Memristive behaviour in inkjet printed graphene oxide thin layers

Abstract

Memristors are passive two-terminal memory devices predicted to have a tremendous impact on many research fields and common applications, paving the way to adaptive electronics and high computing systems. We report on a metal/insulator/metal memristor based on a graphene oxide layer, deposited by inkjet printing at room temperature. The electrical characterization of devices, showing hysteretic characteristics typical of bipolar memristive switching, are discussed and correlated to the structural and compositional analysis of the materials. The electroforming process is ascribed to a lowering in contact resistance due to carbon diffusion in Ag electrode, while the oxygen ion drift is identified as the main physical mechanism for Ag/GO/ITO resistive switching.

---

Introduction

The existence of the memristor was postulated already in 1971, when Leon O. Chua observed that the relationships that linked the four fundamental circuit variables (current, voltage, magnetic flux and charge) were missing one circuit component in order to be symmetric.¹ Putting the memristor along with the other three classical circuit elements (resistor, capacitor and inductor) allowed to define the sixth missing relationship that connected magnetic flux to charge.² Nevertheless, the fabrication of a real memristive device was first achieved only in 2008, exploiting the properties of nano-sized metal oxide layers.³ Since then, many examples of working memristor devices were fabricated using several materials and systems, all of which have in common the nanoscale size of the switching layer, in which the application of relatively low voltages can generate extremely intense fields inducing non-linear electrical characteristics and reversible ionic or electronic transport phenomena.^4,5 Memristors can be exploited for fast, non-volatile and low-energy electrical switching in resistive random access memories for computer electronics architecture.^6,7 In addition, memristors can retain their internal resistance state according to the history of applied input, which make them an ideal candidate for the fabrication of neuromorphic systems and artificial intelligence.^8,9

Among all different materials that can be used for the fabrication of memristive switching layers, the use of graphene oxide (GO) was increasingly studied in recent literature, as highlighted in a previous review paper.¹⁰ The most common configuration for GO-based memristors is the metal/insulator/metal (MIM) structure, where GO is sandwiched between two metal electrodes which can be made of the same material (symmetric) or two different (asymmetric). Nowadays, the synthesis of GO can be achieved quite routinely at low cost and high yield using several techniques¹¹ and GO sheets obtained from chemical routes, such as the exfoliation of oxidized bulk graphite,^12,13 are widely available on the market. The presence of oxygen functionalities on the basal planes and sheet edges makes GO electrically insulating and comparable to other thin-layered oxide materials.¹⁴ The fabrication of GO-based memristors showed some advantages toward other oxide materials, such as the easy processability of GO layers directly from solutions using 2D printing technology. In fact, being GO fully dispersible in common solvents, including water,¹⁵ it is possible to synthesize GO-based printable inks that can be easily processed for the direct deposition of patterned devices.¹⁶ In fact, inkjet printing is one of the most versatile techniques for additive manufacturing of circuits and devices based on graphene, and it is significantly advantageous compared to other subtractive techniques such as lithography and dry/wet etching, in terms of safety, low-cost, rapidity and ease of process, as already demonstrated by several works on composite materials.^17,18 The use of two-dimensional additive technologies like inkjet printing is particularly valuable in the fabrication of passive electronic elements like the memristor, because of the nature of the device itself which is completely relying on the thin-film material's properties without the need of complex microfabrication steps like, for example, in the case of transistors. In fact, memristors fabricated by inkjet printing are ready for integration in microelectronic circuits.

This work reports the fabrication of an asymmetric MIM type memristor based on a GO switching layer deposited by inkjet printing. The electrical characterization of devices, including cycling of hysteretic characteristics typical of memristive switching, are discussed and correlated to the structural analysis of the materials.

