Synergy effect of co-doping Sc and Y in Sb2Te3 for phase-change memory

Shuwei Hu ab, Jiankai Xiao a, Jian Zhou ab, Stephen R. Elliott abc and Zhimei Sun *ab
aSchool of Materials Science and Engineering, Beihang University, Beijing 100191, China. E-mail:
bCenter for Integrated Computational Materials Engineering, International Research Institute for Multidisciplinary Science, Beihang University, Beijing 100191, China
cDepartment of Chemistry, University of Cambridge, Cambridge CB2 1EW, UK

Received 3rd April 2020 , Accepted 25th April 2020

First published on 27th April 2020

Sb2Te3 phase-change material possesses the highest crystallization speed and hence the highest operating speed among investigated phase-change systems. Doping with Y or Sc has been exploited to optimize the performance of Sb2Te3, yet the substituted Y atoms are strongly clustered, while Sc is extremely expensive and thus is unfavourable for commercialization. In this work, we have successfully obtained better-performance and moderate-cost phase-change materials by co-doping Sc and Y based on ab initio calculations and ab initio molecular-dynamics simulations (AIMD). Sc can shrink the lattice while Y expands the lattice, which makes a perfect match between original and co-doped configurations and hence can benefit by maximizing the release of lattice strain. The co-doping increases the band gap to around 0.5 eV, and the concentration ratio of Sc and Y dopants provides an advantageous tool for controlling the electronic structure. Results of calculations using the BoltzTraP code show that co-doping can result in a significant reduction in the electrical conductivity at room temperature. AIMD simulation of amorphous co-doped Sb2Te3 shows that the incorporation of Sc and Y atoms can effectively improve the thermal stability of amorphous Sb2Te3. Overall, co-doping Sc and Y is a feasible way to improve the properties of Sb2Te3 for phase-change memory applications.

1. Introduction

Phase-change memory (PCM) has become one of the promising next-generation storage-device technologies due to the discovery of a fast phase-change property for compositions located on the GeTe–Sb2Te3 pseudobinary line in the Ge–Sb–Te ternary system.1–4 PCM reveals exceptional promise as a mainstream non-volatile memory, with great cyclability, rapid programming/accessing and extended scalability.5–8 The generic design principle of PCM relies on significant electrical-conductivity differences and fast phase transitions between crystalline and amorphous states of phase-change materials to read and write binary data stored as metastable structural states.9,10

The phase-change characteristic is one of the essential properties for phase-change materials, and plenty of studies have focused on the crystallization kinetics of PCMs.11,12 Both experiments and theoretical simulations indicate that crystallization in the chalcogenide GeTe–Sb2Te3 system is nucleation-driven and that crystal nucleation is a stochastic process mediated by thermal fluctuations.13–15 With a high crystallization speed, Sb2Te3 is considered as one of the competitive candidates to exceed the data-rate limit imposed by the crystallization process of other phase-change materials.16 However, its low crystallization temperature, and hence poor thermal stability of the amorphous phase, is detrimental to data retention in the amorphous state at operating temperatures.17 Besides, the relatively low electrical resistivity of Sb2Te3, compared with that of Ge2Sb2Te5, is disadvantageous to reduce the RESET current.17,18 Doping has been exploited to optimize the performance of Sb2Te3, and various dopants, including C,10 N,16 O,19 Al,20 Si,21 Ti,22 Cr,23 Zn,24 Ge,25 Ag,26 have been investigated. In 2016, Li et al. found that Y-doping of Sb2Te3 can effectively reduce the electrical conductivity of the crystalline state and improve the thermal stability of the amorphous state.27 Then, Rao et al. found that Sc-doped Sb2Te3 has excellent phase-change characteristics, yet its crystalline state is metastable cubic Sb2Te3.28 On this basis, we screened all transition-metal-doped forms of rhombohedral Sb2Te3 by a high-throughput ab initio calculation method and finally obtained three potential optimal dopants, namely Sc, Y and Hg, and an in-depth study verified the validity of Sc doping from ab initio calculations and experiments.29 In the works of Zhou et al. and Zewdie et al., in-depth theoretical investigations have been performed to understand the effects of Y and Sc as dopants in the amorphous Sb2Te3, which unravel the mechanism of improving the thermal stability while maintaining the fast phase-change speed by separate dopant.30,31 As an inexpensive dopant, yttrium can make the electrical conductivity of Sb2Te3 decrease significantly, but the substituted atoms are strongly clustered together, which is unfavourable for the uniformity of phase-change materials.27,32 It is worth to mention that our recent experimental work confirms the above theoretical predictions.15 While doping with Sc can overcome the above shortcomings and maintain the improvement, but it is too expensive and thus is unfavourable for commercialization.29 Therefore, it is natural to think of using the co-doping of Sc and Y to obtain a balance between performance and cost.

