Intel (INTC) is concerned about ARM's (ARMH) dominance in low-power processors, which is quite justified. Intel needs ARM's market for running its fabs at full capacity. "We're breaking the myth that ARM can do things that Intel cannot," said Intel's Dadi Perlmutter, when the company introduced Silvermont earlier this month. In this article I will analyze whether Intel can really challenge ARM's low-power lead with its next generation Atoms.
Intel Needs ARM's Market
To produce chips at the lowest possible cost, Intel needs to run its fabs at full capacity. With shifting production to 450mm silicon wafers and shrinking nodes below 22nm, Intel can produce 40% more processors per wafer at full capacity.
Unless there are huge demand for Intel's processors, Intel won't be able to continue production at full capacity. Carrying on production without utilizing full capacity will result in increasing cost. That's why Intel desperately needs ARM's market. You can read this article for more details.
Starting with Silvermont, the new version of Atom, Intel expects it can gradually grab ARM's market. It's a move, according to Intel officials, that will introduce performance and power consumption advantages that can significantly exceed anything ARM and its partners can offer.
According to Perlmutter, Atom SoCs with the Silvermont architecture will offer five times lower power consumption and three times more performance than current Atom SoCs.
Intel's move to 22nm FinFET SoC process using 3D transistors will ensure this high performance. ARM will introduce its 14nm SoCs based on FinFETs in 2014. Investors need to properly judge the true potential of ARM's forthcoming chips with FinFETs, because this will be crucial for Intel's stock.
ARM's 14nm FinFETs Coming
"We're supporting all the independent foundries," said Warren East, CEO of ARM. That includes 20nm planar bulk CMOS and 16nm FinFET at TSMC (TSM); 20nm planar bulk CMOS and 14nm FinFET at Samsung (GM:SSNLF) and 20nm planar bulk CMOS, 20nm FD-SOI and 14nm FinFET at Globalfoundries.
ARM is not just developing a formidable range of processors, their process roadmap is also on track to intersect that of Intel's at 14nm. "I'm no better equipped to judge which of these processes will be more successful than anyone else," says East, "our approach is to be process agnostic."
ARM believes 14nm FinFET process technology will help chip designers to boost performance in cases when it is needed most as well as to cut-down power consumption in idle cases. Ron Moore, director of strategic accounts marketing at ARM's physical IP division, said:
ARM partners design their SoCs with dynamic voltage and frequency scaling (DVFS) to achieve peak performance when needed and power efficiency for lower workload tasks. The lower nominal operating voltage of 14nm FinFET expands the voltage ranges in both underdrive and overdrive conditions. In an ARM Big.Little processing system, we can achieve higher performance on the top end of the range as well as higher efficiency in the middle and lower operating ranges.
It's highly probable that Intel's 22nm FinFETs aren't optimized for mobile SoCs and Intel must have saved some of the tweaks for 14nm, because the real battle for mobility will start at 14nm.
"Common Platform" Working Hard in Developing 14nm FinFETs
Chip making at 14nm and below is not an easy job. The Common Platform Technology Forum, originally set up by IBM (IBM), is an initiative to jointly work for developing processors with smaller geometries. IBM distributes its chip-making technologies to the partners of the group, such as Globalfoundries and Samsung.
Development of 14nm FinFET process technology under the Common Platform version combines FinFET technology on the "front-end" with the same "back-end" as its 20nm process. Earlier this year the members of Common Platform attended a meeting where representatives of both Globalfoundries and Samsung talked about how they were meeting the challenges of moving to 14nm and FinFETs.
Mike Noonen, executive vice president of marketing, sales, quality and design for Globalfoundries said Globalfoundries expects to have early 14nm production this year. Full production of the 14XM process, which uses 14nm FinFETs with a more cost effective back-end, will begin in the first half of 2014. He also said that a dual-core ARM Cortex-A9 shows 62% power reduction or 61% performance improvement on 14XM compared with the foundry's 28SLP process.
K.H. Kim, executive VP of Samsung Electronics, who heads Samsung's foundry operations, said that Samsung has created the first Cortex-A7 designs with FinFETs, and is ready to offer 14nm FinFETs to its customers. He also said that 2013 is mainly a year for validation and design with full production coming next year.
How will Intel's 14nm FinFETs stack up against ARM's?
That's very difficult to answer. But don't forget that Intel is already shipping chips based on FinFETs, although at 22nm. The 14nm version of Silvermont will be known as Airmont. It's noteworthy that starting with Silvermont, Atom will also follow Intel's tick-tock cadence.
While it's true that Intel has huge experience with 22nm FinFETs in mass production, it's also true that 14nm will be the first process at which Intel intends to put mobile SoCs to the front of the node i.e. putting them among the first ICs to be made on a new process. So it won't be wise to assume that Intel's efforts will be more efficient than that of ARM and its partners for winning the mobile battle. Given ARM's impressive track record, I expect ARM's 14nm FinFETs to be quite good.
If Intel fails in mobile, Intel's stock will be under tremendous pressure. Intel has invested a lot of money in its fabs while upgrading from 300mm to 450mm silicon wafers. Intel will not only lose money, it will also lose credibility.
Let's be positive and realistic. In the worst-case scenario for Intel, I expect Intel won't be able to grab ARM's maximum possible market share. Intel will still grow significantly in the low-power processing game. The smartphones and tablets market is growing fast with huge untapped opportunities in the developing nations. Intel can't fail in this market, although the extent of success can be lower than expected. Buying some ARM stocks as a hedge against Intel longs won't be a bad idea.