Fabrication of Resistive Random Access Memory by Atomic Force Microscope Local Anodic Oxidation

Transcription

1 NANO: Brief Reports and Reviews Vol. 10, No. 2 (2015) (8 pages) World Scienti c Publishing Company DOI: /S Fabrication of Resistive Random Access Memory by Atomic Force Microscope Local Anodic Oxidation Je T.H. Tsai*,, Chia-Yun Hsu*, Chia-Hsiang Hsu, Chu-Shou Yang and Tai-Yuan Lin* *Institute of Optoelectronic Sciences National Taiwan Ocean University, Taiwan Graduate Institute of Electro-Optical Engineering Tatung University, Taiwan thtsai@mail.ntou.edu.tw Received 31 July 2014 Accepted 19 November 2014 Published 7 January 2015 The fabrication of gallium, zinc and nickel oxide nanodots for application of resistive random access memory (RRAM) was demonstrated using the atomic force microscopy (AFM) local anodic oxidation technique. Thin metal lms were deposited on indium tin oxide conductive glass substrates. In the atmospheric environment, using AFM equipped with an Ag-coated probe can generate metal oxide nanodots locally on the metal lms. These nanodots act as an insulator layer in a single unit cell of the RRAM. The voltage-biased method allows devices to reset from a lowresistance state (LRS) to a high-resistance state (HRS) at 0.9 V. These results show the ability of the AFM local anodic oxidation to produce 50 nm NiO nanodots on glass substrates for potentially high-density RRAMs. As we developed the characteristics of the structure, we found that a lateral NiO nanobelt RRAM performs very low power operation from such experimental manufacturing process. Using a current-biased method, the lateral device switches from a HRS to a LRS with a low writing voltage of 0.64 V. Keywords: Resistive random access memory; atomic force microscope; anodic oxidation. 1. Introduction In recent years, the area of random access memory (RAM) has faced a challenge in the development of solid-state memory chips with high density and high-capacity memory cells in a stack structure. 1 4 There is a need for a new generation of memory that exhibits fast switching and low power consumption, using commercial CMOS-compatible materials and processes. Several approaches have focused on the development of next-generation nonvolatile memory technologies, 5 and one promising device is the resistive random access memory (RRAM). RRAM o ers great potential in achieving fast programming speed with high density in memory cells. The structure is also relatively simple compared to the conventional RAM. A RRAM cell contains a dielectric layer sandwiched by two metal electrodes. By repeated electric breakdown and reconstruction of the dielectric layer, the variation of resistances can represent the storage of the digital data. Therefore, it is necessary to fabricate the unit memory cell in nanometer scales using conventional

