Tuning of resistive memory switching in electropolymerized metallopolymeric films

Sandwiched electrical devices of an electropolymerized diruthenium metallopolymeric film show excellent resistive memory switching.


Introduction
Thin lms of semiconducting organic and polymeric materials have received tremendous interest for a wide range of optical and electrical applications. 1 Among them, resistive memory devices based on these lms hold great promise for highdensity data storage with a miniaturized device size. 2 A resistive memory, or memristor, operates as an electrical switch between high and low conductivity states (multi-states are possible) and remembers its present resistance when the electric power supply is turned off. 2,3 Organic and polymeric materials have a number of advantages for use as memory elements, such as structural tunability and diversity, good scalability, low cost, low power consumption, exibility, multilevel storage, and large capacity. 2,4,5 The molecular design and lm formation of the active materials are crucial to the performance of memory devices.
Vacuum-deposited or solution-processed lms of small organic molecules (oen with multiple charge-trapping sites) have been reported to exhibit excellent memory behaviour with high ON/ OFF current ratios and multilevel storage. 4 Memory devices based on solution-processed polymeric lms (oen with donoracceptor structural components) have shown promising switching performance. 5 Recently, the incorporation of transition metal complexes into memory devices has also received Fig. 1 Representative known transition metal complexes or metallopolymers as active layers for resistive memory. The thin films of these materials were prepared by spin-coating. much interest (Fig. 1), due to their well-dened and tunable redox properties. 6 For instance, Higuchi has fabricated memristive devices using electrochemically active Co(III) polymers. 7 Poly-N-vinylcarbazoles with on-chain Eu(III) or Ir(III) complexes have been demonstrated to give bistable or ternary memory devices. 8 Goswami reported an azo anion radical complex of Rh(III) as an active layer for molecular memory switching devices. 9 Despite these advances, materials with tailored electronic properties, in conjunction with a good lm formation method, are still in urgent demand for developing highperformance memory devices.
Electropolymerization is a very convenient method for the formation of thin lms, where the polymerization is electrochemically initiated and the polymers are deposited in situ on electrode surfaces to afford adhesive lms. 10 This procedure signicantly shortens the experimental time and avoids the solubility issues oen encountered with other methods. In addition, the equipment needed to carry out electropolymerization is much simpler and cheaper relative to vacuum deposition. Electropolymerized lms of organic or organometallic monomers have been reported to show memory functions with optical outputs. 11 However, both inputs and outputs in the form of electrical signals are preferred for practical data storage technologies. We present herein the rst example of using a metal-containing electropolymerized thin lm as the active layer for promising resistive memory devices.
Complex 1(PF 6 ) 4 shows two cathodic redox waves at À0.31 and À0.79 V vs. Ag/AgCl (Fig. 3a), corresponding to the tppz 0/À and tppz 2À/À processes, respectively. 14 The redox peaks at around À1.50 V are due to the reduction of the tpy ligands. In the initial anodic scan, an oxidation peak at +1.18 V was Scheme 1 Synthesis of (a) poly-1 4+ and (b) poly-2 2+ via the oxidative electropolymerization of 1(PF 6 ) 4 and 2(PF 6 ) 2 , respectively. The counteranions of the polymers are mostly ClO 4 À ions, which were included from the electrolyte during the electropolymerization. observed (Fig. 3b). When the potential was scanned repeatedly between +0.40 and +1.35 V at a Pt disk electrode, the current in the cyclic voltammogram (CV) increased gradually and continuously with the appearance of two new redox waves. This indicated that the oxidative electropolymerization of 1(PF 6 ) 4 in CH 2 Cl 2 proceeded smoothly on the Pt electrode surface. Fig. 3c and d show the CVs of the obtained poly-1 4+ /Pt lm. The cathodic waves are less well-dened with respect to those of the monomer. However, the tppz 0/À and tppz 2À/À processes of the polymeric materials can be clearly recognized and occur at similar potentials (À0.34 and À0.81 V). In the region between À1.0 and À1.8 V, poly-1 4+ displays some irreversible or quasireversible waves. These waves are due to the reduction of tpy ligands and possibly complicated by some charge-trapping peaks. The anodic scan of the polymer shows four well-dened consecutive waves at +0.90, +1.05, +1.44, and +1.74 V, respectively. The former two waves are due to the Nc +/0 processes of the tetraphenylbenzidine segments of the polymers. 