TSMC today announced that it has finally entered risk production on the 16FF+ node. The 16nm FinFET Plus is basically a much more enhanced version of the normal FinFET variant and is approximately 40% faster than the 20nm (SOC) Node, while consuming 50% less power for much greater energy efficiency. Even more interestingly, Nvidia has given an official statement on the matter and talked about using the node in their GPUs.
Nvidia talks 16nm FinFET Plus GPU technology as TSMC reveals 16FF+ risk production
Perhaps the most interesting thing that caught my eye is the fact that TSMC is talking about recently ramped 20nm Planar SOCs, where SOC is the operating keyword. Some would take this as a conclusive hint that no high performance node on 20nm Planar exists, which would mean no 20nm GPUs. TSMC stated that applications of the 16FF+ node will include high end mobile, computing, networking and consumer applications so high end GPUs on this particular node are pretty much a guarantee.
"Our successful ramp-up in 20SoC has blazed a trail for 16FF and 16FF+, allowing us to rapidly offer a highly competitive technology to achieve maximum value for customers' products,", TSMC President and Co-CEO, Dr. Mark Liu said, "We believe this new process can provide our customers the right balance between performance and cost so they can best meet their design requirements and time-to-market goals."
Since we are already talking about the node, let me detail a little on the slightly less interesting ARM applications of the same. 16FF+ can scale the existing ARM 'big' Cortex-A57 core, while allowing it to reach clocks of around 2.3 Ghz. Similarly it can scale the Cortex A53 'LITTLE' core as well, with focus on lowered power consumption (an astonishingly low 75mW in this case). According to the official press release, yield projections and learning curves are excellent. If this is true then this is pretty promising for the next generation of high end ASICs. The process will pass full reliability qualification in November with 60 or so customer designs ready to tape out by the end of 2015. TSMC projects that 16FF+ volume production will begin in July 2015 due to excellent yield projections and node maturity
"NVIDIA and TSMC have collaborated for more than 15 years to deliver complex GPU architectures on state-of-the-art process nodes,.......Our partnership has delivered well over a billion GPUs that are deployed in everything from automobiles to supercomputers. Through working together on the next-generation 16nm FinFET process, we look forward to delivering industry-leading performance and power efficiency with future GPUs and SOCs." - Jeff Fisher, Nvidia
With this latest update from TSMC, I am personally starting to lean towards no 20nm GPUs. It just wouldn't make sense for companies like Nvidia to adopt 20nm just a few months before 16FF+ starts volume production. What would make sense is for Nvidia to have an intermediary core (GM200 anyone?) and then ramp directly to 16nm FinFET Plus. As for AMD, their PR department is currently angling for the 4K Performance fight, where the R9 290X thanks to its 512 bit bus can hold its own against the Maxwell flagships. However, unlike Nvidia, it does make financial sense for AMD to go 20nm. But the question is, does such a node even exist?