A big problem with this is reducing the contact area increases the voltage from larger contact resistance. Also, isn't it standard practice to have design one self-assembled contact hole inside the TiN contact?
A correction and an apology. My quote of current densities of the order 10E8 amps/sq-cm reproduced above was based on my estimate for a sub 20nm device and the reading of a less than clear version of the original paper. I have been in contact with the authors who have drawn my attention to the supplementary information provided with their paper.
From their simulation of a 50 x 50 nm device the current density without the oxide balls is 1.8 x 10E7 A/sq-cm; which is in good agreement with similar sized devices reported elsewhere. More importantly, and again from their simulation, with the oxide balls covering 50% of the area between the GST-TiN heater interface the authors estimate a reset current density of 2.8 x10E6 A/sq-cm. A value that betters by about an order of magnitude any reported so far by any other PCM workers for devices of similar dimensions. Assuming those operating conditions produce reliable overall performance especially with respect to elevated temperature data retention and write/erase lifetime the KAIST paper might mark a significant step forward in PCM development.
The next, and non-trivial step, is reduction to practice with real devices and then to put PCM back into a competitive position, devices scaled to sub 20nm dimensions. For PCM the curve of reset current density J = f(dimensions) follow the general form of those for the ratio of surface area to volume. This would suggest that sub 20nm PCM devices without the oxide balls would have current densities close to 1x10E8 amps and given a similar gain reported above with the inclusion of the balls would result in a device with a reset current density of the order 1 x10E7 A/sq-cm. Much depends on the minimum possible diameter balls and the ability to reproduce very small diameter oxide balls in sub 20nm apertures. A significant challenge remains.
Blog Doing Math in FPGAs Tom Burke 15 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...