According to the 2014 International Technology Roadmap for Semiconductors (ITRS), resistive switching memories (RRAM) are well candidate for next generation nonvolatile memory. Their main properties are fast switching speed, good reliability, low power consumption and CMOS technology compatibility [1, 2], as well as potential scalability beyond NAND flash [3, 4]. They are based on the change in the physical properties of a conductive filament by applying an electric field across a metal-insulator-metal (MIM) or metal-insulator-semiconductor (MIS) structure. A set switching closes the filament and induces a transition from the high-resistance state (HRS) to the low-resistance state (LRS). A reset switching opens the filament and induces the opposite transition. The resistance switching is generally caused by the diffusion of oxygen vacancies, charge carrier trapping and detrapping, and Schottky barrier modulation to produce the memory effect [5]. The implementation in the industry of RRAM memory devices requires a detailed understanding of switching mechanisms, hence significant improvement in the knowledge limits is still needed. To expand the conventional characterization techniques spectrum of RRAM devices, we propose to study the small-signal parameters, namely, conductance (G) and susceptance (B) [6, 7]. They provide information about the physical nature of the switching mechanisms. Moreover, the read of the memory state is carried out without DC power consumption. Here we present comparative results of MIM-RRAM with different insulator and electrode materials.

Admittance parameters were recorded by using a Keithley 4200SCS semiconductor analyzer. The bias voltage was applied to the top electrode with the bottom electrode grounded. A 30 mV rms-ac signal was superimposed to DC bias. A parallel admittance model which provides G and B values was selected to perform the characterization. To obtain the memory maps we apply a return-to-zero voltage pulse sequence as follows: with the sample at the HRS state, we apply a positive voltage pulse during 1 ms, and then the voltage returns to zero. At this moment we measure the admittance at 0 volts (G₀, B₀)_. The pulse amplitude (V_P) is linearly increased until the HRS to LRS state transition (set) occurs. Once the sample is at the LRS state, V_P is linearly decreased to negative values. When V_P reaches negative values enough to take the device back to the HRS state (reset), it is linearly increased again until 0 V. The plots of G₀ and B₀/ω as a function of the programming voltage, V_P, previously applied constitute the memory maps.

In Figs. 1-3 we show the results obtained at 100 kHz for Pt/Ta₂O₅-TiO₂-Ta₂O₅/Ru/TiN, Pt/ZrO₂:Co₂O₃/TiN/Si/Al, and Pt/ZrO₂-Co₂O₃/TiN/Si/Al MIM samples. In all cases, we observe that G₀ and B₀ are very stable in a wide V_P range, so indicating that the conductive filaments are also very stable at both states. Reset and set transitions are very sharp, but reset is more vertical than set. That is, the filament restoring is gradual, whereas it is instantaneously broken. Figs. 2 and 3 show results for two different dielectric layers containing ZrO₂ and Co₂O₃. The first one is doped ZrO₂, i.e., ZrO₂:Co₂O₃, whereas the second one is a double layer stack of 14 nm of ZrO₂ over 12 nm of Co₂O₃. Set occurs in one step at negative polarity in the case of doped ZrO₂ sample (Fig.2); in the bilayer (Fig.3) set occurs in two steps at positive polarity. In the last case, the conductive filaments restoring occurs with different dynamics at each layer.

According to our recently proposed model [6], the conductive filament behaves as an inductance (L₀) in series with a resistive term (R₀), and both are in parallel with the capacitance of the MIM structure (C). Using this equivalent circuit, we have extracted L₀ and R₀ for the Pt/Ta₂O₅-TiO₂-Ta₂O₅/Ru/TiN sample (Fig. 4). Again the signals, and hence the filaments, are very stable at both states, and set and reset transitions occur sharply. The inductive term is negligible at the HRS state, and reaches high values (50 µH) when the filament is formed. This indicates the existence of delays between the current and voltage caused by charge transport mechanisms occurring at the filaments.

[1] R. Waser and M. Aono, Nature Mater. 6 (11), 833 (2007).

[2] J. J. Yang et al. Appl. Phys. Lett. 97, 232102 (2010)

[3] M. J. Kim et al., Int. Electron Devices Meet. 444 (2010)

[4] J. Lee et al. Int. Electron Devices Meet. 452 (2010)

[5] R. Waser et al. Adv. Mater. 21, 2632 (2009)

[6] S. Dueñas et al. Mic. Eng. 178, 30 (2017)

[7] S. Dueñas et al. EDL 38(9), 1216 (2017)