13, 15 The latter two peaks are due to the stepwise Ru III/II processes of the diruthenium segment. 14 Both anodic and cathodic currents are linearly dependent on the scan rate ( Fig. S1 in the ESI †), which is characteristic of redox processes conned to electrode surfaces. The rst three anodic waves are chemically reversible. When the potential was scanned beyond +2.0 V, a fourth wave at +1.74 V appeared, which was much higher in current with respect to the rst three waves. It is possible that further irreversible oxidation of the aminium radical cations is involved in the fourth wave, which causes the return reduction waves to differ signicantly from those scanned at voltages no more positive than +1.6 V. On the basis of the electrochemical results, the LUMO and HOMO energy levels of poly-1 4+ are estimated to be À4.4 and À5.6 eV vs. vacuum, respectively.
The presence of the well-dened Nc +/0 and Ru III/II redox waves indicates the polymer has the expected linear structure with alternating tetraphenylbenzidine and diruthenium structural segments. The possibility of further chain propagation on the phenyl groups of the tetraphenylbenzidine unit to form a cross-linked structure should be low. Otherwise, the Nc +/0 waves would be very complex due to the presence of a strongly-coupled multi-triarylamine structural component. The head-to-tail oxidative electropolymerization mechanism and similar alternating polymer structures have been proposed for other related compounds. 13,16 The FTIR spectrum of 1(PF 6 ) 4 shows an intense peak at 843 cm À1 due to PF 6 À stretching (Fig. S2 †). The poly-1 4+ sample, obtained by scratching the polymeric lm off the electrode surface, shows the disappearance of this signal. Instead, a strong signal at 1089 cm À1 , assigned to ClO 4 À anions, is observed. 17 This indicates that the counteranions of poly-1 4+ are largely ClO 4 À ions, incorporated from the electrolyte ( n Bu 4 NClO 4 ) during electropolymerization. A similar electropolymerization process was performed using 1(PF 6 ) 4 on an indium-tin-oxide (ITO) glass electrode to afford a polymeric thin lm for memory device fabrication. Fig. 4a shows the surface morphology of the obtained thin lm measured by atomic force microscopy (AFM), which shows a Fig. 2 ORTEP drawing of the single-crystal X-ray structure of 1(PF 6 ) 4 at 30% probability. Anions and H atoms are omitted for clarity. Color code: carbon, grey; nitrogen, blue; pink, ruthenium.  homogenous surface texture with a mean roughness, rms, of 5.5 nm. The thickness of the lm was estimated by measuring the step height produced by scanning across a scratching edge (Fig. S3 †). X-ray photoelectron spectroscopy (XPS) of the lm shows bands for O 1s (532.1 eV), N 1s (400.1 eV), C 1s (284.88 eV), Ru 3d 5 (281.28 eV), Cl 2p (207.46 eV), and Si 2p (102.05 eV) (Fig. S4 †), conrming the presence of the ruthenium and perchlorate ions.
The memory device was fabricated by sandwiching the polymeric lm between the substrate ITO electrode and a top layer consisting of an Al electrode (Fig. 4b). The current passing through the polymeric lm with an active area of 6.0 mm 2 was monitored under ambient conditions. Hysteretic currentvoltage (I-V) characteristics were observed for the as-prepared device (Fig. 4c), demonstrating a ash memory function. In the rst voltage sweep from 0 to +5 V, an abrupt increase in the current was observed at a switching threshold voltage of +3.4 V. This indicates that the device was switched from a lowconductivity state (OFF state) to a high-conductivity state (ON state), corresponding to the "write" process. The high-conductivity state was retained during the subsequent positive sweep (the second sweep from 0 to +5 V), implying that the data was memorized. One important feature of the present memory device is that the OFF state can be recovered by simply applying a reverse voltage (the third sweep), where an abrupt drop in current occurs at a switching threshold voltage of À4.3 V. This serves as the "erase" process for the memory device. The device remained in the stable OFF state during the fourth sweep from 0 to À5 V right aer the erase process. These I-V characteristics dene the electrical bistability of the device. The distinct electrical bistates between À4.3 and +3.4 V allow any voltage in this range to read as an OFF or ON signal depending upon the history of the voltage sweep, with an ON/OFF current ratio ranging from 100-1000 (Fig. 4d). Among y devices measured, eight devices displayed such well-dened I-V characteristics with an ON/OFF current ratio over 100. Circuit shortage is one reason for the low success rate. We hope that the device performance will be further improved by the optimization of device fabrication in the future.
