IMEC and Panasonic presented a paper at the recent VLSI conference along with a corresponding press release which is the source of most of what I know. “Highly Reliable TaOx ReRAM with Centralized Filament for 28-nm Embedded Application” by Hayakawa, A. et al. (Panasonic) The premise seems exciting in that the authors state that they have developed a method for keeping the filament formed in their Tantalum Oxide based ReRAM cells in the center of the cell. I’ve no idea how they have done this and even less of an idea how they verified the success of their technology. Nonetheless this is an important advance as it must improve cell to cell reliability and resistance uniformity.
Most field induced diffusion processes tend to increase exponentially (or faster) with the applied field and thus seek out the weakest, i.e. shortest path through a material. The approach taken by Panasonic/IMEC appears to be a subtractive process where the ~100nm diameter cell (between M2 and M3) is etched from a deposited stack of thin films of the bottom electrode, switchable oxide and possibly top electrode although the top electrode could be deposited after the cell definition. One of the problems of this approach is that the films follow the morphology of the underlying layer which in this case has comprises (Cu?) filled vias surrounded by interlayer dielectric. This will have been planarized by CMP and will be flat, but not perfectly flat and can lead to weak spots between top and bottom electrode. While some low spatial frequency roughness van be tolerated, abrupt changes of even a nanometer or less at, for example, the edge of the via, can lead to weal spots in the finished cell.
A second issue is the etch itself where the authors have clearly developed some low damage etch chemistries. The cell is then ‘sealed’ by oxidation which moves the most likely filament path away from the edge of the actual cell followed by encapsulation. The cell is described as Ir(TE)/Ta2O5/TaOx/TaN (BE) but from the TEM image below from IMEC, appears to have an additional layer as I would not have thought that the TEM contrast between Ta2O5 and TaOx would be visible in this type of micrograpgh. Odd!
However the quoted results are impressive and the authors believe this approach can be extended from 40nm to 28nm
I’ve taken so long writing this article that the folks at Panasonic have been good enough to answer my question above about ‘how they verified the success of their technology’ to center the filament in the cell. At the recent Flash Memory Summit in Santa Clara, Zhiqiang Wei presented “Embedded ReRAM is making good progress”. It appears one can see the filaments in a TEM. The sample preparation must be quite daunting but the images in the presentation on slide 6 illustrate the point. There are no further technical details but the TEMs do look different in terms of the cell as well as the position of the filament.
Zhiqiang also highlighted the Radiation Hardness of ReRAM citing a paper from Dakai Chen from NASA Goddard where the Abstract states “… The resistive memory under investigation is a reduction-oxidation random access memory embedded inside a microcontroller. The memory structure consists of Ir(TE), Ta2O5-δ/TaOx metal-oxide, and TaN(BE). The radiation testing focused on the resistive memory array and peripheral circuits, while other portions of the microcontroller were shielded against the ion beam. We found that the resistive memory array is hardened against heavy ion and pulsed-laser-induced bit upsets. However, the microcontroller is susceptible to single-event functional interrupts due to single-event upsets in the resistive memory peripheral control circuits, which comprise of CMOS elements. Furthermore, the resistive memory architecture is not susceptible to functional failures during write, which is problematic for flash memories due to radiation-induced charge pump degradation.”
To me, this would appear to be a more detailed claim than that of ‘radiation tolerance’ made by Adesto (see Blogs passim).*
Christie Marrian, ReRAM Forum Director
*This last sentence has be qualified following some information from Dakai Chen which I hope to expand upon in a future post on the Forum.