Impact of Al Alloying/Doping on the Performance Optimization of HfO₂-Based RRAM

Abstract

Al alloying/doping in HfO₂-based resistive random-access memory (RRAM) has been proven to be an effective method for improving the low-resistance state (LRS) retention. However, a detailed understanding of Al concentration on oxygen vacancy migration and resistive switching (RS) behaviors still needs to be included. Herein, the impact of Al concentration on the RS properties of the TiN/Ti/HfAlO/TiN RRAM devices is addressed. Firstly, it is found that the forming voltage, SET voltage, and RESET voltage can be regulated by varying the Al doping concentration. Moreover, we have demonstrated that the device with 15% Al shows the minimum cycle-to-cycle variability (CCV) and superior endurance (over 10⁶). According to density-functional theory (DFT) calculations, it is found that the increased operation voltage, improved uniformity, and improved endurance are attributed to the elevated migration barrier of oxygen vacancy through Al doping. In addition, LRS retention characteristics of the TiN/Ti/HfAlO/TiN devices with different Al concentrations are compared. It is observed that the LRS retention is greatly enhanced due to the suppressed lateral diffusion process of oxygen vacancy through Al doping. This study demonstrates that Al alloying/doping greatly affects the RS behaviors of HfO₂-based RRAM and provides a feasible way to improve the RS properties through changing the Al concentration.

1. Introduction

RRAM based on transition metal oxides (TMOs) may revolutionize future computing systems and it is considered as one of the most promising candidates for replacing conventional FLASH technology due to its outstanding advantages, including high integration density, low-power operation, high switching speed, and compatibility with the CMOS process [1,2,3,4,5,6]. Among all the investigated TMOs, HfO₂ is one of the most representative candidates because of its high thermal stability, superior endurance, and fast switching speed [7,8,9]. However, it is worth noting that HfO₂-based RRAM operates on the basis of conductive filaments formation/rupture caused by oxygen vacancy migration [10,11]. The uncontrollable ion migration dynamics result in stochastic formation/rupture of filaments, leading to switching variability including resistance fluctuation and even the device failure, and tail-bit retention failure in the LRS, which severely hinders practical applications of HfO₂-based RRAM [12].

In recent years, many reports have demonstrated that Al alloying/doping can effectively improve the endurance, RS uniformity, and LRS retention of RRAM devices [13,14,15,16]. Fantini et al. and Chen et al. found that LRS retention was remarkably enhanced by Al incorporation in HfO₂ [13,14]. Furthermore, Azzaz et al. observed that through inserting a thin Al₂O₃ layer, the reliability of HfO₂-based RRAM devices, including RS uniformity and endurance, could be remarkably improved. However, the effect of Al concentration on RRAM operations and performances, i.e., forming, SET/RESET, endurance, uniformity, and LRS retention, as well as their correlation, has not been thoroughly investigated in previous reported works.

In this study, the electrical properties of HfO₂-based RRAM devices with different Al doping concentrations are systematically investigated. It is found that the forming voltage (V_f), SET voltage (V_SET), and RESET voltage (V_RESET) slightly increase with the increase in Al concentration due to the elevated migration barrier of oxygen vacancy through Al doping. Furthermore, the HfO₂-based RRAM devices exhibit improved RS uniformity and endurance with a higher Al concentration. It is suspected that the elevated oxygen vacancy migration barrier is beneficial for enhancing the stability of oxygen vacancy drift process, resulting in the improvement of uniformity and endurance. More importantly, LRS retention of the device with Al doping is greatly enhanced due to the suppressed lateral diffusion process of oxygen vacancy through Al alloying/doping.

2. Materials and Methods

The fabrication process of the TiN/Ti/HfAlO/TiN devices is described in Figure 1a. To fabricate the TiN/Ti/HfAlO/TiN devices, 10 nm TiN was first deposited on heavily doped Si substrates as a bottom electrode by ALD using ammonia and titanium tetrachloride as precursors. Sequentially, 8 nm HfAlO (9:1), 8 nm HfAlO (4:1), and 8 nm HfAlO (2:1) with different Al concentrations were deposited by ALD at 300 °C. To prepare HfAlO (9:1) film, 9 cycles of HfO₂ were first deposited, followed by 1 cycle of Al₂O₃. Thus, 9 cycles of HfO₂ and 1 cycle of Al₂O₃ can form a complete layer. In addition, HfAlO (4:1) stands for 4 cycles of HfO₂ followed by 1 cycle of Al₂O₃, and HfAlO (2:1) stands for 2 cycles of HfO₂ followed by 1 cycle of Al₂O₃, respectively. Notably, 8 nm Al-free HfO₂ film was used as the reference sample. After the HfAlO deposition, the Ti buffer layer with a thickness of 15 nm was deposited on top of the HfO₂ layer by sputtering. Next, 10 nm TiN was deposited through ALD to protect Ti from oxidation. Finally, the TE layer (TiN/Ti) was patterned via the lithography and inductively coupled plasma etching processes. Figure 1b illustrates the schematic of the TiN/Ti/HfAlO/TiN devices, accompanied by a cross-sectional high-resolution transmission electron microscope (HRTEM) image of the RRAM device. From the energy dispersive spectrometry (EDS) elemental mapping image, one can clearly observe four layers of TiN, Ti, HfAlO, and TiN with clear interfaces between the adjacent layers.

