Node Warfare?

GlobalFoundries unveils 12nm finFET process; foundries jockey for position on way to next full node.

popularity

By Mark LaPedus & Ed Sperling
GlobalFoundries uncorked a 12nm finFET process, which the company said will provide a 15% increase in density and more than 10% improvement in performance over the foundry’s existing 14nm process.

This is GlobalFoundries’ second 12nm process. It announced a 12nm FD-SOI process called 12FDX last September, although it first mentioned a 12nm process back in July of last year.

The announcement adds to the flood of process node numbers in semiconductor manufacturing these days. The reasoning behind all of this activity is that even though 7nm—the next full node—will be introduced next year, it will take time before that process is mature enough for commercial production. Foundries are trying to provide enough options to hold onto existing customers until that happens, while also tapping potential customers in new markets.

The new business involves sectors such as automotive, IoT, communications infrastructure, and artificial intelligence/machine learning. GlobalFoundries, for example, is taking aim at the automotive and RF/analog markets with its new process. The 12LP process is expected to meet Automotive Grade 2 qualification next quarter, which under AEC-Q100 means it will operate at temperatures between -40°C and +105°C, and it is being targeted for transceivers in 6GHz wireless networks.

“It builds on 14nm,” said Sanjay Jha, CEO of GlobalFoundries. “It pushes new designs and constructs.”

He noted that shipments are expected to begin in the first half of next year.

Still, there are now so many process node numbers out there that it’s becoming difficult to sort through them all, a task made all the more difficult because the numbers don’t match up from one foundry to the next. Intel’s 14nm process, for example, has double the number of transistors as TSMC’s 16nm, while Intel’s 10nm process is roughly the same as TSMC’s forthcoming 7nm.

Joanne Itow, managing director for manufacturing at Semico Research, said foundry customers have to do their homework to make sense of all of this.

“You have to do a lot of analysis of what process works with your products,” she said, noting that each foundry appears to be taking a base process and then proliferating it. She described GlobalFoundries move as a small tweak that gives customers new options.

And there is certainly plenty of node tweaking underway. TSMC unveiled a 12nm process this month, while Samsung introduced an 11nm process. In addition, Samsung announced 8nm, 7nm, 6nm, 5nm, and 4nm finFET processes in May, and said it plans to introduce 18nm FD-SOI in 2019.

Related Stories
Inside FD-SOI And Scaling
GlobalFoundries’ CTO opens up on FD-SOI, 7nm finFETs, and what’s next in scaling.
Is 7nm The Last Major Node?
Technical issues increase, costs go up, and not all markets will benefit.
The Race To 10/7nm
Next nodes are expected to be long-lasting, because costs of developing chip after that will skyrocket.



  • Gary Merchant

    So is this actually 12nm, or is it just a marketing term for enhanced 14nm?

    • Mark LaPedus

      Hi Gary. It depends on how you look at it. To me, 12nm is a minor step. (very minor). It buys you some time before jumping to 7nm. Any thoughts?

      • Gary Merchant

        I would like to see millions of transistor per square millimeter included in press releases. It would make node improvements much easier to compare. I’m not entirely sure why they choose to make many small node improvements as opposed to the larger jumps in density like 28nm to 14nm. It seems like that it would be costly to change the equipment so frequently. Especially since we’re so close to the limits of silicon.

        • Mark LaPedus

          12nm is a version of 14nm. Once again, going from 14nm to 12nm is a minor step in terms of pitch, etc. The equipment in the fab is the same for both 14nm and 12nm as well as 10nm and most of 7nm. Chipmakers want to reuse as much of the equipment as possible at these nodes. The exception is EUV litho, which brings a new and expensive cost-of-ownership to the game at 7nm.

  • Kiers

    really? intel has problems with it’s 10nm Finfet.

    any idea what global foundries charges for fab services at say the 20nm node for a 300mm wafer of , say, ARM microprocessors? any idea at all? i’m trying to get basic ball park info?

    • Mark LaPedus

      Kiers. Good questions. Foundries charge different prices for different services and customers. It depends on the volume, mask cost, node, etc. Get an estimate from several vendors.