Nowadays both the volatile and non-volatile memories are under a high development pressure in the current industrial context. Besides the dominant solid-state memory technologies, such as DRAM and FLASH, non-volatile resistive memories (RRAM) are considered adequate candidates to complement the memory landscape [1, 2]. In spite of the interest, no consensus has been reached yet about the selection of appropriate materials for fabrication [3, 4]. The aim of this work is to deepen in the usefulness of metal transition oxides in this field, and specifically to explore the improvements on the memory behavior of metal/ZrO₂/metal structures by stabilizing the metastable phase by doping or nanolaminating the dielectric films with small and controlled amounts of Al₂O₃. The effect on memory window, stability and robustness, as well as the comparative small-signal response are reported in this study.

ZrO₂:Al₂O₃ films were grown on highly-doped conductive Si <100> substrates covered by 10 nm thick TiN fims by means of atomic layer deposition (ALD) at 350 ºC. The reactor was a flow-type hot-wall F120. The used precursors were AlCl₃ (99%, Acros Organics), ZrCl₄ (99.9%, Aldrich), and O₃. For electrical measurements, Al/Ti/ZrO₂:Al₂O₃/TiN/Si/Al capacitor stacks were fabricated. Double-layer 110 nm-Al / 50 nm-Ti top electrodes were beam evaporated. The bottom electrode was provided by evaporating 100-120 nm thick Al layer on etched Si. A similar sample based on HfO₂:Al₂O₃ was also fabricated for comparison, at 350 ºC and by using HfCl₄ (99.9%, Strem) as precursor. Films characteristics of all samples are listed in Table I.

The electrical characterization consisted on recording the DC and small-signal parameters by using a Keithley 4200 semiconductor analyzer. At first, all samples showed insulating behavior, but after a forming step they behaved as memory elements, with two clearly differentiate states: high-resistance state (HRS) and low-resistance state (LRS). The forming is an initial electrical stress necessary to activate the switching property, and consists on applying a DC bias ramp from 0 to ≈ 3-4 V, with a current compliance of 10 mA. In Fig. 1(a) the forming step and subsequent very first I-V memory cycles are depicted for the ZrO₂-based sample. The repetitiveness of the resistive switching is shown in Fig. 1(b), in which 20 consecutive cycles are plotted. After the forming, small-signal parameters were recorded as well. Similarly, to the current, both conductance and capacitance exhibit set and reset loops [5], as it is seen in Fig. 2. In the LRS state G y C are practically constant (ohmic behavior), whereas in the HRS state the linear voltage dependencies of G and C indicate that conduction is dominated by space charge limited conduction (SCLC).

To establish the influence of dielectric composition on the resistive switching performance, a comparison of the I-V loops is shown in Fig. 3 (a). The introduction of very small amounts of Al₂O₃ (Al:Zr = 0.16) extends the voltage range in which the loops appear, and provides narrower windows. However, the current window can be open by increasing the Al:Zr ratio (Al:Zr = 0.30), besides the voltage span diminishes. This way the resistive switching characteristic significantly improves. On the other hand, Fig. 3(b) demonstrates that thicker dielectrics lead to narrower voltage ranges but does not open the current window; finally, samples based in HfO₂ instead of ZrO₂ provide much lower current values.

Admittance parameters sheds more light on this issue. In fact, zero-bias admittance memory maps for different Al:Zr ratios (Fig. 4) clearly show the two different states, as well as the influence of the dielectric composition. These maps represent the conductance and capacitance values measured at 0 V as a function of the programming voltage previously applied [6]. The reading process is carried out at 0 V, so without power consumption. The memory aspect ratio can be modulated by changing the amount of Al₂O₃. The best power consumption behavior is achieved by samples with Al:Zr = 0.30, which has the lowest values of HRS state conductance, and also the lowest values of V_set and V_reset.

[1] A. Bec et al., Appl. Phys. Lett. 77(1), 139 (2000).

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

[3] E. Gale, Semicond. Sci. Technol. 29, 104004 (2014).

[4] H. Nili et al., Adv. Funct. Mater. 24, 6741 (2014)

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

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