In the present work, we have investigated all possible co-doped structures of Sb2Te3 with different concentrations of Sc and Y, and then used the formation energy to confirm that Sc and Y atoms tend to occupy the Sb sites of adjacent Sb atomic layers in the same quintuple layer. Utilizing ab initio calculations and ab initio molecular-dynamics (AIMD) simulations, we performed an in-depth analysis of crystalline and amorphous Sc and Y co-doped Sb2Te3. This work highlights a feasible approach to improve the properties of chalcogenide phase-change materials.

2. Methods

The ab initio calculations in this work are based on density-functional theory (DFT) and were conducted using the Vienna ab initio simulation package (VASP).33,34 The ion-electron interaction is described by the projector-augmented-wave (PAW) method, and the generalized gradient approximation (GGA) of Perdew–Burke–Ernzerhof (PBE) was selected to account for exchange–correlation interactions between electrons.35–37 An energy cutoff of 350 eV was used for the expansion of plane-wave functions, while an automatically generated k-point mesh of 5 × 5 × 1, centred at the Γ point in the Brillouin zone, was adopted and then proved sufficient to achieve convergence. All supercells were fully optimized concerning lattice constants and atomic coordinates by geometrical relaxation, where the criteria for self-consistent iterations were 1 × 10−5 eV for electronic steps and 1 × 10−2 eV Å−1 for ionic steps. In order to take the van der Waals forces into account for better describing the nonbonding interactions, a semi-empirical dispersion potential was added to the conventional Kohn–Sham DFT energy in the scheme of the DFT-D2 method for the calculations.38,39 The Heyd–Scuseria–Ernzerhof (HSE06) hybrid functional was also selected to calculate the electronic density of states to obtain a more accurate band gap.40 The crystal-orbital Hamilton population (COHP) was calculated using the LOBSTER program.41 The transport properties were estimated, based on semiclassical Boltzmann transport theory, with the rigid-band approximation and an assumed constant relaxation time, using the BoltzTraP code.42 For the study of amorphous structures, AIMD simulations used a canonical NVT (constant number, volume, and temperature) ensemble with a Nosé thermostat in constant-temperature processes, and a micro-canonical NVE (constant number, volume, and total energy) ensemble for changing-temperature processes, where the time step was set to 3 fs.43 The statistical analysis of amorphous structures was performed using the R. I. N. G. S. code.44

3. Results and discussion

3.1 Atomic configurations and bonding properties in crystalline co-doped Sb2Te3

In order accurately to reflect the co-doping effect, a reasonable model of Sb2Te3 needs to be established, in which the positions occupied by dopants should conform to the real situation. A 2 × 2 × 1 rhombohedral Sb2Te3 (No. 166 space group R[3 with combining macron]m) supercell containing 60 atoms was utilized to investigate the atomic configurations of Sc and Y dopants. The formation energy was used to judge the optimal doping modes (a detailed calculation method is provided in ESI), and DFT calculations confirmed that both Y and Sc dopants tend to substitute for Sb (denoted as XSb, where X = Y, Sc) among the four possible doping types (shown in Fig. S1, ESI). Accordingly, the co-doped models are denoted by their compositional form as Scx1Yx2Sb2−x1x2Te3. In terms of the calculation method of the formation energy, when the doping concentration of Sc and Y is determined, the difference in formation energy is directly determined by the total energy, so we compare the values of the total energy to find out the optimal configurations. Using ab initio calculations, we have screened out the configuration which has the lowest total energy in the case of co-doping two atoms consisting of one Sc and one Y (denoted as Sc0.083Y0.083Sb1.833Te3), as shown in Fig. 1a. Other optimal configurations with various doping concentrations (up to four dopant atoms in the supercell, x1 + x2 = 0.167, 0.250, 0.333) are provided in ESI (Fig. S2).
image file: d0tc01693d-f1.tif
Fig. 1 (a) The structure with the lowest total energy for Sc0.083Y0.083Sb1.833Te3. (b) The relation between total energy and dopant distance of all structures for Sc0.083Y0.083Sb1.833Te3.

Sc and Y dopants tend to occupy different Sb atomic layers in the same quintuple layer unit, and Sc, Te and Y atoms tend to lie in a straight line. The formula-unit (f.u.) formation energy of 0.07 eV f.u.−1, slightly lower than the sum of only doping with one Sc and doping with one Y, indicates some synergistic effect of co-doping to reduce the total energy of the system. This configuration yields a minimal lattice mismatch of ∼0.1%, which helps to reduce internal stress and to eliminate phase separation. The near-perfect match between original and co-doped lattices originates from the different distortion effects of the two dopants (Sc shrinks the lattice, while Y expands the lattice), thereby neutralizing the effects of each other, and such a distribution implies the most uniform type, maximizing the release of lattice distortion. In order to consider other metastable configurations, we conducted individual calculations for up to 11 symmetrically non-equivalent configurations, the results of which are shown in Fig. 1b. The data show that the total energy is not correlated with the distance between Sc and Y atoms, and that the range of energy values (0.045 eV) is so small that we believe all configurations may exist in real materials. Considering that, when Sc and Y atoms are not in the same quintuple layers, the influence of dopants is almost the same as in single doping systems which have been studied in our previous work,27,29 herein the lowest total energy structure (Fig. 1a), in which both Sc and Y atoms are in the same quintuple layers, was used as a representative for further analysis.

