Research Article www.acsami.org
Using Dopants to Tune Oxygen Vacancy Formation in Transition Metal Oxide Resistive Memory Hao Jiang†,‡ and Derek A. Stewart*,† †
San Jose Research Center, HGST, a Western Digital Company, San Jose, California 95119, United States Department of Materials Science and Engineering, University of Wisconsin-Madison, Madison, Wisconsin 53706, United States
‡
Downloaded via UNIV OF SOUTHERN INDIANA on July 26, 2019 at 14:53:37 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.
S Supporting Information *
ABSTRACT: Introducing dopants is an important way to tailor and improve electronic properties of transition metal oxides used as high-k dielectric thin films and resistance switching layers in leading memory technologies, such as dynamic and resistive random access memory (ReRAM). Ta2O5 has recently received increasing interest because Ta2O5-based ReRAM demonstrates high switching speed, long endurance, and low operating voltage. However, advances in optimizing device characteristics with dopants have been hindered by limited and contradictory experiments in this field. We report on a systematic study on how various metal dopants affect oxygen vacancy formation in crystalline and amorphous Ta2O5 from first principles. We find that isoelectronic dopants and weak n-type dopants have little impact on neutral vacancy formation energy and that p-type dopants can lower the formation energy significantly by introducing holes into the system. In contrast, n-type dopants have a deleterious effect and actually increase the formation energy for charged oxygen vacancies. Given the similar doping trend reported for other binary transition metal oxides, this doping trend should be universally valid for typical binary transition metal oxides. Based on this guideline, we propose that p-type dopants (Al, Hf, Zr, and Ti) can lower the forming/set voltage and improve retention properties of Ta2O5 ReRAM. KEYWORDS: resistive RAM, tantalum oxide, dopant, oxygen vacancy, formation energy
1. INTRODUCTION The use of dopants to tune the electronic properties of semiconductors was essential for the development of key electronic devices (e.g., diodes and field-effect transistors) and the subsequent birth of the vast silicon microelectronics industry. In order to maintain Moore’s law at small scales and also expand the suite of existing memory technologies, numerous groups are now also exploring ways to tailor and improve the electronic properties of oxides by introducing dopants. This has been an active area of research for high-k dielectric oxides used in dynamic random access memory (DRAM)1,2 and an emerging field for oxide-based nonvolatile resistive RAM (ReRAM).3−5 Advances in this area have been challenging due to the presence of multiple defects in deposited films, dopant interactions with oxygen vacancies (VO), and difficulty in characterizing transport mechanisms in oxide stacks. Therefore, it is important to develop a clear atomic scale understanding of the impact of dopants on stoichiometric oxides with a particular focus on VO. Tantalum oxide has © 2017 American Chemical Society
emerged as a frontrunner for the primary switching layer used in ReRAM.6−8 In this study, we perform a systematic firstprinciples study of how various metal dopants affect VO formation in Ta2O5 and discuss how this could impact ReRAM device performance. Oxide-based resistive switching devices or memristive devices have demonstrated the necessary fast switching speed and low power consumption required for future ReRAM.3−5 While some oxide ReRAM devices (e.g., metal-doped SrTiO3 in low current limits) do exhibit interface-based switching (area scaling of the low resistance state),9,10 the vast majority of oxide-based ReRAM devices typically demonstrate a lack of area scaling of the low resistance state that points to a highly localized switching mechanism. The prevailing explanation that has found increasing experimental support is that resistive switching Received: January 4, 2017 Accepted: April 24, 2017 Published: April 24, 2017 16296
DOI: 10.1021/acsami.7b00139 ACS Appl. Mater. Interfaces 2017, 9, 16296−16304
Research Article
ACS Applied Materials & Interfaces
found to occur at a low doping concentration. Although there has been limited work on ReRAM devices based on doped Ta2O5, several groups have explored the use of dopants to increase the dielectric constant and reduce the leakage current in Ta2O5 high-k dielectric films used for nanoscale DRAM applications. The results are overall mixed. Prior studies found that Al-42,43 or Hf-doped1 films typically had higher leakage current than undoped Ta2O5 films. The observed higher conductivity could indicate that Al and Hf promote the formation of VO which leads to defect states and facilitates trapassisted tunneling. However, introducing Ti into Ta2O5 was found to reduce the leakage current.1,2 Lau et al.2 argued that the reduced current was due to Ti forming electrically inactive complexes with existing impurities (C and Si) and oxygen − vacancies (e.g., Si−−V++ O −Ti ). It is important to note that if the deposited Ta2O5 film is substoichiometric, Ti dopants could passivate existing oxygen vacancies in the film and reduce the leakage current.1,2 The ability of some dopants to passivate films could be beneficial in ReRAM applications for establishing more consistent high resistance states (HRS) and also improving the ON/OFF window. While these previous ReRAM and high-k dielectric films studies do show that dopants clearly impact the performance of Ta2O5-based resistive switching devices, a deeper atomistic understanding of how dopants affect VO formation in Ta2O5 is required to establish clear chemical trends and identify optimal dopants for robust ReRAM devices. To address this issue, we perform a systematic study of how metal dopants affect VO formation energies in Ta2O5. Formation energies of oxygen vacancies are closely related to the conductive filament forming process where VO are generated and driven to aggregate by an applied