Excellent resistive switching properties of atomic layer-deposited Al2O3/HfO2/Al2O3 trilayer structures for non-volatile memory applications

We have demonstrated a flexible resistive random access memory unit with trilayer structure by atomic layer deposition (ALD). The device unit is composed of Al2O3/HfO2/Al2O3-based functional stacks on TiN-coated Si substrate. The cross-sectional HRTEM image and XPS depth profile of Al2O3/HfO2/Al2O3 on TiN-coated Si confirm the existence of interfacial layers between trilayer structures of Al2O3/HfO2/Al2O3 after 600°C post-annealing. The memory units of Pt/Al2O3/HfO2/Al2O3/TiN/Si exhibit a typical bipolar, reliable, and reproducible resistive switching behavior, such as stable resistance ratio (>10) of OFF/ON states, sharp distribution of set and reset voltages, better switching endurance up to 103 cycles, and longer data retention at 85°C over 10 years. The possible switching mechanism of trilayer structure of Al2O3/HfO2/Al2O3 has been proposed. The trilayer structure device units of Al2O3/HfO2/Al2O3 on TiN-coated Si prepared by ALD may be a potential candidate for oxide-based resistive random access memory.

However, devices using metal oxides suffer from the dispersion of resistive switching parameters, such as the resistance values of low and high resistance states (LRS and HRS), the required set voltages from the HRS to the LRS, and reset voltages from the LRS to the HRS, which may lead to false programming and readout error [19]. Ruptures of the conducting filaments with various sizes at random locations are thought as the main reason for the non-uniformity of resistive switching parameters [20,21]. Several methods have been attempted to solve this problem, such as minimizing grain boundaries, doping, embedding nanoparticles, device area scaling, and bilayer structures in the memory devices [22][23][24][25][26][27][28]. The bilayer structure devices have been confirmed with evidently improved resistive switching behaviors [24,25]. Several models have been proposed to explain the enhanced performance [29,30]. Moreover, shrinkage of the active memory unit area to sub-100-nm size using a plug-contact-type bottom electrode was also able to obtain a sharp distribution in switching parameters.
The atomic layer deposition (ALD) is a kind of unique and modified chemical vapor deposition (CVD) method [31,32]. The precursor vapors are pulsed into the reactor alternatively and separated by purging with an inert gas.
This results in a stepwise surface-saturated and selflimiting gas-solid reaction mechanism with many advantages such as good reproducibility, excellent conformity and uniformity over large area, low deposition temperature, and simple and precise control of film thickness, especially for deposition of nano-laminated structure. In this work, we fabricated trilayer-structure flexible RRAM units based on Al 2 O 3 /HfO 2 /Al 2 O 3 functional stacks on TiN-coated Si substrate by ALD so as to achieve excellent resistive switching performances with negligible parameter dispersion.

Methods
ALD was performed in a commercial Picosun SUNALE TM R-200 advanced reactor (Picosun, Masalantie 365, FI-02430 Masala, Finland). P-type Si (100) wafers with a resistivity of 1~10 Ω · cm were used as the starting substrates. After the conventional RTA cleaning of the Si wafers without removing native oxide with the diluted HF solution, 30-nm-thick TiN was deposited on Si as bottom electrode at 400°C using TiCl 4 at room temperature (RT) and NH 3 plasma gas as the Ti and N sources by plasma-enhanced atomic layer deposition (PEALD). Liquid NH 3 at room temperature was used as NH 3 plasma source. The plasma power and NH 3 gas flow rate were 2,500 W and 150 sccm, respectively. Subsequently, 6 nm Al 2 O 3 /10 nm HfO 2 /3 nm Al 2 O 3 stacking structures ( Figure 1a) were deposited in turn on TiN-coated Si substrates at 250°C by thermal ALD using Hf[N(C 2 H 5 )CH 3 ] 4 (TEMAH), Al(CH 3 ) 3 , and H 2 O as the Hf, Al, and O sources, respectively, where one oxide cycle consisted of 0.1-s metal source injection, 4-s N 2 purging, 0.1-s H 2 O injection, and 4-s N 2 purging. TEMAH was evaporated at 150°C. Pure N 2 (99.999%) was used as carrier gas and purge gas. Then, 100-nm-thick Pt top electrodes were DC sputtered through a shadow mask with a diameter of 150 μm using the Q150T system. Post-metallization annealing (PMA) was performed at 600°C for 60 s in N 2 using rapid thermal annealing so as to remove slight organic residue in oxide stacking and improve the ohmic contact between the electrodes and metal oxide films.
The growth per cycle (GPC) of pure Al 2 O 3 or HfO 2 on Si was determined by spectroscopic ellipsometer (GES-5, Annecy-le-Vieux, France). The nominal thickness of Al 2 O 3 /HfO 2 /Al 2 O 3 trilayer structure by ALD on TiN-coated Si was evaluated to be about 6.1, 10.3, and 3.0 nm, respectively. The actual thickness and microstructures of Al 2 O 3 /HfO 2 /Al 2 O 3 on TiN-coated Si were examined by a transmission electron microscope (TEM, Tecnai G2F20 S-Twin, FEI, Hillsboro, OR, USA) operating at 200 kV. The composition and chemical state of the samples were analyzed via X-ray photoelectron spectroscopy (XPS, Thermo Fisher K-Alpha, Thermo Fisher Scientific, Waltham, MA, USA) with a monochromatic Al K α source (hν = 1,486.6 eV) for excitation of photoelectrons. The charge effects were calibrated by setting the C 1-s photoemission at 284.6 eV. The XPS depth profile of Al 2 O 3 /HfO 2 /Al 2 O 3 on TiN-coated Si was obtained by Ar ion etching. The resistive switching properties were measured under different modes using a Keithley 4200-SCS semiconductor characterization system at room temperature and 85°C. A current compliance of 10 mA was imposed to protect the fabricated device units from damages of high currents during set processes.

