Chalcogenide phase-change materials (PCMs) are important for nonvolatile memory and reconfigurable photonic technologies. The GeTe – Sb2Te3 mixture system, commonly referred to as GST, is the most well-known PCM family, but new PCMs are needed to broaden the accessible property space while retaining fast switching. We proposed a thermodynamic framework, motivated by Ostwald’s rule, for understanding and identifying PCM materials, since direct modeling of phase-transition dynamics is computationally expensive. Using first-principles calculations, we systematically evaluate the energetics of other ternary chalcogenide mixtures along binary-binary tie lines and their polymorphs. By comparing ground-state and metastable structures, we assess phase stability, miscibility, and the likelihood of GST-like polymorph-mediated crystallization pathways across a broad composition space. This dataset comprises the density functional theory calculation results for ternary chalcogenide structures generated by doping known binary chalcogenide compounds.

To create the ternary crystal structures used for the density functional theory calculations, we interpolated between pairs of binary compositions. We automatically determined the appropriate vacancy defect reaction for charge balancing based on the oxidation states of the relevant species. If the dopant and substituted species have the same oxidation state, no charge compensation is required. If the oxidation state of the dopant species is greater in magnitude than that of the substituted species, then additional vacancies are added on the substituted species sublattice. Otherwise, vacancies are added to the other sublattice of the binary structure. If there are multiple possible substituted species, due to cation multi-valence, we ignore these structures as they would require a more complex ordering scheme. The resulting compositions of the ternary structures were limited to simple stoichiometric ratios with denominators less than 6 (e.g., Ge2Sb2Te5 and not Ge7Sb8Te19) and to compositions achievable with less than 100 atoms.

Given the ternary compositions and parent binary structures, we used EnumLib via pymatgen to generate the ternary structures. The cation and anion sublattices of the parent structures were updated with the fractional occupation of the three species in the ternary composition. Then EnumLib was used to enumerate all the symmetrically distinct atomic orderings with the minimum supercell size. We determined the possible compositions and their required minimum supercell size using the procedure outlined in pseudocode below. We only considered compositions with simple stoichiometric ratios (i.e., with a denominator of less than 6). After ordering, we discarded composition-structure combinations with more than 1,000 ordered structures because they were not computationally practical for density functional theory calculations.

structures = {}
compositions = {}
sizes = {}
for all supercell sizes such that the number of atoms is < 100:
    for all defect reaction multiples that fit in the supercell:
        determine the number of each species
        calculate the corresponding dopant / substituted fraction
        if the fraction is simple and not in compositions:
            assign the fractional occupancy to the parent structure
            add the resulting unordered structure to structures
            add the composition to compositions
            add the supercell size to sizes

We performed the density functional theory calculations using a combination of Perdew – Burke – Ernzerhof (PBE) and r2SCAN calculations. The ordered ternary structures generated by EnumLib were first relaxed using the less expensive PBE functional, then the more expensive r2SCAN functional was used to calculate a more accurate energy of the relaxed structure. This dataset contains just the r2SCAN static calculation results.

The density functional theory calculations were performed with the Vienna Ab initio Simulation Package (VASP). All calculations were performed using automatically generated Materials Project-compatible VASP inputs created by pymatgen. The ternary structure energy values reported in this work were calculated with respect to the entries in the Materials Project, also calculated using pymatgen.