A group of physicists (Naumov, Bellaiche & Fu) at the Univ. of Arkansas are reporting in Nature the exploration of phase transformations in non-volatile ferroelectric memories composed of Pb(Zr,TI)O3. They are finding that the minimum size of nanodiscs/nanorods able to retain a bistable state is of the order of 3.2 nm. That translates to 60×1012 bits/sq. inch (7+ terabytes/sq. in).
Obviously there are a couple of problems here. The first would be how do you read or write a nanorod every 3.2 nm? Of course this may drive the need for improved nanowiring and/or nanomechanical head positioning. The second would be the claim regarding "low-temperature structural bistability". How low does the temperature actually need to be? If one needs a room full of refrigeration equipment to allow a nanoscale memory to retain its data then the applications may be somewhat limited. E.g. you aren't going to see it in an iPod.