Experimental details

A commercial indium-tin oxide (ITO) coated glass slide (square, 25 mm² area, surface resistivity 8–12 Ω sq⁻¹, from Sigma-Aldrich) was cleaned in freshly made piranha solution, rinsed with DI water and dried with nitrogen blow. The slide was used as substrate with ITO bottom electrode for the fabrication of ITO/GO/Ag memristors. Commercial GO with nominal thickness 0.7–1.2 nm was purchased from Cheap Tubes Inc. (US) and used without further purification. A 1.5 mg ml⁻¹ printable GO dispersion in a water/ethanol 1/1 solution was prepared following a procedure similar to the one described elsewhere.^16,19 For the ink design, a relatively low concentration of GO was used to optimize the dispersion. A low GO concentration reduces the risk of nozzle clogging due to particle build-up at the orifice end. As the number of inkjet passes was not an issue for the present test, the implementation of a low concentration/repeated printing setup was preferred to a high concentration/low number of passes in order to achieve GO flakes percolation on the substrate. The dispersion was homogenized by high speed Ultraturrax for 5 minutes, followed by a two-step ultrasonic bath (30 min at 40 kHz plus 30 min at 59 kHz) to further grind and disperse GO agglomerates. Finally, the dispersion was centrifuged at 14 000 rpm for 5 minutes to precipitate residual large and heavy particles. The upper portion of the centrifuged GO dispersion was then inserted into a 3 ml reservoir (discarding the large precipitated particles) and loaded to the inkjet printing system (JETLAB 4-XL from Microfab, US). Printing tests were achieved on the glass/ITO substrate through piezoelectric nozzles made in quartz, with diameter of 80 μm, with a vibration frequency set to 1 kHz. The three-axis movement and the values of voltage and wave form that command the piezoelectric nozzle were computer controlled, and the dimension and speed of ink drops were monitored by a horizontal camera located onto the x–y stage for direct drop observation. The nozzle and substrate holder were kept at room temperature (∼25 °C). The printing parameters were modified from standard printing recipes considering that the characteristics of the formulated ink were outside the conventional range recommended by the inkjet printer manufacturer. In fact, a viscosity of ∼2.3 cP was measured for the GO ink used in this work, while typical fluids jetted by this device are in the range 10–40 cP. The viscosity of the water/ethanol 1/1 solvent solution was increased from a value of ∼2.1 cP to the value reported for the ink by the presence of the dispersed GO flakes. The viscosity measurements were performed at room temperature (25 °C) using an Anton Paar Rheometer (MCR302). Jetting parameters were set as follows, using an asymmetrical pulse: first rise time 25 μs, dwell time 18 μs, fall time 3 μs, echo time 30 μs, second rise time 2 μs, idle voltage 0 V, dwell voltage 40 V, echo voltage −25 V. We found that relatively high values of the dwell voltage (up to 40 V), which is applied to the piezoelectric nozzle during the first segment of the asymmetric pulse controlling the drops jetting, allowed the generation of stable spherical drops, even though the fluid viscosity was lower than the recommended range. Moreover, using a lower echo voltage (−25 V) allowed to control and reduce the formation of satellite drops which are detrimental in the generation of a precise inkjet pattern. The optimization of printing parameters was performed observing the ejected drops by the horizontal camera. Using the described printing setup, ink drops with diameter of ∼70 μm were normally obtained. Consequently, the script used for test tracks printing was set to achieve a drop spacing of 70 μm for complete coverage of the substrate. For the fabrication of devices, the printing process was optimized using a straight line array pattern (with 70 μm interspace between lines) repeated 20 times to achieve a consistent deposition of GO.²⁰ The printed layer was annealed at 80 °C for 30 minutes to eliminate residual solvents and consolidate the GO deposited film. The memristor devices were completed by thermal evaporation of Ag circular top electrodes (diameter 2 mm) through a shadow mask, with 200 nm Ag thickness.

The structural analysis and morphology of the printed GO layer were characterized by optical and scanning electron microscopy (FESEM, ZEISS Merlin). Raman spectroscopy of the printed GO film was performed with a Renishaw inVia Reflex micro-Raman spectrophotometer equipped with a cooled CCD camera. Samples were excited for 10 seconds with an Ar–Kr laser source (wavelength of 514.5 nm, photon flux ∼300 W cm⁻²) using a 10% power filter to avoid layer damage. The chemical composition of pristine and electroformed devices was inspected by X-ray photoelectron spectroscopy (XPS), performed using a monochromatic X-ray beam with an Al K-α source with energy of 1486.6 eV. The depth profile was obtained with a 2 kV Ar ion gun. Current/voltage (I–V) measurements were performed at room temperature using a standard two-point micro-contact setup of a Keithley 2635A multimeter.

