1. Electrical Characterization

DC electrical measurements were performed using an Agilent 4156C semiconductor parameter analyzer with a voltage sweep applied to the TE keeping the BE grounded. Unless otherwise stated, all of the measurements were conducted at room temperature. The SET and RESET compliance currents were fixed at 1 mA and 100 mA, respectively, and all devices were evaluated at the 50th cycle while repeating voltage sweeps (step: 0.05 V) until hard breakdown occurred. Table 1 summarizes the work function values of each TE material, including the effective work function of the Ti/Au (10 nm/100 nm) bi-layer electrode, and their differences with the BE. To further assess switching stability, cycle-to-cycle (C2C) and device-to-device (D2D) variations were statistically analyzed, as shown in Fig. 2.

Fig. 2. (a) C2C variation of HRS and LRS over 50 cycles for four TE devices. (b) D2D variation of SET voltage extracted from 20 devices per TE.

Fig. 3 shows the forming and SET voltages as a function of the work function difference for all TE materials (Pd, Ti, TiN, and Ti/Au). Devices with a larger difference exhibit lower forming and SET voltages, while those with a smaller difference require higher voltages. For instance, the Ti-based device, with the largest difference, showed the lowest potential for forming (3.19 V) and SET (1.42 V), whereas the Ti/Au-based device exhibits the highest values. This behavior arises from the enhanced internal electric field of a larger work function difference, which promotes filament formation and stabilizes subsequent switching events, thereby contributing to more reliable device operation.

Fig. 3. Forming and SET voltages of devices with different TE materials (Pd, Ti, TiN, Ti/Au) and their work function differences.

2. Conduction Mechanism Analysis

Although the TE materials differ, the insulating layers remain identical across all of the devices, suggesting no major variation in switching mechanisms. As Shown in Fig. 4, all devices exhibited Ohmic conduction in the low-voltage region and space-charge-limited conduction (SCLC) at higher voltages. This consistency shows that the work function difference affects the switching voltage rather than the conduction mechanism.

Fig. 4. (a)-(d) I-V characteristics of HfO₂/ZnO devices employing Pd, Ti, TiN, and Ti/Au electrodes, showing switching behavior variations depending on the top metal. (e)-(h) Corresponding SCLC fitting results for devices with Pd, Ti, TiN, and Ti/Au.

The Gibbs free energy of oxide formation represents the thermodynamic driving force for interfacial oxidation, where a more negative value indicates a higher tendency for spontaneous and stable oxide formation. This interfacial energetics directly influences ion transport behavior in oxide-based devices. Ti, with a low formation energy for TiO₂ ($\Delta G^\circ_f \approx -888.8$ kJ/mol) ^[26], enables partial control of oxygen ion migration, suppressing complete filament rupture and leading to a gradual reset transition. This moderated process reduces power dissipation, alleviates stress concentration, and prevents local overheating, as illustrated in Fig. 5. Reported Gibbs free-energy data show that PdO forms near −100 kJ/mol, while gold oxides have positive values and remain unstable under common conditions ^{[29, 30]}. TiN also requires higher energies than Ti to form TiON or TiO₂ ^[31]. These values indicate weaker interfacial oxide formation for Pd, TiN, and Au, leading to less uniform vacancy redistribution and supporting the smoother RESET and stronger retention of Ti-based devices.

Fig. 5. Schematic diagram of oxygen ion/vacancy migration in Ti TE devices during (a) SET and (b) RESET processes based on Gibbs free energy ^{[32- 34]}.

Fig. 6 shows the different reliability behaviors between Ti and Ti/Au TE devices. The Ti TE device Fig. 6(a) maintains high resistance state (HRS) values under prolonged stress, demonstrating stable retention without abrupt resistance collapse.

Fig. 6. Retention characteristics under stress voltage for devices with (a) Ti TE and (b) Ti/Au TE, showing stable operation for Ti and abrupt failure for Ti/Au.

In contrast, the Ti/Au TE device Fig. 6(b) exhibits sudden failure under elevated stress voltages, as indicated by the sharp resistance drop labeled as “failure.” Specifically, failures were observed after approximately 1400 s at V_stress = −1.2 V and after 200 s at V_stress = −1.5 V, confirming the vulnerability of Ti/Au electrodes under high-field stress. In this work, “failure” refers to the abrupt transition from the HRS to the LRS state upon reaching the compliance current during constant voltage stress, indicating uncontrolled filament overgrowth rather than gradual resistance evolution. This discrepancy underscores that the lower Gibbs free energy of Ti facilitates controlled migration of oxygen vacancies, thereby suppressing catastrophic filament rupture and enabling gradual resistance modulation, whereas electrode systems with higher Gibbs free energy, such as Ti/Au, are more prone to abrupt breakdown.

3. Synaptic Characteristics

To evaluate synaptic performance, LTP and LTD measurements were performed by applying 100 consecutive pulses in sequence, as shown in Fig. 7. Both devices were tested using identical pulse schemes consisting of 100 potentiation (LTP) and 100 depression (LTD) pulses with a fixed pulse width of 800 µs. The pulse amplitudes were set to 2.0 V and 2.2 V for the two devices, respectively, and the read voltage was fixed at 0.2 V. All devices exhibited analog conductance modulation, demonstrating stable synaptic switching behavior and acceptable nonlinearity, which is essential for neuromorphic system implementation.

The nonlinearity (NL) was calculated using Eq. (1), where G denotes the normalized conductance and G_Linear is the ideal linear conductance, as defined in ^[35].

The nonlinearity was calculated using the same normalization range and averaging procedure for all devices to ensure a fair comparison.

In particular, the Ti-based device (Fig. 7(a)) exhibited the most gradual and symmetric conductance changes during potentiation and depression. This behavior can be attributed to the controlled migration of oxygen vacancies enabled by the low Gibbs free energy of Ti TE and the moderate work function, which together suppress abrupt filament dissolution and allow fine-tuned weight updates. Consistently, the extracted NL value of the Ti device was 0.51, close to the ideal case, indicating its suitability for stable synaptic weight updates.

Fig. 7. LTP/LTD characteristics under spike stimuli for (a) Ti TE device, (b) Ti/Au TE device, and (c) normalized conductance with extracted NL values.

In contrast, the Ti/Au-based device in Fig. 7(b) showed less gradual modulation with higher NL values (0.90 for potentiation and 0.56 for depression), reflecting the weaker influence of its higher Gibbs free energy. Such quantitative differences highlight that Ti electrodes enable more linear conductance modulation compared to Ti/Au, which is highly desirable for reliable learning behavior in neuromorphic applications.

These results, summarized in Fig. 7, confirm that the choice of top electrode material significantly affects both the resistive switching characteristics and the synaptic plasticity of HfO₂/ZnO-based memristive devices.