TSMCs 20nm fabrication process has always been a strange one. Typically foundries offer several versions of manufacturing technologies aimed at different application: high-performance-chips, low-power chips, low-cost chips, etc. In case of 20nm TSMC only offers only one version of the process aimed at everything from low-cost system-on-chips for ultra-portable devices (such as smartphones) to high-performance processors (such as GPUs). Since there are reports about low 20nm yields at TSMC, high prices and limited production capacities available to AMD and Nvidia, it may make sense not to use this fabrication technology and jump directly to 16nm, which is set to be available for commercial chips in early 2015.
While the 16nm FinFET process technology makes more sense to use in early 2015 than the 20nm manufacturing technology, there is another reason why Nvidia could be more interested in this technology. TSMC has already developed 16nm FinFET+ fabrication process that will be utilised to make chips in volume later in 2015. In fact, TSMC expects the majority of its customers to use the 16nm FinFET+ version of the process due to its noticeable benefits compared to the original. It is interesting to note that there is a rumour about yet another 16nm process technology from TSMC called 16nm FinFET turbo. So far TSMC has not confirmed the existence of the third 16nm FinFET process.
Design rules of the 16nm FinFET and the 16nm FinFET+ are the same, everything is largely compatible. In general, 16nm FinFET designs can easily be produced on the 16nm FinFET+ node with minimal changes (if needed at all). The plus process is expected be fully qualified by September. TSMCs fabs compatible with the companys 20nm process technology are 95 per cent ready to produce chips using various versions of the companys 16nm manufacturing technology, the company said earlier this year/
According to TSMC, 16nm FinFET+ provides up to 15 per cent performance improvement over the 16nm FinFET at the same level of power consumption. At the same clock-rate, chips produced using 16nm FinFET+ are expected to consume 30 per cent less power compared to the same chips made using 16nm FinFET.
In general, once Nvidia designs a chip for the 16nm node, it can speed it up using either the 16nm FinFET+ technology or the rumoured 16nm FinFET turbo technology, which means that the company can easily refresh its 16nm product line in 2016 by spending almost nothing on the development.