Results and discussion

A photograph of the processed glass slide is reported in Fig. 1(a), showing an array of 25 devices. Fig. 1(b) reports an optical microscope image showing a 20 passes inkjet printed GO film on ITO, in which a good distribution of GO flakes can be observed. The Raman spectra acquired on the same sample show the presence of typical carbon bands related to GO (Fig. 1(c)), with the first order G and D peaks at 1358 cm⁻¹ and 1598 cm⁻¹ respectively. In particular, the pronounced D peak reveals the presence of defects, vacancies and distortions of reduced-size sp² carbon domains typical of oxidized graphene layers.²¹ The G peak typical of graphitic carbon appears broader than in non-oxidized graphene and merged with the D′ mode at 1625 cm⁻¹ due to defect scattering.²² The band at higher energy shift can be split in two contributions, respectively due to the G′ mode at ∼2685 cm⁻¹, due a double resonance intervalley Raman scattering process with two phonons at the K point,^23,24 and the D + D′ combination mode at ∼2935 cm⁻¹.

Fig. 1 Photograph of the Ag/GO/ITO 25 devices array (a). Optical microscope image of the 20 passes inkjet printed GO on ITO, revealing a good distribution of GO flakes (b). Raman spectrum of the GO starting material (c).

The microstructure of the deposited GO layer was investigated by FESEM analysis. Fig. 2(a and b) report low magnification images of a GO dispersion deposited by a 3 passes printing, showing the initial stages of printing. At low number of passes, it is possible to observe the generation of the pattern made by ink drops falling on the substrate, and appreciate the dimension of their spreading after the impact. The higher magnification images (Fig. 2(c and d)) show that GO flakes (darker regions) of different dimensions and shapes are randomly deposited over the ITO granular surface (lighter regions), without achieving a continuous layer. FESEM images of the 20 printed passes sample used for the fabrication of devices (Fig. 2(e and f)) show a fairly good coverage of the ITO substrate achieved by GO flakes. The sample printed with 20 passes is composed by GO flakes which are completely overlapping, forming a continuous layer of stacked GO flakes. This is clearly evident comparing the difference in contrast between Fig. 2(c and d) and, Fig. 2(e and f): in the latter, the ITO substrate is still visible, but appears less bright due to electron absorption of the GO coating. At some spots (Fig. 2(f)), the GO flakes stack less uniformly and appear more corrugated and wrinkled, due to random agglomeration of several flakes, as previously reported.²⁰ The surface morphology of the printed devices was thus clearly influenced by the number of printing passes, with an increased roughness in the 20 passes sample due to the presence of wrinkled GO agglomerates. Nevertheless, for the characterization of the electrical properties, the devices printed by 20 passes were used to guarantee a complete coverage of the ITO electrode.

Fig. 2 FESEM images of a sample of GO inkjet printed on ITO after 3 passes (a–d), where the microstructure of GO flakes is appreciable, and of the 20 passes sample used for memristive devices (e and f).

The memristive properties of inkjet printed Ag/GO/ITO devices were studied by I–V cycling characterization. Fig. 3 reports the semi-logarithmic I–V plot of a device sample, in which the typical memristive hysteresis can be observed. The red solid line represents the first sweep, showing that by raising the applied positive voltage from zero the device stays in its pristine state until a forming voltage of 580 mV is reached, and the device switches ON with an abrupt increase of current. The device continues then on a lower resistance state until the current compliance (CC) value is reached. In this experiment, CC was set to 100 mA to prevent Joule heating which can permanently damage the device. It is possible to notice that by decreasing the positive voltage, the device remains in a state of higher conductivity (LRS, low resistance state), which is maintained until a negative bias of about −750 mV is reached. When the negative voltage sweep reaches this value, the device is switched OFF to a relatively high resistance state (HRS). By further cycling of the same device, a memristive hysteresis between the two formed states is observed, with average values of and , respectively. As an example, the fifth and tenth cycles are reported in Fig. 3 (black dotted and dashed lines, respectively). The switching characteristic of this device is defined as bipolar, since switching between ON and OFF states requires different voltage polarity.²⁵ It can be observed in Fig. 3 that the OFF state of the device presents a lower resistance than the pristine state, which is due to an irreversible forming process that occurred before the measurement of resistive switching.

