A graphene integrated highly transparent resistive switching memory device

Resistive switching is a basic physical phenomenon in which resistivity of materials can be electrically modulated between nonvolatile high and low conducting logic states. The potential incorporation of resistive random access memory (RRAM) in the realm of next generation highly scaled electronic and optoelectronic circuits and neuromorphic computing is envisaged owing to its high operation speed (∼100 ps), high-density storage, low power consumption, excellent endurance (10¹² cycles), low cost, and simple cell architecture.^1–9 Recently, extensive interest has risen in visible transparent electronic circuit applications such as integrated solar panel, touch panel, and wearable displays¹⁰ and large-area sensors¹¹ in consumer electronics, defense, space, civilian, and transport sectors. Transparency of the display, logic, and memory units of the integrated circuit is essential to establish the above goal. High optical transparency and high electronic conduction are two mutually exclusive properties. Although researchers have demonstrated see-through transistors,^12,13 high transparency with efficient charge trapping is still a challenging task in mainstream floating gate thin film storage memories in addition to their physical scaling limitation. This issue can be overcome by RRAM that works on a non-charge-based mechanism¹⁴ as long as both resistance-active (oxide) material and electrodes possess more than 80% transparency throughout the visible spectrum.^15–17 Over the past few years, considerable effort has been made to address this existing challenge by developing a variety of new material combinations that include oxides, organics, two-dimensional conductors, and nanostructures in vertical and planar sandwiched architectures and device processing techniques.^15,18–22

With the motivation to design and evaluate transparent, lightweight, and scalable devices, we have studied the switching behavior of a transparent resistive random-access memory (TRRAM) fabricated on a glass platform with semiconductor process friendly Al₂O₃ as a functional oxide layer, intensively researched graphene as a top electrode, and conventional n-type (partly due to oxygen vacancies) ITO as bottom electrodes for non-volatile memory applications. Binary oxides such as HfO₂,²³ TiO₂,²⁴ NiO,²⁵ ZrO₂,²⁶ Nb₂O_5,²⁶ and Al₂O₃²⁷ have been extensively studied for the fabrication of RRAM due to their simple conformation, good compatibility with CMOS back end of line technology, and low-cost of fabrication. Al₂O₃ has drawn the particular attention of scientific community because of its moderately high permittivity of ∼8–10, wide optical bandgap of ∼9 eV with transparency down to deep UV (beneficial for the initial high resistance state), large breakdown electric field of ∼5–30 MV/cm, good thermal stability of as high as 1000 °C, and low cost.¹¹ Based on these properties, Al₂O₃ has found potential applications like high-k gate insulator in thin film transistors,²⁸ tunnel-control oxides in flash memory,²⁹ resistance switching layer in RRAM,³⁰ dielectric layer in metal/insulator/metal (MIM) capacitors,³¹ surface modification at the cathode in Li-ion batteries that improves the cycle stability of the cell,³² and gas diffusion barrier in organic electronic devices,³³ to name a few. Graphene, a two-dimensional crystal made up of just hexagonal covalent bonded carbon atoms, possesses unique electronic properties such as high charge carrier mobility (∼10⁴ cm²/V s), electrical (∼2000 S/cm) and thermal (∼5000 W/mK) conductivities, mechanical strength (∼40 N/m), and flexibility.^34–36 It is a promising candidate material for transparent conductor applications due to its low sheet resistance (∼30 Ω/□) and low light absorbance (only ∼2.3%)^37,38 and has the potential to replace the commercially available transparent conducting electrode, ITO, in near future. Being a good transparent conductor as well as oxygen passivation layer, graphene has shown to be an effective electrode material for various devices. For example, Yang et al.¹⁰ have shown that insertion of graphene layer above the oxide layer helps in reducing resistance variation of high resistance states. Lee et al.³⁹ reported that graphene electrodes improved the performance of pentacene organic FET significantly such as minimized channel resistance, high mobility, and reduced effective barrier height compared to traditional Au electrodes. In this study, in order to figure out the efficiency of graphene as the top electrode in the device, we have compared its performance with an identical device but with conventional platinum as a top electrode.