2 J. T. H. Tsai et al. photolithography skills to achieve high-density memories. RRAMs also attract interest due to other bene ts such as low-voltage operation and low cost per bit. They are suitable for integration in crossbar arrays stacked in multiple levels using compatible CMOS technologies. 6 Atomic force microscopy (AFM) local oxidation has been demonstrated on thin metal lms to fabricate nanometer-sized structures. 7 This technique is based on the spatial con nement of a chemical reaction on substrates within a nanometer-sized region. The AFM tip provides an area de ned by a combination of the probe-sample surface geometry and environmental humidity to form a water bridge. This water bridge acts as an electrolyte and spatially con nes the oxidation of the metal surface or the reaction into metal oxide. Avouris et al. demonstrate AFM nanopatterning on silicon and metals. Local chemistry can be induced using high electric current densities owing through Ti and Cr metal lms in air to generate metal oxides. 8 Dip-pen nanolithography, which can generate metal oxide line patterns in metal with 30 nm of the line width, was rst demonstrated. An advanced nanopatterning technique has been demonstrated, using a similar concept, which can deposit proteins on a surface 9 or generate patterned magnetic nanoparticles. 10 However, this single-tip writing process may limit functional ability for mass production of nanopatterning. An attempt has been made to overcome this problem in order to facilitate a highthroughput manufacturing process using tip-based technology. Salaita has performed parallel multi-tip nanoscale writer using uidic fountain pen nanolithography. This system can generate more complex nanopatterns with higher e±ciency. 11 Scanning tunneling microscope (STM) is another handy tool for building and studying RRAMs, but using a different approach. 12,13 Thermally deposited thin lms are transformed by scanning tunneling microscopy, as demonstrated. The organic complex thin lm can be \recorded" into a nanometer-sized conduction dot for high-density memory usages. 14 We propose to use AFM nanolithography to fabricate metal oxide on a single RRAM cell in order to quickly evaluate the various metals that can be oxidized at room temperature. One example shows that the process can produce high-density gallium oxide nanodots for the construction of nonvolatile memory. 15 Our previous study of AFM local oxidation on Zn has shown exceptional performance in ZnO thin- lm transistors. 16 This room-temperature fabrication process is also necessary to adapt to electronics in both stand-alone and embedded applications. The RRAM memory cells have a simple lateral metalisolator-metal structure and, under electric loading, they exercise nonvolatile resistance switching. 17 This o ers a great opportunity to study the switching mechanism using AFM local oxidation on Ni lms. The cell presents very low writing voltage and low power consumption in a full read-write cycle. 2. Experiment and Results A high-purity Ni target was used for sputtering metal thin lms on Si and glass substrates at 200 C in a vacuum chamber under the base pressure of 5E-7 Torr. The 400 W RF power causes the Ar plasma to generate a 100-nm thick Ni lm within 30 s through this sputtering process. This deposition follows a 350 C thermal treatment in the same vacuum chamber in order to smooth the lm surface. Surface atness is essential when using AFM to perform local oxidation. We found that if the average roughness of the metal lm is greater than 4.5 nm, the AFM local oxidation cannot be completed. To verify the roughness of the sputtered Ni lm, we surveyed the topography and found that the root mean square (RMS) roughness was below 2.26 nm. To evaluate the process performance and verify the nal product by AFM oxidation, we scanned the tip across the area with the applied voltage of þ10 V. Under the high ambient humidity of 85%, the Ni lm will be oxidized. Humidity is a key issue for AFM LAO, as shown in our previous study, 15 and it strongly a ects device operation when oxide materials absorb moisture from the ambient air, causing ionization of the anode metal 18 or charging the reaction by electro-neutrality in cation-transporting dielectric thin lms. 19 Optical microscope observation has distinguished a clear contrast between the anodic oxidation area and the original Ni lm. By maintaining the position of the AFM tip without scanning it, we are able to produce the smallest Ni nanodot of a size of 50 nm. The tip biases strongly a ect the size of the NiO x nanodot that can be realized from the measurement of Ni surface topography using the di erent oxidation voltages of 6 V, 6.5 V and 7 V, shown in Figs. 1(a) 1(c). We also observed the e ect of oxidation time on the size of nanodots, and the results showed that under bias of 6 V, the dot size varied

3 Fabrication of RRAM by Atomic Force Microscope (a) (b) (c) Fig. 1. Tip biases versus size variation of oxide nanodots. (a) 6 V (b) 6.5 V and (c) 7 V

4 J. T. H. Tsai et al. from 60 nm to 250 nm when process time was controlled from 1 s to 18 s, as shown in Fig. 2. We used the micro-raman spectrum to examine the nal product of the nanodot when it grew to over 250 nm. Raman spectroscopy revealed the product of anodic oxidation after room-temperature electro-chemical reaction. The raw material had its signature spectrum at 192 cm 1 ; however, in the oxidation area, both signals disappeared. Instead, a broadened signal with a main peak at 845 cm 1 was found. (a) (b) (c) Fig. 2. Process time versus nanodot size, the topography microscopy image of AFM oxidation areas. (a) 8 s (b) 13 s and (c) 18 s

