A non-destructive reversible resistive switching is demonstrated in single crystals of Cr-doped Mott insulator Ca2RuO4. An applied electrical bias was shown to reduce the DC resistance of the crystal by as much as 75%. The original resistance of the sample could be restored by applying an electrical bias of opposite polarity. We have studied this resistive switching as a function of the bias strength, applied magnetic field, and temperature. A combination of 2-, 3-, and 4-probe measurements provide a means to distinguish between bulk and interfacial contributions to the switching and suggests that the switching is mostly an interfacial effect. The switching was tentatively attributed to electric-field driven lattice distortions which accompany the impurity-induced Mott transition. This field effect was confirmed by temperature-dependent resistivity measurements which show that the activation energy of this material can be tuned by an applied DC electrical bias. The observed resistance switching can potentially be used for building non-volatile memory devices like resistive random access memory.
I. INTRODUCTION
Tuning materials' properties by electrical means, e.g., by applying voltage, remains at the forefront of scientific and applied research because it can generate novel scientific phenomena and has an immense potential for future electronic devices. For instance, in conventional semiconductors like Si, the electronic band structure is fixed by the crystal structure and chemical composition and so defines the materials' transport and optical properties, which ultimately determine the performance of semiconductor devices such as diodes, transistors, and lasers. With an electrical control over the band structure, it would be possible to create a new generation of electronic and optical devices with enhanced functionality and flexibility. Such an electrical control has been previously demonstrated in 2D systems such as electrically gated bilayer graphene1–3 and bulk 3D systems such as Sr2IrO44 and Sr3Ir2O7.5 Electrically driven metal-insulator transitions in transition-metal oxides6,7 are of particular interest for practical use in electronic devices like resistive random access memory (Re-RAM).8–10 Here, metallic (low-) and insulating (high-resistance) states are used to encode information, while its reading and writing involve resistance sensing and electrically stimulated change of the states. It is essential for a functional device that the switching between the high- and low-resistance states could be driven at room temperature by low bias voltages. Recently, such a switching has been demonstrated in the multiband Mott insulator Ca2RuO4.11 However, Ca2RuO4 single crystals were found to disintegrate in the process of metal-to-insulator transition,11 which makes them impractical for applications.
Here, we report a study of electronic transport properties of Cr-doped (2.5%) Ca2RuO4 single crystals under high electrical biases. Our temperature-dependent resistivity measurements unambiguously demonstrate that the activation energy in this material depends on the applied DC electrical bias. This bias dependence allows one to drive continuous variations in the sample resistivity by more than 75% followed by a reversible resistive switching at higher biases. The switching between high- and low-resistance states is reversible, i.e., biases of different polarity can be used to “write” the system into different states and is also non-destructive, i.e., a repetitive switching between the different states can be realized. Our results suggest that Cr doping of Ca2RuO4 can be an effective route toward finding a material with desired switching characteristics for Re-RAM devices based on Ca2RuO4. We also found that switching is not affected by applied magnetic fields, which makes it robust to magnetic perturbations, but depends on the strength of the applied bias, which may potentially be used in devices with multilevel switching capabilities.
II. SAMPLES
Single crystals of Ca2RuO4 doped by Cr (2.5%) were grown by both the flux and the floating zone techniques.12 The Cr-doped Ca2RuO4 crystallizes in the K2NiF4 structure (Pbca space group) with an intergrowth of the perovskite layer and the rock salt layer13 and becomes antiferromagnetically ordered below the Néel temperature TN ≈ 110 K;14,15 the crystal structure was determined by single-crystal x-ray diffraction. Two Ag-paste contacts (area ∼ 0.1 mm2) were made on the opposite edges of a single-crystalline Ca2RuO4 flake with a thickness of about 0.2 mm and the c axis is normal to the flake's surface (area ∼ 0.5 mm2). The two Ag contacts were used to measure the sample resistance in a two-probe geometry; three- and four-probe measurements were performed (with two extra Ag contacts) to distinguish between bulk and interfacial contributions to the measured resistance. In magnetic fields (up to 250 mT) applied parallel and perpendicular to the c axis and at temperatures from 150 to 300 K, we have characterized the DC transport properties of Cr-doped Ca2RuO4 by performing: (i) temperature-dependent resistivity measurements and (ii) current-voltage (I-V) characteristics.