voltage.8,14,15 It has been speculated that lower VO formation energies can lead to lower formation voltage.18−20 The formation energies of VO can also affect the stability of the conductive filament as well as migration of VO near dopants, thereby changing the retention and switching performance of devices. We also explore the doping effect on formation energies of VO in both crystalline and amorphous Ta2O5. In almost all Ta2O5 resistive switching devices, the deposited tantalum oxide layers are amorphous.6−8 Amorphous oxides are easier to deposit using standard techniques (e.g., sputtering at low temperatures), and amorphous oxides are generally isotropic so that the device to device variability is reduced. In contrast, it is more challenging to obtain crystalline oxides at annealing temperatures compatible with CMOS BEOL processing. In addition, anisotropic microstructure features such as crystal orientation or grain boundary distribution can also lead to large variance in the properties for small devices. Although Ta2O5 ReRAM devices are typically based on amorphous films, recent experimental results indicate that locally nanocrystalline regions do appear in the vicinity of the conductive filament in the amorphous Ta2O5.8 While the filament forming process initiates from the amorphous phase, resistive switching mainly involves vacancy dynamics in the local environment of the filament. Therefore, a clear understanding of the doping effect on the defect energy landscape in both crystalline and amorphous phases is essential. To address this issue, we first examine oxygen vacancy formation in crystalline and amorphous Ta2O5 in the absence of dopants. We consider the impact of disorder and local bonding on the distribution of vacancy formation energy at different oxygen sites in amorphous Ta2O5. We then perform a systematic study of how the valence state of dopants affects
is caused by the connection and disruption of a conductive filament inside the oxide under applied bias.5,9−13 The conductive filament consists of a region of high V O concentration with high electrical conductivity, and the formation/connection (disruption) of the conductive filament switches the device into the low (high) resistance state.14−16 The microscopic energy landscape for VO formation and diffusion is critical for determining several key device properties like electroforming, switching voltages, retention, and endurance. Among many proposed memory switching binary oxides like HfOx, TiOx, TaOx, NiOx, and AlOx, tantalum pentoxide (Ta2O5) is a leading candidate material for resistive switching devices. Ta2O5 based ReRAM exhibits superior device characteristics (e.g., high switching speed, long endurance, and low voltage) compared to ReRAM based on other oxides.6−8,17 The robust performance of Ta2O5 based devices has been attributed to the low formation energy and migration barrier of VO in the materials, which facilitate the formation, connection, and disruption of the conductive filament.18 While these devices offer many superlative properties, it is also important to note that Ta2O5 devices can suffer from problems related to broad distributions of high and low resistance states and challenges in achieving low current operation. Introducing dopants into the oxide layer can serve as a powerful tool for device performance optimization. It has been speculated that dopants can not only alter the energy landscape of VO but also modify the electrical conductivity of the filament.19 Both of these changes can potentially lead to promising doped oxides with low operating voltages, good retention, high endurance, as well as high ON/OFF ratio. Many studies have been conducted to explore the doping effect on resistive switching device performance.19−41 For example, Zhao et al.19−21 from Stanford University and Zhao et al.26 from Anhui University systematically studied the trend of VO formation energy in metal-doped TiO2 and HfO2 using first principle-calculations. The authors concluded that a larger difference in the number of valence electrons between the dopant and the host atom can cause more significant reduction in the VO formation energy near the dopants.19 In addition, they also found that as the formation energy of VO decreases, the forming voltage and ON/OFF current ratio also decrease in metal-doped HfO2 resistive switching devices. Besides metal dopants, the doping effect of nonmetal dopants such as H, N, C, and Si in memory switching oxides was also widely studied.24,30−36 Research into using metal dopants to modify the behavior of Ta2O5 ReRAM devices is still a relatively nascent field. To the best of our knowledge, only three references23−25 have reported on the doping effect of metal dopants (Zr, Ti, and Al) in Ta2O5 based resistive switching devices. Ti-doped Ta2O5 devices showed soft breakdown process with higher ON/OFF ratio and higher resistance compared to those of undoped Ta2O5 devices.23 This modification is attributed to the formation of thin conductive filaments due to the suppression of VO − migration caused by inactive Ti−−V++ O −Ti complex. In another study,24 doping Ta2O5−x/TaOy sputtered bilayers with Al for 10 devices resulted in a lower, more consistent reset voltage as well as a tighter distribution of measured low resistance states. However, the distribution of forming voltages was greater than that observed in undoped devices. Most recently, a study25 on Zr-doped Ta2O5 showed decreased forming voltages as the doping concentration increased. The optimal resistance switching performance in terms of ON/OFF ratio and stability was 16297
DOI: 10.1021/acsami.7b00139 ACS Appl. Mater. Interfaces 2017, 9, 16296−16304
Research Article
ACS Applied Materials & Interfaces
Figure 1. Formation energies of neutral VO in amorphous Ta2O5. (a) Cumulative distribution function of VO formation energies; the inset is the radial distribution function of the amorphous Ta2O5 system. (b) Formation energies of VO as a function of coordination index of neighboring Ta atoms, as defined in Supporting Information.