To obtain a visual picture of bonding characteristics between Te and other atoms, the isosurfaces corresponding to electron localization function (ELF) values at 98% of the local maximum ones were employed to pinpoint the positions of lone pairs of electrons.45,46 In electron-rich compounds, lone pairs exert a significant influence on chemical bonding and electronic properties. The ELF isosurfaces generated at the 98% level within one quintuple layer of Sb2Te3 and Sc0.083Y0.083Sb1.833Te3 are shown in Fig. 2a. Around the Te1 atoms (at the interface between neighbouring quintuple layers) are located lone pairs of electrons which are responsible for the existence of geometrical voids and weak bonding, resembling that in Ge2Sb2Te5 and Ge1Sb2Te4.10,47 The accumulation of electrons between bonding atoms implies enhanced localization and hence stronger Sc–Te and Y–Te bonds than Sb–Te bonds. Such stronger bonds may weaken neighbouring Sb–Te bonds to cause electrons to become localized around Sb atoms, and those Sb atoms are more susceptible to be substituted by other dopants at higher doping concentrations. This distinct behaviour of electrons also promises to have an important effect on electronic properties.

image file: d0tc01693d-f2.tif
Fig. 2 (a) ELF isosurfaces (translucent green) generated at the 98th percentile for Sb2Te3 (left) and Sc0.083Y0.083Sb1.833Te3 (right). (b) A COHP bonding analysis for Sb–Te bonds in Sb2Te3 and for Sc–Te and Y–Te bonds in Sc0.083Y0.083Sb1.833Te3. The Fermi-level position is marked by the dashed line.

The crystal-orbital Hamilton population (COHP) was calculated to reveal further the bonding character of Sb–Te, Sc–Te and Y–Te interactions, as shown in Fig. 2b. While the Fermi level falls into nonbonding regions, similar to Ge1Sb2Te4 and Ge2Sb2Te5,48–50 for Sb–Te bonding, energetically unfavourable antibonding states right below the Fermi level are discerned, nonetheless. This implies that a large portion of antibonding states above the Fermi level has been depleted as a result of the optimal stoichiometry Sb2Te3, but an electronic instability rests in the intrinsic Sb–Te bonding characteristics. Meanwhile, these antibonding states are considered to have a considerable degree of association with lone-pair delocalization-induced interactions with such antibonding orbitals, which is a feature of the superfast phase-change property of chalcogenides.50 However, the introduction of Sc and Y dopants reduces the influence of those antibonding states in Sc–Te and Y–Te bonding, which accounts for their stronger interactions as well as the weakness of the phase-change feature.

3.2 Electronic properties of crystalline co-doped Sb2Te3

To understand the origin of the change in electronic properties induced by co-doping, we calculated the band structure of Sc0.083Y0.083Sb1.833Te3 along a commonly used path in reciprocal space using the PBE exchange functional, as shown in Fig. 3a. The band structure displays a larger band gap (0.22 eV) compared with pristine Sb2Te3 (0.13 eV), and the flatter band edges at the conduction-band minimum (CBM) and valence-band maximum (VBM) indicate increased effective masses. The position of the CBM remains at the Γ point, while the position of the VBM has been shifted to the A point, demonstrating the behaviour of an indirect band gap. The band gaps of the other ten symmetrically non-equivalent configurations were also calculated, and the band gap can be increased up to 0.43 eV, maximally. This value is even larger than for single-doping cases with the same overall dopant concentration, because, as the band gap at the Γ point is opened, gaps at other points (especially at the A point) remain rather large, which represents a preferable optimization than for single doping. It is worth mentioning that, from Fig. 3a, Sc-3d and Y-4d orbitals contribute strongly to the CBM, which implies some specific hybridization effect existing in the atomic configuration in Fig. 1a.
image file: d0tc01693d-f3.tif
Fig. 3 (a) The projected band structure of Sc0.083Y0.083Sb1.833Te3. The cyan, blue, red and yellow dots represent the contribution from Sb-5p, Te-5p, Sc-4d and Y-4d orbitals, respectively. The Fermi level is set to 0 eV, and marked by the dotted line. (b) The projected density of states (PDOS) of Sb2Te3 (upper) and Sc0.083Y0.083Sb1.833Te3 (lower). The colour scheme conforms to that in (a); here, the top of the VBM is set to 0. (c) The relation between band-gap value and distance between Sc and Y dopants of all configurations for Sc0.083Y0.083Sb1.833Te3, along with the arithmetic-mean band-gap value, marked by the horizontal dashed line, and the corresponding average inter-dopant distance, marked by the vertical dashed line. (d) The band gaps of Scx1Yx2Sb2−x1x2Te3.