The endurance of the above device as a random-access memory (RAM) device was examined by applying repeated write/ read/erase/read cycles (+5 V/1 V/À5 V/1 V) in pulse mode ( Fig. 5a  and b). Aer a write process at +5 V for 1.5 s, the device was immediately switched to the high-conducting ON state (current ¼ 1.4 Â 10 À3 A), followed by a read process at +1 V for 1.7 s (current ¼ 6.0 Â 10 À4 A). Aer that, an erase process at À5 V was applied for 1.5 s. The device was then switched to the lowconducting OFF state (current ¼ 4.6 Â 10 À5 A), which was again read out at +1 V for 1.7 s (current ¼ 3.8 Â 10 À6 A). No resistance degradation was observed when the device was tested for 500 write/read/erase/read cycles under ambient conditions, with an ON/OFF ratio over 200 (Fig. 5c). This suggests that the device using the poly-1 4+ lm has good stability and reproducibility.
The memory device displays a long retention time (Fig. 6). When the device was turned ON or OFF by applying a voltage greater than the threshold value, the high-or low-conducting state was retained aer 20 min under a small readout voltage (+1 V). During the test of the retention time for the ON state, the current dropped a little in the rst 3 min, and then remained at a steady state. It is possible that current is consumed during the rst few minutes to reach a more balanced conducting state.
The monoruthenium complex 2(PF 6 ) 2 was polymerized by a similar electrochemical oxidation method (Fig. S5 †). Poly-2 2+ displays similar Nc +/0 processes to poly-1 4+ at +0.91 and +1.04 V  vs. Ag/AgCl (the peak at +1.44 V is due to the Ru III/II process, Fig. 7a), which means that these two polymers have similar HOMO levels. Two cathodic redox waves at À1.21 and À1.45 V are observed for poly-2 2+ , associated with two tpy 0/À processes. The LUMO of poly-2 2+ is estimated to be À3.5 eV vs. vacuum, which is 0.9 eV more destabilized with respect to that of poly-1 4+ . A typical AFM height image of the poly-2 2+ /ITO lm is given in Fig. S6. † The sandwiched ITO/poly-2 2+ /Al device shows much poorer memory performance with respect to the poly-1 4+ device. Hysteretic I-V characteristics were observed for the device with the poly-2 2+ lm (Fig. 7b). However, no abrupt decrease or increase in current occurred, and the best ON/OFF ratio achieved was less than 15. This indicates that the presence of the bridged diruthenium structure in poly-1 4+ is crucial for the excellent memory function.
We propose that the mechanism of the eld-induced conductivity of the poly-1 4+ lm probably involves the formation of a charge transfer state. 18 DFT calculations for the basic diruthenium-tetraphenylbenzidine structure of poly-1 4+ show that the HOMO and LUMO energy levels are localized on the tetraphenylbenzidine unit and the tppz bridging ligand, respectively (Fig. 8). The Ru ions and tpy ligands play more important roles in lower occupied orbitals (e.g., HOMOÀ5) and higher unoccupied orbitals (e.g., LUMO+2), respectively. A high electric eld may facilitate intermolecular or intramolecular charge transfer from the tetraphenylbenzidine donor to the tppz acceptor, resulting in a high-conducting state with a high concentration of charge carriers. For poly-2 2+ with a much higher LUMO energy level, the formation of such a charge transfer state is difficult, and the corresponding device displays much poorer memory performance. DFT calculations for the basic monoruthenium-tetraphenylbenzidine structure of poly-2 2+ show that the energy gap between the tetraphenylbenzidine-dominated HOMO and the tpy-localized LUMO is 2.15 eV (Fig. S7 †), which is much larger relative to that of the basic diruthenium-tetraphenylbenzidine structure of poly-1 4+ (0.84 eV). This also suggests that the formation of a charge transfer state for poly-2 2+ is much more difficult with respect to poly-1 4+ .

