I marked the last graph up to better illustrate what happened:
Here I show a red line that is approximately parallel to the magenta line: i.e. the area scaling for Intel from 22nm to 14nm is about the same as for TSMC from 28nm to 20nm. Thus as Intel admitted, since their 22nm was about the same at TSMC's 28nm, so now Intel's 14nm is about the same as TSMC's 20nm (for the metric of area density only!)
When TSMC (and Samsung) go to 16/14nm, they will in fact also pick up some additional area scaling, plus of course FinFET transistors. The further significant factor is that while Intel will have 14nm in products next Qtr, the Foundries will have them about 6-9 months later. Apple will likely have the A9 next year in 14nm from Samsung, and that will actually be prior to Intel have an integrated smartphone SoC out in their own 14nm process, so the Foundries will have essentially closed the process gap with Intel, as far as mobile goes. This has largely been driven by the intense competition for Apple's huge SoC volumes, which has pitted Samsung and TSMC in a fierce battle, and has provided them with a vehicle for rapidly ramping a leading edge node, something which has never existed for the Foundries prior to Apple's creation of the mobile segment.
TSMC recently noted that 20nm is the fastest ramp they have ever had (courtesy of Apple). They also basically acknowledged that they will lag the next node vs their competitor, which has lead to the general conclusion that Samsung will get the bulk of A9 next year for Apple. |
