2.1 Crystal Structure of α-Al2S3 and Polarization Switching Behavior
The crystal structures of cation-vacancy ordered wurtzite α-Al2S3 and conventional wurtzite ZnS are summarized in Fig. 1. α-Al2S3 shows a polar and chiral P61 space group symmetry. Two-thirds of the tetrahedral sites in wurtzite-type structures are occupied by the Al atoms, while the remaining one-third are vacant 22,23. In contrast to conventional wurtzite-type compounds, α-Al2S3 has two kinds of crystallographically inequivalent cation sites, both of which are located at Wyckoff positions 6a. These sites can be readily distinguished based on the position of the nearest-neighbor Al vacant sites, i.e., whether it locates either in the sulfur vertex or basal-triangle side along the c axis. Figure 1e and f extract a column of the coordination tetrahedra along the c axis from the whole structure of α-Al2S3 and ZnS, respectively. Hereafter, the tetrahedral sites with a nearest-neighbor vacant site in the vertex or basal-triangle side are referred to as T1 and T2, respectively. The suffixes "u" and "d" indicate the upwards-oriented or downwards-oriented tetrahedra, respectively.
The calculated lattice constants (a = 6.41 Å, c = 17.80 Å) for α-Al2S3 are in good agreement with the experimental ones (a = 6.44 Å, c = 17.90 Å) 23, substantiating that the GGA-PBEsol functional used in this study well reproduces crystal structures. The calculated electric polarization of α-Al2S3, 66 µC/cm2, is comparable to or smaller than those of wurtzite ferroelectrics (e.g., 65 and 135 µC/cm2 for ZnS and AlN, respectively) 13,15. An MEP between the two oppositely polarized states with all the tetrahedra oriented upwards or downwards was calculated by using SS-NEB methods to elucidate the switching barrier and behavior (Fig. 2a). Interestingly, α-Al2S3 has four local energy minima in the MEP, referred to as + HP, +LP, –LP, and –HP in the order of the magnitude of polarization. Here, the states with high, low, and zero absolute polarization values are denoted as HP, LP, and ZP, respectively, and the prefix "+" and "–" symbols indicate positive and negative polarization values, respectively. Figure 3 illustrates these five structures providing the local energy minima and the saddle point. (An animation of the structural evolution associated with the switching is available in the Supporting Information and would help to understand the switching behavior.) Both + HP and + LP states possess P61 space group symmetry. The P61 symmetry is preserved during the switching pathway, except in the saddle-point ZP state, which has zero polarization and nonpolar P6122 symmetry. The 61 symmetry element is preserved during the entire polarization switching process, indicating that the polarization direction remains along the c axis. In contrast to quadruple-well ferroelectrics such as BiFeO3, which have multiple independent polarization axes 24, the four local-minimum states in the MEP of α-Al2S3 are polarized along the c axis, revealing that α-Al2S3 is an unusual ferroelectric with a uniaxial quadruple-well potential as observed in CuInP2S6 25.
Figure 2b illustrates the structural evolution of α-Al2S3 during the polarization switching as the columns of coordination tetrahedra along the c axis for simplicity. All the six columns included in the unit cell are related to each other by the 61 symmetry operation, which is preserved during the polarization switching. The atomic displacement in the column reflects the entire polarization switching process. In the initial high-polarization + HP state, the two inequivalent Al atoms, denoted as Al1 and Al2, occupy the T1u and T2u sites, respectively, enclosed by the upwards-oriented tetrahedra. Beyond a small potential hill, there emerges the + LP state with lower total energy and polarization (+ 30 µC/cm2), where the Al1 atoms occupy the T1u sites whereas the Al2 atoms occupy 5-fold coordinated bipyramidal sites, referred to as B1 sites. In the structural evolution from the + HP to + LP states, the Al2 atoms move from the T2u sites towards the vacant sites and migrate to the bipyramidal B1 sites, whereas the Al1 atoms remain at the T1u sites. Upon transitioning from the + LP to ZP states, the Al2 atoms move from the bipyramidal B1 sites to the T1d sites in the downward-oriented tetrahedra, whereas the Al1 atoms still remain at the T1u sites, resulting in a half-switched state where half of the tetrahedra and the other half are oriented upward and downward with the polarization canceled out.
