The immediate fall out from the announcement was an assessment of 'who leads who' in process technology – 'my fab is better than your fab', if you like. Despite the huge investments needed and the enormous technical challenges, all four main players – TSMC, Globalfoundries, Samsung and Intel – are pushing to get leading edge processes into production as soon as possible, mostly driven by the needs of smartphone companies. However, while Intel seems to be stuck for the moment at 14nm, Samsung says it's going straight to 10nm shortly. TSMC also has 10nm plans, but ambitions for 7nm; some say it's possible it might try to do 10nm and 7nm in the same year.
The rewards are high: Apple and Qualcomm, for example, need leading edge processes to support the low power consumption and higher performance aspirations for their various smartphone chips. Winning their business is important for the foundries.
As many commentators have pointed out, there's a big jump between making a 7nm test chip and that technology entering volume production. One of the major hurdles is to get extreme ultraviolet (EUV) lithography to the point where it makes commercial sense. While IBM has used EUV for its test chip, none of the systems or research milestones so far announced hit the mark – source power is still too low, which means throughput is also too low.
EUV remains the great hope for semiconductor manufacturers: the problems have been known for some years, but the technology remains elusive. Because of this, manufacturers have developed double, triple – and now quadruple – patterning as another way of making 10nm features from light with a wavelength of 193nm. But multiple patterning means more process steps and longer to get a wafer through the fab. Intel's 14nm process is said to have about 70 steps; a few more steps for additional patterning may not seem like much, but they introduce further chance for error.
While the likes of Apple and Qualcomm have a demonstrable need for 7nm and smaller, the rest of the industry isn't so keen. Unless you have the volumes provided by the likes of the iPhone or other smartphones, then leading edge processes will not make sense. Now, many designs are being targeted at 28nm processes, which have rapidly become cost effective, and commentators would not be surprised to see that node being used for a long time to come.
