In an unusual move for Intel, the chip giant has ever so slightly taken the wraps off of one of their future generation Core architectures. Basic information on the Ice Lake architecture has been published over on Intel's codename decoder, officially confirming for the first time the existence of the architecture and that it will be made on Intel's 10nm+ process.

The Ice Lake processor family is a successor to the 8th generation Intel® Core™ processor family. These processors utilize Intel’s industry-leading 10 nm+ process technology.

This is an unexpected development as the company has yet to formally detail (let alone launch) the first 10nm Core architecture –  Cannon Lake – and it's rare these days for Intel to talk more than a generation ahead in CPU architectures. Equally as interesting is the fact that Intel is calling Ice Lake the successor to their upcoming 8th generation Coffee Lake processors, which codename bingo aside, throws some confusion on where the 14nm Coffee Lake and 10nm Cannon Lake will eventually stand.

As a refresher, the last few generations of Core have been Sandy Bridge, Ivy Bridge, Broadwell, Haswell, Skylake, with Kaby Lake being the latest and was recently released at the top of the year. Kaby Lake is Intel’s third Core product produced using a 14nm lithography process, specifically the second-generation ’14 PLUS’ (or 14+) version of Intel's 14nm process.

Meanwhile when it comes to future products, back at CES Intel briefly showed a device based on post-Kaby Lake designs, called Cannon Lake and based on their 10nm process. Since then Intel has also confirmed that the 8th Generation of processors for desktops, called Coffee Lake, will be announced on August 21st (and we recently received promotional material to that effect). Ice Lake then, seems poised to follow both Coffee Lake and Cannon Lake, succeeding both architectures with a single architecture based on 10nm+.

Working purely on lithographic nomenclature, Intel has three processes on 14nm: 14, 14+, and 14++. As shown to everyone at Intel’s Technology Manufacturing Day a couple of months ago, these will be followed by a trio of 10nm processes: 10nm, 10nm+ (10+), and 10++,

On the desktop, Core processors will go from 14 to 14+ to 14++, such that we move from Skylake to Kaby Lake to Coffee Lake. On the Laptop side, this goes from 14 to 14+ to 14++/10, such that we move from Skylake to Kaby Lake to Coffee Lake like the desktops, but also that at some time during the Coffee Lake generation, Cannon Lake will also be launched for laptops. The next node for both after this is 10+, which will be helmed by the Ice Lake architecture.

Intel's Core Architecture Cadence
Microarchitecture Core Generation Process Node Release Year
Sandy Bridge 2nd 32nm 2011
Ivy Bridge 3rd 22nm 2012
Haswell 4th 22nm 2013
Broadwell 5th 14nm 2014
Skylake 6th 14nm 2015
Kaby Lake 7th 14nm+ 2016
Coffee Lake 8th 14nm++ 2017
Cannon Lake 8th? 10nm 2018?
Ice Lake 9th? 10nm+ 2018?

The way that the desktop and laptop markets will be diverging then converging is confusing a lot of people. Why is the laptop market splitting between 14++ and 10, and why is the desktop market not going to 10nm but straight to 10+? What lies beyond is a miasma of guess work, leaked slides, and guessing Intel’s strategy, but I believe the answer lies in Intel’s manufacturing technologies and the ability to move to newer lithographic nodes.

(We should interject here that the naming of a lithographic node has slowly lost its relevance between the features of the process and the actual transistor density and performance, such that TSMC’s improved 16FF+ is called 12FFN, but relies on similar transistor sizes with enhanced attributes. But 12 is a smaller number than 14, which is the marketing angle kicking in. By all accounts, Intel has typically been considered the more accurate foundry when it comes to numerical lithographic naming of the process, which others consider is to their detriment.)

Intel originally predicted that they would move to 10nm almost a year ago, at the end of 2016 and 2 years after the launch of their 14nm process. But the challenge in managing the technology required to advance to their version of 10nm has been fraught with difficulty. In all cases it can depend on external equipment, fine tuning a process, or getting acceptable yields – while one manufacturer might be satisfied with an 80% yield, another might consider that a failure. Being able to obtain high yields (ramp up) will also be a function of die size, and so the newest nodes are typically launched with smaller mobile parts in mind first, as the yields for smaller parts are better than larger parts at the same defect rate.

Simply put, the first generation of 10nm requires small processors to ensure high yields. Intel seems to be putting the smaller die sizes (i.e. anything under 15W for a laptop) into the 10nm Cannon Lake bucket, while the larger 35W+ chips will be on 14++ Coffee Lake, a tried and tested sub-node for larger CPUs. While the desktop sits on 14++ for a bit longer, it gives time for Intel to further develop their 10nm fabrication abilities, leading to their 10+ process for larger chips by working their other large chip segments (FPGA, MIC) first.

From a manufacturing standpoint, Intel has been using multiple patterning techniques in its 14nm processes, and the industry is looking to when the transition to EUV will take place. Anton has some great writeups of the state of EUV and how different companies are transitioning to smaller nodes - they are well worth a read.

The crux of the matter is that EUV would shorten time to market and arguably make the process easier (if only more expensive), and several fab companies are waiting for Intel to jump onto it first. With EUV not ready, Intel has had to invest into deeper multi-patterning techniques, which raise costs, decrease yields, and increase wafer process times considerably.