5 Fig. 3. Schematic of AFM LAO process for memory cell fabrication. Fig. 4. NiO nanodot RRAM I V characterization. The device set at 7 V from HRS to LRS, then reset at 0.9 V reappearance to HRS showing the bipolar switching behavior. After we con rmed the e ectiveness of the process for fabricating NiO by AFM, local Ni oxide was then tested by locating the AFM tip in contact mode without scanning. We applied a voltage on a Pt-coated AFM tip at di erent times to perform the fabrication of the NiO x nanodot. The experimental schematic is shown in Fig. 3. From Fig. 2, it is clear that the sizes of the nanodots are Fabrication of RRAM by Atomic Force Microscope proportional to the process time. The observation of current versus processing time is in line with our previous study on GaO nanodot. Therefore, the current constantly decreases when owing from tip to substrate via this just-formed dielectric layer. The resistance increases by four orders of magnitude in just 10 s. This explains how the e ective process to generate anodic oxides at room temperature can be very fast. We test this structure when we re-scan the surface and x the probe at the top of the nanodot. This creates a two-thermal structure (top electrode: AFM probe, bottom electrode: ITO glass) to test current voltage measurements. We found that this vertical RRAM cell has a set-voltage of 7 V with breakdown situation current of 1 A. This cell has a fairly low reset voltage when the reverse bias is applied (Fig. 4). Early study of the switching mechanism from NiO determined the formation and rupture of a nickel lament in the NiO matrix based on the thermochemical reaction. 20 This creates the RRAM in unipolar switching. The coexistence of the bipolar and unipolar resistive-switching modes in NiO cells, realized using an optimized oxidation process, was also observed. 21 It was also found that bipolar switching from NiO-based devices requires a di erent switching mechanism as the active materials are single crystal NiO in a nanowire structure. 22,23 From our results, the bipolar switching behavior may relate to the electrode material, as we use an Ag-coated AFM tip. The Ag can dissolve into NiO x when positive voltage is applied. This may cause the AFM tip electrode to release active ions into the dielectric layer to form a conduction lament to sustain the low-resistance state (LRS). These ion chains require a negative voltage to retract from the contact of the electrode to reset the RRAM in the high-resistance state (HRS). Hence, it Fig. 5. The AFM topography of NiO nanobelt RRAM device in 2D (left) and 3D (right) view

6 J. T. H. Tsai et al. Fig. 6. performs the bipolar switching behaviors in Fig. 4 when switching between LRS and HRS. Using the same technique, we are then able to produce the lateral NiO nanobelt on Ni lm when we scan the AFM tip and form oxidation material across a patterned Ni line. The AFM micrograph shows the device in detail when we apply voltage on the AFM tip while scanning across the 10 m line, as shown in Fig. 5. Then, observing this nanobelt under SEM, we found that it is constructed by the chaining of nanodots in the polycrystalline chain structure, as shown in Fig. 6. Although the SEM observation shows that the dots are not connected precisely, the electric measurement still shows the electric isolation formation. The conductance drops by ve orders of magnitude when the nanobelt forms (from 10 5 Ato10 10 A). 3. Results and Discussion The advantage of using AFM to oxidize Ni is that this process is e ective at room temperature. For RRAM applications, this requires a dielectric with a metal-rich composition. Such a material can easily form a conducting lament (CF), as the operation The SEM micrograph shows the detail structure of NiO nanobelt. model shows. 24 Most oxidation processes require high temperatures, and the nal products are fully oxidized materials. In AFM local oxidation, however, the nanodot presents a metal-rich composite that may enhance the ion penetration to form a CF. To operate the nanobelt formed RRAM single-bit memory, we control current in steps of 10 pa using a semiconductor analyzer equipped with a probe station. The test ambient is then controlled for a low humidity of 35% in order to minimize the probe-topad point contact resistance from the W probe tip to the Ni electrodes. The response voltage increases slowly due to the high resistance of the dielectric. After the current reaches 10 7 A, the responding voltage decreases while CFs form. A breakdown occurs at 0.8 V, while resistance drops from 4 M to 0.8 k a change of four orders of magnitude. Hence, this memory cell acts as a \set" or \write" to digital data \1". To erase the data, we use the voltage control mode to step the voltage in 10 mv increments. The response current rises from 10 6 A to 10 3 A. It suddenly changes from conduction to resistance and the current drops to 10 7 A. This forces the device to \reset" or write digital data to \0". Figure 7 shows the I V characterization of set

