Dynamics and parameters of resistive switching in metal oxides
Currently, metal nanooxide-based resistive switching memory is being studied extensively as one of the most competitive candidates for non-volatile memory applications because of its simple structure, rapid switching and excellent scalability. The mechanisms of resistive switching phenomena in oxides can be quite diverse. One possible mechanism is electronic switching, originating from a trap/de-trap process through the defects of oxides. Another possible mechanism is ionic switching, which is usually attributed to the formation/rupture of conductive filaments (CFs) which may consist of oxygen vacancies or metal precipitates. It is illustrated in Fig. 1.
Figure 1. Dynamic growth and formation/rupture processes of filament-type resistive switching mechanism in unipolar (a) and bipolar switching (b) modes.
The process of CFs formation and their transition from an initial high resistance state (HRS) to a low resistance state (LRS) are interpreted as a dielectric soft breakdown associated with the migration of oxygen ions toward the anode, and the formation of CFs in the bulk oxide connecting both electrodes. In the unipolar reset process where the reset occurs at the same polarity as the set, joule-heating-assisted diffusion of oxygen ions from the anode occurs and the surrounding oxides rupture the CFs by recombining with oxygen vacancies or by the re-oxidation of metal precipitates. Cooling down and formation of a functional region in the conductive filament occur during the set process (Fig. 1). In the bipolar reset process that occurs at the opposite polarity as the set, electric-field-assisted drift of oxygen ions from the oxygen reservoir ruptures the CFs.
Electrochemical ion transport may be considered to be a driving mechanism of the slow switching. Fast electronic switching is provided by the synchronized process of the capture and release of mobile charge carriers on the bi-stable trapping states in the regions of the formation/rupture CFs. These states are cause by the presence of oxygen vacancies.
The length of CFs was found to grow exponentially (Fig. 2a). The magnitude of the external bias determines the characteristic time of growth. A significant increase of the filament growth rate is reached at the external bias of 2-2.5 V.
Figure 2. Dynamics of conductive filament growth in HfO2 of 5 nm thickness under various applied voltages (a). RRAM dynamics during pulse programming (5 V) with an operating temperature of 300 K (b).
Modeling of the dynamics of CFs growth in HfO2 during programming voltage pulse of 5 V that depends on the operating temperature shows that the RRAM cell doesn't switch on immediately, but it takes approximately 4-6 ns. The switching event is determined by a sharp increase in the conductivity and is marked by the dotted line (Fig. 2b). Thus, the time may be determined by the difference between the time in which the programmable pulse is applied and the moment of switching. ON time depends exponentially on the external bias. This result shows the importance of the dynamic effects involving a range of transient processes.
Modeling of electronic switching of bi-stable trap states was carried out for hafnium dioxide with the following parameters: thermal ionization energy of traps of 0.5 eV, oscillation frequency of 10-12 GHz, trap concentrationof1019 cm−3, and noise intensity of 0.08-0.15. The bi-stable potential of trap centers under the influence of a weak periodic modulation was confirmed to make a transition from one state to another only under the effect of noise.
Fig. 3a shows that the impact of noise leads to the switching of the trap state in hafnium dioxide from one metastable state to another in a few nanoseconds. An increasing intensity of the noise increases variations in the output signal y(t), causing formation of metastable states. The switching frequency from one state to another grows with an increase of the periodic influence amplitude. The switching frequency increases with the increasing frequency of the periodic impact, and the residence time of the trapping center increases with the increasing phase in the metastable state.
Figure 3. Switching dynamics (a) and the switching time vs noise intensity at different amplitudes of the periodic force (b).
Modeling studies of the electron switching of trapping states from one metastable state to another showed that the switching time is about few nanoseconds. It decreases becoming less than 1 ns with the increasing amplitude of periodic influence and noise intensity (Fig. 3b).
The modeling shows that the complete mechanism of resistance switching in memory cells should include consideration of ion migration, the growth of conductive filaments and their rupture, as well as the capture and release of electrons including metastable trapping states.