late 2015 will probably be closing in on 0.1 J/GH.
How? Is CT's 16nm design, when downclocked, 3 times more efficient? I doubt that. So how is someone gonna archive 0.1J/GH on 16nm within one year?
I personally think that it will be easier to optimise farm locations, rather than hardware from say 0.25J/GH to 0.15J/GH.