Flexible memristive devices based on polyimide:mica nanosheet nanocomposites with an embedded PEDOT:PSS layer

Flexible memristive devices with a structure of Al/polyimide:mica/poly(3,4-ethylenedioxythiophene) polystyrene sulfonate/indium-tin-oxide/polyethylene glycol naphthalate showed electrical bistability characteristics. The maximum current margin of the devices with mica nanosheets was much larger than that of the devices without mica nanosheets. For these devices, the current vs. time curves showed nonvolatile characteristics with a retention time of more than 1 × 104 s, and the current vs. number-of-cycles curves demonstrated an endurance for high resistance state/low resistance state switchings of 1 × 102 cycles. As to the operation performance, the “reset” voltage was distributed between 2.5 and 3 V, and the “set” voltage was distributed between −0.7 and −0.5 V, indicative of high uniformity. The electrical characteristics of the devices after full bendings with various radii of curvature were similar to those before bending, which was indicative of devices having ultra-flexibility. The carrier transport and the operation mechanisms of the devices were explained based on the current vs. voltage curves and the energy band diagrams.


Results and Discussion
(a) shows a typical TEM image of the exfoliated mica nanosheets. Characterization by using AFM detected a sheet-like morphology with a thickness ranging from 1 to 4 nm and with a layer number ranging from 1 to 3, as shown in Fig. 1(b) 14 . The thinnest sheets corresponded to monolayer nanosheets. A schematic of the memristive devices is shown in Fig. 1(c,d) shows a cross-sectional SEM image of the memristive device without top electrodes. The SEM image shows that the PI:mica and the PEDOT:PSS layers were precisely deposited on the ITO-coated PEN substrate. The thicknesses of the PEDOT:PSS layers and the PI:mica layer, as determined form the SEM image, were 338 and 424 nm, respectively. Note that after an ultrasonic treatment for 4 h, the exfoliated muscovite-type mica was uniformly mixed with the PI layer in the NMP solvent. As a result, a homogeneous solution of exfoliated mica and PI without any precipitate was obtained, as shown in inset of Fig. 1(e). Because the exfoliated muscovite-type mica was uniformly distributed in the PI solution, the mica nanosheets should be uniformly distributed in the PI:mica layer, as shown in the SEM image in Fig. 1(e). The AFM images in Fig. 1(f) show the roughness of the PI:mica layer. The value of the root-mean-square (RMS) roughness of the PI:mica layer was 5.377 nm.
Because the PI:mica nanocomposite forms a uniform structure, the memristive device exhibits stable memory switching characteristics. The I-V curves in Fig. 2(a) show the switching characteristics of the memristive device. The black and the red circles represent the I-V curves of the Al/PI:mica/PEDOT:PSS/ITO device and the Al/PI/PEDOT:PSS/ITO device, respectively. The I-V curve of the Al/PI:mica/PEDOT:PSS/ITO device shows hysteresis characteristics, which is an essential feature of memristive memory devices 23 . Two different resistance states, the high resistance state (HRS) and the low resistance state (LRS), were clearly observed. Initially, the device in its initial state stayed in the HRS. When the negative bias voltage reached −0.3 V, the current increased from 1.03 × 10 −6 to 1.22 × 10 −2 A, indicative of the switch from the HRS to the LRS. The HRS-to-LRS transition corresponds to a set process in a memory cell. After the device had switched from the HRS to the LRS, it stayed in the LRS. However, when the positive bias voltage was increased to 3 V or more, the current decreased from 4.15 × 10 −2 to 3.29 × 10 −5 A, as shown in Fig. 2(a), indicative of a switch from the LRS to the HRS. The LRS-to-HRS transition corresponds to a reset process in a memory cell. The LRS/HRS current ratio of the device containing mica at a reading voltage of 1 V was approximately 4.28 × 10 3 , as shown by the black circles in Fig. 2(a). However, the device without the mica nanosheets did not clearly show any the memory characteristics, as shown by the red circles in Fig. 2(a). Therefore, one can conclude that the mica nanosheets play an important role in the appearance of memory characteristics for the Al/PI:mica/PEDOT:PSS/ITO device. One should note that mica concentration of 1 wt% used in this work is the concentration required for optimal device performance. When the mica concentration was lower than 1 wt%, the ON/OFF ratio of the device gradually increased with increasing mica concentration. This behavior indicated that the mica was acting as carrier trap sites in the device, resulting in an increase in the number of electrons trapped in the active layer. However, the ON/OFF ratio decreased with increasing mica concentration when the mica concentration was higher than 1 wt%, which might results from the aggregation of mica nanosheets in the active layer.
