Intel’s silicon superiority may slip as shrinking chips gets harder

Back in 2007, Intel set an ambitious but achievable roadmap for future chip releases. Each new generation of Intel processor would alternate between shrinking the previous generation’s architecture – a tick – and presenting a new architecture design – a tock. Haswell, a tock, was introduced in mid-2013 as a new 22 nanometer (nm) architecture with expanded functionality and better performance. The Haswell design was shrunk down to 14nm, tick, where it was known as Broadwell. What happened in between is what’s really interesting.

In the middle of 2014, Intel was having trouble shrinking the Haswell die, and released the Haswell Refresh, aka Devil’s Canyon lineup. As something of a dessert course for Haswell, it included just three chips, one for each of the Core i3, i5, and i7 lines, all unlocked and based on the Haswell architecture.

Kaby Lake sits in the same sort of awkward position. It follows hot on Skylake’s heels, perhaps too close for a proper architecture redesign, but likely with a wider selection of offerings than the Haswell Refresh. Sources indicate the Kaby Lake update is mostly focused on bringing some new connectivity and optimization benefits to the line. Among the changes, we expect to see native USB 3.1 and HDMI 2.0 support, plus expanded encoding and PCIe support. Importantly, Windows 10 is the only version of Windows that will run on a Kaby Lake chip, a major change for Intel and Microsoft.

Further down the road, Cannonlake will be the first 10nm chip from Intel, but there’s another surprise there. News from the same investor meeting suggests the 10nm die shrink will result in three distinct architectures, as with the 14nm production node.

Cannonlake will be the first, based off the Skylake architecture. In 2018, Ice Lake will replace Skylake’s architecture on the 10nm die. Tiger Lake will follow in 2019, and again will be an architecture update based the 10nm production process.

That, of course, is three products based on a 14nm production process, and three products based on 10nm, which ruins the traditional tick-tock cadence.

It’s unclear what this ultimately means for Intel’s Tick-Tock model, but it’s certainly not looking good. While architecture changes will continue to bring about new features and improvements, shrinking the die is only going to get more difficult. Tiger Lake’s 2019 release might seem far off to the consumer, but it’s right around the corner for a massive design and production company like Intel, and you can bet work on it is already well under way.

This slowdown in Intel’s leap to new, smaller production nodes may also give competitors a chance to keep up. TSMC, a major foundry that products chips for many companies including Apple and Nvidia, plans 10nm production for late 2016 — beating Intel, if it meets its goals.