Results and discussion
The schematic of the RRAM device structures of Al 2 O 3 / HfO 2 /Al 2 O 3 trilayer structure by ALD on TiN-coated Si with Pt top electrode is illustrated in Figure 1a. Figure   For almost all the samples, larger forming voltage is needed to form conductive channels before the switching test. The forming voltage of device unit is about −2 V. The initial resistance state of the memory unit in the first cycle (black curve) is higher than that of the second and third cycles (blue and red curves). Moreover, as-prepared original device unit is in the LRS and an excess positive voltage of 2 V is needed to reset the device unit from LRS to HRS (denoted by arrows 1 and 2 in Figure 3a). The I-V curves from the second and third cycles are almost in superposition with similar set and reset voltages.
For high-density memory application, uniformity of both set and reset voltages is very important. Figure 3b,c plots the statistical results of distribution and cumulative probability of the set and reset voltages measured from a single device unit for 400 times tests. The set and reset voltages show narrow distribution from −1.21 to −0.85 V and from 1.24 to 1.38 V in a device unit, respectively. Their corresponding average values are −0.96 and 1.31 V, respectively. Moreover, 50 randomly selected device units also exhibit less deviation of set and reset voltages, as seen in Figure 3d. Especially, the reset voltage has better monodispersion than the set voltage in Figure 3b-d.
The switching endurance and data retention characteristics of the device unit of Pt/Al 2 O 3 /HfO 2 /Al 2 O 3 /TiN/Si have been examined. The stable and reproducible switching properties have been achieved in Figure 4a,b. The sweeping voltage was applied from 0 to −2 V for set and 0 to 1.5 V for reset with a reading voltage of 0.1 V at room temperature. The endurance test shows a consistent 10 3 switching cycle with a stable resistance ratio of OFF/ ON states above 10 ( Figure 4a). Retention characteristics of the device unit were measured at room temperature and 85°C, as seen in Figure 4b. Both HRS and LRS were read at 0.1 V for cumulative waiting time of 10 4 s. At room temperature, no obvious degradation of the memory window is observed with slightly increasing LRS and HRS. At 85°C, the memory window exhibits better thermal stability with the resistance ratio of LRS/HRS >10, indicating a better retention endurance of the memory unit over a 10-year lifetime based on the extrapolation method. It is interesting that in the retention test, both HRS and LRS increase with time at room temperature but decrease at 85°C on the contrary. This phenomenon can be explained by the change of oxygen vacancy concentration with time at room temperature and 85°C. As known, the oxygen vacancy concentration of some oxide thin film samples stored in air ambient is not constant [33]. The air oxygen may slowly diffuse into the Al 2 O 3 /HfO 2 /Al 2 O 3 -based samples at room temperature, leading to the gradual decrease of oxygen vacancy concentration, i.e. the increase of HRS and LRS in oxide thin films with retention time. When raising the measuring temperature to 85°C, on one hand, the oxygen diffusion into the device unit reduces the oxygen vacancy concentration; on the other hand, the enhanced temperature may produce more oxygen vacancies in Al 2 O 3 /HfO 2 /Al 2 O 3 trilayer structures. Evidently the increased carriers of oxygen vacancies predominate, so both HRS and LRS of the device unit decrease with time at 85°C.   Evidently, the above results confirm that the device unit of trilayer structure of Al 2 O 3 /HfO 2 /Al 2 O 3 on TiN-coated Si exhibits excellent resistive switching performances, such as stability of resistance ratio of on/off states, uniformity of set and reset voltages, and reliability and durability of switching between the LRS and the HRS.
In order to obtain in-depth understanding on switching mechanism of trilayer structure of Al 2 O 3 /HfO 2 /Al 2 O 3 on TiN-coated Si, we performed the XPS analyses on Al 2 O 3 / HfO 2 /Al 2 O 3 structures. XPS spectra were fitted with Gaussian-Lorentzian (G-L) functions after smart-type background subtraction.
The narrow-scan XPS spectra of Al 2p, Hf 4f, and O 1-s peaks in Al 2 O 3 and HfO 2 layers are shown in  diffusion between Al 2 O 3 and HfO 2 films. Defect equation of the interfacial diffusion can be expressed as: Where Al ′ Hf means that Al 3+ occupies the position of Hf 4+ in HfO 2 lattice, with singular negative charge. Therefore, based on Equations 1 and 2, HfO 2 layer produces and stores more oxygen vacancies than Al 2 O 3 layer. The physical size of the conductive filament in oxygen vacancy-deficient layer is narrower than in the oxygen vacancy-rich layer [19], so it can be inferred reasonably that the filament in HfO 2 layer is thicker and more intensive. Based on the bilayer-structure model, the oxygen exchange through the interface between two layers played an important role in improving resistive switching characteristics [19]. In this work, the two interfacial layers (IL) between trilayer-structure of Al 2 O 3 /HfO 2 /Al 2 O 3 on TiN-coated Si have also a similar effect on connected or broken filaments in memory units, as indicated in Figure 6a,b.
The two interfacial layers between Al 2 O 3 /IL/HfO 2 /IL/ Al 2 O 3 could affect the growth position, direction, and overlapping of conductive filaments. The linkage (Figure 6a) or rupture (Figure 6b) of the conductive filaments corresponds to the SET from HRS to LRS or the RESET from LRS to HRS, respectively, which more easily occur in two interfacial layers. Further, the shape and position of the conductive filaments in Al 2 O 3 and HfO 2 layers change less in the SET and RESET processing. The resistive switching mechanism is mainly dominated by formation and rupture of conducting filaments of oxygen vacancies in interfacial layers. The thermal redox near the interface between the oxides or electrochemical migration of oxygen ions may be also the main driving force [4,37,38]. Further work is needed to understand this mechanism well.

