundai/Kia took the world by storm seemingly overnight. It’s pretty amazing. I remember the SK government deliberately weakened their currency in order to assist with this. I don’t understand the economics of it but suffice to say that it worked.

          • JustAnEngineer
          • 11 months ago

          [url<]https://en.wikipedia.org/wiki/Chaebol[/url<]

    • DPete27
    • 11 months ago

    I assume that since ASML makes the EUV scanner, they can sell the same product to Intel/ TSMC/ GloFo/ etc and all those companies can also produce 7nm?

      • Jeff Kampman
      • 11 months ago

      EUV is simply one cog in an outlandishly complex machine. If 7-nm chip fabrication was as easy as buying the EUV scanner, we’d have a much more competitive silicon fabrication landscape today.

      The fact that GlobalFoundries has exited the 7-nm race, that Intel is struggling to get a process of similar density working even without EUV, and that TSMC is able to produce its first generation of 7-nm silicon just fine without EUV speak to the diversity of approaches that are possible here and the fact that a vast range of supporting development is required above and beyond the lithography tools needed. Samsung is just aggressively biting the bullet that all leading-edge chip fabs are eventually going to have to bite if they want to keep advancing densities.

        • psuedonymous
        • 11 months ago

        On top of that: Intel are tackling the inevitable metal transition (Cu to Co/Ru) internally, and letting others take the hit on ‘Early Adopter’ EUV rollout to production. Let TSMC and Samsung spend [i<]their[/i<] dosh working out how to get an EUV compatible pellicle working.

      • Stonebender
      • 11 months ago

      Intel has had ASML EUV tools in their factories since at least 2011. There are two reasons they’re not being used yet: 10nm is still a focus, and EUV still doesn’t have the throughput that Intel wants.

    • Wirko
    • 11 months ago

    I’m feeling small next to all these mighty nanometres.

    • tsk
    • 11 months ago

    Intel went from 1st to last place in the bleeding edge race real fast.

      • DeadOfKnight
      • 11 months ago

      GloFo is definitely last, and we have no idea what the race really looks like beyond what they’re all telling their investors.

        • Goty
        • 11 months ago

        Is it last place if you’re not participating?

        • tsk
        • 11 months ago

        GloFo dropped out of that race earlier this year, it’s only TSMC, Samsung and Intel now.

        We know what that race looks like by products, TSMC is shipping 7nm chips, and we know the density of those chips.

    • WhatMeWorry
    • 11 months ago

    Look on my Works, ye Mighty, and despair!

    • Krogoth
    • 11 months ago

    Intel looks at its “superior” 10nm process : “MY PRECIOUS!”

      • chuckula
      • 11 months ago

      Samsung is making 16 core Zen 2 chips this year…
      CONFIRMED!

        • Krogoth
        • 11 months ago

        Intel: “NASTY LITTLE THIEVES! STEALING MY PRECIOUS!”

        • freebird
        • 11 months ago

        Intel’s 10gen & 11gen Core series produced at Samsung & TSMC …
        CONFIRMED!

        😛

      • freebird
      • 11 months ago

      -Intel gets mad that “others” are playing with their toys first!

      [url<]https://www.reuters.com/article/us-asml-holding-intel-stake/intel-reduces-stake-in-asml-to-below-3-percent-idUSKCN1MM0QD[/url<] -reduces investment in making of said toys...

        • Bobs_Your_Uncle
        • 11 months ago

        This is actually quite a practical move. As Jeff pointed out above, “[i<]EUV is simply one cog in an outlandishly complex machine.[/i<]" By purchasing a stake in ASML back in 2012 Intel was able to provide significant funding to support development of a technology that they (as well as other Fabs) would directly benefit from . Now that the EUV process they (desperately) need has been proven and ASML has it in production, it makes sense for Intel to reallocate capital funding of ASML to other (presumably in-house) R&D needed to advance their own 10nm and 7nm processes. Given Intel has firmly positioned itself betwixt a quite large rock and an exceedingly hard place, this move seems to be an exercise in pretty good judgement.

          • freebird
          • 11 months ago

          Yeah, but sometimes a joke is just a joke…
          (such as my previous comment.)

Pin It on Pinterest

Share This