Company quietly announces chips built on “10nm+” process even before 10nm ones are out.

Intel has given an unusual insight into the road ahead for its mainstream desktop and laptop processors, confirming the existence of a new processor family called Ice Lake.

Once upon a time, the company planned to follow up Skylake, built on a 14nm process, with Cannon Lake, built on a 10nm process and shipping in late 2016. But that plan was derailed. The 14nm process took longer than expected to bed down and start working properly. Our understanding is that Intel moved engineers that were developing 10nm to help with fixing 14nm. This had a few knock-on effects. First, it required Intel to produce additional designs built on 14nm: last year's Kaby Lake uses the second-generation 14nm+ process, and this year's Coffee Lake will use a third-generation 14nm++ process.

Second, it delayed 10nm. 10nm parts aren't now expected until 2018, when Cannon Lake finally materializes. The newly confirmed Ice Lake will use a second-generation 10nm process, 10nm+.

Intel's current plan is to split the desktop and laptop chips up. Desktop chips will stick with 14nm variants, currently Kaby Lake and soon Coffee Lake. Laptop chips will diverge; there will be not only 14nm++ Coffee Lake laptop parts but also 10nm Cannon Lake parts. Ian Cutress at AnandTech speculates that the split will be driven by core size and power; the smaller 15W parts will be Cannon Lakes because small chips will maximize the yields of the new 10nm process. Larger processors, from 35W and up, will stick with 14nm++ and Coffee Lake.

Ice Lake, built on 10nm+, may re-unify things. In principle, the second-generation, more mature 10nm process should offer better yields for larger chips and hence be suitable for a wider range of Intel's processors.

Underpinning all these delays and difficulties with new processes is the continued difficulty of developing production-ready extreme UV (EUV) lithography techniques. The circuit and gate patterns are transferred to the silicon wafer using an optical process called "photolithography." Currently, that uses ultraviolet light with a wavelength of 193nm. While this large wavelength can be used to create chips with much smaller features, including the 14nm processors today and 10nm parts imminently, doing so requires complex, multistage manufacturing in a technique called "multipatterning." EUV, which uses 13.5nm light, would make that aspect of the manufacturing much simpler—but it presents the problem that the EUV light itself is hard to generate and hard to manipulate. EUV systems can't generally uses lenses (most lens materials absorb EUV), only mirrors.

EUV is one of those technologies that has been just around the corner for years. Its arrival has been anticipated since the 1990s. In 2013, it was hoped that commercially viable systems would ship in 2015, but they didn't. Intel is, however, continuing to invest in the technology. The development of working EUV will make new processes easier to introduce, at least for a while.