ARM promised us the first range of ARM64 products by 2015 but nothing seems to be ready why not.

Well you have to go back a good few years where AMD ran into production problems when they went down to 28nm this meant that now you still have AMD working with 28NM and Intel rocketting ahead at 14NM.
Intel use an older wafer technology but have made many improvments like using Hafnium and introducing tri-gates.
AMD thought their SOI wafer fab would get them further and got stuck at 28NM awaiting fin fet at lower NMs

This all lead to ARM companies having to wait for 16NM fin fet technologies for their ARM64 products to be more viable as they've had no end of trouble with SOI 28NM

So this all means that by early 2016 your ARM64 products will finally start to arrive and by 2017 should be abundant.

By late 2016 early 2017 Intel plan to go down to 11NM will Intel manage it.
Well they are still relient on old wafter technology so they might run into similar problems with Overheating and Quantum leakage down at 11 and below who knows Intel may have to as they've suggested start using fully depleted SOI.

Will ARM go down further by say 2018 - 2019 If they use fully depleted SOI and better Nano wiring they should do ok they may not have to for 14nm but probably will by 11nm or below.

So anyway don't worry by early next year ARM64 will more than lightly be the best thing since sliced bread.

What might all this mean 11nm and below.
Well this depends on progress in nanotechnology.

The faster processor option.
One senario is where they jump straight down to 5 nm and use tiny thermo optic pins at higher speed. In this case senario you could fit 2 small ARM64 cores on a tiny die and rev up the speed to about 8 - 12 GHz compensate for the low transistor count.
+ more speed for many calculations and the ability to use multiple pins on an optical grid, Long term production cost could be far lower than today.
- less CPU complexity means a reduction in the instruction set meaning that many usful extensions would have to be reduced or excluded. Going this small might mean even excluding backwards compatability and using emulation instead for 32bit integer coding. Also intial R and D costs might be higher.

The Layered processor option.
Years ago intel produced a 2 layed Pentium example but the idea at the time was not very cost effective for a CPU like it is now for DRAM and Flash Memory.
down at 11nm and below this option might be revisited.
+ you could have more cores even use small Cellular computing as opposed to the standard von nueman architecture.
- you might suffer thermal issues which might leave your first chips only running at 1GHz or less and possibily higher long term production costs.

The standard course option.
Intel Chips havent gone up as much on the GHz as AMD have using SOI or Fully Depleted SOI finfet technology might get an 8nm chip close to 5.5 or 7 GHz on a chip with 28 cores.
+ More estabilish than the other 2 options, probably speed benefits in the short term at least compared to the pin option.
- Pin technology can be spread accross an optical bus avoiding thermal issues, you can have as many cells on the standard tech as you can on the layed approach.

So where are we heading in the 20's.
My guess is that although cellular technologies are ready to make use of layered chips and although the standard CPU system is well estabilished the benefits eventually from precion engineering at a such a small scale on the pins will bare good fruit by at least 2025 if not before and the temtation to perfect such nano technology will tempt engineering down this path. However memory might lag behind but if you think all watson needed was a few gigs of data to play jepordy but masses of processing power this issue might not matter to much only with professional video and such even games could contain far more detail with less data given enough processing power.

Overall in the short term its cheaper to have more slower cells than it is to have less faster cells and such faster cell technology aimed at compensating for the low cell count with faster processing hasen't been developed yet. However RISCy VonNueman Architectuere still has at least some milage on it.