An interesting two part article appeared recently in the IEEE Transactions on Electron Devices (full references below). The papers describe the modeling and simulation of an RRAM cell along with a comparison with results from experiment. The simulation engine incorporates a trap-assisted-tunneling current solver and the stochastic generation and recombination of oxygen vacancies which are responsible for the conduction filaments in the low resistance state (see figures). Various features of the IV characteristics of the SET and RESET process are identified in both simulation and experiment including seemingly ‘random’ jumps in current values. The authors go on to discuss the problem of resistance variation in both low and high resistance states resulting from a number of simulated program/erase cycles. Again there is a remarkable agreement between their simulation and experimental results. ‘Tail’ bits are visible in both type of distributions and the authors go on to propose a methodology to reduce the presence of tail bits based on results from their simulator. You’ll have to read the paper for more details as the point that struck me was not so much the details of the proposed remedy but the fact that that the authors thinking was guided by their simulator rather than experiment. While the ultimate validation of these ideas awaits further study and experiment, it is a clear sign of progress that the complex processes that occur during the operation of an RRAM cell are beginning to be abstracted and modeled so that the key features of RRAM cell operation can be replicated in simulation. Further, the types of seemingly random variations and variations apparent in cycle to cycle variation (such as high resistance state resistance variation) are being replicated by simulation. This opens the way to virtual experimentation for RRAM cell optimization in which direction the authors have taken some first steps.
One should perhaps sound a slight word of caution in that just because a model/simulator reproduces the characteristics seen in experiment doesn’t necessarily validate the model (or for that matter the experiment!). It clearly is a necessary step but not the ultimate proof. I remember a project in which I was involved earlier in my career where experiment and modeling were apparently providing clear agreement and giving us clear insights into the apparent mechanisms responsible for our observations. Unfortunately, it turned out that this was somewhat fortuitous and something completely different was taking place. Nonetheless these papers represent an impressive set of results and an important step forward in RRAM understanding and development. The modeling and simulation appear (to this observer) to incorporate the relevant physics at an abstraction level at which simulation becomes computationally tractable. Thanks to Professor Philip Wong from Stanford for alerting me to these papers and I look forward to learning more from the authors as they continue their thorough approach to providing insights on how RRAM cells function and can be improved.
Christie Marrian, www.ReRAM-Forum.com, Moderator
On the Switching Parameter Variation of Metal-Oxide RRAM—Part I: Physical Modeling and Simulation Methodology, Ximeng Guan, Member, IEEE, Shimeng Yu, Student Member, IEEE, and H.-S. Philip Wong, Fellow, IEEE, IEEE Transactions on Electron Devices, Vol. 59, No. 4, April 2012, p1172
On the Switching Parameter Variation of Metal Oxide RRAM—Part II: Model Corroboration and Device Design Strategy , Shimeng Yu, Student Member, IEEE, Ximeng Guan, Member, IEEE, and H.-S. Philip Wong, Fellow, IEEE, IEEE Transactions on Electron Devices, Vol. 59, No. 4, April 2012, p1183