Conclusions
In summary, we have demonstrated an electropolymerized lmbased single-layer electrical device that displays excellent resistive memory performance, including a high ON/OFF ratio, low operational voltage, good stability, and long retention time. We believe that the performance of the device can be further improved by capsulation and device optimization. Compared to spin-coating and vacuum deposition, lm formation by electropolymerization is a much more convenient and cheaper method. Considering that a large number of redox-active electropolymerized lms of both organic and organometallic monomers are available to date, 10 our work demonstrates an important alternative for the development of high-density memristor materials and devices with respect to those based on spin-coating or vacuum deposition of small molecules or organic polymers. In addition, this work represents another successful yet rare example of using transition metal complexes as the active layer for memristor devices. [7][8][9] The polymeric lms with the diruthenium complex exhibit much better memory performance with respect to those with the monoruthenium complex. The high HOMO energy level and low LUMO energy level arising respectively from the tetraphenylbenzidine and ruthenium-chelated tppz units in poly-1 4+ are crucial to the successful memory performance. This highlights the critical roles of the molecular design and electronic properties of materials in developing excellent molecular devices.

Electrochemical measurements
All electrochemical measurements were taken using a CHI 660D potentiostat under an atmosphere of nitrogen. All measurements were carried out in 0.1 M n Bu 4 NClO 4 in denoted solvents. The potentials are referenced to a Ag/AgCl electrode in saturated aqueous NaCl, ignoring the liquid junction potential. The working electrode was a home-made Pt disk electrode (d ¼ 2 mm) or a transparent ITO glass electrode (<10 U per square). The ITO glass was pre-cleaned with water, acetone, and then  2-propanol in an ultrasonic bath (15 min each), and dried in a nitrogen airow before use. A large area platinum wire coil was used as the counter electrode. A three-compartment electrochemical cell was used in the electropolymerization experiments. The working electrode (ITO glass) was positioned parallel to and opposite the counter electrode.

X-ray crystallography
The X-ray diffraction data were collected using a Rigaku Saturn 724 diffractometer on a rotating anode (Mo-K radiation, 0.71073Å) at 173 K. The structure was solved by the direct method using  and rened with Olex2. 20 A single crystal of 1(PF 6 ) 4 was obtained by slow diffusion of hexane into a solution in dichloromethane. Crystallographic data for 1(PF 6 ) 4 (CCDC 1023945): C 90 H 64 N 14 Ru 2 F 24 P 4 , M ¼ 2123.57, triclinic, space group P 1, a ¼ 13.811 (3)

XPS measurements
XPS spectroscopy data were obtained with an ESCALab220i-XL electron spectrometer from VG Scientic using 300 W Al Ka radiation. The base pressure was about 3 Â 10 À9 mbar. The binding energies were referenced to the C 1s line at 284.8 eV from adventitious carbon.

AFM images
AFM was carried out with a Brucker Multimode 8 using tappingmode with a scan speed of 1 Hz.

Fabrication and characterization of memory devices
An 80 nm-thick aluminum top electrode was thermally evaporated onto the polymer lm on ITO glass at a pressure of around 10 À6 Torr. The active area of the lm sandwiched between two electrodes was 2.0 mm Â 3.0 mm in size. The devices were characterized under ambient conditions, using a Keithley 4200 SCS semiconductor parameter analyzer.

Computational methods
DFT calculations were carried out using the B3LYP exchange correlation functional 21 and implemented in the Gaussian 09 package. 22 The electronic structures of the complexes were determined using a general basis set with the Los Alamos effective core potential LANL2DZ basis set for ruthenium and 6-31G* for other atoms. 23 Solvation effects in CH 2 Cl 2 were included by using the conductor-like polarizable continuum model (CPCM). 24 No symmetry constraints were used in the optimization (nosymm keyword was used). Frequency calculations have been performed with the same level of theory to ensure that the optimized geometries were local minima. All orbitals have been computed at an isovalue of 0.02 e per bohr 3 .

Synthesis
NMR spectra were recorded in the designated solvent on a Bruker Avance 400 MHz spectrometer. Spectral shis are reported in ppm values from the residual protons of the deuterated solvent. Mass data were obtained using a Bruker Daltonics Inc. Apex II FT-ICR or Autoex III MALDI-TOF mass spectrometer. The matrix for MALDI-TOF measurement was acyano-4-hydroxycinnamic acid. Microanalysis was carried out using a Flash EA 1112 or Carlo Erba 1106 analyzer at the Institute of Chemistry, Chinese Academy of Sciences.