Since band gaps of Sb2Te3 calculated by the PBE functional are underestimated, the HSE06 hybrid functional was then used to calculate the projected density of states (PDOS), from which we can derive a more accurate value of the band gap. Fig. 3b shows the PDOS of Sb2Te3 and Sc0.083Y0.083Sb1.833Te3, respectively. It shows an increase in the band gap from 0.35 eV to 0.53 eV by means of co-doping. Sc-3d and Y-4d orbitals contribute a lot to the conduction band, which reflects a mixed feature of Sb2Te3, Sc2Te3 and Y2Te3, and the magnified density at the CBM corresponds to the flatter band edge and hence increased effective mass. The enlargement of the band gap can be understood by an increased difficulty in transferring valence electrons, induced by the accumulation at bonding regions and the transformation from delocalization to localization, as previously shown in Fig. 2a.

Considering that dopants may be randomly distributed, the band gaps of all possible Sc0.083Y0.083Sb1.833Te3 configurations were calculated using the HSE06 functional, as shown in Fig. 3c. Although we assume that the mean value may represent the real material's behaviour, a slight positive correlation between band gap and dopant distance implies that real materials will possess better properties if the dopants are widely dispersed, and the uniformity is thus improved. The cases of three and four doping atoms in the 60-atom supercell were also considered, for which all the band gaps calculated using the HSE06 functional are shown in Fig. 3d. It is obvious that co-doping increases the band gaps to about 0.5 eV. Furthermore, the concentration ratio of Sc and Y dopants can effectively regulate the band gap, because a dispersed dopant distribution in Sc-doped Sb2Te3 is related to larger band gaps, which provides an advantageous tool for controlling the electronic structure.

3.3 Electrical conductivity of crystalline co-doped Sb2Te3

Since the phase-change process needs enough electrical power to heat the phase-change materials above their melting point in the RESET process, such a melting operation is always limited by the power, so the electrical conductivity becomes a critical determinant of the RESET current. We can estimate the electrical conductivity using the expression σ = (neμe + peμh), where the electron and hole mobilities are respectively given by μe = /me* and μh = /mh*, and such expressions demonstrate that the effective mass, m*, the collision time, τ, and the carrier density, n, are three tunable elements in determining the electrical conductivity. Within the k·p perturbation theory,51image file: d0tc01693d-t1.tif, where c and ν denote the conduction- and valence-band edge states, respectively; we can see that flatter band edges and an enlarged band gap increase the effective mass. In most doped systems, dopants usually act as electron-scattering centres, and this situation is such a case, in which Sc and Y dopants can reduce the collision time. Based on the considerations as mentioned above, the electrical conductivity of co-doped Sb2Te3 is believed to be reduced effectively. Here, the electrical conductivity of co-doping configurations is calculated within the framework of the semiclassical Boltzmann transport theory. In order to simplify the calculation, we use the experimental value of the collision time τ of pristine Sb2Te3 for calculation,52 and the results are shown in the inset of Fig. 4. Both single doping or co-doping systems can lead to a significant reduction in the electrical conductivity at room temperature, and the conductivity has little relationship with the doping concentration ratio (x1[thin space (1/6-em)]:[thin space (1/6-em)]x2) at a given doping concentration (x1 + x2). The values for n = 1020 cm−3 with the same concentration are averaged to represent this doping effect, and the conductivity decreases continuously as the doping concentration increases, as shown in Fig. 4. The continuous decrease in electrical conductivity turns out to be promising in reducing the power consumption of PCMs. The interchangeability between Sc and Y dopants also enables engineers to replace any proportion of dopants with the other element in industrial applications, which provides a new idea for reducing costs, and the mechanism of adjustable ratios between them also renders controlled modification possible.
image file: d0tc01693d-f4.tif
Fig. 4 The calculated average electrical conductivity as a function of overall doping concentration (n = 1020 cm−3). Positive and negative carrier concentrations in the insets represent p- and n-type doping, respectively. The insets show the electrical conductivity as a function of carrier concentration for Scx1Yx2Sb2−x1x2Te3, where x1 + x2 is: (a) 0.083; (b) 0.167; (c) 0.250; (d) 0.333.

3.4 Local structure and thermal stability of amorphous co-doped Sb2Te3

A 4 × 4 × 1 supercell containing 240 atoms was utilized to construct amorphous models for Sb2Te3 and Sc0.083Y0.083Sb1.833Te3via a simulated melt-quenching process. The ensemble was instantaneously melted and the temperature was held at 3000 K to generate an utterly disordered arrangement, then quenched to 1000 K (the melting point of Sb2Te3 is 900 K) to relieve internal stress, and eventually quenched to 300 K to produce amorphous models. The first quenching process ran for 2000 steps and the second ran for 15000 steps with a rate of cooling gradient of −15 K ps−1. This quenching rate was verified to establish reasonable models that no longer preserved liquid character nor resulted in crystallization.