Scientific REPORTS | (2018) 8:12275 | DOI:10.1038/s41598-018-30771-5 Figure 2(b) shows the endurance reliability results for the Al/PI:mica/PEDOT:PSS/ITO device. The endurance reliability measurements were carried out at room temperature. The memristive device could be repeatedly switched between the LRS and the HRS for 100 cycles without any noticeable deterioration in its performance. Figure 2(c) shows the retention performance of the device. The retention performances show the variations of the currents in the LRS and in the HRS as functions of time. As can be seen, a stable LRS/HRS current ratio was maintained for a time longer than 1 × 10 4 s, indicating that the device remained stable even during long operations. Figure 2(d) shows the probability distributions of the reset voltage (V reset ) and the set voltage (V set ) for the device. The values of the V reset for the device were dominantly distributed between 2.0 and 3.5 V, and those of the V set were mainly dispersed between −1.0 and −0.1 V. Therefore, the V set and the V reset for the devices could be defined as −1.0 V and 3.5 V, respectively.
The I-V curves for the Al/PI:mica/PEDOT:PSS/ITO devices before and after bending with bending radii of 10 and 20 mm are shown in Fig. 3(a). The I-V results demonstrate that the memory performances of the devices bent at different radii were similar to those of the device in its flat state, indicative of strong bending stability. The LRS/HRS current ratios of the devices after bending at radii of 20 and 10 mm were approximately 7.67 × 10 2 and 1.08 × 10 2 , as shown by the orange and the green circles in Fig. 3(a), respectively. Because the bending of the device might break the ITO electrode due to its fragility, resulting in a remarkable decrease in the LRS current, the LRS/HRS current ratio for the curving device should be reduced in comparison with that before bending 24 . The PEDOT:PSS film was used to improve the stability of the device when suffered to mechanical bending. The PEDOT:PSS thin film was used to improve the stability of the device while it was being subjected to repeated mechanical bending. As is well known, the poor wettability of polymer layers on ITO electrodes can lead to poor stability of the device because of an unfavorable disparity in the surface energy 25 . Especially, the repetitive mechanical curving could cause the device's performance to deteriorate. Therefore, a modified layer between the ITO electrode and the active film is important to the improvement of the stability of flexible devices. The LRS/ HRS ratio for the device without the PEDOT:PSS layer decreased under repeated bending, as shown in Fig. 3(b). However, the device with the PEDOT:PSS layer could be stably operated under repeated bending without any noticeable degradation. Figure 3(b) shows the endurance reliability of the device after bending. The device could be repetitively operated without any noticeable variation, regardless of bending, indicating that the electrical characteristics of the device were very stable. Figure 3(c) show the retention performances of the device after bending. The device maintains the LRS/HRS ratio for at least 1 × 10 4 s after bending, indicating that, regardless of bending, the device was very reliable, even for lengthy operations.
The device was repeatedly bent 30 times at different angles in order to further investigate characteristics of the flexible memristive device, and the I-V characteristics after bending were compared with those before bending (flat state). Figure 4(a) presents the geometry of the device when it is bent, where R is the bending radius, D is the thickness of the device, θ is the central angle, and L is the distance of the line connecting the two endpoints, which is the chord. The bending angle can be determined from the value of L after bending, as shown in Fig. 4(b,c) shows that the electrical properties of the device did not vary significantly as a result of its being bent. Figure 4(d) shows the distribution of the LRS/HRS ratios for 30 devices with bending different bending angles.
