# Fileset

[Computational material screening for electrode materials of BaSi2 solar cells.pdf](https://mdr.nims.go.jp/filesets/8a316d81-4c14-4387-873f-e369b2ff6e31/download)

## Creator

Tomoaki Yazaki, Keisuke Arimoto, Junji Yamanaka, Kosuke O. Hara

## Rights

[Creative Commons BY Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/)

## Other metadata

[Computational material screening for electrode materials of BaSi2 solar cells](https://mdr.nims.go.jp/datasets/6e716375-d5f0-477b-b68e-c6c453c3b858)

## Fulltext

Microsoft Word - AMO_TSTM_A_2617671.docxScience and Technology of Advanced Materials: MethodsISSN: 2766-0400 (Online) Journal homepage: www.tandfonline.com/journals/tstm20Computational material screening for electrodematerials of BaSi solar cellsTomoaki Yazaki, Keisuke Arimoto, Junji Yamanaka & Kosuke O. HaraTo cite this article: Tomoaki Yazaki, Keisuke Arimoto, Junji Yamanaka & Kosuke O. Hara (28 Jan2026): Computational material screening for electrode materials of BaSi solar cells, Science andTechnology of Advanced Materials: Methods, DOI: 10.1080/27660400.2026.2617671To link to this article:  https://doi.org/10.1080/27660400.2026.2617671© 2026 The Author(s). Published by NationalInstitute for Materials Science in partnershipwith Taylor & Francis GroupView supplementary material Accepted author version posted online: 28Jan 2026.Submit your article to this journal Article views: 65View related articles View Crossmark dataFull Terms & Conditions of access and use can be found athttps://www.tandfonline.com/action/journalInformation?journalCode=tstm20https://www.tandfonline.com/journals/tstm20?src=pdfhttps://www.tandfonline.com/action/showCitFormats?doi=10.1080/27660400.2026.2617671https://doi.org/10.1080/27660400.2026.2617671https://www.tandfonline.com/doi/suppl/10.1080/27660400.2026.2617671https://www.tandfonline.com/doi/suppl/10.1080/27660400.2026.2617671https://www.tandfonline.com/action/authorSubmission?journalCode=tstm20&show=instructions&src=pdfhttps://www.tandfonline.com/action/authorSubmission?journalCode=tstm20&show=instructions&src=pdfhttps://www.tandfonline.com/doi/mlt/10.1080/27660400.2026.2617671?src=pdfhttps://www.tandfonline.com/doi/mlt/10.1080/27660400.2026.2617671?src=pdfhttp://crossmark.crossref.org/dialog/?doi=10.1080/27660400.2026.2617671&domain=pdf&date_stamp=28%20Jan%202026http://crossmark.crossref.org/dialog/?doi=10.1080/27660400.2026.2617671&domain=pdf&date_stamp=28%20Jan%202026https://www.tandfonline.com/action/journalInformation?journalCode=tstm20ACCEPTED MANUSCRIPT1     We present a high-throughput screening workflow integrating device simulations and property computations, enabling rapid identification of promising electrode materials for BaSi₂ and broader photovoltaic technologies.  Computational material screening for electrode materials of BaSi 2  solar cells  Tomoaki Yazakia, Keisuke Arimotoa, Junji Yamanakaa, and Kosuke O. Harab  aUniversity of Yamanashi, 7-32 Miyamae, Kofu, Yamanashi, Japan; bNara Institute of Science and Technology, 8916-5 Takayama, Ikoma, 630-0192, Nara, Japan    ARTICLE HISTORY Compiled January 8, 2026  ABSTRACT In this study, we developed a computational material screening workflow for metallic electrodes of BaSi 2  solar cells. Elemental and binary metallic materials in the Materials Project database were screened for four device models with different charge transport layer materials contacting the electrode. The screening criteria included chemical stability, melting point, and work function. In contrast to conventional screening approaches that rely solely on materials descriptors, the present workflow explicitly incorporates device-level performance constraints by using the relationship between the work function and the simulated power conversion efficiency. For melting point evaluation, a linear regression estimation from the cohesive energy and a machine learning model were compared to assess their accuracy, which revealed the higher accuracy of the machine learning model. For work function evaluation, first-principles calculation and another machine learning model were compared, which showed similar accuracies. Considering the computational cost, the machine learning model was used for screening. The threshold of work function screening was determined by device simulations. As a result, promising materials for metallic electrodes were successfully identified. Moreover, the developed screening workflow with high versatility will be applicable to material discovery for other solar cells and semiconductor devices.   KEYWORDS Solar cells; Metallic electrode; High-throughput virtual screening; Work function; Melting point; Density functional theory         CONTACT T. Yazaki Author. Email: tyazaki617@g.chuo-u.ac.jp CONTACT K. O. Hara Author. Email: hara.kosuke@naist.ac.jp https://crossmark.crossref.org/dialog/?doi=10.1080/27660400.2026.2617671&domain=pdfACCEPTED MANUSCRIPT2     1.  