III. RESULTS AND DISCUSSION
Figure 1(a) shows the temperature dependence of Ca2RuO4 2-probe resistance R(T) at a small bias (I = 0.1 mA). The inset to Fig. 1(a) shows the corresponding Arrhenius plot – ln(R) vs 1/T. From the plot's slope, we extracted [Fig. 1(b)] the activation energy at low bias (Δ0 ≈ 100 – 240 meV for T = 150 – 240 K). By performing a similar analysis at different applied biases, we were able to reconstruct the bias dependence of the activation energy Δ at a fixed temperature. Figure 1(c) shows an example of such a reconstruction (Δ vs V) at T = 200 K. The activation energy in Ca2RuO4 decreases with increasing bias similar to the behavior observed previously in Sr2IrO44 and Sr3Ir2O7.5 In the case of iridates, the decrease of the gap has been attributed to the electric field-induced lattice distortions and their effect on electronic states and transport properties.4,5 Therefore, it is tempting to suggest that in experiments with Ca2RuO4, high local electric fields can also alter the equilibrium positions of oxygen ions with respect to ruthenium and stimulate distortions of the corner-shared RuO6 octahedra, thus provoking modifications of the localized states and electronic structure. Such an electric-field effect can be taken into account with the field-effect model successfully used to fit similar temperature-dependent data in iridates4,5
where Δ0 is the activation energy at zero bias as extracted from Fig. 1(b), kB is Boltzmann constant, and R0 and A are fitting parameters.
Figure 2(a) shows the variation of Ca2RuO4 resistance R = V/I as a function of the applied bias voltage V at room temperature; the inset shows the corresponding I vs V characteristic. Arrows indicate the V-sweep directions: starting from zero bias (high-resistance state; solid red circle), the sample resistance decreases from about 2 kΩ down to ∼200 Ω as the positive bias increases up to 1.5 V. When the bias is then decreased back to zero, the sample remains in the low-resistance state (solid blue square). Only after the bias is reversed, increased to −1.5 V, and then decreased to zero, the resistance goes back into the original high-resistance state (R ≈2 kΩ; solid red circle). The R(V) curve thus exhibits a large clock-wise hysteresis. Such a hysteresis is consistent with a first-order Mott transition16,17 during a V sweep observed previously in undoped Ca2RuO4.11 We should note, however, that there is a significant difference between the switching effects observed in Ref. 11 and in Fig. 2(a). In Ref. 11, the switching between high- and low-resistance states occurs at the same polarity of the applied voltage bias, i.e., the “flowing current” at a nonzero (e.g., positive) bias voltage plays a key role in maintaining the induced metallic state that disappears at zero bias. In contrast, opposite bias polarities are required to switch from high-to-low and low-to-high resistance states in Fig. 2(a); i.e., both high- and low-resistance states can be stabilized at zero bias without the “flowing current.”
Three R vs V curves [Fig. 2(a)] measured in magnetic field (applied in the ab plane) of 0 mT (open squares), 67.5 mT (triangles), and 135 mT (circles) illustrate the reproducibility of our measurements and the absence of any significant magnetic-field effects on the transition. The opening of the R(V) hysteresis depends on the maximum applied bias V. Four R vs V curves in Fig. 2(b) (open down-triangles, up-triangles, circles, and squares) show R(V) loops for four different values of the maxim positive bias applied to the sample (0.8 V, 1 V, 1.2 V, and 1.5 V) while the maximum negative bias remains the same (−1.5 V). The larger the applied bias, the larger the hysteresis and the corresponding change in resistance. Here, the low-resistance state has four different values (0.45, 0.8, 1.2, and 1.5 kΩ) as indicated by solid blue squares in Fig. 2(b). This finding may be potentially used for multilevel switching, e.g., in multistate Re-RAM.