For the amorphous state of phase-change materials, the crystallization kinetics have different essential characteristics. Some phase-change materials are nucleation-driven (for example, the Ge–Sb–Te system), while others are growth-driven (for example, Ag4In3Sb67Te26 (AIST)).1 Therefore, there are plenty of methods to control the crystallization process of phase-change materials.53–56 In forming the bonding network, amorphous chalcogenides possess a certain proportion of wrong-bonding configurations, i.e., Sb–Sb.57 Lee et al. found that ultrafast crystal growth of phase-change material is intimately linked to the presence of distinctly coexisting weak covalent and lone-pair interactions,50 and wrong bonds can help create these conditions. Fig. 5a shows diffusely existing homopolar Sb–Sb bonds in the pristine Sb2Te3 model but fewer such bonds in the co-doped model due to the depletion of wrong bonding configurations around dopants (shown in Fig. 5b). The dopants prefer to bond with surrounding Te atoms and afterwards reduce wrong bonds, which is appropriately conducive to control the phase-transition speed. To more accurately describe the chemical environment of Sb–Sb wrong bonds, contour plots of ELF projected on the (0 0 1) plane are shown in Fig. 6a. Distinct lone pairs are formed near the Sb atoms of Sb–Sb bonds; meanwhile, the Sb–Te covalent bonds are significantly weakened. These phenomena can help to reduce the energy barrier for bond breakage and reorganization during the phase-change process.

image file: d0tc01693d-f5.tif
Fig. 5 The distribution of homopolar Sb–Sb bonds in: (a) amorphous Sb2Te3; (b) amorphous Sc0.083Y0.083Sb1.833Te3, along with heteropolar Sc–Te and Y–Te bonds. Translucent spheres depict Sb and Te atoms not involved in the bonds mentioned above.

image file: d0tc01693d-f6.tif
Fig. 6 (a) ELF contour plot of Sb–Sb wrong bonding configuration on the (0 0 1) plane. (b) Projected density of states (PDOS) of amorphous Sb2Te3 and Sc0.083Y0.083Sb1.833Te3 calculated by the HSE06 functional.

More detailed insights can be obtained from the DOS. Here, we adopted the more computationally intensive HSE06 functional rather than the PBE functional for more accurate band-gap estimations. Unlike in the crystalline state, the spatial fluctuations of dihedral angles, bond lengths and bond angles in the amorphous state broaden the edges of the conduction and valence bands to form band tails. In the band tails, the electronic states can have a localized character, which appears as a spike at the edge of the band gap in the DOS. From Fig. 6b, unlike for the crystalline state of Sb2Te3, the p-orbital electrons of both Sb and Te show a strong localizability tendency like band tails, which confirms the generation of lone-pair electrons. However, the dopants do not show this band-tail phenomenon, and the electron localizability is significantly decreased after doping with Sc and Y atoms. Hence, reducing Sb–Sb wrong-bonding configurations can be beneficial for controlling the otherwise too easy phase-change process of Sb2Te3 and hence for improving its thermal stability. Meanwhile, the value of the band gap of the amorphous state of Sc0.083Y0.083Sb1.833Te3 is 0.48 eV, which is slightly lower than the crystalline state, while the value of amorphous Sb2Te3 increases to 0.43 eV from 0.35 eV for the crystalline state. This change is closely related to the change of localized character before and after adding dopants.

In order to further study the mechanism of better performance with co-doping, statistical data for the amorphous configurations were calculated. From Fig. 7a, the partial pair correlation functions show a stronger propensity of Sc and Y than Sb to coordinate with Te, which identifies six strong Y–Te bonds, compared with three weak Sb–Te bonds. A rather weak peak for Sb–Sb correlations reveals the origin of homopolar bonds, while neither Sc nor Y seems to form any wrong bonds with Sb. This distinction in configurations can regularize the neighbouring arrangement around dopants to get rid of wrong bonding in such regions. From Fig. 7b, the bond-angle distribution of Te–Sb–Te shows a defective octahedral environment for Sb atoms in the liquid and amorphous phases, reminiscent of the geometry in the crystalline phase.58 However, both Te–Sc–Te and Te–Y–Te present broadened bimodal distribution peaks, shifting from ∼90° to ∼75° and from ∼170° to ∼135°, and the deviation of peaks manifests the transformation of the neighbouring environment around dopants. Moreover, the total average coordination numbers of Sc and Y are 7.13 and 7.69, respectively, larger than the value of 5.39 for Sb, showing a stronger coordination effect. Here, we take the cutoff radius corresponding to the minimum after the first peak in the pair distribution function. Based on the combination of the above facts, Sc and Y atoms are no longer in a defective octahedral environment, and their coordination configurations are conducive to a distortion in geometry, which are spatially unfavourable for crystal-nucleus appearance and growth.

image file: d0tc01693d-f7.tif
Fig. 7 (a) Partial pair correlation functions: Sb-centered in amorphous Sb2Te3; Sc-centered and Y-centered in amorphous Sc0.083Y0.083Sb1.833Te3. Solid (dashed) lines correspond to the coordination with Te (Sb) atoms. Vertical dotted lines mark the positions of the first maximum concerning solid lines. (b) Bond-angle distribution: Te–Sb–Te in amorphous (solid), liquid (dotted), crystalline (shaded) Sb2Te3; Te–Sc–Te and Te–Y–Te in amorphous Sc0.083Y0.083Sb1.833Te3. Vertical dotted lines mark the positions of peaks concerning solid lines.