7 Fabrication of RRAM by Atomic Force Microscope (a) (b) and reset. We test this device for over 120 cycles, and from the 120 results that show very limited variations from data to data. This indicates good performance and very low writes voltages from our nanobelt NiO RRAM cell. However, there are still about four data errors, as we de ne the high level state at 10 5 A, compared to the low-level state of A, which gives an ON/OFF ratio in We can read the stored data by xing the voltage at 0.2 V in both the set and reset states. The stable current through the device shows the stability of the memory ability when the device is in the \set" condition. However, to measure resistance in the reset state, the current approaches the level of background noise of pa ranges. Compared to the set and reset state, the variation of resistance is nearly We can therefore conclude that such a characterization is suitable for digital memory. In this work, we produce a device of practical size, and we used room temperature to create a metal-oxide thin lm. From the power consumption aspect, we use low power at the writing and erasing steps. It is also indicated that, by such a process, a low-cost (such (c) Fig. 7. RRAM I V characterization, (a) the rst cycle (b) the 120th cycle (c) an overlap of 120 cycles (d) statistic data of currents from 120 cycles. (d) as ITO glass) substrate is compatible with roomtemperature oxidation. 4. Conclusion Ni oxide is an ideal candidate for RRAM applications. We successfully demonstrated the use of AFM local oxidization to fabricate NiO x nanodots and nanobelts on glass substrates. This process performs a simple and practical oxidation, at room temperature, and is feasible in generating nanometer-size patterns on low-cost substrates without an expensive conventional photolithography tool. The lateral type of RRAM characterization shows very low write voltage of 0.8 V in the write cycle, and a stable current output from the read cycle. Acknowledgments The authors gratefully acknowledge the National Science Council and National Nano Device Laboratory, Taiwan under the Grants NSC M MY

8 J. T. H. Tsai et al. References 1. Y. Wu, B. Lee and H. S. P. Wong, IEEE Electron Device Lett. 31, 1449 (2010). 2. J. Shin et al., IEEE Electron Device Lett. 32, 958 (2011). 3. J. Park et al., IEEE Electron Device Lett. 33, 646 (2012). 4. T. V. Kundozerova, A. M. Grishin, G. B. Stefanovich and A. A. Velichko, IEEE Trans. Electron. Devices 59, 1144 (2012). 5. D. Ielmini, F. Nardi and C. Cagli, IEEE Trans. Electron Dev. 58, 3246 (2011). 6. L. M. Yang et al., IEEE Electron Device Lett. 33,89 (2012). 7. S. Shingubara, Y. Murakami, K. Morimoto and T. Takahagi, Surf. Sci. 532, 317 (2003). 8. P. Avouris, R. Martel, T. Hertel and R. Sandstrom, Appl. Phys. A Mater. Sci. Process. 66, S659 (1998). 9. K. B. Lee, J. H. Lim and C. A. Mirkin, J. Am. Chem. Soc. 125, 5588 (2003). 10. X. G. Liu, L. Fu, S. H. Hong, V. P. Dravid and C. A. Mirkin, Adv. Mater. 14, 231 (2002). 11. K. Salaita, Y. H. Wang, J. Fragala, R. A. Vega, C. Liu and C. A. Mirkin, Angew. Chem. Int. Ed. 45, 7220 (2006). 12. K. Terabe, T. Hasegawa, T. Nakayama and M. Aono, Nature 433, 47 (2005). 13. I. Valov et al., Nat. Mater. 11, 530 (2012). 14. L. P. Ma et al., Appl. Phys. Lett. 69, (1996). 15. J. T. H. Tsai, C. H. Hsu, C. Y. Hsu and C. S. Yang, Electron. Lett. 49, 554 (2013). 16. J. T. H. Tsai, B. H. B. Lee and M. S. Yang, Phys. Rev. B 80, (2009). 17. F. Nardi, D. Deleruyelle, S. Spiga, C. Muller, B. Bouteille and D. Ielmini, J. Appl. Phys. 112, (2012). 18. T. Tsuruoka, K. Terabe, T. Hasegawa, I. Valov, R. Waser and M. Aono, Adv. Funct. Mater. 22, 777 (2012). 19. S. Tappertzhofen, I. Valov, T. Tsuruoka, T. Hasegawa, R. Waser and M. Aono, Acs NANO 7, 6396 (2013). 20. J. F. Gibbons and W. W. Beadle, Solid-State Electron. 7, 785 (1964). 21. C. Kugeler et al., Thin Solid Films 518, 2258 (2010). 22. S. I. Kim, J. H. Lee, Y. W. Chang, S. S. Hwang and K. H. Yoo, Appl. Phys. Lett. 93 (2008). 23. K. Oka, T. Yanagida, K. Nagashima, H. Tanaka and T. Kawai, J. Am. Chem. Soc. 131, 3434 (2009). 24. B. Gao et al., IEEE Electron Device Lett. 30, 1326 (2009)