At the NCCAVS meeting on 11/15 (see 11/20 Blog) Rene Meyer gave a detailed technical presentation on ‘Tunnel RRAM’ devices and arrays. I was unable to attend the presentation but the chair (Yi Ma from Adesto) passed on the presentation. This and the other presentations from the meeting will soon be available on line. There is a wealth of detail in the slides which I certainly can’t do full justice to so I recommend downloading the slides even if you were able to attend. The presentation gives a more in depth look at the cell, materials and characteristics and so complements the broader view given at the Flash Memory Summit in August (see 10/28 and 11/26 Blogs). A couple of things struck me as particularly interesting and serve to highlight the Unity/Rambus approach to ReRAM.
The cell is based on a bilayer where one layer is a deposited oxide which acts as a tunnel barrier and so allows the current through the cell to be controlled (by for example, depositing a thicker layer to reduce the current). The other layer is a complex metal oxide (PCMO) layer and is the origin of the memory effect (see schematic at top of Blog). Unlike the filamentary model, this is seen as a uniform change in the characteristics of the tunnel layer as a result of oxygen diffusion at the interface between the two layers. Current (set, reset and read) scales with area, i.e. the current density through the device remains constant over a three orders of magnitude change in device area. Further the IV characteristics do not show the abrupt transition between the two resistance states. In fact, were it not for arrows on the IV characteristics marked set and reset, it would not be obvious what voltages and currents are required for set/reset! Thus the external current compliance level is apparently not critical. This is in contrast to ReRAM cells (often described as filamentary) where the current compliance determines the value of the low resistance state of the cell. An alternate view would be that the Unity/Rambus cell has a built in current compliance set by the deposited oxide layer.
A second highlight in the presentation is the discussion and comparison of various select devices. Or in the authors case the lack of need for an external select device. For the Unity/Rambus bilayer cell, the IV characteristic of the cell itself is sufficiently non-linear that the current of ‘half selected cells’ is much lower than half of the current through the selected cells. This analysis is quantified by a parameter corresponding to the ratio of the currents at the full voltage (for read or set for example) and half of that voltage. Clearly, the larger this parameter the more non-linear the IV characteristic. The ‘best’ select device using this criterion is the MIEC (Mixed Ionic Electronic Conductor) select device described by IBM (hopefully the subject of a future Blog). Taking into account the presence of the actual cell reduces this parameter somewhat. However, it allows the size of a ‘crossbar’ type array to be defined within the limits set by the acceptable level of leakage current through the sneak paths. In Unity/Rambus’ case this naturally leads to the multiple tile architecture described here and previously.
There is some small array data presented along with an MLC demonstration (see above). While impressive, I am always a little dubious of the presentation of this type of data as it suggests that it is the best result obtained to date. (For example if ‘better’ data from a ‘larger’ array existed, presumably that would have been presented!). Indeed, the author finishes by admitting that issues remain in terms of data retention and processing the CMO materials. Nonetheless, the results and level of detail in the presentation are impressive. I will update the Blog with the link for the presentation when it is available.
Christie Marrian, www.ReRAM-Forum.com Moderator