All of which leads to a miasma of increased delays, much to the potential chagrin of investors but also customers who had banked on the power improvements that a typical new lithography node brings. Intel is still keeping spirits high, by producing numbers that would suggest that their methodology is still in tune with Moore’s Law, even if the products seem to be further strung out. Some analysts concur with Intel’s statements, while others see it as hand-waving until 10/10+ hits the market. Intel would also point out that it is developing other technologies such as Embedded Multi-Die Interconnect Bridges (EMIB) to assist in equipping chip with high-speed fabric or glue-logic.

Given its position as a post-8th gen architecture, Ice Lake is likely to hit sometime in 2018, perhaps 2019, depending on Intel’s rate of progress with larger chips and the 10+ process. Intel’s other market segments, such as FPGAs (Altera), Xeon Phi (MIC) and custom foundry partners, are also in the mix to get into some 10nm action.

(Note that Intel’s next generation of Xeon Scalable Processors is called Cascade Lake, a 2018 refresh of the Skylake generation launched this year.)

Source: Intel

Comments Locked

106 Comments

View All Comments

  • eddman - Tuesday, August 15, 2017 - link

    It is possible that intel might end up using both coffee lake and cannon lake chips in their 8th gen, 8xxx lineup; perhaps CFL for desktop parts and CNL for low-power mobile parts.
  • name99 - Tuesday, August 15, 2017 - link

    The new model is Tick Tock Sproing Thud. It's funny because it's true.
    HT to SemiAccurate.
  • watzupken - Tuesday, August 15, 2017 - link

    You can certainly sense the urgency from Intel's camp of late. But to be honest, there is no point looking 2 generations ahead since thing may change over the next couple of years. Bring out Cannon Lake, then we talk about the next gen.
  • name99 - Tuesday, August 15, 2017 - link

    The issue is what you're interested in.
    If your interest is ONLY in "what CPU can I buy in the near future" or "how can I support my corporate tribe", then what you say is correct. But look at the big picture beyond those.
    That picture shows an Intel that is breaking down before our eyes.

    It started with Broadwell which shipped late and lame (in spite of Intel's protestations beforehand that yields were the most awesome ever). Lying about how well their process is doing seems to be the new normal (witness Optane, still MIA from the market where it actually matters, and being sold into markets where flash is a much better choice).

    Meanwhile architectural innovation seems to have ground to a halt with Skylake (it actually stopped with Sandy Bridge, since then they've been treading water, but now they're not even trying --- hence once Lake after another).
    Now we have this process-based split across Kaby, Coffee, Cannon, and Ice that looks like nothing so much as mad scrambling. It's not that having separate designs and processes for server vs desktop vs laptop is a bad idea; but these do not look anything like separate DESIGNS, more like just separate "WTF can we do to get SOMETHING, ANYTHING that works?". There's no evidence here of design in terms of things like giving Cannon a differently balanced microarchitecture that's a better fit for its target space as compared to say Coffee.

    So yeah, it's interesting from the point of view of "How does Intel's monopoly end?"
    Ten years ago was just before iPhone, just before Nehalem. A lot can change in ten years. So the question is what changes over the next ten years? What all this naming nonsense suggests to me is that Intel is even less prepared for the future than they were for mobile. At the time I suggested that Intel's mobile attempts were a pointless waste of time, based not on dick-measuring how one Intel core stacked up against a competitor ARM core, but based on the fact that Intel's STRATEGY made no sense, it revealed a corporation that was either (perhaps both) deeply incompetent or deeply deluded.
    I see the same fundamental strategic failures now at the mainstream level. Sure things look fine right now, but we're not talking now, we're talking about 10 yrs from now and how well Intel is poised to cope with the future.
  • Hurr Durr - Tuesday, August 15, 2017 - link

    Oh look, usual crapple shill yapping about corporate tribes of all things.
  • name99 - Tuesday, August 15, 2017 - link

    Seriously? I comment that some people's primary interest in this news is "how can I support my corporate tribe" (not that there's anything wrong with that, simply pointing out that that that's not the ONLY lens of analysis) and your response is to talk about "crapple".
    I kinda rest my case.
  • twtech - Tuesday, August 15, 2017 - link

    My understanding is that several years back, Intel decided to adopt a focus on hiring & promoting for reasons other than being the best at what you do. For a company that relies on continued superior engineering to maintain its advantage, that doesn't seem like a good thing - at least in terms of the quality of the output product.
  • name99 - Tuesday, August 15, 2017 - link

    And do you have a source for this beyond Fox News and James Damore?

    If we're throwing out wild theories, my wild theory is that the problems all started when America First sentiment mean Intel could no longer hire the best possible engineers from the entire world.

    Se how that works?
  • mak1977 - Tuesday, August 15, 2017 - link

    How does Intel Monopoly End.

    It ends when arm support on windows is passable . Microsoft is working on ARM for windows with some kinda of emulation in hardware by the ARM parts to do x86/win32 apps.
    https://www.xda-developers.com/microsoft-qualcomm-...

    Once the performance is good enough and More ARM vendors come on board, it will be the end of Intel , they can't compete on price with arm.
  • BillBear - Tuesday, August 15, 2017 - link

    Intel's going to have to either advance the state of the art (something it certainly did not do with Kaby Lake) or it's going to have to learn to live without a 60% profit margin.

    You can't price gouge and stand still at the same time.

Log in

Don't have an account? Sign up now