Several conduction models, as determined from fittings of the I-V curves on a log-log scale, can be used to clarify the carrier transport and memory mechanisms of the memristive device 26 , and Fig. 5(a,b) show the results of such fittings. When a set bias voltage was applied to the device, the device switched to the LRS, as shown in Fig. 5(a). The carrier transport mechanism of the device under negative bias voltage exhibited a linear relationship between I and V, indicative of Ohmic conduction. Thus, one can speculate that a conduction filament was formed in the PI:mica layer when the device stayed in the LRS 27 . However, the carrier transport mechanisms in the HRS were found to be related to both Ohmic conduction and space-charge-limited-current (SCLC) conduction. The slope of the fitting curve at low voltage for the HRS was approximately 1, as shown in Fig. 5(b), indicative of Ohmic conduction. However, the charge transport mechanism under low applied bias voltage was controlled by thermally generated free electrons, with the number of electrons injected from the electrode increasing with increasing electric field. Therefore, the carrier transport mechanism is dominantly affected by the injected electrons, resulting in their capture in trap sites. As a result, the current increased rapidly in a manner consistent with Child's square law (I~V 2 ). From the Ohmic region to the Child's square region, the threshold voltage coincides with a transition from a trap-unfilled state to a trap-filled state 28,29 . Thus, immediately after the transition to the   HRS, the charge transport mechanism exhibits a SCLC conduction behavior, with a slope of approximately 2, because some electrons remain trapped.
The set/reset operation mechanisms of the memristive device based on the charge-trapping mechanism are illustrated in Fig. 6(a-d). One should note that the mica nanosheets employed in this work had layer numbers ranging from 1 to 3 and bandgap energies ranging from 2.5 to 3.2 eV 14 . The trapping procedure corresponds to the memory process. The mica nanosheets act as electron trapping sites because the energy level of the conduction band for the mica nanosheets is lower than the lowest unoccupied orbital (LUMO) level of the PI layer, as shown  in Fig. 6(a). The electrons emitted from the Al electrode in the HRS are partly captured in the mica nanosheets as trapped electrons, so the HRS current of the device at low voltages can be attributed to the SCLC process, as shown in Fig. 6(b) [30][31][32] . When higher voltages are applied, the number of injected electrons increases, and more electrons are captured by the mica nanosheets. As a result, the electron occupation probability for the mica increases as the Fermi level shifts to the LUMO level of the PI layer [30][31][32] . When the trap sites are fully occupied by electrons, the device switches from the HRS to the LRS, as shown in Fig. 6(c) 33 . The LRS current is attributed to the Ohmic process because of the trap-filled state and the possible presence of filamentary channels 34 . Even after the applied voltage is removed, the device remains in the LRS, indicative of non-volatility. When a positive voltage is applied, the device remains in the LRS until the reset voltage approaches 3 V. When that happens, the electrons trapped in the mica are emitted due to the positive voltage, as shown in Fig. 6(d), which corresponds to the erase process in the memory 33 .

Conclusion
Flexible memristive devices based on PI:mica nanocomposites were investigated due to their superior mechanical flexibility. The bistable characteristics of the memristive devices were investigated in order to understand their stabilities and conduction mechanisms. The mica nanosheets embedded in the PI layer enhanced the resistive switching characteristics. The LRS/HRS current ratio for the devices before bending was 4.28 × 10 3 , and the LRS/ HRS current ratios after bending at radii of 20 and 10 mm were 7.67 × 10 2 and 1.08 × 10 2 , respectively. The endurance and the retention results before and after bending demonstrated that the flexible devices were stable and reliable. According to the fittings of the I-V curves, the carrier transport mechanism in the LRS could be attributed to Ohmic conduction and those in the HRS to Ohmic and SCLC behaviors. Furthermore, the mechanisms of the devices could be explained by using an energy band diagram. These results indicate that mica nanosheets should be useful for potential applications in flexible memristive devices.