This is only as you near the theoretical limit of density, though, right? At easy-to-achieve densities that we would be using in the foreseeable future, if we p

This is only as you near the theoretical limit of density, though, right? At easy-to-achieve densities that we would be using in the foreseeable future, if we plotted their access times in these different configurations, wouldn't it--practically speaking--still scale at the cube root? I guess that was covered in the comment. > This applies even without hitting theoretical information density limits, whatever your best process size and latency. But now I guess I don't understand why... :(. I feel like we can almost trivially show this isn't true by starting with a downright primitive "best process size"--maybe a system where every bit is so large I can see it and access it using a pair of tweezers--but maybe I am fooling myself?