GOING BELOW 20mnTruth vs Vapor (time to tell the ugly truth about it)
Intel always seems to want to sell vapor for at least a year before designing and prototyping anything -- which is why their chip names change so many times before something actually ships.
Vapor pushing is how Intel does their product planning after all .... only every third one actually makes it to reality.
The ARM world has been prototyping FinFeet 16nm traces for a while now trying to come up with a production capable decent yielding process. Both TSMC and Global Foundaries can do it, but they are being honest about what they propose to do in order to make it really workable for processing cost and scrap wise.
Global Foundaries is the less capable of the pair of ARM foundaries and this is what GF says they can do for 16nm using double pass (or perhaps triple pass) EUV lithography which will cost like 2-3 times as much processing time to do a wafer. The scrap rate will be 3-4 times higher than simple 20nm flat lithography which is where they are today.
Here is the GF ARM tricks laid out in all its glory -- first off all the flat straight traces are simple flat 20nm wide traces to provide for decent current carrying ability without making the layer too thick -- then changing over to 14-16nm FinFeet fins on the curves and transistors to provide more area/current capacity and cut down on the electrical losses that takes place in every curve/transistor.
Why just paste in tighter FinFeet corners and transistor bridges on to a standard 20mn 5 volt design and pay for all that extra processing time and scrap rate? You'd only get like a 10-15% boost in speed and practically no extra energy savings for all the extra passes and doping and effort. It would still be a 5 volt chip.
So why do it? Answer is they really aren't going to -- this is why the ARM world shows a flat section between 20nm and 16nm because there isn't really real overall die shrink there to be had.
(20nm flat traces take up the same room in both processes and the die size stays the same).
The ARM world will stay unchanged on the die size between 20nm and 16nm. So, they will actually
most likely make the hop from 20mn all the way down to 14nm/10mn as the next step down.
Why? With simple long straight 14nm flat traces and 10nm FinFeet curves/transistors that make some processing sense this is the downsize that
actually yields enough improvement to make the trip worth while.
14nm/10mn will yield a 40-50% speed increase and a solid 25-30% efficiency boost for doing it. The chip's guts will run at 2.5 volts instead of 5 volts, so a lot of other stuff needs to be redesigned at this same point in time (lots of vendor involvement in batteries, screens, etc. etc.)
Intel is still caught up in all their last 3 years of vapor pushing and outright lying -- so they will delay and fiddle and most likely make the same larger leap down, but still only to 14nm/10nm when they do go.
They might do it a year (or part of a year) before ARM actually does it, assuming they can disperse their own fumes well enough to see that far out (and assuming all the 2.5 volt vendor bits and pieces are ready to go that early).
Now that ARM is saying 14nm flats and 10nm FinFeets make some pretty good processing sense to them, likely Intel will do something very similar whenever their vapor fog bank lifts and they actually go do it.
Now, am I really saying that Intel is really starting just to copy the ARM world's ideas and
really are just trying to do it a little quicker to maintain their corporate image as the great inventor?
Yep. Look at their Merrifield chipset (which will really actually happen long long before Broxton ever becomes real) -- it
ACTUALLY IS an ARM standard design (using ARM standard vendors) except Intel is plugging in some Intel customized CPU cores in place of the stock ARM CPU cores.
Intel's cost will likely be still be significantly higher because they are carrying WAY too much USA overhead (lots of people, really big salaries).
So they will still likely have to loss leader the new chip in order to make it "price competitive".
.... and since it actually is the similar same, is it really going to run any better?