The previous PCMO scaling issues, from what I gather, were...

The previous PCMO scaling issues, from what I gather, were associated with filamentary PCMO - they couldn't scale the cells once you reach a minimum size with the filament (the filament begins to cause stray programming of the cell materials).

non filamentary unlocks the scaling issues. What is the lowest 4DS geometry - who knows.

but it's good to remember while dram can scale lower, it seems to be most commercially viable at 14nm and fabs may not tool up for lower geometries, instead looking at 3D stacking... So smaller isn't always better (or at least not commercially).

if 20nm is proven here, 4DS are already using methods that are fab ready, and thus is likely very quickly will be commercially viable. there may not be an immediate need to go smaller straight away. The other benefits of the memory may outweigh the cell size, which they can tackle later.