A capacitance-coupled Ga₂O₃ memristor

A memristor is a nonlinear resistance switching device, regarded as a promising electronic component for neuromorphic computing as a scheme for realizing artificial intelligence (AI).¹ The device was predicted by Leon Chua as the fourth fundamental electronic component in addition to resistor, capacitor, and inductor in 1971, when he investigated the relationship between each two out of the four circuit variables, namely, current i, voltage v, charge q, and magnetic flux φ.² The first solid-state memristor was realized by Strukov et al. from HP labs in 2008.³ Subsequently, memristors have been developing rapidly, including single devices with various working mechanisms and neuro-network systems.^4–8 The typical feature of a memristor is that the device remembers its previous resistance based on the history of the voltage applied,⁹ reflected by a hysteresis loop in the current–voltage (IV) characteristics.

Many materials have been developed to fabricate memristors, including binary oxides, perovskites, and two-dimensional materials.¹⁰ Among them, binary oxides (e.g., TiO_x, ZrO_x, and ZnO_x) are the largest material group due to their easy fabrication and strong current hysteresis.^11–13 Ga₂O₃ as a new binary oxide semiconductor material (fourth-gen semiconductor) demonstrates its great potential in various fields, including power electronics, radiation detectors, and memristors.^14–18 So far, nearly all published studies on Ga₂O₃ memristors focus on the changing of resistivity, while the effects of capacitance have yet to be understood.

In this work, we fabricated an ITO/Ga₂O₃/ITO coplanar interdigital device, which exhibits current hysteresis assisted with a coupled variable capacitance. By using an equivalent circuit, the behavior of the device can be quantitatively analyzed. Based on this, the device working mechanism can be well explained with a conductive filament model.

A 150 nm undoped Ga₂O₃ film was deposited on a quartz substrate by radio frequency sputtering with a Ga₂O₃ target under a power of 60 W. The sputtering pressure was 0.4 Pa, and the substrate was not intentionally heated. Details of growth can be found elsewhere.^19–21 The surface of the film shows a mirror-like flatness, with a measured roughness <1 nm in a 3 × 3 μm² scan area by atomic force microscopy (Fig. S1 of the supplementary material). As indicated by x-ray diffraction measurement (Fig. S2 of the supplementary material), the material is amphoteric Ga₂O₃ (a-Ga₂O₃).^19,22 The transmittance/reflectance measurements show that the bandgap of the material is around 4.8 eV (Fig. S3 of the supplementary material).²² X-ray photoelectron spectroscopy (XPS) shows oxygen binding energy at 530.8 and 531.7 eV [Fig. 1(a)], corresponding to the lattice oxygen of Ga₂O₃ and oxygen vacancy (VO) related defects. It should be noted that the peak at 531.7 eV indicates the existence of oxygen vacancies.²² The device was fabricated by depositing ITO interdigital electrodes on the film with standard UV lithography. The gap between two ITO interdigital electrodes is 5 μm, with 75 pairs in total [Fig. 1(b)]. The electronic property of the device was measured using a probe station in an electronically shielded box equipped with a Keithley 2636B source meter (SMU), with a current resolution down to 2 fA. The alternative current (AC) frequency response was measured with a function generator (Rohde and Schwarz HMF2525) and a digital oscilloscope (Rohde and Schwarz RTB2002).

When V = 0, the current is solely dependent on capacitance and the sweeping speed, which will be reflected by the difference between I_F and I_R (defined as “opening” in this paper). To prove this prediction, IV measurements with various sweep speeds have been carried out. The near V = 0 IV plots are shown in Fig. 2. Obviously, the opening decreases with decreasing sweeping speed (or increasing step time in the measurement), indicating that the total value of $CfΔVΔt$ is decreasing. At the same time, an oscillation of current with the applied voltage can be observed, with variable peak-to-peak amplitude and period related to the voltage sweeping speed (Fig. S5 of the supplementary material).