To test the reliability and reproducibility of switching between high- and low-resistance states, we have set a repetitive switching experiment between different states. From a R(V) curve (like in Fig. 2), we choose two states (solid red circle and blue square) that can be clearly distinguished at a low bias of 0.1 V; here, high- and low-resistances represent “0” and “1” states, respectively, of our Ca2RuO4 memory cell. A positive bias (1.5 V) is then used to switch the system from “0” to “1,” while a negative high bias (−1.5 V) is used to switch it back from “1” to “0”. Both “0” and “1” states can be read out by probing its resistance at a low bias of 0.1 V. In Fig. 3(a), an alternating voltage profile is applied to the sample to simulate the writing on a memory device. The applied voltage goes continuously through the sequence: +1.5 V, 0.1 V, −1.5 V, and 0.1 V. High biases are used to switch (write) the system between high- and low-resistance states, while low biases are used to read out the states. We routinely perform >200 write-read cycles in our experiments as illustrated in Fig. 3(a). Here, open circles and left scale show the applied voltage sequence, while solid symbols and right scale show the measured resistance of the device. Figure 3(b) shows the resulting distributions of the measured resistance values of the two states, which indicate a larger uncertainty in the high-resistance state. High- and low-resistance states are clearly separated with the high/low resistance ratio of about 400%. These results show that switching in our Cr-doped Ca2RuO4 crystals is robust and makes this material a promising candidate for Re-RAM applications.
In pure Ca2RuO4, the insulator-to-metal transition (IMT) was previously associated with an electric field-induced structural transition11 that results in a pronounced change in its transport properties. Here, the structural transition refers to a change in the material structure of Ca2RuO4 which was probed in Ref. 11 by X-ray diffraction. A valid alternative to this picture is the impurity-driven IMT where the Mott insulator becomes a metal via excitation of carriers from an impurity bound state (indirect band) to its conduction band.18,19 In Ca2Ru1-xCrxO4, the Cr doping is expected to induce holes in the lower Hubbard band. It is possible that the local hole concentration is altered by the electric field (applied bias) that triggers IMT. This impurity-driven mechanism suggests an enhanced effect near interfaces because they have more impurities compared to the bulk. Next, we examine whether the transition is associated with the bulk or interfacial resistance switching, i.e., whether the switching occurs in the bulk of a sample or near the interfaces between the sample and electrodes (Ag-paste contacts in our experiments). In order to distinguish between the bulk and interfacial contributions, we carried out 2-, 3-, and 4-probe resistance measurements using four Ag-paste contacts to another piece of Ca2RuO4 single crystal.
Figures 4(a)–4(d) show R(V) curves recorded in 2-, 3-, and 4-probe measurements at room temperature; insets show the corresponding wiring schemes. In a two-probe configuration of Fig. 4(a), the resistance is measured across the same two contacts used to supply current and is expected to include resistance contributions from the sample bulk and two interfaces between the sample and Ag contacts #1 and #4. The three-probe configurations of Figs. 4(b) and 4(c) involve a bulk contribution and only one Ag contact: #1 for (b) and #4 for (c). Finally, the four-probe configuration in Fig. 4(d) is expected to provide only the bulk resistance without any contributions from interfaces. Note that the different configurations have different resistances that defines different V-scales for Figs. 4(a)–4(d). This difference is particularly noticeable in the four-probe configuration of Fig. 4(d), where the resistance is about two orders of magnitude smaller than in the other three configurations. Here the measured small variations in resistance [grey data points in Fig. 4(d)] have a relatively high noise level. To increase the signal-to-noise ratio in this configuration, we have averaged 100 R(V) curves [solid black curve in Fig. 4(d)].
