TSMC Begins Volume Production of 16nm FinFET Process – Nvidia Pascal GP100 GPU Among the Products in Production
The TSMC 16nm FinFET node is probably the most notable process, that is of interest to PC enthusiasts. This is the node that will house Nvidia's next generation lineup of graphic cards (specifically the "16FF+" variant) and is one of the most authentic indicators of their time-frame. Taipei Times, in accordance with everything we heard in the past, has confirmed that TSMC has (finally) started mass production of 16nm FinFET products. However, It is expected that the initial run will be dedicated for Apple SoCs.
TSMC starts mass production of products on the 16nm FinFET node
We had previously reported some months ago that TSMC will enter into volume production in the time-range of Q3 2015, specifically July, so this news doesn't really come as much of a surprise. TSMC and Nvidia have also confirmed on more than one occasion that the next generation (Pascal) GPUs will be produced on the 16nm FinFET+ node, with initial confirmation dating back approximately 9 months. AMD's next generation Radeon graphics processor on the other hand, codenamed Arctic Islands, was not on the official list of products released by TSMC, so while their CEO have confirmed the use of a FinFET node (14/16) the exact specifics remain to be seen.
16nm FinFET tech entered into risk production and approached mature yields a while back, and now full fledged production has begun full steam ahead. More than 60 projects are underway, with known products in development including Avago, Freescale, LG, MediaTek, NVIDIA, Renesas and Xilinx. The list is obviously not exhaustive in nature but like i mentioned above, the initial short list did not include AMD. I must admit however that this could have changed in the meantime since the press release is fairly dated.
As I mentioned a long time back, GM200 is an intermediary product with Pascal GP100 graphics processor finally ushering in the era of the sub 28nm utopia. With the initial production being more or less fully utilized by Apple, we should see the Pascal GP100 GPU in Q1 2016 by the earliest, with Q2 being the more conservative estimate. 2016 is also the date that AMD has confirmed products involving their anxiously anticipated Zen micro-architecture and Arctic Islands GPU on a FinFET node. Needless to say, it is gearing up to be a pretty interesting year.
TSMC’s 16FF+ (FinFET Plus) technology can provide above 65 percent higher speed, around 2 times the density, or 70 percent less power than its 28HPM technology. Comparing with 20SoC technology, 16FF+ provides extra 40% higher speed and 60% power saving. By leveraging the experience of 20SoC technology, TSMC 16FF+ shares the same metal backend process in order to quickly improve yield and demonstrate process maturity for time-to-market value.
The Pascal GP100 GPU allegedly has 17 Billion transistors and around 32 GB of HBM2 based vRAM. If these reports are correct, then the GP100 will be an absolute beast in terms of graphics processing power. In comparison, the current Nvidia flagship, the TITAN-X has 8 Billion transistors, so we are looking at an alleged jump of more than double. There are however, two major problems with the theory. We have about 1 Billion transistors more unaccounted for, and we are assuming a die size equal to that of the TITAN-X GM200 (601mm^2) which frankly speaking is just not going to happen on a brand new node. Not to mention that Nvidia will want to keep leg room for future Pascal variants on the same node.
No, a die size of 601mm^2 is very unplausible. Assuming the report is true, I believe (caution: speculation) we might be looking at a combined transistor count. In the meantime however, there is finally cause to rejoice. The era of the 28nm GPU is over, and the age of sub 20nm FinFETs has begun.
Stay in the loop
GET A DAILY DIGEST OF LATEST TECHNOLOGY NEWS
Straight to your inbox
Subscribe to our newsletter