Intel Now Content To 'Match' TSMC's Process Technology

(Image credit: Shutterstock)

At a fireside chat with Credit Suisse at their 23rd annual technical conference, Intel’s CEO Bob Swan said that its 7nm process is expected to match TSMC’s 5nm process. He also noted that Intel's 5nm process is also expected to match TSMC’s 3nm process. 

However, what Swan didn’t mention is that Intel is no longer in the lead in terms of process technology and that its 7nm process is expected to arrive about a year later, in 2021, compared to TSMC’s 5nm which will produce device chips by the second half of 2020.

Intel's Recent Process Roadmap

When Intel announced the 22nm Tri-gate (FinFET) process, it was more than a generation ahead compared to TSMC and other third-party foundry competitors. For one, it was on a smaller 22nm process compared to others who were just moving to 28nm/32nm process nodes. And second, the switch to FinFET alone granted its own generational boost in performance and efficiency. Intel’s process leadership was uncontested for years after that. 

The only exception was in mobile chips, where its 22nm FinFET Atom chip could barely match the latest high-end Arm chips on a 28nm planar node, and at a higher chip cost to boot. It’s why Intel eventually attempted to license the Atom design to Chinese fabless semiconductor companies so that they build their own “Atom” chips more cheaply on TSMC’s 28nm process. 

However, that didn’t work out either, as few device makers were interested in making the trade-off compared to the Arm chips they’ve come to rely on for each device launch.

Intel then switched to 14nm, and that wasn’t a very smooth ride either. The company experienced some delays with Broadwell chips, which were the first to use the 14nm process. Intel also ended-up replacing the consumer Broadwell generation on the market with Skylake rather quickly.

The reason the switch to 14nm was quite bumpy, too, was because Intel set out to increase the chip density by 2.4x, which ended-up a little too difficult to implement, but the company succeeded in achieving that eventually.

However, instead of taking this lesson to heart, Intel attempted to increase the density even more aggressively with the 10nm process, by 2.7x. After years and years of delays, the company recently admitted that the goal was too aggressive for the company.

This is why for its move to 7nm EUV, Intel will scale back the density increase to 2.0x. The switch to an extreme ultraviolet (EUV) lithography process is difficult enough as it is. It’s also Intel’s first attempt to implement EUV, following in Samsung and TSMC’s footsteps.

From a Leader to a Follower?

Although the claims of Intel’s CEO are meant to reassure both customers and partners that the company is “still” as competitive as TSMC, this statement doesn’t come from a position of leadership in this aspect.

Even if what Intel claims is true and the process is indeed more or less equal performance, efficiency, and density-wise, the fact is that devices with chips made on TSMC’s 5nm process are expected to come out in the second half of 2020. Those devices will include the next-generation iPhones, too. Meanwhile, Intel has recently said that it will start launching 7nm chips sometime in 2021.

Intel has also fallen behind AMD in terms of having secure processors and firmware, going by the number of recent vulnerabilities (surpassing AMD 15:1). The company also says that it’s no longer interested in large market share, coincidentally in a time when AMD’s Zen 2 chips are expected to make the biggest dent in Intel’s sales, both in the consumer and data center markets in more than a decade.

Although none of this means Intel still isn’t in a much stronger position both brand and financially-wise compared to AMD and TSMC and that it can’t recover, it’s still a trend that doesn’t do Intel any favors and that Intel will have to reverse before it’s too late.

Lucian Armasu
Lucian Armasu is a Contributing Writer for Tom's Hardware US. He covers software news and the issues surrounding privacy and security.
  • JayNor
    Intel and TSMC are both waiting on ASML EUV equipment that meets the requirements for high volume production. Good luck if you're betting high volume production on a limited resource.

    "The big news from ASML Q3 report is the shipment of the first NXE:3400C systems. We consider the 3400C to be the first fully HVM-oriented systems."

    "At the current production rate, orders placed in Q4 won’t be fulfilled any earlier than mid-2021. It’s currently unclear what impact, if any, the large backlog in EUV machines will have on foundry roadmaps."
  • JayNor
    Intel's 22FFL process enabled the ultra low leakage io layer of their new Lakefield chip that is designed into Surface Neo next year. That 10nm chip also demonstrates Intel's Foveros 3D manufacturing capability for logic, which Intel claimed to be production ready in Dec 2018. I don't see any similar product on the roadmap from AMD/tsmc.
  • JayNor
    Intel's 22FFL process also enabled a very high retention automotive temperature STT-MRAM product, which this article states is a leading feature.

    "As far as retention goes, Intel reported 10-year retention not only at the Grade 2 automotive temperature range (-40 to 105 °C) but also at 200 degrees C for 10 years which is much higher than anything else we are aware reported to date."
  • JayNor
    Intel and Micron together developed a process for 3DXpoint. I haven't seen any response from AMD/TSMC.
  • JayNor
    ASML leaked Intel's processing roadmap at IEDM. No indication they are waiting for tsmc.