Conclusions
In summary, reliable and uniform RRAM units based on trilayer structure of Al 2 O 3 /HfO 2 /Al 2 O 3 on TiN-coated Si have been prepared by ALD. The cross-sectional HRTEM image and XPS depth profile of Al 2 O 3 /HfO 2 / Al 2 O 3 on TiN-coated Si confirm the existence of interfacial layers between the trilayer structure of Al 2 O 3 /HfO 2 /Al 2 O 3 . The memory units of Pt/Al 2 O 3 / HfO 2 /Al 2 O 3 /TiN/Si exhibit a typical bipolar, reliable, and reproducible resistive switching behavior, such as stable resistance ratio (>10) of OFF/ON states, sharp distribution of set and reset voltages, better switching endurance up to 10 3 cycles, and longer data retention at 85°C over 10 years. The possible switching mechanism of trilayer structure of Al 2 O 3 /HfO 2 /Al 2 O 3 has been proposed, and two interfacial layers between Al 2 O 3 /IL/HfO 2 /IL/Al 2 O 3 play an important role in improving resistive switching characteristics. The oxygen ion migration and the redox reactions at the tip of the localized filament lead to the formation/rupture of the conducting filaments mainly in two interfacial layers during the SET and RESET processing. The Al 2 O 3 /HfO 2 /Al 2 O 3 trilayer structure device units by ALD may be a potential candidate for oxide-based RRAM.