Introduction The selection of appropriate electrode materials is crucial for improving the performance of electronic and energy conversion devices, such as solar cells, thermoelectric devices, and MOSFETs [1–7]. In particular, the metallic electrode of a solar cell significantly affects carrier transport efficiency, thereby influencing the overall power conversion efficiency (PCE). Additionally, it is essential to have a high melting point ( mT ) to withstand high temperature growth and annealing of upper layers and to prevent unintended chemical reactions at the interface. It is challenging to identify optimal electrode materials by comprehensively considering band alignment, mT , and chemical reactivity. The conventional selection of metallic materials has been based on empirical rules and trial-and-error methods, making it challenging to comprehensively evaluate multidimensional material properties. Additionally, the material search space was limited, restricting the number of materials considered. Computational material screening is a powerful approach that screens a huge set of data in databases based on computed properties [8]. This approach can accelerate the discovery of ideal materials in a rational way. The focus of this study is the metallic electrode of BaSi 2  solar cells. BaSi 2  has a band gap ( gE ) of 1.3 eV, high optical absorption coefficients ( 4 110 cm−≥  for photon energies over gE ) [9–11], and long minority carrier lifetime (up to 27 μ s) [12–14], making it a promising light-absorbing material for solar cells. However, the maximum PCE of BaSi 2  solar cells reported to date remains around 10% [15], which is partly attributed to the lack of an optimized device structure. In particular, BaSi 2  has a relatively low electron affinity of 3.2 eV [16] compared to other solar cell absorbers, making it difficult to use the device architectures of other solar cells. In previous studies, we adopted a double heterojunction structure and searched for materials for the electron transport layer (ETL) and hole transport layer (HTL) by computational material screening, successfully identifying optimal materials [17]. The most promising materials among them include Sm 2 O 3 , 6H-SiC, and 4H-SiC for ETL and BaS for HTL because of the best band alignment, compositional simplicity, and synthesizability. Our screening workflow is unique in that it relies not only on material descriptors but also on device-level performance constraints. This allows us to estimate the expected PCE with the selected materials. The estimated PCE with the identified materials exceeded 30%. The thin film deposition processes for these layers are currently being developed [18, 19]. However, suitable materials for the transparent conductive electrode and back metal electrode have not been explored. In particular, the selection of an appropriate back electrode material poses a challenge due to the large number of candidate materials compared to transparent conductive materials. In this study, we develop a device-simulation-coupled computational material screening procedure for metallic electrodes to identify the optimal electrode materials for double heterojunction BaSi 2  solar cells. Particular focus is on mT  and work function ( mφ ). mT  screening is desirable for BaSi 2  solar cells because the optical properties of BaSi 2  are improved by annealing at 850–1000  C [20], requiring materials with high mT . Two methods are investigated for rapidly and accurately estimating mT  and mφ . Based on the result, an effective procedure of computational material screening is proposed for electrode materials of solar cells. Because the ETL and HTL materials contacting the electrode affect the screening results, we studied four different device structures with Sm 2 O 3 , 6H-SiC, and 4H-SiC as ETL materials and BaS as the HTL material in this study. Promising candidate materials are identified for different device structures.  ACCEPTED MANUSCRIPT3     2.  Methods   2.1.  Device simulations  Device simulations were conducted using the console version of wxAMPS [21, 22] under AM 1.5 illumination with an intensity of 1000 W/ 2m  to investigate the required conditions for the mφ  of the electrode materials. Four device models shown in Fig. 1 were investigated. All four models consist of transparent conductive film (TCF, 100 nm), HTL (100 nm), BaSi 2  (2 μ m [17]), ETL (100 nm), and metallic electrode. Three of the models have ETL at the back side, contacting with metallic electrodes, while the other one has HTL at the back side. The ETL materials selected were Sm 2 O 3 , 4H-SiC, and 6H-SiC, while BaS was chosen for HTL. The material properties were basically referenced from literatures, as summarized in Table 1. Absorption coefficients of BaSi 2  were taken from Ref. [23] with neglecting the values for photon energies below gE . However, the absorption coefficients for ETL, HTL, and TCF were assumed for simplicity to be step functions, which were 510  cm 1−  for photon energies over gE , and zero otherwise. Electron densities in BaSi 2  and ETL were assumed to be 161 10×  cm 3−  [24–26] and 141 10×  cm 3− , respectively, while hole density in HTL was assumed to be 141 10×  cm 3− . The conductivity type of TCF was assumed to be either n- or p-type when contacting with ETL or HTL, respectively, with carrier density of 211 10×  cm 3− . For BaSi 2 , the effective density of states in the conduction band ( CN ), the effective density of states in the valence band ( VN ), and the radiative recombination coefficient ( B ) are 18C = 6.76 10N ×  cm 3− , 19V = 1.07 10N ×  cm 3− , and 11= 1.09 10B −×  cm 3  s 1−  [17]. These values were also used for ETL, HTL, and TCF. All carrier mobilities were assumed to be 100  cm 2 V 1− s 1− . The dielectric constant of TCF was assumed to be the same as BaSi 2 . The gE  and electron affinity of TCF were set to the optimal values that maximize PCE for given models. The front- and back-side reflectances were set to 0 and 1, respectively. The effective surface recombination velocities for both the front and back sides were assumed to be 710  cm/s. Bulk defects and band tailing were not considered, and the intraband tunneling mode was employed as the simulation mode. Metal electrodes were not explicitly included in the models. Instead, effects of different mφ  were considered by different potential barrier heights (PBH) at the back surfaces. For the ETL/metal electrode interface, PBH corresponds to the difference between the mφ  of the electrode and the electron affinity of ETL, while for the HTL/electrode interface, PBH corresponds to the difference between the mφ  of the electrode and the ionization potential of HTL.  2.2.  