Data-Driven Discovery of Novel Chalcogenides for Photovoltaics

The efficiency of CIGS and CZTS solar cells is significantly limited by defects arising from cation disorder which could potentially act as nonradiative recombination centers for carriers, reducing the overall efficiency of the solar cells [1-2]. Composition engineering offers a promising approach to discover new semiconductors which in addition to showing attractive optoelectronic properties, are also defect-tolerant [3]. In this study, we explored the chemical space of quaternary A₂BCX₄ and ternary ABX₂ chalcogenide semiconductors where X = [S, Se, or Te], focusing on their thermodynamic stability, optoelectronic properties, and defect behavior, by implementing a high-throughput density functional theory (DFT) workflow. We define the A₂BCX₄ chemical space as A = [Na, K, Rb, Cs, Ag, Cu], B = [Mg, Ca, Sr, Ba, Zn, Cd], and C = [Ge, Sn], and the ABX₂ chemical space as A = [Na, K, Rb, Cs, Cu, Ag] and B = [Al, Ga, In], with each compound simulated in both the Stannite and Kesterite phases.<br/>For a total of 540 compounds, we performed geometry optimization using the GGA-PBEsol [6] functional, and the relaxed geometries were used as input to a hybrid HSE06 calculation with spin-orbit coupling (SOC) to accurately calculate the decomposition energy, defined as the tendency to decompose into competing phases A-X, B-X, or C-X phases. 358 compounds were found to be stable to decomposition, and we performed additional HSE06+SOC calculations to determine their electronic band gaps and spectroscopic limited maximum efficiency (SLME) [5] based on the computed optical absorption spectra. This eventually led us to a list of 16 stable compounds that show an SLME &gt; 32%, suggesting that they could be highly effective absorbers for single-junction solar cells. Finally, we performed detailed point defect calculations on the top candidates to confirm their tolerance to the spontaneous formation of potentially harmful vacancies, self-interstitials, or anti-sites, as well as the nature of equilibrium conductivity and dopability in a given semiconductor [4].<br/>Finally, we expanded our exploration by training random forest regression (RFR) models for each property using descriptors that one-hot encode composition and phase, and additionally include well-known elemental properties of the A/B/C/X species [7]. Using rigorously optimized predictive models, we made property predictions over &gt; 200,000 hypothetical compositions with cation- or anion-site mixing allowed and established an active learning loop to perform new calculations for iteratively improving the models. This exercise led us to hundreds of new alloys that show low decomposition energy, band gap between 1 and 2 eV, and SLME &gt; 30%, at the HSE+SOC level of theory. Defect calculations are performed on the best candidates to evaluate defect tolerance and dopability. Overall, the integration of high-throughput DFT calculations, RFR models, and defect calculations offers significant insights into the design and identification of new, defect-tolerant materials for solar applications.<br/><b>REFERENCES</b><br/> <br/>[1] W. Chen, D et al., <i>Energy & Environmental Science</i>, vol. 14, no. 6, pp. 3567–3578, 2021<br/>[2] S. Kim et al., <i>Journal of Materials Chemistry. A</i>, vol. 7, no. 6, pp. 2686–2693, Jan. 2019<br/>[3] Mannodi-Kanakkithodi, Computational materials science, vol. 243, pp. 113108–113108, Jul. 2024<br/>[4] D. Broberg et al., npj Computational Materials, vol. 9, no. 1, May 2023<br/>[5] M. Bercx et al., Physical Chemistry Chemical Physics/PCCP. Physical Chemistry Chemical Physics, vol. 18, no. 30, pp. 20542–20549, Jan. 2016<br/>[6] G. I. Csonka et al., Physical Review. B, Condensed Matter and Materials Physics, vol. 79, no. 15, Apr. 2009<br/>[7] J. Yang et al., <i>Journal of Chemical Physics</i>, vol. 160, no. 6, Feb. 2024