In the latter half of polarization switching from the ZP state to the –LP state to the final –HP state, the Al1 atoms at the T1u sites migrate to the T2d sites via the bipyramidal B2 sites. It should be noted that the Al1 (Al2) atoms sitting at the T1u (T2u) sites in the initial + HP state occupy the T2d (T1d) sites in the final –HP states, indicating the swapping of the crystallographically inequivalent sites during the polarization switching between the T1 and T2 sites. This highlights the nonconventional ferroelectricity of α-Al2S3.
Our first-principles phonon calculation revealed that both the HP and LP states were dynamically stable (Figure S1 and S2). Notably, Al2S3 with the LP structure has been synthesized by chemical vapor transport 23, while solid-state reaction methods yield α-Al2S3, which adopts the HP structure. These facts imply that these isosymmetric polymorphs are energetically antagonized at room temperature or above, although our first-principles calculations predict that the LP state is more stable than the HP state at 0 K.
2.2 Comparison of Switching Behavior and Barriers in Wurtzite Compounds
In this study, the switching barriers for quadruple-well potential surfaces are defined as the highest energy barrier between a valley and its neighboring peak towards the switching direction 18. The switching barrier of α-Al2S3 corresponds to the total energy difference between the LP and ZP states. The switching barrier of α-Al2S3, 51 meV/cation, is approximately one-tenth that of AlN (523 meV/cation) and one-third that of Al15/16B1/16N (150 meV/cation) 26. The moderate switching barrier is anticipated to enable polarization switching prior to electric breakdown. We compare the switching behavior in α-Al2S3 and other wurtzite ferroelectrics to understand the moderate switching barrier in α-Al2S3 below.
In the binary wurtzite ferroelectrics such as pristine AlN, all the cations displace collectively from the upwards-oriented to the downwards-oriented tetrahedra during polarization switching 12,14. In ternary wurtzite systems such as Li2SiO3, for which two-step polarization switching has been predicted by first-principles calculations, more electronegative atoms move first, followed by the migration of less electronegative atoms, with the first switching barrier being the highest 18. In contrast to these cases, in α-Al2S3, consisting of only one type of cations, the Al atoms situated at the crystallographically different sites exhibit individual motion; the Al2 atoms exhibit displacement preceding that of the Al1 atoms, as shown in Fig. 2b, predominantly because the Al2 atoms can displace towards the vacancy sites apart from the Al1 atoms, thereby mitigating cation-cation electrostatic repulsion.
In the saddle-point structures for MEPs in binary and ternary wurtzite-type compounds, cations locate at the trigonal bipyramidal sites 10,12,26. The basal triangles of the bipyramids do not offer enough space for accommodating the cations, resulting in the expansion of the anion triangles. The resultant in-plane lattice expansion destabilizes the saddle-point structures. In sharp contrast, α-Al2S3 has the lowest total energy when half of the Al atoms locate at the bipyramidal sites (i.e., the LP states), highlighting the switching behavior quite different from that for other wurtzite ferroelectrics. Upon transitioning from the + HP to + LP states, the Al2 atoms move to the center of S triangles, accompanying an expansion of the triangle (See Figure S3). In conventional wurtzite ferroelectrics, such triangle expansion increases the in-plane lattice constants typically by 5% 14,16, destabilizing the h-BN-like saddle-point structures. In α-Al2S3, the in-plane lattice constant for the LP state is only 1% larger than that of the HP state (See Figure S4). Figure 3b and d depict the LP structure, in which the AlS5 and AlS4 polyhedra tilt with respect to the c axis, resulting in the buckling of close-packed sulfur basal planes, i.e., non-flat sulfur basal planes. The buckling alleviates the in-plane lattice expansion. In conventional wurtzite-type compounds, four anion tetrahedra are connected to each other via an S atom, "locking" the anion polyhedral network. Meanwhile, in α-Al2S3, just two or three tetrahedra are connected to each other due to the Al vacancies, leading to a flexibility of the polyhedral network.
Thus, in contrast to conventional wurtzite-type ferroelectrics, the cation vacancies mitigate the electrostatic repulsion between cations in α-Al2S3 (Fig. 2b). Also, the disconnection of polyhedral network induced by the Al vacancies produces structural flexibility alleviating the in-plane lattice expansion. These features are considered as the primary reasons why the switching barrier of the defective wurtzite α-Al2S3 is much smaller than those of "filled" wurtzite-type compounds.
