Impact of Bilayer CeO₂−x/ZnO Heterostructures and Electroforming Polarity on Resistive Switching Performance of Non‑Volatile Memory Devices

Background

Conventional flash memories are approaching their physical limits, prompting the search for next‑generation non‑volatile memories. Resistive random‑access memory (RRAM) is a leading candidate due to its high scalability, long data retention, sub‑nanosecond switching, low power consumption, and simple metal‑oxide‑metal (MOM) architecture.

RRAM cells exhibit bipolar resistive switching (BRS) or unipolar switching (URS), toggling between a low‑resistance state (LRS) and a high‑resistance state (HRS) under applied voltage sweeps. The reliability of these transitions depends on uniform SET/RESET voltages and current levels, which are governed by dielectric properties, electrode work functions, and fabrication processes. Switching mechanisms are broadly classified as bulk‑limited (dielectric permittivity‑driven) or interface‑limited (electron correlation at the metal‑dielectric interface).

Ceria (CeO₂) is attractive for RS applications because of its large dielectric constant (~26), low Gibbs free energy (−1024 kJ mol⁻¹), dual oxidation states (Ce⁴⁺ ↔ Ce³⁺), and a high concentration of oxygen vacancies in non‑stoichiometric CeO₂−x. Zinc oxide (ZnO) is a wide‑bandgap, optically transparent semiconductor with high resistivity (10⁵ Ω cm) and a rich defect chemistry, making it a popular choice for RRAM dielectrics.

Bilayer structures, such as CeO₂−x/ZnO or ZnO/CeO₂−x, have been shown to reduce electroforming and SET/RESET voltages, improve switching uniformity, and enhance endurance. However, systematic studies examining how the heterostructure order and electroforming polarity affect RS behavior remain scarce.

In this study, we present a comparative analysis of Ti/CeO₂−x/ZnO/Pt and Ti/ZnO/CeO₂−x/Pt devices, highlighting how device architecture and electroforming polarity dictate RS performance, endurance, and retention.

Methods

Two device stacks were fabricated: Ti/CeO₂−x/ZnO/Pt and Ti/ZnO/CeO₂−x/Pt. For the first stack, a 10 nm ZnO film was sputtered onto commercial Pt/Ti/SiO₂/Si substrates at room temperature (75 W RF power, ~10 mTorr Ar/O₂ 6:18, 24 sccm). A 5 nm CeO₂−x layer was then deposited atop the ZnO under identical conditions. A 150 µm diameter circular Pt/Ti top electrode was patterned by DC sputtering using a shadow mask. The second stack was fabricated identically, with the layer order reversed. All devices were characterized using an Agilent B1500A parameter analyzer, and cross‑sectional HRTEM imaging was performed on a JEM‑2001F microscope.

Results and Discussion

Device Architecture and Electroforming
Figure 1 illustrates the schematic of both heterostructure devices. Current‑voltage (I‑V) measurements reveal classic bipolar RS behavior. In Ti/CeO₂−x/ZnO/Pt, a +2 V sweep induces a sudden current jump at 0.6 V, indicating filament formation and an ON‑state that persists after the voltage is removed. A subsequent negative sweep restores the HRS, and a positive sweep returns the device to LRS.

When a negative electroforming voltage is applied, the device requires a higher magnitude (−5.6 V) to form conductive paths and fails to switch back to HRS, implying irreversible filament formation. Thus, Ti/CeO₂−x/ZnO/Pt devices must be electroformed positively for reliable non‑volatile operation.

In contrast, Ti/ZnO/CeO₂−x/Pt devices can be electroformed at both polarities. Positive electroforming (0 → +4 V) produces a filament at +3 V; subsequent SET/RESET cycles occur at ±2 V. Negative electroforming (0 → −3.5 V) similarly yields reversible switching, with SET at −2.5 V and RESET at +1.5 V. A 1 mA compliance current was used during electroforming and SET to prevent permanent breakdown.

Switching Uniformity
Cumulative probability plots of SET and RESET voltages (Figure 3) show narrower distributions for Ti/CeO₂−x/ZnO/Pt compared to Ti/ZnO/CeO₂−x/Pt, indicating superior voltage uniformity. Average electroforming voltages are significantly lower for the CeO₂−x/ZnO stack, while SET/RESET voltages differ only marginally. This behavior aligns with the lower Gibbs free energy difference between ZnO and CeO₂−x, facilitating oxygen vacancy migration and reducing operation voltages.

Endurance
Endurance tests at 0.2 V demonstrate that positively electroformed Ti/CeO₂−x/ZnO/Pt devices retain a memory window of ~10 over 10² cycles, thanks to Schottky barrier formation at the Ti/CeO₂−x interface. Negatively electroformed Ti/ZnO/CeO₂−x/Pt devices exhibit superior endurance (~10² cycles) compared to their positively electroformed counterparts, which suffer from poor ON/OFF distinction due to the absence of a Schottky barrier at the Ti/ZnO interface.

Retention
Both heterostructures maintain stable HRS and LRS resistances for at least 10⁴ s at room temperature under a 0.2 V read pulse, indicating robust data retention without power.

Conduction Mechanisms
High‑field I‑V curves fit a Schottky emission model (ln I vs. √V linearity) for both stacks, confirming interface‑limited transport. Temperature‑dependent measurements (200–300 K) show metallic behavior in LRS (resistance increases with T) and semiconducting behavior in HRS (resistance decreases with T). Activation energies (~0.092 eV) extracted from Arrhenius plots match the ionization energy of oxygen vacancies, reinforcing the vacancy‑filament switching model.

Band Alignment and Switching Dynamics
Energy band diagrams (Figure 7) reveal a 1.02 eV difference between the work functions of ZnO (4.35 eV) and CeO₂ (3.33 eV), driving electron transfer from CeO₂ to ZnO. In Ti/CeO₂−x/ZnO/Pt, positive bias attracts oxygen ions to the CeO₂−x/Pt interface, forming filaments that rupture under negative bias. In Ti/ZnO/CeO₂−x/Pt, the absence of a Schottky barrier at the Ti/ZnO interface leads to poorer endurance under positive bias, whereas negative bias restores reliable switching.

Overall, the heterostructure order and electroforming polarity critically affect RS performance, with CeO₂−x/ZnO stacks offering lower electroforming voltages and better endurance when electroformed positively, and ZnO/CeO₂−x stacks performing best when electroformed negatively.

Conclusions

We have shown that bilayer CeO₂−x/ZnO and ZnO/CeO₂−x heterostructures, together with the sign of electroforming polarity, significantly influence resistive switching characteristics. Ti/CeO₂−x/ZnO/Pt devices benefit from a TiO interfacial layer under positive bias, while Ti/ZnO/CeO₂−x/Pt devices rely on CeO₂−x as an oxygen reservoir under negative bias. Both configurations exhibit Schottky‑emission‑limited conduction in HRS and vacancy‑filament switching mechanisms, offering promising pathways toward high‑endurance, low‑voltage RRAM cells.

Abbreviations

BRS

Bipolar resistive switching

DC

Direct current

HRS

High resistance state

HRTEM

High‑resolution transmission electron microscopy

LRS

Low resistance state

MOM

Metal‑oxide‑metal

RRAM

Resistive random access memory

RS

Resistive switching

TE

Top electrode

URS

Unipolar resistive switching