Computational material screening The Materials Project database [27] was screened using the mp-api and pymatgen [28] Python library (access in June 2023). The search space was limited to single and binary compounds free from rare (less abundant than Ag in the Earth’s crust), highly toxic (Be, Cd, As, Hg, Pb, Tl), and radioactive elements. Only metallic materials were included by setting g = 0E  eV. Additionally, to ensure the stability of the materials, we applied the condition that the energy above the convex hull = 0  eV. First, we evaluated interfacial chemical reactivity. The interfacial chemical reactivity was evaluated using the InterfacialReactivity class [30] of pymatgen. We excluded the materials ACCEPTED MANUSCRIPT4     that can thermodynamically react with the adjacent layer. mT  was evaluated by two methods to find out a rapid and accurate method that is suitable for high-throughput screening. One is based on the relationship between the mT  and cohesive energies ( cohE ). The cohE  of compounds was calculated by Miedema’s theory using Miedema.py [31, 32]. Linear regression was analyzed between the cohE  and the “temperature for congruent melting” data from the Materials Platform for Data Science (MPDS) for randomly selected 93 binary compounds. The intercept was set to zero. This regression line allowed us to estimate the mT . The other method uses a machine learning model developed by Hong  et al. [33]. This machine learning model was trained on approximately 10,000 compound data [33], and predicts a mT  using a graph neural network and a residual neural network. The accuracies of these methods were evaluated by the prediction errors for different randomly selected data from MPDS. Appropriate estimation methods for mφ  were also investigated. We compared the density functional theory (DFT) and machine learning methods using the experimental work function data [34] for elemental metals as reference. We used Quantum Espresso [35, 36] for DFT calculations. mφ  was estimated as the difference between the Fermi and vacuum levels using slab models of thicker than 20 Å with 20 Å vacuum regions. Slab models were generated using the pymatgen library [37, 28]. If slabs with several different surface terminations were prepared, their mφ  were averaged. All calculations were performed with the Perdew–Burke–Ernzerhof functional tuned for solids (PBEsol) [38] in the generalized gradient approximation (GGA). mφ  were estimated after structural relaxation. Gaussian smearing with the spreading of 68 meV was used for occupations. We followed our previous papers [26, 17] for other parameter settings. The machine learning model for mφ  investigated was the random forest model developed by Schindler  et al. [39]. This model was trained on a dataset comprising 58,332 mφ  and 33,631 exfoliation energies obtained from high-throughput DFT calculations. In this study, we averaged mφ  over different surface terminations to obtain a representative mφ  for a given crystal plane. This approach was adopted because identifying the most probable termination for each material is impractical, even though actual surface terminations depend on crystal growth conditions. mφ  of 33 elemental metals were evaluated by both methods, and the accuracy was evaluated using the experimentally reported mφ  [34] as reference values.  3.  Results and discussion  3.1.  Device simulations Figures 2(a) and 2(b) show the band alignments of the model (C) in Fig. 1 under dark conditions with PBH = 0.0 and 0.5 eV, respectively. In the figures, the metallic electrode is assumed to be present on the right side of the Sm 2 O 3  ETL. Obviously, this band alignment with PBH = 0.0 eV allows efficient charge separation with small potential barriers for electrons toward the back electrode through Sm 2 O 3  ETL (0.26 eV) and for holes toward the front TCF through BaS HTL (0.15 eV). On the other hand, with PBH = 0.5 eV, the band energy levels of ETL and BaSi 2  near ETL rise toward the back electrode. This band bending obstructs electron transport and reduces PCE. To identify the mφ  range where smooth charge transport is assured, PCE was simulated with different PBH values. Figures 3(a)–(d) show the PCE as functions of mφ  of metallic electrodes for the models (A)–(D) in Fig. 1, respectively. The horizontal dashed line in each panel of Fig. 3 indicates a ACCEPTED MANUSCRIPT5     PCE of 30%. With the models (A)–(C), PCE decreases with increasing mφ . This is because the band energy levels of the ETL and BaSi 2  near ETL increase, as shown in the insets. The thresholds of the mφ  where PCE starts to decrease are 3.90, 3.61, and 3.32 eV for the models (A)–(C), respectively. These thresholds corresponding to PBH of 0.3 eV, which is presumably the highest barrier that electrons can overcome. On the other hand, with the model (D), PCE decreases with decreasing mφ , because the band energy levels are lowered and the barriers for holes are generated. The threshold where PCE starts to decrease is 4.42 eV, which again corresponds to the PBH of 0.3 eV for holes. This 0.3 eV PBH threshold is a useful criterion that could be applicable to various types of solar cells. For the purpose of electrode materials selection of this study, the mφ  thresholds serve as criteria for computational material screening.  3.2.  