2.3 Stabilization Mechanism of AlS5 Trigonal Bipyramidal Coordination
We have discussed above how the Al vacancies stabilize the intermediate structures including the LP states from the structural aspects. The half of Al atoms occupy the trigonal bipyramidal sites in the most stable intermediate LP structures for α-Al2S3, whereas, in binary and ternary wurtzite ferroelectrics, cations occupy the trigonal bipyramidal sites in the unstable saddle-point structures. It remains unclear why the LP states are stable in α-Al2S3 from the viewpoint of chemical bonding. To elucidate the underlying stabilization mechanism of the AlS5 bipyramids in terms of chemical bonding, bond valence (BV) and crystal orbital Hamilton population (COHP) between the Al and S atoms were calculated. Here, BV and COHP describe the strength of chemical bonding based on the bond length and the interactions between atomic orbitals, respectively. The negative and positive values of COHP indicate bonding and antibonding interactions, respectively. We utilize negative COHP (–COHP) integrated with respect to energy (–ICOHP) within the valence bands, which indicates the magnitude of net energy gain due to bonding and anti-bonding interactions. Figure 4a labels the S atoms composing of AlS4 tetrahedra or AlS5 bipyramids as follows: the axial S atoms located at the + c and –c sides with respect to the Al atoms as ax1 and ax2, respectively, and the equatorial S atoms composing of basal triangles as Eq. 1, Eq. 2, and Eq. 3. Figure 4b and c present the BVs and –ICOHP, respectively, for the Al1 and Al2 atoms against the S atoms for the ZP, +LP, and + HP states.
Transitioning from the + HP to + LP states, the Al2 atoms migrate from the tetrahedral T2u sites to the bipyramidal B1 sites (Fig. 2b), as described above. In the + HP state, the BV sums and –ICOHP of both Al1 and Al2 atoms are primarily contributed to by four S atoms, ax1, Eq. 1, Eq. 2, and Eq. 3, with minimal contribution from ax2, confirming 4-fold tetragonal coordination (Fig. 4b and c). In contrast, in the + LP state, the BV sum and –ICOHP of Al2 atom are significantly contributed to by the five S atoms, corroborating 5-fold coordination rather than 3-fold coordination.
Let us pay attention to the chemical bonding evolution for the Al2 atoms, as it is crucial to understand the stabilization mechanism of the + LP state. Figure 4b reveals that the BV sum of the Al2 atoms is significantly smaller for the + LP state than for the + HP state, indicating that the ionic bonding between Al2 and S atoms become weaker upon transitioning from the + HP to + LP states. As shown in Fig. 4c, however, the sum of –ICOHP for the Al2 atom is comparable in the + LP and + HP states, revealing that the energy gain due to covalent bonding remains upon the structural evolution. This motivates us to unravel the Al2-S covalent bonding that compensates for the poor ionic bonding in more detail.
Figure 4d and e depict the –COHP for the Al2-S bonds as a function of energy in the + LP and + HP states, respectively. A remarkable difference is seen just below the Fermi level; the axial and equatorial S atoms show bonding and antibonding contributions, respectively, in the energy range from − 1 to 0 eV for the + LP state, yielding a net bonding contribution, whereas negligible –COHP is found in this energy range for the + HP state, indicating no bonding contribution. This remarkable difference is considered as a major factor stabilizing the LP states. Figure 4f and g illustrate the wavefunctions of eigenstates just below the Fermi level for the + LP and + HP states, respectively, both of which mainly consist of S 3pz orbitals. In the eigenstate of the + LP state, the 3pz orbitals of the ax2 S atoms elongate towards the Al2 atoms to overlap with the Al2 3pz orbitals, clearly indicating a s-like bonding interaction between these orbitals. Meanwhile, the S 3pz orbitals of the ax2 S atoms are localized and nonbonding in the eigenstate for the + HP state. The ax2 S atoms are 2-fold coordinated by Al atoms in the + HP state, rendering one of the 3p orbitals nonbonding. The Al2 displacements to the bipyramidal sites yield an additional Al coordination to the ax2 S atoms, causing the ax2 S 3pz orbitals to take part in the bonding states with the Al 3pz states. The formation of the bonding states in the LP states is considered as another factor stabilizing the LP state with respect to the HP state.
2.4 Effects of Epitaxial Strain and Chemical Doping on Switching Barriers
For wurtzite ferroelectrics, lowering the switching barrier is a key priority for wider practical applications. It has been demonstrated that epitaxial biaxial strain and chemical doping lower the switching barriers for ZnO and AlN by theory and experiments 16,17,27. Here, the effects of biaxial strain and chemical doping on MEPs are investigated for Al2S3 to give insights into the experimental control of ferroelectric switching.
