(Invited) In Situ Filamentary Resistance-Switching Experiments Using Bulk Oxide Single Crystals and Polycrystals

Abstract

This presentation will describe how in situ testing of bulk oxide specimens can provide useful information about nanofilament formation and resistance switching in oxide memristor devices. Past in situ experiments performed in transmission electron microscopes using electron-transparent thin films are known to be subject to artifacts because of radiolysis. Past ex situ experiments inspecting filaments in memistor devices are also problematic because of poor statistics statistics: finding a filament in a resistance-switching device is like finding a needle in a haystack. Two contrasting model oxides in single crystal and polycrystal forms were used in our study. (a) Yttria stabilized cubic zirconia (YSZ) is a fast oxygen ion conductor with a very wide electrolytic range and a very small contribution from hole conductivity especially at low temperatures. (b) Undoped as well as acceptor doped strontium titanate (STO) is a p-type semiconductor with a small amount of quenched-in oxygen vacancies that move very slowly especially at low temperatures. These oxides at the composition studied have no structural or electronic phase transition, unlike V2O5or VO2. Especially at low temperature, their cations are so immobile that they cannot migrate to form metallic filaments, unlike Ag or Cu. Therefore, they may be regarded as two extremes of prototypical oxides, and any observation common to both are likely to be general and applicable to other oxides and nitrides as well. This is important, for already there are a large number of materials—in many structures and chemistries—made into memristors and they all behave similarly. Our goal is to find the common element that explains their behavior, namely filament formation and filamentary resistance-switching. To achieve this goal, our in situ experiment on bulk specimens must specifically demonstrate at least the following: (1) a localized region can form in a virgin oxide, which results in an overall resistance that is orders of magnitude smaller, obtaining the “low resistance state” (LRS); (2) in this LRS, the overall resistance can next be increased by orders of magnitude to reach the “high resistance state” (HRS), meanwhile some features of the localized region should still be retained. These have been successfully demonstrated in our experiment providing highly reproducible observations. Having achieved (1-2) and delineated the conditions for doing so, the in situ experiment should also answer the following question: what is the thermodynamics, kinetics, geometry and electrostatics/electrochemistry of filament formation and resistance switching. We attempt to answer the above question by comparing the experimental observations with the solution of the field equation for the distribution of oxygen and electric potential, the concentrations of defects and electrons/holes, and the temperature. Given the appropriate boundary conditions and our knowledge of YSZ and STO, this is solvable although the solution procedure is quite different for two oxides because they are two rather different conductors with rather different distributions of oxygen vacancies. It is important to note that the boundary conditions, which include the flux of oxygen, the ratio of ionic current to the electronic current, and the form of the electric potential/field, are themselves observables that can be ascertained from the experiment in situ and ex situ. Moreover, varying the boundary conditions provides a further means to tune the conditions for (1-2), thus lending further support to our explanation. Despite the macroscopic size of our samples, the findings can be transferred to nanoscale devices using scaling arguments. For example, in the field equation, one scaling parameter is jL, the product of current density and the distance between the top and bottom electrodes, and in the normalized form the same distribution is obtained in different problems when their jL is the same. Since j scales with (filament area)-1and L with (filament length), it follows that the switching current is much smaller in a nanoscale device than in a bulk sample. These considerations have provided further insight to the switching parameters, such as critical voltage, of nanoscale memristors.