Your browser is not supported. Please update it.

9 October 2014

Foundries race to invest in low power process technologies for IoT

The dawn of the internet of things is spurring huge investments in new chip technologies and manufacturing processes. Hard on the heels of Intel’s planned spend of $6bn on a new Israeli plant to support 10nm geometries, Samsung say it will spend $15bn (KRW15.6 trillion) on a new plant in South Korea, with a focus on processors for smart connected devices such as wearables and cars.

As some players point out, the IoT will involve a wide variety of chips, many analog and mixed-signal (see separate item on Cortus), and those rely on coarser geometries. So the race to fewer nanometers will not be the only one in the new wave of chips – but it will certainly be the most visible, symbolic and expensive. Most observers would say Intel has a significant lead at the current time, and its CEO Brian Krzanich plans to turn that lead into foundry revenues for non-conflicting customers, rather than purely harnessing it for internal advantage, as in the past.

Samsung also needs to invest in new processes and plants for that dual purpose – to give its own chips an edge, and to bolster its foundry business. It has to build up its semiconductor business to offset the slowdown in its largest activity, mobile devices (see separate item). As Lee Min Hee, an analyst at I’M Investment and Securities in Seoul, told Bloomberg: “Semiconductor has been Samsung’s long time key business, while the company can no longer expect strong growth from its mobile unit. The company is now making a huge investment for the future as its key mobile business isn’t likely to contribute to profit as much as it used to.”

Memory chips are an important revenue stream for Samsung but is also aims to build up its strength in mobile and IoT processors and in its foundry services. At the ARM TechCon event last week, it showed off a working 14nm chip using the FinFET process. The company says it has multiple foundry customers for this early-stage 14nm process, though it has not named any, and it will also launch its own chips. It demonstrated a mobile apps processor which was assumed to be the next generation of its Exynos family, and which was seen decoding HD video streams.

Samsung knows it has to keep up with foundry rivals, which increasingly may include Intel, as its customers, and its own processor division, race to reduce the size and power of their chipsets to target the growth in IoT devices. TSMC held its own event in the same week as TechCon and described its 16nm FinFET process. Kelvin Low, a foundry marketing manager at Samsung, told EETimes: “We are locked in a huge race to 16/14nm FinFET production.” Samsung is working in partnership with GlobalFoundries on the 14nm process and both will have volume production next year.

This will be one boost to the foundry business – in which Samsung needs to hold on to as much of Apple’s business as it can. It is now sharing the spoils with TSMC, but is thought to be supplying about 40% of Apple’s A8 chips and to have won a contract for the upcoming A9, using 14nm process. Though Apple would have political reasons to divorce Samsung altogether, the Korean vendor has repeatedly been able to leap ahead in quality and reliability terms and keep itself indispensable to this important customer.

The new manufacturing plant will expand capacity for IoT chips and other categories driven by the new silicon-efficient processes. It will start operations in 2017 and will be located in Gyeonggi province, south of South Korean capital Seoul. Profit at Samsung’s chip division is projected by analysts to be KRW2.14 trillion, which would be up from KRW2.06 trillion a year before.

The plant will be highly strategic, because TSMC is generally agreed to have overtaken Samsung in the race to get FinFET technology into commercial production. “Recent developments suggest TSMC has taken the lead over Samsung at developing [16/14nm] FinFET manufacturing technology,” Mehdi Hosseini, an analyst at Susquehanna Financial, wrote in a client note quoted by EETimes. Hosseini said that ecosystem checks suggested Samsung had suffered a setback with its 14nm FinFET project, while TSMC’s 16nm pilot line was going smoothly and volume commercial production could start a quarter earlier than expected, in Q215. Early customers are likely to be HiSilicon, China’s largest fabless chip provider, Apple, Qualcomm, Mediatek and, later, Xilinx and Altera. Samsung also makes chips for Apple and Qualcomm but it may lag its rival because of the “steep learning curve” associated with FinFET.