Commercial low cost 180 nm thick ITO coated glass was used as a substrate for Al₂O₃ thin film deposition. About 15 nm thick amorphous Al₂O₃ layers were deposited on these substrates by plasma-enhanced atomic layer deposition (PE-ALD) in a Picosun R200 system at an optimized growth rate of 1.1 Å/cycle. Trimethyl aluminum (TMA) was used as a precursor of aluminum, and oxygen plasma acted as an oxidizing agent. The substrate temperature was maintained at 200 °C, a low device processing temperature. The thickness of the deposited film was measured using a JA Woollam spectroscopic ellipsometer. X-ray photoelectron spectroscopy (XPS) was performed on the Al₂O₃ layer with a monochromatized Al K_α X-ray source operating at a base pressure of 4.9 × 10⁻⁹ Torr. Monolayer graphene (G) used in the device as the top electrode was grown on copper in a hot filament chemical vapor deposition (HFCVD) reactor at a substrate temperature of 1000 °C and a filament temperature of 1600 °C by flowing 1 sccm of methane for 80 min in a hydrogen atmosphere of 35 Torr. The graphene layer was then transferred from copper to Glass-ITO/Al₂O₃ by a standard poly methyl methacrylate (PMMA) assisted wet transfer method. About 500 nm thick layer of PMMA (MicroChem 950 A9) was spin-coated on the copper/graphene sample and, subsequently, baked on a hot plate at a temperature of 150 °C for 10 min. The copper substrate was then etched out by 20% ammonium persulfate solution. The obtained G/PMMA sample was rinsed several times in de-ionized water and placed onto the ITO/Al₂O₃ sample that was placed overnight on a hot plate set at 45 °C. Finally, the top PMMA layer was removed by selectively dissolving it in hot acetone to obtain a layer of graphene on ITO/Al₂O₃. To obtain graphene as a top electrode first, we evaporated ∼200 nm thick square gold layer on the 2D material through a metal shadow mask. The device was then subjected to oxygen plasma etching at a power of 80 W under a pressure of 200 mTorr to remove the graphene which was uncovered by gold. The gold layer covering the non-reacted graphene was removed by a reclamation process to obtain patterned graphene top electrodes. For this purpose, the sample was kept in a solution of 4 gm of KI, 1 gm of I₂, and 40 ml of de-ionized water for 10 h and then dried by N₂. The gold removal can be explained by the following series of reactions.⁴⁰ $I2+KI→K++I3−$⁠, iodine reacts with potassium iodide to form potassium ion and tri-iodide anion. These products can react with gold to form potassium auric iodide, $3K++3I3−+2Au→2KAuI4+K++I−$⁠. Hence the gold completely reacts with potassium and iodine ion leaving the ITO/Al₂O₃/G sample gold free. The device fabrication process is schematically shown in Fig. 1(a).

The surface topography of the graphene electrodes and active oxide layer was analyzed in vacuum using a scanning electron microscope operating at 70× magnification. Phase purity of processed graphene was further confirmed by Raman spectroscopy. The laser line at 514.5 nm from a Coherent argon ion laser (Innova 70-C) was focused on the sample. A liquid nitrogen-cooled CCD device collected the Raman scattered signal through a 50× objective. The transmittance spectra of the device were recorded by a ratio recording, computer controlled UV-visible spectrometer (Shimadzu UV-2450) in the spectral range of 280–800 nm. I-V measurements were done at room temperature using a Keithley electrometer (model #2401) with the top electrode DC biased and the bottom electrode grounded. For comparative study, ∼50 nm thick and ∼90 μm broad square platinum dots were deposited on an identical ITO/Al₂O₃ sample by dc-magnetron sputtering at a power density of 1 W/cm² using a metal shadow mask to form the ITO/Al₂O₃/Pt capacitor structure (control device) as shown in Fig. 1(e).

Figure 1(b) shows the SEM image of well-defined graphene top electrodes having an area of ∼80 × 80 μm². The Raman spectra showed in Fig. 1(c) provided an intensity ratio of the two prominent peaks; G-band (1580 cm⁻¹) originated from the in-plane vibration of sp² carbon atoms and 2D-band (2689 cm⁻¹) due to the two-phonon double resonance as 0.5. This figure along with a full-width at half maximum (FWHM) of ∼12 cm⁻¹ for G-band and ∼32 cm⁻¹ for 2D-band confirmed the monolayer nature of graphene. Figure 1(d) shows that the fabricated device is highly transparent in the visible region (370-700 nm) with an optical transparency of >82%. It is apparent that the transparency of the device is not significantly affected by the process of graphene electrode formation. Optical transmittance of ITO/Al₂O₃ structure is reduced slightly (only by 1.2%) because of low optical absorbance in graphene.³⁷ Note that here graphene square dots serving as top electrodes occupy less area resulting in less transmittance reduction as compared to reported 2% reduction for the graphene layer.⁴¹ 