To summarize the outcome of these measurements, (i) a hysteresis in R(V) is observed in all four configurations, although the hysteresis opening and the corresponding difference between the high- and low- resistance states (Rhigh and Rlow) probed at 0.1 V (see solid red circles and blue squares in Fig. 4) vary significantly in different configurations. To quantify the difference, we use the ratio (Rhigh – Rlow)/Rlow, which is the largest (2%–2.5%) in three-probe measurements and lowest (0.16%) in four-probe measurements. The two-probe geometry shows an intermediate effect (0.7%). We thus conclude that interfaces contribute more to the resistance change, which may be partially associated with higher electric fields inevitably present at interfaces due to band bending. Another observation that can be made from data in Figs. 4(a)–4(d) is (ii) a different sense of circulation of the hysteresis loops: clockwise in (a), (b), and (d), and anti-clockwise in (c). In all three-probe measurements we have performed on our Ca2RuO4 crystals [like those in Figs. 4(b) and 4(c)], the direction of the loop circulation is different for the two interfaces that supply electrical bias to the sample. This is consistent with the opposite polarity of electric fields near the two interfaces and supports our previous conclusion of the predominant interfacial contribution to the resistance change. The exact magnitude of the resistance change from any given interface/contact is ultimately defined by local properties of the contact of interest and, unfortunately, is not readily controlled in our experiments with Ag-paste contacts. This problem calls for further investigations with other contact types to maximize the switching effect. In our experiments, we found that the resistance change measured in two-probe geometry, like the one in Fig. 4(a), is a simple algebraic sum of individual contributions from the two involved contacts/interfaces [Figs. 4(b) and 4(c)] and the bulk [Fig. 4(d)]. Since the two interfaces produce resistance changes of opposite polarities, they partially compensate each other and result in a smaller (intermediate) effect in the two-probe geometry.
Finally, we studied the effects of temperature on the resistive switching effect. Red and blue traces in Fig. 5 show the temperature dependencies of Rhigh and Rlow states, respectively. The resistance states were extracted from R(V) measurements like those in Fig. 2. Inset to Fig. 5 shows the temperature dependence of the (Rhigh – Rlow) difference. It is obvious that low temperature suppresses the switching; the switching effect is present only above 260 K. At low temperatures, there is no switching, and the sample resistance increases exponentially with decreasing temperature in agreement with a simple picture of thermally activated transport [see Eq. (1) and Fig. 1(a)]. We tentatively attribute the suppression of the switching at low temperatures to suppressed lattice distortions/structural transition; the latter could be associated with the L-Pbca to S-Pbca phase transition which in a pure Ca2RuO4 occurs at about 360 K but may be lowered to 260 K by Cr doping. A change in the slope of R(T) around 260 K in Fig. 5 is consistent with this hypothesis. However, further studies are needed to verify whether the anomaly at 260 K is indeed associated with the structural transition.
IV. CONCLUSION
To conclude, electrically tunable band-gap and non-destructive resistive switching were all found in the Cr-doped (2.5%) Mott insulator Ca2RuO4. We observed a continuous reduction in the resistivity of Ca2RuO4 single crystals as a function of increasing bias followed by an abrupt switching at higher biases. These results follow a chain of observations in other oxide systems such as Sr2IrO44 and Sr3Ir2O75 and corroborate the idea of an electric-field effect on electronic states in these materials. This field effect was confirmed in Ca2RuO4 by temperature-dependent resistivity measurements that show that the activation energy of this material can be tuned by an applied DC electrical bias. Our two-, three-, and four-probe resistance measurements suggest that the resistive switching observed in Ca2RuO4 is primarily associated with the interfacial regions between the sample and contact (Ag-paste) electrodes. The switching was tentatively attributed to electric field-driven lattice distortions that accompany the impurity-induced Mott transition.18,19 The reversible resistive switching can potentially be used as a switching mechanism in industrial applications, e.g., in non-volatile memory devices like a resistive random access memory (Re-RAM). Our observation of the non-destructive switching in Cr-doped Ca2RuO4 opens a viable pathway to solve the disintegration problem in resistive switching oxides and, with further optimization of dopant material and doping level, can provide optimal performance (including resistance ratio and repetition characteristics) for such applications.
ACKNOWLEDGMENTS
This work was supported in part by C-SPIN, one of six centers of STARnet, a Semiconductor Research Corporation program, sponsored by MARCO and DARPA, by NSF Grant Nos. DMR-1712101 and DMR-1122603, and by the King Abdullah University of Science and Technology (KAUST) Office of Sponsored Research (OSR) under Award No. OSR-2015-CRG4-2626.