Whatever one might say about Intel’s position in the high-mobility market, where rivals currently dominate, the company has been driving the pace of process node development — the science of reducing the size of chip features via an arcane brew of physics, chemistry, and metallurgy — for a long time.
Intel is keeping up a particularly fast pace in its system-on-a-chip (SoC) business, based on its low-power Atom core. In three short years, Intel has pushed its factories to produce ever-smaller features on SoCs, jumping from 32nm to 22nm to 14nm at a cadence of one node per year, twice the usual pace. A nanometer (nm) is a billionth of a meter. Intel will ship 22nm SoCs in volume in 2013 and is expecting to hit volumes of 14nm chips in 2014. By 2014, Intel expects to align the schedules its Atom-based SoCs and its flagship Core processor line, which, in the past, has reached new process nodes more quickly. For example, the company is shipping volume production of 22nm Core products this year.
Advanced Micro Devices (AMD), ARM Holdings, Samsung, and even IBM could only watch with mouths agape as the Santa Clara juggernaut hit the power-performance-cost problem with a tour de force of scientific ingenuity.
But hark! What is that sound? It is the hoofbeats of Globalfoundries 14nm process node, right around the corner. Wait! How did it do that? How did it pirouette from just barely producing 28nm chips in volume to nailing a 14nm node in less than a year?
The answer: a combination of different ways. 3D is a big one. Although Globalfoundries was not able to bring a FinFET transistor to market as fast as Intel, it has benefited from a decade of work on these 3D structures carried out by IBM and contributed to the Common Platform, which includes Globalfoundries and Samsung as well as IBM.
As a result, Globalfoundries was able to announce earlier today its 14XM (14nm eXtreme Mobility), which successfully grafts 14nm FinFET features onto the company’s existing 20nm scaffolding. Essentially, Globalfoundries has been able to produce a stable tri-gate structure at 14nm for the chip’s transistors, which can be blended with existing 20nm planar (flat) structures for the interconnection areas of the chip. Globalfoundries 20nm parts are just ramping up now.
Another piece of magic is high-K metal gate (HKMG). HKMG greatly reduces electrical leakage, which normally increases as chips get smaller. It was critical to master this technique, which Intel has had since 2007. It was only in 2010 that AMD was able to get HKMG up and running in its 32nm processors, which have both processing and graphics on the same piece of silicon. (AMD calls these mixed units Accelerated Processing Units or APUs.)
Globalfoundries’ architecture might seem like a bit of a mishmash at first glance. Putting different size features on different chip planes raises questions as to whether such a design can reap all the cost benefits of a fully 14nm part. For example, the 20nm planes will determine the size of the chip, which will affect both its cost and its power-savings.
However, the smaller transistors will make the part more efficient than a 20nm-only unit would be, and the design choice meets all of Globalfoundries’ objectives, and, I daresay, those of its customers.
These objectives include rapid time to market. The foundry expects to offer the 14XM to chip makers in volume in 2014. Meanwhile, in 2013, it will roll out its 20nm planar process for customers that need to move more quickly.
Another goal that the 14XM meets is ultra-low power. The 20nm process already ironed out by Globalfoundries is optimized for mobile SoCs. These chips are designed for smartphones that need to operate all day without recharging. This technology is preserved and carried forward in the 14XM.
Reusing the methods learned at 20nm has also enabled Globalfoundries to reduce the risk in moving to 14nm high-volume manufacturing. Customers will be able to migrate easily from the 20nm to the 14nm node.
And, of course, all this would be useless if the parts didn’t meet the competition on the dimensions of cost and performance. Globalfoundries has indicated that, in internal bake-offs, its 14XM leads in highly constrained tests in which performance, power (battery life), and cost are all factored in.
Finally, the company, being a foundry, has put together an ecosystem that will enable its customers to produce truly world-class parts. It has a full suite of process design kits, a range of multicore graphics solutions, other libraries and intellectual property (IP) suites available to chip designers, and a variety of packaging options.
Most importantly, all of these efforts are focused on the mobile SoC business, which is arguably the sweet spot. Intel may lead in process node technology and dominate the PC market, but a whole new crop of companies has arisen to rule in high mobility (smartphones, tablets, Ultrabooks).
In this market, Intel (an integrated design manufacturer) is a Johnny-come-lately, with a few phones, a tablet or two, and whatever Ultrabooks are selling. The leaders are companies like Apple (chips and systems), Samsung (chips and systems), Google (operating system), Qualcomm (chips), ARM (chip designs), TSMC (foundry), and Globalfoundries (foundry).
Globalfoundries has come from way behind to nip at Intel’s heels. The 14XM will hit the market in volume in mid-2014, right on top of Intel’s 14nm processors. And Globalfoundries and its customers have the pole position in high mobility, offering an ultra-low power mobile SoC with a whole ecosystem around it rather than just a processor.
And as a bonus, Globalfoundries is doing all of the work on the 14XM at its new factory in Upstate New York, joining Intel in creating good jobs in the United States.
A race that had almost gotten boring is suddenly exciting again.
Disclosure: Endpoint has a consulting relationship with Globalfoundries.
© 2012 Endpoint Technologies Associates, Inc. All rights reserved.
Twitter: RogerKay
