Two recent articles [1,2] demonstrated the use of atomic force microscope (AFM) and slice-by-slice conductance mapping to build a tomographic picture of ReRAM filaments.

EDNAOL 2016JUN24 TA 01Fig1 Figure 1: Two different types of ReRAMs were found to have very similar structures.

Two different types of ReRAMs, one based on silicon oxide (SiOx) and the other on hafnium oxide (HfO) were found to have very similar structures. The form is illustrated in figure 1. Both structures have a conical structure terminating in the active or region where switching occurs, shown in red in figure 1. A second common feature is what appears to be many possible conduction path options that initially started to grow but failed to develop into the main filaments. The similarity with lightning also shown in Figure 1 is interesting and may or may not be related. Included in figure 1 is the actual AFM image of a SiOx filament from the work at University College London.[1] The degree to which this is a common feature for all ReRAMs/RRAMs is unknown; if it is a common feature, it carries some important implications. In "ReRAMs: Forming Scaling and Quantised Conductance"[2] I raised the possibility that the need for this structure, or the passive part of it, to form the active region and to act as some sort reservoir could have implications on scaling, such as casting doubt on many of the claims for the possibility of ReRAM/RRAM scaling based just on the size of the active region. I would like to raise a discussion about the effect of a common ReRAM filament structure on claims for write/erase endurance and scaling. Figure 2 illustrates the way in which write endurance might require and use the filaments-in-waiting.

EDNAOL 2016JUN24 TA 01Fig2Figure 2: The way in which write endurance might require and use the filaments-in-waiting.

From left to right in figure 2 initially shows the filament with the switching volume in red. If the initial active or switching region fails, shown in white with an F, then one of the filaments-in-waiting can be used to form a new active region and write/erase operation can continue. The need forming is an anathema for ReRAMs/RRAMs because of the higher voltage it requires. In this case, the forming of a new active region would only require a fraction of the voltage that was used to form the original filament. This means it would not be apparent to the user or the components with which the ReRAM is embedded. This leads to the possibility that the write/erase endurance claims by those offering ReRAMs may include this low voltage reforming effect. There's nothing wrong with that, right? Yes... and no. If it is even actually possible to scale ReRAM/RRAMs (memristors) to just the size of the active region, the absence of the filaments-in-waiting compromises the write/erase endurance. Is it possible to obtain an AFM conductance topographical image of a ReRAM device at the end of its write/erase lifetime to see if there is any evidence of the use of the filaments-in-waiting? If anybody has a view or has already carried out such an experiment, I would certainly like to hear about it. What must be considered is the possibility that the filaments-in-waiting have another role to control and keep a constant current density in the switching region. If the voltage across the device increases some leakage, current will flow from the tips and minimise any possible damage to the active region. If they are not present, write/erase endurance might well be compromised for a completely different reason. I will discuss the subject of the ReRAM/RRAM filaments-in-waiting acting like the "arcing horns" familiar to high voltage engineers in another article.
References

  1. Neale, Ron. "ReRAMs: 3D Filaments and Brain-like Functions", EE Times, 3/2/2016 04:58 PM EST.
  2. Neale, Ron. "ReRAMs: Forming Scaling and Quantised Conductance", EE Times, 3/30/2016 07:01 PM EDT.


About the author
Ron Neale is an independent electrical/electronic manufacturing professional.