Oct 05

ReRAM Cell Switching and Trends

Professor Daniele Ielmini was kind enough to point me to a couple of publications by his group that have just appeared in the IEEE’s Transactions on Electron Devices. Confused by the way a ReRAM cell behaves? Looking for a clear definition of the various terms used? Then I recommend you study this paper as well as the recent publications by Professor Philip Wong’s group at Stanford. The papers repay careful reading as they cover a lot of ground.

In part 1, a detailed experimental study of Hafnium Oxide ReRAM cells is reported. The work was performed in collaboration with Sematech with David Gilmer being a co-author and responsible for the ReRAM process technology. The main focus of the paper is on the set and reset operation and, in particular, the interaction of the preceding reset/set operation on a subsequent set/reset operation. The devices require a forming step (first figure, device current vs voltage). While the devices are by today’s standards relatively large, and exhibit set/reset voltages and set/reset resistances that do not vary with device size, a key characteristic of a conduction filament (CF) based mechanism. A key conclusion is that the device resistance following the set process (which is abrupt, first figure) can be controlled with the compliance current (the maximum current through the device) during the set process. This is illustrated in the second figure (set resistance vs compliance current) showing the measured device resistance after a set process as a function of the compliance current. The data is a compilation of the authors’ own work and results from the literature for many different materials and device geometries. As you can see, the points lie close to a line which corresponds to an inverse dependence of set resistance and compliance current corresponding to a constant voltage of ~0.4V. Quite remarkable, especially when one considers the diversity of device thickness, size and materials included. The authors come up with a plausible (to me at least) understanding of this seemingly non-intuitive result by pointing out that the set resistance is essentially determined by the geometry of the resulting conducting filament which is determined by the compliance current rather than the device structure. I wonder what is the physical reason is for the actual value of 0.4V?

For the reset process (which is gradual) the key parameter is the maximum voltage applied during the reset sweep. Again taking a series of measurements from the literature as well as their own results, the authors point out that the reset process is initiated at the voltage at which the current is close to the compliance current for the preceding set process. The authors interpret their results with the aid of a model which is described in Part 2. I plan to review this once I have solved a riddle related to an embedded movie!

In summary, this paper represents another step in the maturity of our understanding of the behavior of ReRAM cells. I was particularly struck by the authors identifying trends that apply to both their own work and that results from the open literature that cover a lot of differing materials, device geometries and testing methodologies.

Christie Marrian, www.ReRAM-Forum, Moderator

ps Frederico Nardi, the lead author on the paper is now with Intermolecular in San Jose which I presume is further indication of that company’s involvement with ReRAM. See Blog on the 2nd International Workshop on Resistive RAM.

Resistive Switching by Voltage-Driven Ion Migration in Bipolar RRAM—Part I: Experimental Study, p 2461, IEEE Transactions on Electron Devices, Vol. 59, No. 9, September 2012, Federico Nardi, Stefano Larentis, Simone Balatti, David C. Gilmer, and Daniele Ielmini.


  1. Peter Newman

    I wish people would stop characterising memristive systems with charts showing current as a function of voltage. That’s appropriate for non-memory functional transfer functions (as for resistors). But memristors are integrators. Charts of resistance against integrated energy through the device would be more useful.

  2. admin

    Hi Peter
    Thanks for the comment. I’m not sure I fully agree but I contacted Prof Ielmini who made the following remarks (minor editing by the Moderator). “There must be a reason why all works on RRAM/CBRAM show at least one I-V curve. That’s because the I-V curve gives the most important parameters about the memory, namely high/low resistances, switching voltages and switching currents. From these data, one can figure out if a certain material/device is interesting for a certain application, i.e. low-power, low-voltage, multilevel operation. Representation in terms of the energy would instead fail to give this breadth of information. Conversely, you can get information about the energy simply from the voltage and current in the I-V curve, plus the time-scale of the experiment (usually in the 1 s range).”

  3. Peter Newman

    Thanks, Chris, for following this up, and thanks to Prof Ielmini for comments. Frequency-swept I against V curves may indeed gives the most important parameters about the memory for those seeking supra-threshold binary switches.

    My interest is in sub-threshold analogue memory so I would welcome integration-oriented parameters rather than the V-I-F (frequency) differentials. The change in integrated value (resistance for a memristor, voltage for a capacitor) for per injection of energy is more appropriate for integration applications. The equivalent swept V-I-F curves for a capacitor – the other fundamental integrator and hence also a Lissajous figure – similarly makes it hard to determine the capacitance.

    But all power to those making material advances in memristor tech – there are many applications waiting!

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>