TSMC 20nm will offer limited performance improvements over TSMC 28nm, and a higher cost/transistor. Also note that AAPL and likely QCOM will be hogging up the initial capacity for quite a while
No, 28nm is the right choice for a budget part until Intel can move it in-house. Then, on a cost/transistor basis (and performance/watt basis), it makes more sense to build on Intel 14nm.
Why is that ironic? By the time 20nm is out, there will be 14nm Atom parts. Also there are NO 20nm parts coming soon. Q1 is ramp, so we're expecting first 20nm products at Q3 at the earliest.
Also, the focus on density is not just on process, but on design. The Silvermont core scaled better than 22nm density indicates.
Correct me if I am wrong but I don't know of any ARM shop that is projecting any volume shipments of any ARM 20nm processor products in 2014.
20nm has not yet arrived. There are no volume shipments. We have no idea how it will ramp. And when it does finally arrive all of the high-end ARM shops are going to be competing for scarce capacity.
We don't even know what the cost will be.
Why on earth would Intel want to deal with any of that uncertainty?