ARM and TSMC are gearing up to shift to a 10nm manufacturing process for 64-bit ARMv8-A SoCs in 2015, a move which should improve performance and energy efficiency compared to current-generation chips. The announcement of the shift to 10nm continues ARM and TSMC's manufacturing partnership, which has been strong for many years.
Specifically, the two companies are looking to bring their FinFET technology down to the 10nm node. FinFET transistors are what other companies (such as Intel) call 3D or tri-gate transistors, which perform better than their standard counterparts. TSMC has already had success producing FinFET transistors on a 16nm manufacturing node, and they're hoping the success will continue in the shift to 10nm.
In fact just recently, ARM and TSMC announced that they had managed to fabricate a 16nm FinFET SoC using ARM's big.LITTLE multi-core architecture. The chip joined together 64-bit Cortex-A57 CPU cores with Cortex-A53s, acheiving solid performance at 2.3 GHz on the A57s, plus power consumption as low as 75 mW on the low-power A53s during "common workloads".
Most current ARM-based mobile SoCs are manufactured on a 28nm process, with some of the newest chips - such as Samsung's bleeding-edge Exynos SoCs - having shifted to the smaller 20nm node. TSMC has previously suggested that performance gains could be somewhere around 40% moving from 28nm to 16nm FinFET, meaning we can expect an even greater jump when the technology is scaled down to 10nm.
The TSMC/ARM roadmap points to 10nm FinFET designs taping out in Q4 2015, which indicates these chips will not hit the market until 2016 at the earliest, possibly even 2017. We're currently still transitioning to 20nm in available products, with 16nm chips likely to hit the market in 2015 after TSMC successfully taped-out Cortex-A57 CPUs on the smaller node in early 2013.
Meanwhile, Intel will jump the gun and launch 14nm Broadwell products at the end of this year. This is typical for Intel, whose advanced fabrication technology allows them to release chips on smaller nodes on a consistently earlier basis.