Oct 1, 2012Science and Technology
GlobalFoundries’ 14-nm chip saves fabless-foundry model from doomsday

FinFET is a multiple-gate transistor technology where the source and drain are wrapped in a semiconductor "fin."In a surprise attack against Intel, GlobalFoundries accelerated its semiconductor roadmap by announcing plans for a 14-nanometer chip using FinFET technology. The breakthrough chip is due by 2014, just one year after the foundry’s 20-nanometer offering hits production. A bold move because roadmaps rarely get tested or altered, even bolder because FinFET designs, headed for mobile devices, represent a radical transformation from planar to 3-D transistors. More telling, GlobalFoundries’ ambition rejuvenates the fabless-foundry production model many experts believe is broken, and destined for failure.

At a time when fabless designers like Qualcomm and Nvidia question whether their foundry partners are keeping pace with Moore’s Law, GlobalFoundries’ expedited 14-nm chip moves the California company to the head of the foundry pack, ahead of both Taiwan Semiconductor Manufacturing Co. (TSMC) and United Microelectronics Corp. (UMC). Being revered as the technological champion of the foundry ranks is important as mobile requirements for power and performance change the semiconductor landscape, but running down Intel, the undisputed leader of FinFET design, would shock the industry back into thinking the fabless-foundry model still works.

Optimized for mobile system-on-chip (SoC) designs, GlobalFoundries’ 14-nm nonplanar architecture delivers 40 percent to 60 percent more battery life compared to 2-D planar transistors at the 20-nm node. FinFET features low-power attributes, leveraging a 3-D transistor design operating at a lower voltage with less current leakage. In practice, FinFET means longer battery life for mobile applications. And less power consumption for plugged-in applications, such as chips driving datacenters.

“The company’s 14nm-XM offering will give customers the performance and power benefits of three-dimensional FinFET transistors with less risk and a faster time-to-market, helping the fabless ecosystem maintain its leadership in mobility while enabling a new generation of smart mobile devices.” That’s the GlobalFoundries market promise. And it’s not all talk as 14nm-XM development is underway in Fab 8, GlobalFoundries’ fabrication in Saratoga County, New York. According to plan, sample chips will be ready in 2013, with market-ready products due no later than the beginning of 2015.Ivy Bridge, Intel's series of 22nm FinFET microprocessors, hit the market in 2011.

If GlobalFoundries can move 14nm at this pace, they would push and pressure TSMC, a much larger competitor. Crossing the three-dimensional 14-nanometer gap is tricky and expensive, and GlobalFoundries has hatched a good plan. In fact, industry pundits argue 14-nm nodes will put some foundries out of business, playing further into Intel’s hands. But aspiring to be another massive foundry isn’t GlobaFoundries’ mission. And the target isn’t Taiwan. Speaking of Intel, all this 14-nm ballyhoo is aimed at them. Aimed at Intel’s stranglehold on leading-edge nano production, and esteemed position as the world’s only integrated provider.  

However, as clever as GlobalFoundries’ 14-nm revelation may be in fabless circles, Intel set the FinFET standard way back in 2011, introducing their 3-D “Tri-Gate” design for 22-nanometer transistors. “More than 6 million 22-nm tri-gate transistors could fit in the period at the end of this sentence,” Intel claimed. Codenamed “Ivy Bridge,” Intel’s 22-nm pioneer jumped the market, now powering multiple Core processor models targeting both desktops and mobile devices.

According to Intel, Tri-Gate technology should “double the 3-D graphics and HD media processing performance compared with Intel’s previous generation of chips.” Ideal for serious computing, or mobile power savings, Ivy Bridge fuels the second-generation ultrabook, a hybrid device Intel hopes will harpoon iPad’s runaway success.

And like Intel clockwork, the 14-nm fabrication node launches in 2013, with 10nm following in 2015. Boorish as all the Moore’s Law talk may be, Intel sticks to the roadmap and continues to showcase the design, development, marketing and distribution making their rich integrated model so scary.

Scary to the fabless-foundry model facing escalating barriers to entry as vendors leap from planar technology, at today’s 20nm, to FinFET at 14nm and on and on down the road. To play at this level, a fab runs in the $6 billion range, and process research and development over $1 billion. Conservatively, a five-fold increase over 130-nm production costs. And VLSI Research reports moving to 450mm fabs will require a $32 billion investment in process and tool research and development. The more feature dimensions get squeezed, the higher the cost of process development.

So did GlobalFoundries simply take advantage of a slow news day to punch other foundries in the stomach, just for good 14-nm humor? Intel prints money and has proven FinFET architecture in working devices. Sounds like game over for the fabless-foundry model.

Although, according to IHS, the $29 billion pure foundry market still enjoys double-digit growth - projected to grow 14 percent in 2013 -- the price of poker skyrockets when shifting from planar to FinFET. Especially when fabless designers have grumbled about delays in past nodes built for lesser technologies. As transistors get smaller, process variables become more unpredictable while affecting yields, leakage and power consumption. Given these economic and technological hurdles, obsolescence for stodgier foundries can be charted along Mayan doomsday calendars.

Not so fast. GlobalFoundries embraces the new challenges presented by advanced chips, and speaks openly about evolving the current fabless-foundry model to meet customer demands for better collaboration, and product demands for deeper integration of design and manufacturing. They also appear more flexible when considering different business models.

One idea offers dedicated modules within a single fab. Or, on a per-customer basis, GlobalFoundries would build a dedicated fab for customers willing to share rising development costs. Samsung built Apple a dedicated fab in Austin, Texas. An important detail, because GlobaFoundries’ competition, in TSMC, UMC and Samsung, refuse to stand idly by and wait for Intel to steamroll them. They have ideas and workarounds, too.

Lucrative partnerships with big players will help satisfy the huge capital investment foundries need to survive the next few years of the semiconductor roadmap. Here, GlobalFoundries starts with ARM, recently announcing a multi-year agreement to deliver optimized SoC solutions for the biggest designer of mobile chips in the world. “In the growing era of extreme mobility, FinFET technology will be a critical enabler to the next generation of smart mobile devices,” said Dipesh Patel, deputy general manager of the Physical IP Division at ARM.

And for deep pockets required by ownership, GlobalFoundries was fully absorbed by the Abu Dhabi government in March. Despite a cloudy economic climate, and a dwindling PC market, the company has committed $11 billion to plants and equipment since 2010. No cost cutting here.

GlobalFoundries is bankrolled and supports a SoC design well positioned to feed our mobile hunger. According to EE Times, the pure foundry market will be worth $63 billion by 2018. And with Intel really just breaking into the mobile market, GlobalFoundries forces analysts and customers to rethink the fate of the fabless-foundry model. If 14nm symbolizes future success, then Intel may not be in a semiconductor league of their own.  

Share
Be the first to comment.