The reduction of switching barrier in α-Al2S3 requires the stabilization of the ZP states and/or the destabilization of the LP states. Without the constraint of strain, the in-plane lattice constant of the LP state is larger than that of the ZP state, as shown in Figure S4, implying that the MEP can be controlled by harnessing biaxial strain. Figure 5a shows calculated MEPs under biaxial strain. The total energies of the LP states become higher with respect to those of the ZP states under compressive (negative) biaxial strain, which is consistent with the in-plane lattice constant of the LP state larger than that of the ZP state (Figure S4). Figure 5b shows the switching barriers under strain, which correspond to the total energy difference between the LP and ZP states. The switching barriers decrease by 8.5% with 2% compressive strain. The biaxial strain dependence of switching barriers for α-Al2S3 is contrary to that for wurtzite ferroelectrics such as ZnO and AlN, where the switching barriers are reduced by tensile biaxial strain 14. This is because the in-plane lattice constant decreases in α-Al2S3 and increases in typical wurtzite ferroelectrics when climbing the potential hills for the polarization switching.
Next, we consider the chemical doping effects on the switching barriers. It has been reported that the switching barrier is successfully reduced for AlN by doping B atoms, which favor planar triangle coordination. It is not likely that B-atom doping helps lower the switching barrier for α-Al2S3, because it can stabilize the LP states with trigonal bipyramidal coordination. Here, we focus on doping of Ga atoms. Ga is considered as a prime candidate for the following three reasons; (1) Ga atoms are isovalent to Al. (2) α-Ga2S3 adopts γ-In2Se3 structure similarly to α-Al2S3 28. (3) The ionic radius of Ga3+ is bigger than that of Al3+. The cation-anion radius ratio of Al to S (rAl/rS = 0.212) is less than 0.225, which is the minimum value for 4-fold tetrahedral coordination according to the Pauling's third rule. Meanwhile, rGa/rS is 0.255, indicating that Ga atoms favor tetrahedral coordination of S atoms 29. It is expected from these facts that Ga-doping stabilize the ZP and HP states and destabilize the LP states, reducing the switching barrier.
Figure 5c shows the MEPs for representative structural models of Al(12-x)/6Gax/6S3 with x = 0, 2, 4, 6, 8, 10, 12. An increase in x stabilizes the ZP and HP states with respect to the LP states, as expected. In the low-x range (0 \(\:\text{≤}\) x \(\:\text{≤}\) 6), the pathway from the +LP to ZP to –LP states corresponds to the highest potential hill, whereas, in the high-x range (6 \(\:\text{≤}\) x \(\:\text{≤}\) 12), the +HP-+LP pathway includes the highest potential hill. Figure 5d shows the box-and-whisker plot of the switching barriers against doping concentration x in Al(12-x)/6Gax/6S3. The switching barrier decreases with an increase in x, and shows a minimum at x = 6, followed by an increase in the barrier above x = 6. The barrier is about 40% smaller for x = 6 compared to pristine α-Al2S3. Thus, our calculations predict that Ga-doping facilitates the polarization switching for α-Al2S3. Furthermore, Ga-doping modulates the relative energy relationship of the four different polarization states, which enables the control of stable phases.
2.5 Piezoelectric Constants
α-Al2S3 was found to be a quite rare example of quadruple-well ferroelectrics. Its polarization is always oriented along the c axis in the MEP with four local energy minima. The four polar states (\(\:\text{±}\)HP and \(\:\text{±}\)LP) can be distinguished by PFM if they have distinct piezoelectric constants. Here, we calculated the piezoelectric constants for the +HP and +LP states. Table 1 summarizes the piezoelectric stress constants (e33), elastic stiffness coefficients (C33), and piezoelectric constants (d33) for the + HP and + LP states. Their piezoelectric constants are comparable to that of pristine AlN (~5 pC/N) 30,31. The piezoelectric constants of the LP and HP states are distinctly different so that these two states are distinguished for the c-plane cleaved single crystal samples by PFM, as in CuInP2S6 25. The four polar states can be detected by PFM since the positively and negatively polarized states show piezoresponse with opposite signs.
Table 1
Piezoelectric stress constants (e33), elastic stiffness coefficients (C33), and piezoelectric constants (d33) for the + HP and + LP states of α-Al2S3.
State | e33 (C/m2) | C33 (GPa) | d33 (pC/N) |
+HP | 0.654 | 90 | 7.3 |
+LP | 0.246 | 65 | 3.8 |