Sadly Bulldozer has nothing else to do...
Micro-op queue, 2 BTB branch prediction(branch calculated per core queue), CX/AX; complex execution(iMUL/iDIV/CRC/Branch+ALUs) and address execution(AGUs+ALUs), more flexible FPU, L0 caches, etc. However, most of that stuff is more ideal on 12FDX which is to 14nm FinFET as 22FDX is to 28nm.
not even 25% more performance by node shrink can save it.
The 22FDX design wouldn't be a 25% more performance node. It would be closer to a 50% or higher frequency node. The 22FDX node re-uses the 20FD/14FD node transistor, which is faster than 20BLK strained(20SHP to GlobalFoundries). What isn't gained in density is gained in absolute and nominal performance which directly translates to frequency.
1V is the safest voltage to expect a 50% increase and ~0.75V with a ~75% increase at minimum;
A9-9420e => 2.7 GHz @ 1V || +50% = 4.05 GHz // 1.8 +75% = 3.15 GHz
A6-9220e => 2.4 GHz @ 1V || +50% = 3.6 GHz // 1.6 +75% = 2.8 GHz
A4-9120e => 2.2 GHz @ 1V || +50% = 3.3 GHz // 1.5 +75% = 2.63 GHz
^This boost is achievable through a minimally redesigned port of 28HPA/9T/114CPP/90Mx to 22FDX/8T/104CPP/80Mx. The lower variability of 22FDX over 28BLK especially strained BLK(28SHP/28A/28HPA) can allow higher frequency at lower power do to a better product sign-off.
The biggest gain 22FDX brings over is a more scalable LVT and a much faster RVT transistor. Which is important for the CPU and GPU. I think the UVD/VCE would also benefit from the opposite with RVT being less leaky at same speed.
However, it only works if AMD gives us a Family 15h + Graphic Core Next + UVD/VCE SoC on 22FDX in the frame of Stoney. If AMD decides to do post-Family 15h(still CMT) + Single-CU WGP RDNA(still 3 CU) + VCN, somewhat like Kaveri(Stoney) to Carrizo(ex: Ernst). Then, some of it has to be reworked to the potential differences in Vt selection, ratio-ed channel lengths, different work units, etc.
However, when has AMD ever not hoisted up a GloFo node themselves. We obviously should be looking at a post-Family 15h w/ (single-cu ver)RDNA+VCN2 /etc. on 12FDX. With an absurd die size of less than 100 mm squared; between 50mm2 and 100mm2. For a price-sensitive, cost-effective, power-optimized, tiny high-performance APU SoC for the pocketbook market, or whatever. (GPD/Chuwi/etc but with AMD as the product focus point and below $399, instead of these $799 intel monstrosities.)
====
Edit, Sep 17;
With the
http://www.rock-chips.com/a/en/products/RK18_Series/2019/0529/989.html
Which is on GlobalFoundries' 22FDX, at any month can the AMD Stoney revision feasibly launch.
CES2019 - Rockchip, a leading Chinese semiconductor company, today released the RK1808, an AIoT solution with built-in high performance NPU. In terms
www.businesswire.com
Which no one told me popped up around the same time period as Synpatics;
CES LAS VEGAS, SAN JOSE, Calif.
www.synaptics.com
Oof~
Edit: Octo. 17
2020 is looking like another Stoney Refresh on 28nm.