An Agilent B1500A semiconductor parameter analyzer conducted electrical measurements in the atmospheric ambient. The voltage bias was applied to the top electrode while the bottom electrode was grounded.

All calculations were performed using the projector augmented-wave (PAW) method of the DFT as implemented in the Vienna ab initio simulation package (VASP) [17]. The exchange-correlation potential was used within the local density approximation (LDA) with U corrections (LDA + U), which was proven to achieve accuracy in HfOx similar to hybrid functionals such as HSE06. Periodic boundary conditions, and an energy cutoff of 220.431 eV, were employed for the plane-wave expansion of electron wave functions. All ions were relaxed to an energy convergence of 10⁻⁶ eV/atom and forces less than 0.01 eV/Å per ion. The simulations were carried out on a 3 × 3 × 3 supercell of monoclinic HfO2 with 108 Hf and 210 O atoms for HfO2, 94 Hf, 210 O, and 14 Al atoms for HfAlO (9:1), 80 Hf, 210 O, and 32 Al atoms for HfAlO (4:1), respectively.

3. Results

3.1. Al Concentration-Dependent RS

Table 1. Extracted physical parameters of the studied samples.

Split	Al%	Density (g/cm3)	Band Gap (eV)
HfO2	0%	9.62	5.6
HfAlO (9:1)	6.7%	8.79	5.7
HfAlO (4:1)	15%	8.2	5.8
HfAlO (2:1)	30%	7.8	5.9

summarizes the main physical parameters of the HfO₂ with Al alloying/doping used in our study. Material density was measured by X-ray reflectometry (XRR), Al% was measured by Rutherford backscattering spectroscopy (RBS), and the optical band gap was measured by ellipsometry. It was found that HfAlO with a higher Al concentration showed both a smaller density and a larger band gap.

For the TiN/Ti/HfAlO/TiN RRAM devices, a forming process was required to achieve a stable switching behavior, as shown in Figure 2a. At the beginning, the current of the present devices was low due to its high resistance. After the application of a relatively large positive voltage, the device switched to the LRS. Subsequently, stable bipolar resistive switching was obtained when the voltage swept between −1.5 V and 1.4 V, as shown in Figure 2b. During a continuous sweeping of a bias voltage from 0 V→1.4 V→−1.5 V→0 V, a pinched hysteresis loop was obtained, in which SET (the switching from a high-resistance state (HRS) to LRS) occurs at about 0.9 V and RESET (the switching from a LRS to HRS) happens at about −0.9 V. A compliance current of 0.7 mA was applied at the SET process to prevent irreversible breakdown, while no compliance current was applied at the RESET process.

To reveal the effect of Al doping concentration on the RS performance, electrical properties, including V_f, V_SET, and V_RESET of the TiN/Ti/HfAlO/TiN devices with different Al concentrations, are depicted in Figure 2. Note that more than 10 devices were tested for each structure to compare. As shown in Figure 2a, the mean values of V_f of our prepared devices with three different Al concentrations (0, 6.7%, and 15%) are 2.88, 3.11, and 3.16 V, respectively. It is worth noting that the devices with 30% Al are prone to hard breakdowns and cannot be reset due to high breakdown voltage. It is clearly observed that the V_f slightly increases with the increase in Al concentration. Figure 2b,c show the V_SET and V_RESET distribution of the TiN/Ti/HfAlO/TiN devices with different Al concentrations. It is found that the absolute values of V_SET and V_RESET shows a slight increase when the Al doping concentration increases from 0% to 15%. The higher V_f, V_SET, and V_RESET for HfAlO with higher Al concentration may be related to a higher migration barrier of oxygen vacancy, which will be discussed in detail below.

To further investigate the effect of Al concentration on the RS uniformity, the CCV of the TiN/Ti/HfAlO/TiN devices with different Al concentrations is displayed in Figure 3. Figure 3a,b present the statistical distribution of V_SET, V_RESET, LRS, and HRS, obtained by 100 dc sweep cycles. It can be clearly observed that the sample with higher Al doping concentration shows an improved control of tail bits in the distribution of V_SET, V_RESET, LRS, and HRS, indicating that the device with 15% Al doping concentration exhibits the minimum CCV. The above results demonstrate that the device with 15% Al doping concentration shows the best RS uniformity. In addition, it is found that the resistance values of HRS and LRS increase with the increase in Al doping concentration due to a lower leakage current in HfAlO with a higher migration barrier of oxygen vacancy.