Computational material screening The workflow of computational material screening for metallic electrode materials is shown in Fig. 4. The search space was defined by g = 0E  eV, the energy above the convex hull = 0  eV, and elemental and binary substances without rare, toxic, and radioactive elements described in Methods. At this stage, 1,830 metallic materials were included in the candidate list. The interface reactivity screening decreased the number of candidates to 567, 567, 878, and 1483 for models (A)–(D), respectively. To further screen the materials by mT , we compared two estimation methods to select an appropriate method: linear regression using Miedema’s cohE  and the machine learning model. The physical principle of melting indicates that a high cohE  leads to a high mT , which is the basis of estimating mT  from cohE . Actually, mT  is reported to be roughly proportional to cohE  in Laves phases [31]. We determined the regression line between cohE  and mT  using the data for random binary metallic compounds in MPDS. cohE  was calculated using Miedema’s theory [40, 31]. Figure 5 shows the mT  plotted against cohE  together with the regression line. A positive correlation is confirmed between mT  and cohE . The coefficient of determination was 0.85 according to Kvålseth’s 82R  definition [41]. The obtained regression equation is 1m coh( / K) = 5.27 ( / kJmol ).T E −×  Figure 6 compares the mT  estimated using the regression line derived from cohE  and those predicted by the machine learning model developed by Hong  et al [33], with the reference values obtained from MPDS. Different material data were used from those used for the linear regression. The accuracy of the mT  estimation was evaluated using the mean absolute error (MAE). The MAE of the linear regression was 616.7 K, while that of the machine learning model was 69.6 K, showing the higher accuracy of the machine learning model than the linear regression method. The estimation based on cohE  is reported effective for materials with specific structures, such as Laves phases [31]. The current evaluation indicates that the simple linear regression is not as effective as machine learning for materials with various crystal structures. Therefore, we employed the machine learning model [33] for mT  estimation. We screened the candidate materials with the mT  criterion of 1000  C because annealing at 1000  C is reported to enhance the photoresponsivity of BaSi 2  [20], and hence, the metallic electrode is possibly exposed to high temperatures up to 1000  C during solar cell fabrication. After the mT  screening step, 388, 388, 624, and 1136 candidates were retained for models (A)–(D) in Fig. 1, respectively. ACCEPTED MANUSCRIPT6     The last screening step involved mφ , which is the essential parameter influencing the PCE, as shown in Fig. 3. DFT is the most reliable method to computationally estimate the mφ  of materials, while Schindler  et al. recently reported a machine learning model using random forest for mφ  prediction [39]. We investigated the prediction accuracy of these two methods by estimating the mφ  of 33 elemental metals using the reported experimental values [34] as reference. Figure 7 shows the mφ  of elemental metals estimated by DFT and the machine learning model. The (100) surface was selected for all metals. The average mφ  of different surface terminations is shown. The MAEs between the estimated and reference values was 0.30 eV for DFT and 0.31 eV for the machine learning model. This result indicates that both methods estimate the mφ  with similar accuracy. This is possibly because the machine learning model was trained on the DFT calculation data, leading to a prediction error comparable to that of DFT. The data used to train the machine learning model were prepared with ionically unrelaxed slabs, while our DFT calculations involved ionic relaxations of slab models. Regardless of this difference, the machine learning model worked well. Additionally, computational costs are greatly lower with the machine learning model than DFT. Therefore, screening was conducted using the machine learning model [39]. To represent the work function of a polycrystalline surface, work functions calculated by the machine learning model for inequivalent surface planes were averaged. Miller indices up to 1 were considered. In principle, a weighted average that accounts for the exponential dependence of thermionic emission on the work function should be used, with the fractional areas as weights [42]. However, these fractional areas cannot be reliably estimated. Therefore, we adopted a simple arithmetic average over inequivalent crystal planes. To assess the validity of this approach, we evaluated the work functions of 43 polycrystalline elements, for which experimental polycrystalline values are available [34]. The resulting MAE was 0.26 eV, comparable to that obtained for the (100) surface. This result indicates that averaging over inequivalent planes provides sufficient accuracy for screening purposes. The mφ  screening was performed with the criterion that the PCE should exceed 30%, which was derived from the simulation results shown in Fig. 3. This step resulted in 197, 89, 41, and 233 candidates for the models (A)–(D) in Fig.1, respectively. These final candidates and their properties are summarized in Table 2–5. When the ETL was placed on the back side [models (A)–(C)], the number of candidate materials was smaller than otherwise [model (D)]. This is due to the imposed conditions of high mT  and low mφ . It is generally known that transition metals with high mT  tend to have high mφ  [43–45]. In contrast, rare-earth elements tend to exhibit both high