Thermal stability is an important performance metric of non-volatile memory. Here, we used a simulated annealing process, and a higher temperature of 600 K was adopted to accelerate the phase-change process. We measured the degree of crystallization from two aspects, energy and residual pressure. The amorphous phase lies in a higher-energy state compared with the crystalline phase, and the process of crystallization therefore results in a drop in the total energy. Fig. 8a shows a steep drop of the total energy at 40 ps in the case of pristine Sb2Te3, while the total energy of co-doped Sb2Te3 is almost unchanged. Residual pressure is also a commonly used indicator to measure the degree of crystallization. The higher the degree of crystallization, the lower the residual pressure. Fig. 8b shows the variation of residual pressure with simulated annealing time. Pristine Sb2Te3 has a steep pressure release around 50 ps, indicating that it has undergone a phase-change process. In contrast, co-doped Sb2Te3 shows no obvious changes, which indicates a good improvement in thermal stability. Overall, co-doping of Sb2Te3 effectively improves thermal stability, which is very beneficial for commercial applications.

image file: d0tc01693d-f8.tif
Fig. 8 (a) The total energy evolutions for amorphous Sb2Te3 and co-doped Sb2Te3 (Sc0.083Y0.083Sb1.833Te3) from the AIMD calculations at 600 K for 100 ps. (b) The residual pressure evolutions for amorphous Sb2Te3 and co-doped Sb2Te3 (Sc0.083Y0.083Sb1.833Te3) from the AIMD calculations at 600 K for 100 ps.

4. Conclusions

In this work, using ab initio calculations, we have successfully screened for all optimal Sc and Y co-doped crystalline Sb2Te3 configurations. Sc and Y dopants substitute for Sb atoms in different Sb atomic layers within the same quintuple layer unit, and Sc, Te and Y atoms tend to lie in a straight line. Sc can shrink the lattice while Y expands the lattice, which makes for a perfect overall match between undoped and co-doped configurations and has the benefit of maximizing the release of lattice strain. ELF and COHP results provide evidence for stronger Sc–Te and Y–Te bonds compared with Sb–Te bonds, which reduce the contribution of antibonding states and account for stronger interactions and a strengthened structure. Using PBE and HSE06 functionals, we particularly analyzed the effect of co-doping on the band gap; co-doping increases the band gap to about 0.5 eV. Furthermore, the concentration ratio of Sc and Y dopants provides an advantageous tool for controlling the electronic structures. By use of the BoltzTraP code to calculate the electrical conductivity, both single doping and co-doping can result in a significant reduction in the electrical conductivity at room temperature, and this is not sensitive to the concentration ratio of Sc and Y dopants. AIMD simulations of amorphous co-doped Sb2Te3 show that Sc and Y atoms can suppress Sb–Sb wrong bonding configurations, which helps to control the speed of the phase-change process. Meanwhile, Sc and Y atoms are no longer in a defective octahedral environment, and their coordination configurations are conducive to distortions in geometry, which are spatially unfavourable for crystal-nucleus growth and hence beneficial to improve the thermal stability of amorphous Sb2Te3. Finally, it is worth mentioning that co-doping is a feasible approach to improve the properties of Sb2Te3 for phase-change applications.

Conflicts of interest

The authors declare that they have no conflicts of interest.


This work was financially supported by the National Key Research and Development Program of China (Grant No. 2017YFB0701700), and the National Natural Science Foundation of China (Grant No. 51872017).


