Thermoelectric CoGeTe with an Orthorhombic Crystal Symmetry and Balance of the Electrical and Thermal Properties

Abstract

Applying crystal symmetry to discover and optimize the performance of thermoelectric (TE) materials has attracted much attention. Here, we report CoGeTe with a middle-class crystalline system as a novel n-type TE material. Density functional theory indicates that orthorhombic CoGeTe shows multiband dispersion near the bottom of the conduction band, which is mainly occupied by the Co 3d states. Through Ni doping, these multiple bands can be activated, leading to a maximum power factor of 1.14 mW/m K 2 @786 K for Co 0.95 Ni 0.05 GeTe. In addition, phonon-dispersion calculations reveal that CoGeTe possesses relatively strong harmonic properties, including sound velocity and Debye temperature. Furthermore, the local distorted CoGe 3 Te 3 octahedron in the matrix is beneficial for anharmonic phonon scattering. In particular, the Gruneisen parameter of Te in the crystal structure is clearly larger than those of Co and Ge. The observed thermal conductivity of Co 0.95 Ni 0.05 GeTe is between 6.50 and 5.38 W/m K in the temperature range 300−860 K. Owing to the combination of the enhanced power factor and reduced thermal conductivity, the maximum zT value reaches 0.18 at 860 K. This study suggests that TE materials with orthorhombic structures provide an ideal platform to balance the power factor and thermal conductivity in search of high-performance thermoelectrics.

Artificial Intelligence for Learning Material Synthesis Processes of Thermoelectric Materials

“…Figure presents two cases of the generated synthesis recipes with the smallest or the largest Levenshtein distances to the ground truth synthesis recipes; i.e., the figure shows the best and worst results of the AI-based generation of the synthesis recipes. As shown in the most similar case of Figure a, SPENDE accurately predicted the synthesis operations and engineering conditions of the synthesis recipe of CoGeTe, and the GPT model generated a paragraph similar to the ground truth synthesis recipe extracted from the scientific literature . For the most different case of Figure b, although SPENDE made a wrong prediction of the synthesis sequence, the GPT model generated a synthesis recipe that briefly describes the synthesis process of Gd 2 Se 2.92 from the scientific literature …”

“…Ground truth and generated synthesis recipes of the thermoelectric materials on the TMSR dataset. (a) A case where the ground truth synthesis recipe from the scientific literature and the synthesis recipe generated by GPT-SPENDE are most similar. (b) A case where the ground truth synthesis recipe from the scientific literature and the synthesis recipe generated by GPT-SPENDE are most different.…”

“…As shown in the most similar case of Figure 6a, SPENDE accurately predicted the synthesis operations and engineering conditions of the synthesis recipe of CoGeTe, and the GPT model generated a paragraph similar to the ground truth synthesis recipe extracted from the scientific literature. 41 For the most different case of Figure 6b, although SPENDE made a wrong prediction of the synthesis sequence, the GPT model generated a synthesis recipe that briefly describes the synthesis process of Gd 2 Se 2.92 from the scientific literature. 42 Thermoelectric Performance Prediction with Predicted Material Synthesis Processes.…”