What's sure is that they won't go further than GPU manufacturers, so if those only passed the 14nm mark, we'll wait a long time before we see 10nm or so.
but a better 14-16nm is possible.
I think maybe an improved s9+ next spring or an s11 but either more of the same chips on a lower clock. or a better chip that is still 14-16nm
to see a 10nm maybe 2020 at best.
I can see the possibility of somewhat better than the current BM1387 being possible, but I doubt it's going to be a major improvement like Bitmain pulled off on the 28nm generation.
30-50% better efficiency should be possible, but given that the 14/16 nm node chips aren't running at much lower voltages than 28nm was capable of, there's some serious limit on how much more efficiency is available to achieve.
We've already had at least one early 14/16nm design that proved to be little or no better than the best 28nm designs (LK-1401, the apparently-abandoned Innosilicon A3 might be another given the final efficiency ratings on the A4) and BW already claimed some major improvements going to the LK-1402.
Under 0.10 watts/GH? Possibly. Under 0.06? Looking MIGHTY iffy at best, even the Bitfury demo chip didn't manage that at it's lowest power setting.
I'm a bit supprised we haven't seen SOI move down to the newer node range before now - did I miss something, or did SoI completely skip the 22nm genration?