Moving the HBM PHY and DDR PHYs to the HBM base dies is crazy and then in the very next slide they propose logic too. The parallelism you can gain by hooking this up to SIMD running on the HBM logic elements blows my mind.

See also recent Asianometry video "SK hynix and the HBM Revolution":

* https://www.youtube.com/watch?v=Cg5tAujp6Go

And from a few years ago "The Special Memory Powering the AI Revolution":

* https://www.youtube.com/watch?v=yAw63F1W_Us

Great resource, thank you.

[dead]