Wasn't baytrail 6watts and now cherry trail is 2 watts. it seems to have increased in speed while they were at it
Luckily there was some gains while moving from 22nm to 14nm.
Now it seems there are no significant architectural change and performances were more an issue than power. Actually Intel SOC were doing quite well under sustained load (and I guess even better now).
The issue is from their last 22nm SOC (embarking PowerVR GPU) I suspect that the user experience is mostly the same.
Things could get improved, multi-core scaling ain't that great from 2 to 4 cores, SIMD could see an update, Intel could now afford a slightly wider cores, etc.
At the other hand of the spectrum there are those low power Broadwell cores. They can ( and do) throttle a lot but Intel pushes the top clocks pretty high, yet their performances are unmatched by competing designs.
So I see 2 options Intel should have made better that those Cherry Trail (so improve on Silvermont micro architecture) or simply end that line of CPU:
I know it sounds crazy BUT they do have now others CPU for extremely low power operations AND their main line (unmatched) CPUs are getting a reasonable option (not to mention that if one wants Windows... sadly AMD offering is lacking... so no choice).
Core M are pretty impressive yet they are a tiny package 81mm2 iirc, there are no SOC though. In my opinion to really make a strong showing they have design a SOC based on their main line CPU (so Broadwell) something like that:
High-end: 1 Broadwell core, 2MB of L3, 16EU GPU, a 64 bit memory interface (single channel) supporting DDR4.
Intel "needs" to create further segmentation with its higher end Core M line (generating higher margins) I believe they should have done it through any combination of the options below (as usual I would add):
1) significantly lower max turbo clock speed (north of 2GHz instead of south of 3GHz).
2) Varying level of of SIMD support (/crippling).
3) On/Off X86_64 support.
4) On/Off SMT support.
5) On/Off DDR4 support