mT  and low mφ  [46, 34, 47], whereas alkaline earth metals, despite having low mT , also tend to have low mφ  [48, 34]. As a result, many of the final candidates for models (A)–(C) contain one of rare earth and alkaline earth elements. Among these models, in particular model (A), the ETL in contact with the electrode has a relatively large electron affinity, so the allowable upper limit (threshold) of the electrode work function is higher than in the other models. As a result, more materials satisfy this threshold, and the number of candidates that passes the screening criteria is correspondingly larger. Then, to further screen materials, the candidates were arranged in the order of electrical conductivity (accurately, electrical conductivity divided by the relaxation time), as shown in Fig. 8. The electrical conductivity was referenced from Ref. [49], where a large dataset of transport data were obtained by DFT and BoltzTraP calculations assuming a constant relaxation time. Materials without available data were not included in the figure. This figure shows that ScGa 3 , CaAl 2 , GdS, and PrSi 2  are especially promising among the final candidates for model (A). ACCEPTED MANUSCRIPT7     Although they are promising from the viewpoints of physical and chemical properties, expensive and air-sensitive characters of these materials possibly make their usage practically difficult. On the other hand, in model Fig. 1 (D), a relatively large number of candidate materials were obtained. This is because the screening conditions required high mT  and high mφ , leading to the selection of many transition metal compounds. The large number of candidate materials including transition metal compounds suggests that model (D) is promising for practical implementation. Figure 9 shows the electrical conductivity of the candidates. The high rank materials include elemental metals: Cu, W, Fe, and Ni. They are accessible stable materials and therefore promising for practical implementation. Thus, the computational screening workflow proposed in this study is useful for finding out promising electrode materials as well as for selecting versatile device models. As a final supplementary investigation, we also applied the proposed screening workflow to conventional crystalline Si solar cells. The screened candidates ranked practical materials, such as Al and Ag, among the top candidates, demonstrating the effectiveness of the workflow. Detailed procedures and results are provided in the Supplemental Material.  4.  Conclusions We developed a computational material screening workflow for electrode materials of BaSi 2  solar cells. In particular, the accuracy of mT  and mφ  estimations was evaluated by comparing conventional methods with machine learning predictions, demonstrating that efficient screening can be performed for a wide variety of materials with machine learning models. Device simulations revealed the thresholds of mφ  for four different device models, which were used as one of the screening criteria. Based on the large number of final candidate materials that satisfied the screening conditions, two device configurations were identified as particularly promising for practila implementations: model (A) with 6H-SiC ETL and model (D) with BaS HTL on a metallic electrode. Among the screened candidates, ScGa 3 , CaAl 2 , GdS, and PrSi 2  were identified as promising electrode materials for model (A) due to their high electrical conductivity, while Cu, W, Fe, and Ni were selected for model (D). Notably, some of these electrode materials have not yet been reported for BaSi 2  solar cells. The screening workflow developed in this study is highly versatile and is applicable not only to BaSi 2  solar cells but also to other solar cells and semiconductor devices.  Acknowledgment This work was partly supported by the Murata Science Foundation.  Disclosure statement The authors report there are no competing interests to declare..  Data availability The data will be available on request from the corresponding authors.  References    [1]  Zhang L, Liu F. High-throughput approach to explore cold metals for electronic and thermoelectric devices. npj Comput Mater. 2024 Apr;10(1). doi:10.1038/s41524-024-01267-w.  ACCEPTED MANUSCRIPT8     [2]  Choi HC, Yun H, Yoon JS, et al. Neural approach for modeling and optimizing si-mosfet manufacturing. IEEE Access. 2020;8:159351–159370. doi:10.1109/ACCESS.2020.3019933.  [3]  He R, Schierning G, Nielsch K. Thermoelectric devices: A review of devices, architectures, and contact optimization. Adv Mater Technol. 2018;3(4):1700256. doi:10.1002/admt.201700256.  [4]  Ferreira-Teixeira S, Pereira AM. Geometrical optimization of a thermoelectric device: Numerical simulations. Energ Convers Manage. 2018;169:217–227. doi:10.1016/j.enconman.2018.05.030.  [5]  Omer S, Infield D. Design optimization of thermoelectric devices for solar power generation. Sol Energy Mater Sol Cells. 1998;53(1):67–82. doi:10.1016/S0927-0248(98)00008-7.  [6]  Brabec CJ, Sariciftci NS, Hummelen JC. Plastic solar cells. Adv Funct Mater. 2001;11(1):15–26. doi:10.1002/1616-3028(200102)11:1<15::AID-ADFM15>3.0.CO;2-A.  [7]  Deepthi Jayan K, Sebastian V. Comprehensive device modelling and performance analysis of MASnI3 based perovskite solar cells with diverse ETM, HTM and back metal contacts. Sol Energy. 2021;217:40–48. doi:10.1016/j.solener.2021.01.058.  [8]  Luo S, Li T, Wang X, et al. High-throughput computational materials screening and discovery of optoelectronic semiconductors. WIREs Comput Mol Sci. 2021;11(1):e1489. doi:10.1002/wcms.1489.  [9]  Toh K, Saito T, Suemasu T. Optical absorption properties of BaSi2 epitaxial films grown on a transparent silicon-on-insulator substrate using molecular beam epitaxy. Jpn J Appl Phys. 2011;50(6R):068001. doi:10.1143/JJAP.50.068001.  [10]  Kumar M, Umezawa N, Imai M. BaSi2 as a promising low-cost, earth-abundant material with large optical activity for thin-film solar cells: A hybrid density functional study. Appl Phys Express. 2014;7(7):071203. doi:10.7567/APEX.7.071203.  [11]  Trinh CT, Nakagawa Y, Hara KO, et al. Photoresponse properties of BaSi2 film grown on Si (100) by vacuum evaporation. Mater Res Express. 2016;3(7):076204. doi:10.1088/2053-1591/3/7/076204.  [12]  Hara KO, Usami N, Nakamura K, et al. Determination of bulk minority-carrier lifetime in BaSi2 earth-abundant absorber films by utilizing a drastic enhancement of carrier lifetime by post-growth annealing. Appl Phys Express. 2013;6(11):112302. doi:10.7567/APEX.6.112302.  [13]  Xu Z, Shohonov DA, Filonov AB, et al. Marked enhancement of the photoresponsivity and minority-carrier lifetime of BaSi2 passivated with atomic hydrogen. Phys Rev Mater. 2019;3(6):065403. doi:10.1103/PhysRevMaterials.3.065403.  [14]  Shaalan NM, Hara KO, Trinh CT, et al. Simple method for significant improvement of minority-carrier lifetime of evaporated BaSi2 thin film by sputtered-AlOx passivation. Mater Sci Semicond Process. 2018;76:37–41. doi:10.1016/j.mssp.2017.12.015. ACCEPTED MANUSCRIPT9      [15]  Fujiwara M, Takahashi K, Nakagawa Y, et al. Improved conversion efficiency of p-type BaSi2/n-type crystalline Si heterojunction solar cells by a low growth rate deposition of BaSi2. AIP Adv. 2022 04;12(4):045115. doi:10.1063/5.0083812.  [16]  Suemasu T, Morita K, Kobayashi M, et al. Band diagrams of BaSi2/Si structure by Kelvin probe and current–voltage characteristics. Jpn J Appl Phys. 2006;45(6L):L519. doi:10.1143/JJAP.45.L519.  [17]  Hara KO. Designing limiting-efficiency BaSi2 solar cells by device simulation and computational material screening. Sol Energy. 2022;245:136–145. doi:10.1016/j.solener.2022.08.044.  [18]  Hara KO, Nakanishi T, Ueda R, et al. Inert interlayers and evaporation techniques for high-quality BaSi2 heterostructures. Jpn J Appl Phys. 2025;64(2):02SP31. doi:10.35848/1347-4065/adb167.  [19]  Yazaki T, Kunieda S, Arimoto K, et al. Development of flat monoclinic Sm2O3 films by sputtering and post-deposition annealing for solar cell applications. Mater Sci Semicond Process. 2025;194:109514. doi:10.1016/j.mssp.2025.109514.  [20]  Haku Y, Aonuki S, Yamashita Y, et al. Effect of post-annealing on the significant photoresponsivity enhancement of BaSi2 epitaxial films on Si(111). Appl Phys Express. 2021 jan;14(2):021003. doi:10.35848/1882-0786/abdc9f.  [21]  Liu Y, Sun Y, Rockett A. An improved algorithm for solving equations for intra-band tunneling current in heterojunction solar cells. Thin Solid Films. 2012;520(15):4947–4950. doi:10.1016/j.tsf.2012.03.012.  [22]  Liu Y, Sun Y, Rockett A. Batch simulation of solar cells by using Matlab and wxAMPS. In: 2012 38th IEEE Photovoltaic Specialists Conference; IEEE; 2012. p. 000902–000905.  [23]  Vismara R, Isabella O, Zeman M. Organometallic halide perovskite/barium di-silicide thin-film double-junction solar cells. In: Photonics for Solar Energy Systems VI; Vol. 9898; SPIE; 2016. p. 54–62.  [24]  Du W, Baba M, Toko K, et al. Analysis of the electrical properties of Cr/n-BaSi2 Schottky junction and n-BaSi2/p-Si heterojunction diodes for solar cell applications. J Appl Phys. 2014;115(22). doi:10.1063/1.4882117.  [25]  Takabe R, Deng T, Kodama K, et al. Impact of Ba to Si deposition rate ratios during molecular beam epitaxy on carrier concentration and spectral response of BaSi2 epitaxial films. J Appl Phys. 2018;123(4). doi:10.1063/1.4994850.  [26]  Hara KO. Discovering low-electron-affinity semiconductors for junction partners of photovoltaic BaSi2 by database screening and first principles calculations. J Phys Chem C. 2021;125(44):24310–24317. doi:10.1021/acs.jpcc.1c06338.  [27]  Jain A, Ong SP, Hautier G, et al. Commentary: The Materials Project: A materials genome approach to accelerating materials innovation. APL Mater. 2013;1(1):011002. ACCEPTED MANUSCRIPT1     doi:10.1063/1.4812323.  [28]  Ong SP, Richards WD, Jain A, et al. Python Materials Genomics (pymatgen): A robust, open-source python library for materials analysis. Comput Mater Sci. 2013;68:314–319. doi:10.1016/j.commatsci.2012.10.028.  [29]  Zur A, McGill TC. Lattice match: An application to heteroepitaxy. J Appl Phys. 1984 01;55(2):378–386. doi:10.1063/1.333084.  [30]  Richards WD, Miara LJ, Wang Y, et al. Interface stability in solid-state batteries. Chem Mater. 2016;28(1):266–273. doi:10.1021/acs.chemmater.5b04082.  [31]  Li C, Hoe JL, Wu P. Empirical correlation between melting temperature and cohesive energy of binary laves phases. J Phys Chem Solids. 2003;64(2):201–212. doi:10.1016/S0022-3697(02)00267-6.  [32]  Kirklin S, Saal JE, Meredig B, et al. The Open Quantum Materials Database (OQMD): assessing the accuracy of DFT formation energies. npj Comput Mater. 2015;1(1):1–15. doi:10.1038/npjcompumats.2015.10.  [33]  Hong QJ, Ushakov SV, van de Walle A, et al. Melting temperature prediction using a graph neural network model: From ancient minerals to new materials. Proc Natl Acad Sci. 2022;119(36):e2209630119. doi:10.1073/pnas.2209630119.  [34]  Lide DR, editor. CRC handbook of chemistry and physics, internet version 2005. CRC press; 2005.  Available from:  http://www.hbcpnetbase.com .  [35]  Giannozzi P, Baroni S, Bonini N, et al. QUANTUM ESPRESSO: a modular and open-source software project for quantum simulations of materials. J Phys: Condens Matter. 2009;21(39):395502. doi:10.1088/0953-8984/21/39/395502.  [36]  Giannozzi P, Andreussi O, Brumme T, et al. Advanced capabilities for materials modelling with Quantum ESPRESSO. J Phys: Condens Matter. 2017;29(46):465901. doi:10.1088/1361-648X/aa8f79.  [37]  Sun W, Ceder G. Efficient creation and convergence of surface slabs. Surf Sci. 2013;617:53–59. doi:10.1016/j.susc.2013.05.016.  [38]  Perdew JP, Ruzsinszky A, Csonka GI, et al. Restoring the density-gradient expansion for exchange in solids and surfaces. Phys Rev Lett. 2008;100(13):136406. doi:10.1103/PhysRevLett.100.136406.  [39]  Schindler P, Antoniuk ER, Cheon G, et al. Discovery of stable surfaces with extreme work functions by high-throughput density functional theory and machine learning. Adv Funct Mater. 2024;34(19):2401764. doi:10.1002/adfm.202401764.  [40]  Miedema AR, De Chatel PF, De Boer FR. Cohesion in alloys—fundamentals of a semi-empirical model. Physica B. 1980;100(1):1–28. doi:10.1016/0378-4363(80)90054-6.  [41]  Kvålseth TO. Cautionary note about R2. Amer Statist. 1985;39(4):279–285. doi:10.1080/00031305.1985.10479448. ACCEPTED MANUSCRIPT1      [42]  Kawano H. Effective work functions for ionic and electronic emissions from mono- and polycrystalline surfaces. Prog Surf Sci. 2008;83:1–165. doi:10.1016/j.progsurf.2007.11.001.  [43]  Methfessel M, Hennig D, Scheffler M. Trends of the surface relaxations, surface energies, and work functions of the 4d transition metals. Phys Rev B. 1992;46(8):4816. doi:10.1103/PhysRevB.46.4816.  [44]  Richter S, Gutschka C, Danner D, et al. High-throughput phase exploration of ternary transition metal carbide TM-X-C (X= Al/Si) thin films. Acta Mater. 2025;288:120839. doi:10.1016/j.actamat.2025.120839.  [45]  Tran R, Li XG, Montoya JH, et al. Anisotropic work function of elemental crystals. Surf Sci. 2019;687:48–55. doi:10.1016/j.susc.2019.05.002.  [46]  Nikolić MV, Radić SM, Minić V, et al. The dependence of the work function of rare earth metals on their electron structure. Microelectron J. 1996;27(1):93–96. doi:10.1016/0026-2692(95)00097-6.  [47]  Mae Y. Schematic interpretation of anomalies in the physical properties of Eu and Yb among the lanthanides. Int J Mater Sci Appl. 2017;6(4):165–170. doi:10.11648/j.ijmsa.20170604.14.  [48]  Michaelson HB. The work function of the elements and its periodicity. J Appl Phys. 1977;48(11):4729–4733. doi:10.1063/1.323539.  [49]  Ricci F, Chen W, Aydemir U, et al. An ab initio electronic transport database for inorganic materials. Scientific data. 2017;4(1):1–13. doi:10.1038/sdata.2017.85.  [50]  Migas DB, Shaposhnikov VL, Borisenko VE. Isostructural BaSi 2 , BaGe 2  and SrGe2: electronic and optical properties. Phys Status Solidi B. 2007;244(7):2611–2618. doi:10.1002/pssb.200642556.  [51]  Chang IYK, Hwang YR, Juan PC, et al. Temperature dependence of the current conduction mechanisms in Sm2O3 thin films. J Electrochem Soc. 2008 oct;155(12):G265. doi:10.1149/1.2979141.  [52]  Zollner S, Chen JG, Duda E, et al. Dielectric functions of bulk 4H and 6H SiC and spectroscopic ellipsometry studies of thin SiC films on Si. J Appl Phys. 1999;85(12):8353–8361. doi:10.1063/1.370682.  [53]  Kaneko Y, Koda T. New developments in IIa–VIb (alkaline-earth chalcogenide) binary semiconductors. J Cryst Growth. 1988;86(1-4):72–78. doi:10.1016/0022-0248(90)90701-L.  [54]  Stewart AD, Gerger A, Gila BP, et al. Determination of Sm2O3/GaAs heterojunction band offsets by x-ray photoelectron spectroscopy. Appl Phys Lett. 2008;92(15):153511. doi:10.1063/1.2911726.  [55]  Gobeli GW, Allen FG. Photoelectric properties of cleaved GaAs, GaSb, InAs, and InSb surfaces; comparison with Si and Ge. Phys Rev. 1965;137(1A):A245. ACCEPTED MANUSCRIPT1     doi:10.1103/PhysRev.137.A245.  [56]  Davydov SY. On the electron affinity of silicon carbide polytypes. Semiconductors. 2007;41:696–698. doi:10.1134/S1063782607060152.  [57]  Hu CY, Hensley EB. Electron affinities of the barium chalcogenides. J Appl Phys. 1969;40(8):3346–3351. doi:10.1063/1.1658186.   Figure  1. Device models used for simulations.  Figure  2. The calculated band alignments of the model (C) in Fig. 1 with PBH = (a) 0.0 and (b) 0.5 eV under dark conditions. Upper blue line denotes the conduction band minimum while the lower orange line denotes the valence band maximum.  Figure  3. (a)–(d) Simulated PCE of solar cells as functions of the mφ  of metallic electrodes in models (A)–(D), where the electrodes are in contact with 6H-SiC, 4H-SiC, Sm 2 O 3 , and BaS, respectively. The red dashed line indicates a PCE of 30%. The insets show the band alignments near the back electrode for specific work functions. CB and VB denote the conduction and valence bands, respectively.  Figure  4. Workflow of computational material screening for metallic electrodes of solar cells proposed in this study.  Figure  5. mT  plotted against cohE  of randomly selected binary metallic compounds, along with the regression line.  Figure  6. Comparison of the mT  of binary metallic compounds estimated by the linear regression (circles) and the machine learning model [33] (squares) with the reference values.  Figure  7. Comparison of the mφ  of elemental metals estimated by DFT and the machine learning model [39]. The surface orientation selected was (100) for all data.  Figure  8. Electrical conductivity of the candidate materials for model (A). The electrical conductivity divided by the relaxation time is shown.  Figure  9. Electrical conductivity of the candidate materials for model (D). The electrical conductivity divided by the relaxation time is shown.   Table 1.  Material properties used for device simulations.     