Figure 2 shows the resistive switching characteristics of the TRRAM devices with graphene and Pt as top or SET electrodes. Electroforming is the primary step and is an essential phenomenon in resistive switching to commence switching in the active material. This process is usually carried out by applying a high voltage or current compliance to the as-prepared switching device^5,42 that in turn leads to the controlled decrease in resistance. In this study, electroforming was observed at 4.8 V and 6.8 V for ITO/Al₂O₃/G and ITO/Al₂O₃/Pt devices, respectively, which was revealed by a sudden increase in current flow and are shown as insets (b) and (c) in Fig. 2. To avoid any electric breakdown due to high current flow, compliance current I_cc was set at 1 mA in both cases. In the case of ITO/Al₂O₃/G, the device turned to low resistance state (LRS), also called ON-state with a resistance of ∼330 Ω (read at 0.1 V) after electroforming. On applying a negative bias voltage, the current suddenly decreased from ∼1.55 mA to ∼2.1 μA at ∼−0.65 V, turning the device from LRS to HRS (High Resistance State) also called OFF-state, and the process is called the “RESET” process. Subsequently, applying the positive voltage bias from 0 V onwards, a rapid increase in current was observed at ∼0.8 V, switching the device back to LRS again. This switching from HRS to LRS is called the “SET” process. Hence, the device is found to switch with alternate polarity and is called bipolar switching. In the case of the ITO/Al₂O₃/Pt device, SET and RESET voltages were observed at 2.6 V and −2.4 V (a drop in RESET or operation current from ∼18.9 mA to ∼89 μA), respectively, which is higher than the corresponding figures obtained for ITO/Al₂O₃/G.

Endurance and retention properties were measured for a better evaluation of the reliability and functional characteristics of the devices. Resistance at the ON-state is denoted by R_ON measured under an I_cc of 1 mA and that at the OFF state is denoted by R_OFF (both read at +0.1 V). The endurance test of the ITO/Al₂O₃/G transparent device depicted in Fig. 3(a) demonstrates no signs of memory state deterioration. For retention tests, the resistance at the ON state was read out for 10⁴ s at +0.1 V, subsequently, the device was switched to the OFF state, and the resistance was recited for 10⁴ s. The ratio of R_OFF/R_ON was observed to be ∼5 × 10³. Replotting the I-V curve in log-log scale as shown in Fig. 3(c), the conduction mechanism in the ON state of the ITO/Al₂O₃/G device was concluded to be of Ohmic nature as the slope of the plot was determined to be ∼1. This can be due to the formation of conductive filaments in the device during the SET process. In the OFF state, the I-V plot showed a linear behavior at low voltages (0 to +0.4 V). However, at high voltages, the device showed a nonlinear behavior dominated by the electrode/interface-limited Schottky charge transport which can be elucidated via the room temperature linear current density (J) vs E^1/2 plot shown in the inset of Fig. 3(d). The Ohmic behavior at a lower voltage regime might be due to the remnant minor concentration of tiny conductive filaments even after rupture of majority of filaments during the SET process.

Endurance and retention tests conducted on the ITO/Al₂O₃/Pt device are shown in Figs. 4(a) and 4(b), respectively. The obtained R_OFF/R_ON ratio was ∼10⁴ and is in fairly good agreement with the study by Wu et al.⁴⁴ Both ON and OFF states were observed to be stable, and no significant degradation was seen. The switching ratio of approximately 10⁴ in the retention of bistable resistive states may be prolonged to more than 10 years.⁴⁴ The ON state leakage curve replotted in logarithmic scale [Fig. 4(c)] can be approximated by a straight line with a slope of ∼1 and confirmed Ohmic conduction. I-V characteristics for the OFF state showed a nonlinear behavior. The conduction mechanism can be identified as bulk limited Pool-Frenkel emission^44,45 as the ln J/E versus E^1/2 plot shows a linear relationship as demonstrated in Fig. 4(d). The difference between metal work-function and Fermi level of the wide bandgap semiconductor (insulator) determines the type of contact⁴³ during the formation of the metal-insulator interface and explains the asymmetry observed in the I-V data (Fig. 2). The work-function of graphene (∼4.9 eV) and platinum (∼5.2 eV) is higher than that of ITO (∼4.7 eV). This higher work-function can be attributed to the formation of the Schottky barrier at the G or Pt/insulator active interface with an expected shorter barrier height for G.