Fig. 3 Semi-logarithmic plot of the I–V characterization of an inkjet printed Ag/GO/ITO device, showing memristive hysteresis. The red solid line represents the first sweep. The black dotted and dashed lines refer to further cycling of the same device (fifth and tenth cycles respectively).

The forming and switching process was further investigated by comparing the XPS depth profile of a pristine Ag/GO/ITO device with the same measurement performed after electroforming and cycling (HRS state). The plots in Fig. 4 reports the structure of the device, showing the presence of Ag top electrode, the switching layer made of partially oxidized carbon, and the tin oxide bottom electrode.

Fig. 4 XPS depth profile of a Ag/GO/ITO memristor before (a) and after (b) electroforming.

Two different mechanisms are commonly identified to explain electroforming and resistive switching behaviour when nanoscale oxides are subjected to an electric field: one is the formation of a metal filament due to the diffusion of species from the top electrode, the other is related to the drift of oxygen vacancies under a bias voltage.^10,26 Our findings differ substantially from results previously reported for GO-based MIM structures, which rely either on the formation of metal filaments within the insulating layer by migration of metal ions from a suitable electrode (typically Cu),²⁷ or on oxygen ion diffusion.²⁸ In particular, for the latter case, some works reported that the GO layer did not even participate directly in the switching phenomena, which depended on a metal oxide insulating layer generated at the interface with the electrode, with the role of GO relegated to oxygen ions reservoir.^25,29 Another work ascribed the switching characteristics to the presence of reversible highly resistive regions of the GO layer near the metal interface, due to local rearrangement to heavily sp³ hybridized carbon lattice.²⁷ Our compositional profile reported in Fig. 4 shows no appreciable presence of metal (Ag) inside the GO layer, thus we can deduce that the formation of a metal filament can't play a main role in our samples, differently from what discussed in previous works.^27,30 On the other hand, XPS data clearly show a remarkable presence of carbon atoms in the Ag electrode. Therefore we can argue that the application of a sufficiently high positive voltage to the device during electroforming causes the diffusion of carbon from the GO layer into the Ag electrode. Such modification obviously changes the contact resistance, thus incrementing the current in the HRS from the pristine to the electroformed device. It should be remarked that the so-formed state is not perfectly stable, as it can be deduced by the relative large variability during cycles of (more than 10%), respect to (less than 1%). Similar large instability (also reported in terms of HRS/LRS ratio or V_SET and V_RESET) is found for most of GO-based memristive devices and it is typically attributed to the intrinsic variability of the GO chemical structure.¹⁰

We can finally postulate that the resistive switching in our samples is dominated by oxygen ions drift that can change the energy barrier at the interface (as suggested elsewhere³¹) or reduce the GO to a more conductive graphene-like structure (as reported in a previous work³²).

Conclusions

The memristive behaviour of MIM devices based on inkjet printed GO was proved. After electroforming, I–V characteristics show a bipolar switching behaviour with positive set (580 mV) and negative reset (−750 mV). Thanks to compositional analysis by means of XPS depth profile, it was possible to ascribe the electroforming process to a lowering in contact resistance due to carbon diffusion in Ag electrode and to identify the oxygen ion drift as the main physical mechanism for Ag/GO/ITO resistive switching.

While the relatively small ON/OFF ratio would be a limit for the use of a similar structure in RAMs, the reported low voltages (<1) make such device appealing for adaptive electronics and low power applications.

Acknowledgements

The support by M. Raimondo in helping with FESEM and S. Guastella with XPS analyses is gratefully acknowledged.