  1. W. Zhang, R. Mazzarello, M. Wuttig and E. Ma, Nat. Rev. Mater., 2019, 4, 150–168 CrossRef CAS.
  2. J. Hegedüs and S. R. Elliott, Nat. Mater., 2008, 7, 399–405 CrossRef PubMed.
  3. Z. Sun, J. Zhou and R. Ahuja, Phys. Rev. Lett., 2007, 98, 055505 CrossRef PubMed.
  4. A. Lotnyk, M. Behrens and B. Rauschenbach, Nanoscale Adv., 2019, 1, 3836–3857 RSC.
  5. Z. Sun, J. Mater. Sci., 2012, 47, 7635–7641 CrossRef CAS.
  6. Y. Chen, N. Chen, B. Chen, Q. Zhang, X. Li, Q. Deng, B. Zhang, S. Zhang, Z. Zhang and X. Han, J. Appl. Phys., 2018, 124, 145107 CrossRef.
  7. A. L. Lacaita, Solid-State Electron., 2006, 50, 24–31 CrossRef CAS.
  8. B. Liu, S. Hu, J. Zhou and Z. Sun, AIP Adv., 2019, 9, 035131 CrossRef.
  9. S. Raoux, W. Welnic and D. Ielmini, Chem. Rev., 2010, 110, 240–267 CrossRef CAS PubMed.
  10. G. Wang, J. Zhou, S. R. Elliott and Z. Sun, J. Alloys Compd., 2019, 782, 852–858 CrossRef CAS.
  11. D. Loke, T. H. Lee, W. J. Wang, L. P. Shi, R. Zhao, Y. C. Yeo, T. C. Chong and S. R. Elliott, Science, 2012, 336, 1566–1569 CrossRef CAS PubMed.
  12. I. Ronneberger, W. Zhang, H. Eshet and R. Mazzarello, Adv. Funct. Mater., 2015, 25, 6407–6413 CrossRef CAS.
  13. M. Salinga, E. Carria, A. Kaldenbach, M. Bornhöfft, J. Benke, J. Mayer and M. Wuttig, Nat. Commun., 2013, 4, 1–8 Search PubMed.
  14. D. Loke, J. M. Skelton, T. H. Lee, R. Zhao, T. C. Chong and S. R. Elliott, ACS Appl. Mater. Interfaces, 2018, 10, 41855–41860 CrossRef CAS PubMed.
  15. B. Liu, W. Liu, Z. Li, K. Li, L. Wu, J. Zhou, Z. Song and Z. Sun, ACS Appl. Mater. Interfaces, 2020 DOI:10.1021/acsami.0c03027.
  16. X. Li, F. Rao, Z. Song, M. Zhu, W. Liu and Z. Sun, J. Appl. Phys., 2011, 110, 094318 CrossRef.
  17. X. Li, Z. Sun, Z. Song, F. Rao, L. Wu and W. Liu, Solid State Sci., 2011, 13, 131–134 CrossRef CAS.
  18. L. Zhang, N. Miao, J. Zhou, J. Mi and Z. Sun, J. Alloys Compd., 2018, 738, 270–276 CrossRef CAS.
  19. Y. Yin, S. Morioka, S. Kozaki, R. Satoh and S. Hosaka, Appl. Surf. Sci., 2015, 349, 230–234 CrossRef CAS.
  20. C. Peng, L. Wu, Z. Song, F. Rao, M. Zhu, X. Li, B. Liu, L. Cheng, S. Feng and P. Yang, Appl. Surf. Sci., 2011, 257, 10667–10670 CrossRef CAS.
  21. X. Li, F. Rao, Z. Song, K. Ren, W. Liu and Z. Sun, Appl. Surf. Sci., 2011, 257, 4566–4568 CrossRef CAS.
  22. M. Zhu, M. Xia, F. Rao, X. Li, L. Wu, X. Ji, S. Lv, Z. Song, S. Feng and H. Sun, Nat. Commun., 2014, 5, 1–6 Search PubMed.
  23. Q. Wang, B. Liu, Y. Xia, Y. Zheng, R. Huo, M. Zhu, S. Song, S. Lv, Y. Cheng and Z. Song, Phys. Status Solidi RRL, 2015, 9, 470–474 CrossRef CAS.
  24. G. Wang, Q. Nie, X. Shen, R. Wang, L. Wu, Y. Lv, J. Fu, T. Xu and S. Dai, Mater. Lett., 2012, 87, 135–138 CrossRef CAS.
  25. J. Yu, B. Liu, T. Zhang, Z. Song, S. Feng and B. Chen, Appl. Surf. Sci., 2007, 253, 6125–6129 CrossRef CAS.
  26. J. Xu, B. Liu, Z. Song, S. Feng and B. Chen, Mater. Sci. Eng., B, 2006, 127, 228–232 CrossRef CAS.
  27. Z. Li, C. Si, J. Zhou, H. Xu and Z. Sun, ACS Appl. Mater. Interfaces, 2016, 8, 26126–26134 CrossRef CAS PubMed.
  28. F. Rao, K. Ding, Y. Zhou, Y. Zheng, M. Xia, S. Lv, Z. Song, S. Feng, I. Ronneberger, R. Mazzarello, W. Zhang and E. Ma, Science, 2017, 358, 1423–1427 CrossRef CAS PubMed.
  29. S. Hu, B. Liu, Z. Li, J. Zhou and Z. Sun, Comput. Mater. Sci., 2019, 165, 51–58 CrossRef CAS.
  30. Y. Zhou, L. Sun, G. M. Zewdie, R. Mazzarello, V. L. Deringer, E. Ma and W. Zhang, J. Mater. Chem. C, 2020, 8, 3646–3654 RSC.