The observed significant difference in SET and RESET voltages is probably the result of the electrode-oxide interaction.^41,43 Graphene shows a low sheet resistance (in-plane) and much lower contact resistance with an insulator than other conventional electrodes like Pt. G/Al₂O₃ is a Schottky contact⁴¹ that helps us to improve the electric field intensity in the switching layer and hence a faster conductive filament growth. Also, nonpolar graphene exhibits a weak interaction with other materials resulting in an out-of-plane (6 kΩ/□) (Ref. 8) leading to a high series resistance and a low RESET current in the device. Owing to its higher work function than G, Pt is expected to have a higher Schottky barrier height with Al₂O₃ which in turn results in a larger voltage drop prior to the filament formation. The filament formation is associated with the Joule heating generated during the Forming/SET/RESET processes.⁴⁶ The inert and atomically thin graphene has a much lower out-of-plane thermal conductivity (∼6 Wm⁻¹K⁻¹) than its in-plane counterpart (2000-4000 Wm⁻¹K⁻¹)⁴⁷ and can act as an interfacial thermal resistance that can considerably reduce the heat dissipated making the programming processes more energy-efficient.⁴⁸ It can be noted that the programming power given by the product of RESET voltage and the peak in RESET current was ∼1 mW for the graphene electroded device which is only ∼2% of that consumed by the Pt electroded device. This indicates the possibility of reducing the programming power further by lowering the SET current compliance in scaled devices without compromising the stability in switching. Additionally, it has been proved that graphene can remove an undesired surface effect better that increases switching yield of the RRAM device.^10,41 However, a detailed study is yet to be done in this direction.

Though not well understood, various models have been proposed to illustrate the mechanism of resistive switching, such as trap charging and discharging,⁴⁹ formation and rupture of conduction filaments,⁵⁰ and modulation of Schottky barrier.⁵¹ Several studies on metal-oxide based RRAM have shown that the generation and recombination of oxygen vacancies are responsible for switching.^22,52–54 In most dielectric layers, oxygen vacancy is the common defect which is responsible for switching. From an analysis of XPS survey spectrum shown in Fig. 5(a), the elemental ratio of Al:O:C (here C represents carbon in graphene electrode) was determined to be 0.265:0.257:0.477. An asymmetric O 1s core level spectrum [Fig. 5(b)] obtained was Gaussian deconvoluted into two peaks after Shirley background correction. The binding energy of 530.4 eV (FWHM ∼ 2.23 eV) could be attributed to the oxygen that was bound to Al in the Al₂O₃ lattice, while that at 531.47 eV (FWHM ∼ 2.29 eV) belongs to the oxygen deficiencies (oxygen vacancies), a common defect found in the dielectric layer. Hence, we assume that oxygen vacancies play a key role in switching mechanism as explained below.

Figure 5(c) shows the schematic of the switching process mechanism that undergoes at OFF and ON states. During the SET process, under positive bias, oxygen anions were extracted from the lattice, accumulating oxygen vacancies in the disordered Al₂O₃ layer starting from the ITO/Al₂O₃ interface resulting in a considerable reduction of the Schottky barrier height at the G or Pt/Al₂O₃ interface. The Kroger-Vink notation given below expresses the formation of the oxygen vacancies,^55,56

where O_o represents an oxygen ion in a regular site and $Vo2+$ is an oxygen vacancy with a double positive charge in the regular lattice site. Negatively charged non-lattice oxygen anions (⁠$O2−→12O2+2e−$⁠) thus formed are dragged toward the top electrode⁵⁷ leaving behind a more oxygen deficient Al₂O_3−n active layer. This process leads to the formation of a conical shaped filament made up of oxygen vacancies in the oxide layer as shown in Fig. 5(c) which acts as the low resistance path for the current flow. Hence current flows steadily through the filament resulting in a low resistance state in the Al₂O_3−n layer. The interface between the electrode and the oxide performs a vital role in the switching process.⁵⁷ Under the application of a negative bias voltage, non-lattice oxygen anions start to migrate from the interface, injecting into the lattice. While recombining oxygen ions with vacancies, the oxidation reaction occurs that results in a more insulating Al₂O₃ state. During this process, the conductive filaments in the Al₂O_3−n layer split causing a sudden drop in current. Thereafter the device gets switched to HRS and the process is called RESET.

In conclusion, we fabricated a graphene integrated see-through ITO/Al₂O₃/G TRRAM device using a hybrid process and investigated its anticlockwise bipolar resistive switching characteristics. We found that the ON/OFF resistance ratio in the ITO/Al₂O₃/G device was slightly reduced compared to that in the ITO/Al₂O₃/Pt device, but the SET/RESET voltages significantly decreased, and the RESET current was noticeably reduced which are important features for memory operations. The observed resistive switching mechanism is understood to be due to the formation and rupture of conductive filaments formed out of oxygen vacancies. Excellent endurance and retention characteristics and good contrast in the ON/OFF resistance ratio make the fabricated transparent device a potential candidate for TRRAM.

Financial support from DOD Grant No. FA9550-16-1-0295 is acknowledged. This project was supported in part by PR Space Grant (NASA Training Grant Number NNX15AI11H) and PR NASA EPSCoR (NASA Cooperative Agreement NNX15AK43A). S.P.P. thanks the American Society for Engineering Education for fellowship.