In addition to RS uniformity, the reliability of the TiN/Ti/HfAlO/TiN devices with different Al concentrations is systematically investigated, as shown in Figure 4. Figure 4a compares the endurance of the devices with different Al concentrations of 0%, 6.7%, and 15%. To ensure the reliability and accuracy of the experimental results, the same electrical condition with fixed SET pulse (1.7 V/700 ns)/RESET pulse (−1.2 V/7 μs) was applied. It is clearly observed that the ON/OFF ratio is over 10 and also that the endurance increases from 200 to 2.4 × 10⁶ when the Al concentration increases from 0% to 15%, indicating that Al doping can effectively promote the improvement of endurance.

The evolution of the LRS retention profile for HfAlO with different Al doping concentrations is also compared, as shown in Figure 4b. Notably, the same initial resistance level was required before backing all the tested devices to ensure the reliability of the data. It can be clearly observed that HfAlO with 15% Al doping concentration shows the best LRS retention at 200 °C. The improvement in LRS retention with Al doping is in agreement with the results of [13,14]. In this study, the bit failure time was defined as the point where the initial LRS drifted toward 10⁵. The extracted failure times based on the above criteria are illustrated in an Arrhenius failure time plot with respect to the inverse of the temperature, showing that the LRS retention failure process in HfAlO-based RRAM follows an Arrhenius type of law. Higher Ea is extracted for HfAlO with 15% Al doping concentration (∼1.7 eV), which is consistent with its better thermal stability, as shown in Figure 4b. Failure temperatures of 149 °C, 142 °C, and 135 °C at 10 years are extracted for HfAlO with Al doping concentrations of 0%, 6.7%, and 15%, respectively. The improvement of LRS retention is attributed to the suppression of the lateral diffusion of oxygen vacancies, which will be discussed in Section 3.2.

3.2. Mechanism Analysis

As an attempt to have a deeper microscopic understanding of the improvement of RS properties, which are associated with the oxygen vacancy migration process, the migration barriers of oxygen vacancy in HfAlO and HfO₂ are calculated using DFT calculations. Figure 5 shows the diffusion profile of oxygen vacancy in HfO₂, and an Ea of 2.16 eV is consistent with the 2.19 eV of [18]. Moreover, Ea for HfAlO (9:1) and HfAlO (4:1) is 2.33 eV and 2.53 eV, respectively, indicating that Al alloying/doping increases the oxygen vacancy migration barrier, which is consistent with the high Ea for HfAlO (4:1) extracted from this experiment. The reason for the higher migration barrier of HfAlO (4:1) is the bond shrinkage in HfO₂ with Al incorporation [19]. Therefore, larger operation voltages for the devices with higher Al concentration can be explained by the increased migration barrier in HfO₂ with Al doping (see Figure 2).

As shown in Figure 6a, the migration barrier of the oxygen vacancy increases with the increase in Al doping concentration, leading to a slower oxygen vacancy drift/diffusion process under the same applied electric field or concentration gradient. To clarify the improved RS properties of HfO₂ with a higher Al concentration, schematic diagrams of the generalized model, to understand the role of the oxygen vacancy migration process in the RS performance during SET and RESET operations, are depicted in Figure 6b,c. For the TiN/Ti/HfAlO/TiN devices, the RS behaviors originate from conductive filaments formations/ruptures, and the longitudinal drift of oxygen vacancies at the HfAlO/TiN interface dominates the SET/RESET processes, as shown in Figure 6b. Although a higher migration barrier results in an increase in operating voltage, it can also enhance the stability of the oxygen vacancy migration process, thereby improving the endurance and uniformity of the devices. Figure 6c illustrates the oxygen vacancy diffusion process in the LRS, where green dots stand for oxygen vacancies. Lateral oxygen vacancy diffusion is an important origin of tail-bit retention loss. In the LRS, the filament is formed when oxygen vacancies drift from the HfAlO to the bottom TiN under the external electric field. Oxygen vacancies at the edge of the filament can diffuse out laterally and thus low resistance increases. Al doping can increase the height of the oxygen vacancy migration barrier, and thus oxygen vacancy lateral diffusion is hindered in tail bits, leading to the enhancement of tail-bit retention in the LRS.

4. Conclusions

In this study, the effect of the Al doping concentration on the RS performance (operation voltage, uniformity, endurance, and LRS retention) has been systematically investigated. It was found that the device with a higher Al concentration showed higher operation voltage, improved uniformity, and endurance. According to DFT calculations, the Al alloying/doping plays a significant role in the engineering of the migration barrier of oxygen vacancy, which contributes to enhancing the stability of the oxygen vacancy migration process. In addition, LRS retention of the TiN/Ti/HfAlO/TiN devices with different Al concentrations was compared. It was found that the device with 15% Al exhibited the best LRS retention due to the suppressed lateral diffusion process of oxygen vacancy through Al alloying/doping.