Notes and references

1. L. O. Chua, IEEE Trans. Circuit Theory, 1971, 18, 507.
2. L. O. Chua and S. M. Kang, Proc. IEEE, 1976, 64, 209.
3. D. B. Strukov, G. S. Snider, D. R. Stewart and R. S. Williams, Nature, 2008, 453, 80.
4. D. S. Jeong, R. Thomas, R. S. Katiyar, J. F. Scott, H. Kohlstedt, A. Petraru and C. S. Hwang, Rep. Prog. Phys., 2012, 75, 076502.
5. A. Sawa, Mater. Today, 2008, 11, 28.
6. L. O. Chua, Appl. Phys. A, 2011, 102, 765.
7. J. J. Yang, D. B. Strukov and D. R. Stewart, Nat. Nanotechnol., 2013, 8, 13.
8. S. H. Jo, T. Chang, I. Ebong, B. B. Bhadviya, P. Mazumder and W. Lu, Nano Lett., 2010, 10, 1297.
9. M. P. Sah, K. Hyongsuk and L. O. Chua, IEEE Circ. Syst. Mag., 2014, 14, 12.
10. S. Porro, E. Accornero, C. F. Pirri and C. Ricciardi, Carbon, 2015, 85, 383.
11. W. S. Hummers and R. E. Offeman, J. Am. Chem. Soc., 1958, 80, 1339.
12. D. R. Dreyer, S. Park, C. W. Bielawski and R. S. Ruoff, Chem. Soc. Rev., 2010, 39, 228.
13. J. N. Coleman, Acc. Chem. Res., 2012, 46, 14.
14. A. Lerf, H. He, M. Forster and J. Klinowski, J. Phys. Chem. B, 1998, 102, 4477.
15. J. P. Rourke, P. A. Pandey, J. J. Moore, M. Bates, I. A. Kinloch, R. J. Young and N. R. Wilson, Angew. Chem., Int. Ed., 2011, 50, 3173.
16. R. Giardi, S. Porro, A. Chiolerio, E. Celasco and M. Sangermano, J. Mater. Sci., 2013, 48, 1249.
17. S. Bocchini, A. Chiolerio, S. Porro, D. Accardo, N. Garino, K. Bejtka, D. Perrone and C. F. Pirri, J. Mater. Chem. C, 2013, 1, 5101.
18. A. Chiolerio, S. Bocchini and S. Porro, Adv. Funct. Mater., 2014, 24, 3375.
19. S. Porro, R. Giardi and A. Chiolerio, Appl. Phys. A, 2014, 117, 1289.
20. L. T. Le, M. H. Ervin, H. Qiu, B. E. Fuchs and W. Y. Lee, Electrochem. Commun., 2011, 13, 355.
21. I. Roppolo, A. Chiappone, K. Bejtka, E. Celasco, A. Chiodoni, F. Giorgis, M. Sangermano and S. Porro, Carbon, 2014, 77, 226.
22. W. Xu, N. Mao and J. Zhang, Small, 2013, 9, 1206.
23. T. Gokus, R. R. Nair, A. Bonetti, M. Böhmler, A. Lombardo, K. S. Novoselov, A. K. Geim, A. C. Ferrari and A. Hartschuh, ACS Nano, 2009, 3, 3963.
24. A. Kaniyoor and S. Ramaprabhu, AIP Adv., 2012, 2, 032183.
25. H. Y. Jeong, J. Y. Kim, J. W. Kim, J. O. Hwang, J. Kim, J. Y. Lee, T. H. Yoon, B. J. Cho, S. O. Kim, R. S. Ruoff and S. Choiet, Nano Lett., 2010, 10, 4381.
26. R. Waser and M. Aono, Nat. Mater., 2007, 6, 833.
27. C. L. He, F. Zhuge, X. F. Zhou, M. Li, G. C. Zhou, Y. W. Liu, J. Z. Wang, B. Chen, W. J. Su, Z. P. Liu, Y. H. Wu, P. Cui and R. W. Li, Appl. Phys. Lett., 2009, 95, 232101.
28. G. N. Panin, O. O. Kapitanova, S. W. Lee, A. N. Barano and T. W. Kang, Jpn. J. Appl. Phys., 2011, 50, 070110.
29. H. W. Nho, J. Y. Kim, J. Wang, H. J. Shin, S. Y. Choi and T. H. Yoon, J. Synchrotron Radiat., 2014, 21, 170.
30. F. Zhuge, B. Hu, C. He, X. Zhou, Z. Liu and R. W. Li, Carbon, 2011, 49, 3796.
31. L. H. Wang, W. Yang, Q. Q. Sun, P. Zhou, H. L. Lu, S. J. Ding and D. W. Zhang, Appl. Phys. Lett., 2012, 100, 063509.
32. H. Q. Wei, P. Zhou, Q. Q. Sun, L. H. Wang, Y. Geng and D. W. Zhang, IEEE Nanotechnol. Mater. Devices Conf., 2012, 54.

---