Yeah on mobile screen and radio dominates the compute part by quite a bit. Not that it's not still worth optimizing.

Yes, It is just lower prioritises.

100% of the Mobile Phone Energy Usage during Internet Browsing, 50% goes to Display, 30% goes to Radio Transmission and 5G Encode / Decoding. Your Three SoC + NAND + Memory only uses 20%, I just think title is a little misleading.

Not to mention the Software we are running could be anywhere between 2 to 10x from maximum efficiency. It is just the human cost involves to get to the point might not be worth it.

There is also Moore's Law, while we might be near or already at the end of it, we still have few more nodes to go and could reduce the energy usage of SoC + NAND + RAM by 30% to 50%.

While on the Display and Radio is isn't so clear how far we could further reduces it.

Screen and radio are both data movement tools, so I don't understand the OP's classification process.

The next step after in-memory computing: in-screen computing!

Maybe not computing, but cathode ray tubes were used as RAM in the forties: https://www.radiomuseum.org/forum/williams_kilburn_williams_...

We're already moving towards self refreshing panels and the like. And it's not a stretch to imagine in-screen cursor movement.

I agree, sorry. The original paper used the phrase "total energy" but although the paper text does mention powering the screen, their figures just talk about computational energy, so that is likely what they meant. It's too late for me to edit the title, but it should be "62.7% of computational energy".

Ok, we've clarified that in the title above.