  31. G. M. Zewdie, Y. Zhou, L. Sun, F. Rao, V. L. Deringer, R. Mazzarello and W. Zhang, Chem. Mater., 2019, 31, 4008–4015 CrossRef CAS.
  32. Z. Li, N. Miao, J. Zhou, H. Xu and Z. Sun, J. Appl. Phys., 2017, 122, 195107 CrossRef.
  33. W. Kohn, Rev. Mod. Phys., 1999, 71, 1253 CrossRef CAS.
  34. G. Kresse and J. Furthmüller, Phys. Rev. B: Condens. Matter Mater. Phys., 1996, 54, 11169 CrossRef CAS PubMed.
  35. P. E. Blöchl, Phys. Rev. B: Condens. Matter Mater. Phys., 1994, 24, 17953 CrossRef PubMed.
  36. J. P. Perdew and Y. Wang, Phys. Rev. B: Condens. Matter Mater. Phys., 1992, 45, 13244–13249 CrossRef PubMed.
  37. J. P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. Lett., 1996, 77, 3865–3868 CrossRef CAS PubMed.
  38. Z. Li, N. Miao, J. Zhou, Z. Sun, Z. Liu and H. Xu, Nano Energy, 2018, 43, 285–290 CrossRef CAS.
  39. B. S. Sa, J. Zhou, R. Ahuja and Z. M. Sun, Comput. Mater. Sci., 2014, 82, 66–69 CrossRef CAS.
  40. J. Heyd, G. E. Scuseria and M. Ernzerhof, J. Chem. Phys., 2003, 118, 8207–8215 CrossRef CAS.
  41. R. Dronskowski and P. E. Blöchl, J. Phys. Chem., 1993, 97, 8617–8624 CrossRef CAS.
  42. G. K. H. Madsen and D. J. Singh, Comput. Phys. Commun., 2006, 175, 67–71 CrossRef CAS.
  43. N. Shuichi, Prog. Theor. Phys. Suppl., 1991, 103, 1–46 CrossRef.
  44. S. Le Roux and P. Jund, Comput. Mater. Sci., 2010, 49, 70–83 CrossRef CAS.
  45. A. Savin, R. Nesper, S. Wengert and T. F. Fässler, Angew. Chem., Int. Ed. Engl., 1997, 36, 1808–1832 CrossRef CAS.
  46. G. Gibbs, D. Cox, M. Boisen Jr, R. T. Downs and N. Ross, Phys. Chem. Miner., 2003, 30, 305–316 CrossRef CAS.
  47. J. Zhou, Z. Sun, Y. Pan, Z. Song and R. Ahuja, EPL, 2011, 95, 27002 CrossRef.
  48. M. Wuttig, D. Lusebrink, D. Wamwangi, W. Welnic, M. Gillessen and R. Dronskowski, Nat. Mater., 2007, 6, 122–128 CrossRef CAS PubMed.
  49. J. M. Skelton and S. R. Elliott, J. Phys.: Condens. Matter, 2013, 25, 205801 CrossRef CAS PubMed.
  50. T. H. Lee and S. R. Elliott, Adv. Mater., 2017, 29, 1700814 CrossRef PubMed.
  51. L. C. L. Y. Voon and M. Willatzen, The k·p method: electronic properties of semiconductors, Springer, Berlin, 2009 Search PubMed.
  52. N. F. Hinsche, S. Zastrow, J. Gooth, L. Pudewill, R. Zierold, F. Rittweger, T. Rauch, J. Henk, K. Nielsch and I. Mertig, ACS Nano, 2015, 9, 4406–4411 CrossRef CAS PubMed.
  53. J. Orava, A. L. Greer, B. Gholipour, D. W. Hewak and C. E. Smith, Nat. Mater., 2012, 11, 279–283 CrossRef CAS PubMed.
  54. J. Orava, D. W. Hewak and A. L. Greer, Adv. Funct. Mater., 2015, 25, 4851–4858 CrossRef CAS.
  55. P. Zalden, F. Quirin, M. Schumacher, J. Siegel, S. Wei, A. Koc, M. Nicoul, M. Trigo, P. Andreasson, H. Enquist, M. J. Shu, T. Pardini, M. Chollet, D. Zhu, H. Lemke, I. Ronneberger, J. Larsson, A. M. Lindenberg, H. E. Fischer, S. Hau-Riege, D. A. Reis, R. Mazzarello, M. Wuttig and K. Sokolowski-Tinten, Science, 2019, 364, 1062–1067 CrossRef CAS PubMed.
  56. B. Chen, Y. Chen, K. Ding, K. Li, F. Jiao, L. Wang, X. Zeng, J. Wang, X. Shen, W. Zhang, F. Rao and E. Ma, Chem. Mater., 2019, 31, 8794–8800 CrossRef CAS.
  57. Y. Chen, T. Xu, X. Shen, R. Wang, S. Zong, S. Dai and Q. Nie, J. Alloys Compd., 2013, 580, 578–583 CrossRef CAS.
  58. S. Caravati, M. Bernasconi and M. Parrinello, Phys. Rev. B: Condens. Matter Mater. Phys., 2010, 81, 014201 CrossRef.


Electronic supplementary information (ESI) available. See DOI: 10.1039/d0tc01693d

This journal is © The Royal Society of Chemistry 2020