Professor Philip Wong of Stanford University was kind enough to point me to a review article he and his collaborators have recently published in the Proceedings of the IEEE (full reference below). This must be one of the most comprehensive reviews of binary metal oxide based RRAM for non-volatile memory. The paper is comprehensive (155 references!) and provides a wealth of detail on the current thinking on resistive switching mechanisms, materials, RRAM arrays and device metrics. The authors point to the evidence for conductive filaments related to oxygen vacancies being the conduction mechanism in the low resistance state. Oxygen drift and/or diffusion during reset blocks the conduction pathways and the cell reverts to a high resistance state.
Of particular note is the identification of the memory cell properties that are desirable for large scale integration. These include low current and low (but not too low) voltage for set and reset, unipolar operation (makes the selection device simpler), scalability and material and processing compatibility with existing Back End of Line (BEOL) semiconductor manufacturing. The paper also describes some of the trade-offs that are necessary. For example, in metal-insulator-metal cells it is desirable to have a very thin oxide layer to minimize the necessity for a forming process, however, the thinner the oxide, the lower the memory window. The authors further point out that the memory window improves with decreasing cell sizes as the low resistance state is roughly independent of cell size and the high resistance state increases as the cell size decreases. However, the current required to reset the cell (transition from low to high resistance) does not appear to scale with cell size. The authors discuss ways that trade-offs can be made to ameliorate this potential roadblock. With current devices, the reset current is typically rather greater than that available from devices than can be conveniently fabricated in the BEOL of a semiconductor manufacturing process. (This suggests that either non-traditional materials are required to fabricate, for example, a diode select device or a FEOL (crystalline Si) select device is needed). While the challenges facing RRAM may appear daunting, it is a mark of a degree of maturity in the technology (and the thoroughness of this paper) that they are being described in the open literature. Further, this allows the considerable innovative talents of scientists and engineers in academia and industry throughout the world to be applied to overcoming these challenges.
While RRAM has the potential for stand-alone memory, companies face the problem that new RRAM products would, initially at least, be competing with their existing Flash products. In their paper, the authors suggest that embedded applications look particularly attractive for RRAM as it has a clear advantage over the existing technologies of Flash (faster and lower voltage programming) and DRAM (non-volatility). Further, there is the potential that the RRAM cells can be fabricated in the BEOL without major changes to the FEOL.
Metal–Oxide RRAM, H.-S. Philip Wong, Fellow IEEE, Heng-Yuan Lee, Shimeng Yu, Student Member IEEE, Yu-Sheng Chen, Yi Wu, Pang-Shiu Chen, Byoungil Lee, Frederick T. Chen, and Ming-Jinn Tsai, p1951 Proceedings of the IEEE, Vol. 100, No. 6, June 2012
Christie Marrian, www.ReRAM-Forum.com Moderator