and-quench classical molecular dynamics approach55,56 (see Supporting Information for further details). The quenched amorphous structure was then further optimized by DFT structural relaxation at fixed volume. The radical distribution function (g(r)) of the amorphous Ta2O5 system is plotted in the subplot of Figure 1a. Short-range order is maintained in the amorphous system as the first-nearest-neighbor Ta−O peak (1.94 Å) and O−O peak (2.71 Å) occur at the same positions as in crystalline phase. However, long-range order no longer holds as g(r) converges to 1 when the radius is larger than 4 Å, confirming the amorphous character of the system. To understand the vacancy formation energy distribution in the amorphous system, 40 VO sites are randomly selected out of the total 270 oxygen atom sites in the supercell. Each of the selected VO sites is structurally optimized in neutral state by DFT calculations. The formation energies of neutral VO in an amorphous system are calculated as
oxygen vacancy formation for both crystalline and amorphous Ta2O5. We compare our results to other theoretical and experimental data from other oxides, and we consider how the charge state of the vacancy affects vacancy formation near dopants. Finally, we discuss how this will impact general ReRAM device operation and provide some suggestions for optimal dopants in Ta2O5 based ReRAM devices.
2. EXPERIMENTAL METHODS Density functional theory (DFT) electronic-structure calculations were done using the Quantum Espresso package44 with ultrasoft pseudopotentials. Calculation parameters used in this work are similar to those in our previous paper on crystalline Ta2O5,18 and key parameters are reiterated here. Plane wave and charge density energy cutoffs were 110 and 880 Ry, respectively. We used the Perdew− Burke−Ernzerhof (PBE) generalized gradient approximation (GGA) for exchange and correlation. Force convergence threshold of 0.05 eV/ Å was used for all structural optimizations. A 3 × 3 × 3 supercell (378 atoms) was employed throughout the study in order to capture the infinitely adaptive structure of λ-Ta2O545 as well as to obtain a reasonable amorphous structure. The Brillouin zone was sampled at the Γ point for the large supercell used in this study. Molecular dynamics simulations were conducted using the Atomistic Toolkit (ATK) developed by Quantum Wise.46−48 The empirical potential developed by Trinastic et al.49 was selected to describe the interatomic interactions in Ta2O5 for simulated annealing runs.
a a Ef = E Vac − Estoi − μO
(1)
where EaVac is the supercell energy of the amorphous system with one VO, Eastoi is the supercell energy of the stoichiometric amorphous system (without VO), and μO is the oxygen chemical potential. The cumulative distribution function of Ef is plotted in Figure 1a. The statistical average formation energy of VO in amorphous system is 4.86 eV with a standard deviation of 0.94 eV. The formation energy falls into a range between 2.37 and 6.49 eV, which is much wider than the window between 4.99 and 5.74 eV in the λ phase.18 A similar wide distribution range was also reported in ref 57 and can be attributed to the deviation of local atomic structures in amorphous phase from the crystalline phase. In addition, we found a strong correlation between the VO formation energy and the coordination number of neighboring Ta atoms of VO, as shown in Figure 1b. The definition of the coordination index in Figure 1b is discussed in detail in the Supporting Information. Here an intuitive physical interpretation of the index is that positive (negative) coordination index means neighboring Ta atoms of a VO are initially overcoordinated (undercoordinated) with respect to the coordination number of 6 in the crystalline phase.45 A correlation factor of −0.68 is found between coordination index and VO formation energies (value of −1 indicates a perfect negative correlation and value of 0 indicates no correlation). Therefore, in general, overcoordinated Ta atoms promote lower vacancy formation energies at neighboring oxygen sites.