In order to understand the working mechanism, we further analyzed the capacitance at different sweeping speeds. Figure 3 shows the capacitance obtained at V = 0 from Eq. (1), plotted as a function of $ΔVΔt$⁠. The result indicates that the capacitance is not a constant, with a decreasing value with increasing sweeping speed. A conducting filament model schematically shown in Fig. 3(b) is used to understand our device.^24–27 When there is no bias, the oxygen vacancies are distributed randomly inside Ga₂O₃, where no conductive filament forms. Under an external bias, the oxygen vacancies are driven toward the direction of the electrode, and adjacent vacancies subsequently form a conductive filament. Different from the existing reports where a very thin Ga₂O₃ film (<100 nm) was employed to facilitate the filament connecting two electrodes,^28–30 the distance between two electrodes is 5 μm in our device. Therefore, the conductive filaments do not span the entire device. Instead, only the length of the non-filament region (d) is reduced. As a result, the resistance (R ∝ l − d, where l is the length of the material between two electrodes) will decrease and capacitance will increase (⁠$C∝1d$⁠). Regarding the current oscillation, it is attributed to temporal polarization and depolarization processes [Fig. 3(c)].³¹ 

Finally, we measured the AC response of our C-memristor. The top panel (black curve) of Fig. 4(a) shows an applied square wave at 7 kHz, while the bottom one (red curve) shows the measured response of the C-memristor. By fitting the time decay into $I=A⁡exp−tτ+B$ (where A, B, τ are fitting parameters), a time decay of τ = 5.9 μs was obtained (blue curve), corresponding to a maximum working frequency of ∼169 kHz. Figure 4(b) shows the measurements up to 30 kHz with our current setup, exhibiting a stable time constant.

In summary, we have realized a C-memristor based on a-Ga₂O₃ film with ITO interdigital electrodes. The device exhibits a clear current hysteresis with coupled capacitance, where the latter decides the current differences between forward and reverse scanning (speed). We established an equivalent circuit for quantitatively studying the details of electrical properties. Together with a conductive filament model, the results and device working mechanism can be well explained. In addition, the device offers facilitated readout binary electronic states (polarity) and a fast-working frequency of up to 169 kHz. We envisage that coupled capacitance will offer an additional degree of freedom for building memristor-based neural networks.^1,32,33

See the supplementary material for the surface flatness measured by AFM (Fig. S1), XRD scans (Fig. S2), transmittance/reflectance measurements (Fig. S3), a full-image of the device (Fig. S4), and detailed IV measurements at various voltage sweeping speeds (Fig. S5).

This work was supported by the EPSRC under Grant No. EP/T019085/1, Royal Society under Grant No. IEC\NSFC\242145, and SACEME Seedcorn funding from Swansea University. H.L. and Z.M. thank the National Natural Science Foundation of China for the support under Grant Nos. 12174275 and 62174113.

The authors have no conflicts to disclose.

Alfred Moore: Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – original draft (equal); Writing – review & editing (equal). Lijie Li: Conceptualization (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Resources (equal); Supervision (lead); Writing – original draft (equal); Writing – review & editing (equal). Hang Shao: Methodology (equal). Xiaoyan Tang: Methodology (equal). Huili Liang: Funding acquisition (equal); Methodology (equal); Resources (equal); Supervision (equal); Writing – original draft (equal); Writing – review & editing (equal). Zengxia Mei: Funding acquisition (equal); Methodology (equal); Resources (equal); Supervision (equal); Writing – original draft (equal); Writing – review & editing (equal). Yaonan Hou: Conceptualization (lead); Data curation (equal); Formal analysis (lead); Funding acquisition (equal); Methodology (equal); Project administration (lead); Resources (equal); Supervision (equal); Writing – original draft (lead); Writing – review & editing (lead).

The data that support the findings of this study are available from the corresponding author upon reasonable request.