3. RESULTS Among several proposed crystalline structures,45,50−53 we choose the λ phase45 to represent crystalline Ta2O5. This is because the λ phase not only shows the lowest energy per formula unit45 but also agrees well with experimentally measured bond length distribution and bandgap.54 In λTa2O5, there are three distinct VO sites: 2 coordinated inplane sites (2f), 3 coordinated in-plane sites (3f), and 2 coordinated sites between the Ta−O planes (bwp). The formation energies of oxygen vacancies in these sites without dopants have been investigated in our previous study.18 In ReRAM devices, sputtered amorphous Ta2O5 layers are typically used6−8 rather than crystalline films. Therefore, it is important to examine whether any conclusions drawn from the crystalline Ta2O5 study would still be valid for the amorphous phase. To test this, we have generated an amorphous Ta2O5 supercell to study how dopants affect vacancy formation at different bonding sites. We created the amorphous Ta2O5 supercell from a 3 × 3 × 3 crystalline supercell using a melt16298
DOI: 10.1021/acsami.7b00139 ACS Appl. Mater. Interfaces 2017, 9, 16296−16304
Research Article
ACS Applied Materials & Interfaces
Figure 2. Formation energies of neutral VO next to isolated metal dopants in Ta2O5 with respect to the number of valence electrons of dopants: (a) in λ-Ta2O5; (b) in amorphous Ta2O5. (c) Comparison of VO formation energies at different oxygen sites are shown for doped λ-/amorphous-Ta2O5 in the current study (black squares), doped monoclinic-HfO219,20 (red circles) and doped rutile TiO219,20 (blue triangles) by researchers from Stanford University, and doped monoclinic-ZrO238 (green diamonds) by researchers from Peking University. Relative valence state shows the differences in the number of valence electrons between dopants and host atoms. VO formation energies are scaled between 0 and 1 with respect to the maximum and minimum values reported in each study. Substitutional metal dopants are considered in these studies. The wide distribution window at each valence column comes from different VO formation energies calculated for different oxides.
This finding agrees with the statement in ref 58 that VO tends to form at the position where the O atom and its first neighboring Ta atom have high coordination numbers. This implies that in the initial forming process, oxygen vacancies are generated first in locally oxygen-rich regions as the formation energies in these regions are lower. As more vacancies form and the oxide becomes oxygen deficient, the formation energy increases, and it becomes thermodynamically more difficult to form new vacancies. We now consider the effect of various dopants on oxygen vacancy formation in crystalline and amorphous Ta2O5. Ten metal dopants including Mg, Al, Ga, Ti, Zr, Hf, Nb, Mo, Re, and Ru are selected to cover dopants with valence electrons ranging from 2 to 8. Our calculations show that substitutional sites are more energetically favorable than interstitial sites for these dopants in λ phase (see Supporting Information for details). Given that all selected dopants prefer to occupy substitutional sites, each single dopant is introduced into the crystal and amorphous supercell by replacing one Ta atom in the first-nearest-neighbor Ta site of VO. Three vacancy sites (2f, 3f, and bwp) in λ-Ta2O5 and three VO sites in amorphous Ta2O5 were selected to investigate how formation energies are modified by dopants. The three VO sites in amorphous Ta2O5 were chosen to represent local coordination environments that are lower, equivalent, and higher than that found in crystalline Ta2O5. The local bonding environments for the three vacancies are shown in the Supporting Information. Although ideally one would analyze the impact of dopants at numerous vacancy sites in the amorphous oxide, given the high computational cost associated with these calculations, the current comparison should be sufficient to suggest key trends. The formation energy of VO is calculated as doped doped Ef = E Vac − E lattice − μO + q(εVBM + εF + ΔV )
deposition takes place under oxygen-rich conditions. In this case, oxygen atoms that leave the vacancy sites will go on to form oxygen molecules in the ambient atmosphere. It is important to note that even if the oxygen chemical potential changes (e.g., oxygen poor environment) the trend of vacancy formation energy Ef with respect to various dopants will be preserved since changing μO only provides a constant shift to the trend. For our case of a single dopant in a 378 atom supercell, the doping concentration in our calculations is approximately 2 × 1020 cm−3 or 0.9 atom %. This dopant concentration is comparable to those used in some experimental investigations.24,25,37 In general, a low concentration of dopants is preferable for promoting oxygen vacancies in ReRAM oxide layers. A high dopant concentration could lead to the formation of separate stoichiometric oxide regions (e.g., large regions of Al2O3 embedded in Ta2O5) and a potential reduction in the overall resident vacancy population. The formation energies of neutral VO next to metal dopants in λ-Ta2O5 and amorphous Ta2O5 are plotted with respect to the valence charge state of dopants in Figure 2a,b. These figures highlight a strong dependence of VO formation energy on the number of dopant valence electrons. We find that as the difference in the number of electrons between p-type dopants and host atom (Ta in this case) increases the formation energies of VO near dopants is reduced. Nb has the same number of valence electrons as Ta, and the formation energy of VO in a Nb-doped system at all 3 sites is comparable (