Material  Parameter   BaSi 2    BaS   Sm 2 O 3    6H-SiC   4H-SiC   Dielectric constant   14 [50]   15 [27]   15 [51]   8 [52]   8 [52]  gE  (eV)   1.3 [9, 10, 11]   3.81 [53]   5.16 [54, 55]  3.0 [56]   3.2 [56]  Electron affinity (eV)   3.2 [16]   0.84 [57]   2.94 [54, 55]  3.5 [56]   3.2 [56]  ACCEPTED MANUSCRIPT1        Formula   mT  (K)   mφ  (eV)   Formula   mT  (K)   mφ  (eV)   Formula   mT  (K)   mφ  (eV)  Ba 3 Ge 4    1423   3.67   EuSi   1597   3.73   PrS   2478   2.60  Ba 3 Ge 5    1418   3.48   EuSi 2    1756   3.73   PrSb   2384   3.30  Ba 3 Si 4    1412   3.90   EuSn   1373   3.53   PrSi   1934   3.69  Ba 5 Sb 3    1405   2.90   GdAl 2    1774   3.69   PrSi 2    1942   3.56  BaAl 4    1325   3.50   GdAl 3    1761   3.70   PrSn 2    1437   3.55  BaGa 2    1290   3.08   GdGa 2    1635   3.54   PrSn 3    1434   3.33  BaGa 4    1277   3.56   GdGa 3    1642   3.78   ScAl 3    1654   3.82  BaGe   1426   3.41   GdGe   2041   3.52   ScGa 3    1393   3.73  BaSi   1413   3.38   GdN   2994   3.23   ScGe   2276   3.82  BaSn   1416   3.29   GdP   2996   3.52   ScN   2747   3.44  Ca11 Sb10    1310   3.50   GdS   2530   2.87   ScP   2569   3.75  CaAl 2    1342   3.70   GdSb   2421   3.51   ScSb   2488   3.70  CaGa 2    1277   3.79   GdSi   2091   3.81   ScSi   2259   3.86  CaGe   1432   3.83   GdSn 2    1396   3.62   Sm11 Sn10    1758   3.43  CaGe 2    1308   3.85   Ho 3 Ge 4    2089   3.75   Sm 3 Ge 5    1743   3.75  CaSi   1524   3.67   Ho 3 Ge 5    1788   3.84   SmAl 3    1737   3.70  CaSn   1374   3.25   Ho 3 Sn 7    1362   3.82   SmGa 2    1640   3.47  Ce 3 Al11    1530   3.76   HoAl 3    1763   3.82   SmGa 3    1646   3.76  Ce 3 Ge 5    1755   3.78   HoGa 2    1583   3.57   SmGe   1790   3.56  Ce 3 Sn 7    1434   3.74   HoGa 3    1596   3.84   SmS   2288   3.80  Ce 4 Ge 7    1752   3.67   HoGe   2161   3.73   SmSi   2064   3.44  Ce 4 H11    1333   3.66   HoP   2713   3.60   SmSn 2    1373   3.58  Ce 7 O12    2587   3.77   HoS   2550   3.07   SmSn 3    1381   3.40  CeAl 3    1591   3.85   HoSb   2398   3.59   Sr 5 Sb 3    1309   3.27  CeGa 3    1707   3.73   HoSi   2132   3.71   SrAl 4    1327   3.51  CeGa 6    1591   3.87   HoSn 2    1357   3.85   SrGa 2    1283   3.27  CeN   3025   2.97   La 2 H 5    1320   3.42   SrGa 4    1276   3.63  CeP   2986   3.31   La 2 H 5    1320   3.58   SrGe   1426   3.58  CeS   2652   2.65   La 2 Sn 3    1445   3.46   SrGe 2    1405   3.71  CeSb   2073   3.29   La 3 Al11    1524   3.82   SrSi   1421   3.85  CeSb 2    1829   3.89   La 3 Ge 5    1748   3.66   SrSn   1394   3.49  CeSn 3    1433   3.33   La 3 Sn 7    1436   3.58   Tb 3 Ge 4    2059   3.61  Dy 3 Ge 4    2095   3.69   LaGa 2    1725   3.34   Tb 3 Ge 5    1888   3.76  Dy 3 Ge 5    1868   3.75   LaGe   1758   3.25   Tb 3 Sn 7    1374   3.78  Dy 5 S 7    1944   3.86   LaP   2727   3.27   Tb 5 S 7    1963   3.81  Dy 5 Sn 11    1358   3.80   LaS   2429   2.47   TbAl 3    1762   3.71  ACCEPTED MANUSCRIPT1     DyAl 3    1760   3.72   LaSb   2051   3.29   TbGa 2    1573   3.55  DyGa 2    1589   3.57   LaSb 2    1801   3.86   TbGa 3    1551   3.72  DyGa 3    1608   3.74   LaSn 3    1433   3.27   TbGe   2137   3.68  DyGe   2138   3.36   Lu 11 Ge10   2253   3.65   TbS   2602   3.90  DyP   2690   3.60   Lu 3 Ge 4    2107   3.88   TbSi   2107   3.41  DyS   2557   3.01   LuAl 3    1741   3.73   TbSn 2    1372   3.69  DySb   2383   3.58   LuGa 2    1572   3.80   TiF 3    1707   3.55  DySi   2119   3.86   LuGa 3    1530   3.67   TiN   3225   3.55  Er 3 Ge 4    2162   3.75   LuGe 2    1574   3.85   Tm 3 Ge 4    2125   3.82  Er 3 Sn 7    1351   3.83   LuP   2534   3.62   TmAl 3    1758   3.79  ErAl 3    1758   3.65   LuSb   2394   3.61   TmGa 2    1573   3.76  ErGa 2    1578   3.61   LuSn 2    1335   3.87   TmGa 3    1565   3.81  ErGa 3    1584   3.79   Nd 2 Sn 3    1437   3.52   TmGe   2211   3.75  ErGe   2211   3.84   Nd 3 Ge 5    1802   3.74   TmP   2636   3.64  ErP   2595   3.62   Nd 3 Sn 7    1430   3.63   TmSi   2139   3.83  ErS   2394   3.10   NdAl 3    1721   3.63   TmSn 2    1334   3.74  ErSb   2348   3.60   NdGa 2    1732   3.39   Y 3 Ge 4    2146   3.60  ErSi   2141   3.85   NdGa 3    1716   3.75   Y 3 Ge 5    1928   3.77  ErSn 2    1346   3.83   NdGe   1797   3.50   YAl 3    1754   3.66  Eu16 Sb11    1811   3.49   NdP   2842   3.36   YGa 2    1630   3.56  Eu 3 Ge 5    1330   3.61   NdS   2525   2.64   YGa 3    1631   3.80  Eu 3 P 4    1609   3.89   NdSb   2384   3.29   YGe   2193   3.71  Eu 5 Sb 3    1791   3.24   NdSi   1941   3.59   YP   2697   3.59  EuAl 4    1366   3.64   NdSn 2    1433   3.54   YSi   2124   3.79  EuGa 2    1292   3.51   NdSn 3    1428   3.35   YSn 2    1356   3.85  EuGa 4    1286   3.77   Pr 2 Sn 3    1440   3.50   Zr 5 Zn 39    1303   3.81  EuGe   1423   3.56   PrAl 3    1659   3.85   ZrAl 3    1860   3.81  EuO   2246   3.03   PrGa 2    1736   3.37   ZrC   3534   3.85  EuP   2256   3.38   PrGa 3    1725   3.73   ZrGa 3    1650   3.90  EuS   2534   2.85   PrGe   1768   3.50        Table  2.: List of candidate metallic materials for model (A): 6H-SiC contacting layer. The estimated mT  and mφ  are also summarized.      Formula   mT  (K)   mφ  (eV)   Formula   mT  (K)   mφ  (eV)   Formula   mT  (K)   mφ  (eV)  Ba 3 Ge 5    1418   3.48   EuS   2534   2.85   NdSn 2    1433   3.54  Ba 5 Sb 3    1405   2.90   EuSn   1373   3.53   NdSn 3    1428   3.35  ACCEPTED MANUSCRIPT1     BaAl 4    1325   3.50   GdGa 2    1635   3.54   Pr 2 Sn 3    1440   3.50  BaGa 2    1290   3.08   GdGe   2041   3.52   PrGa 2    1736   3.37  BaGa 4    1277   3.56   GdN   2994   3.23   PrGe   1768   3.50  BaGe   1426   3.41   GdP   2996   3.52   PrS   2478   2.60  BaSi   1413   3.38   GdS   2530   2.87   PrSb   2384   3.30  BaSn   1416   3.29   GdSb   2421   3.51   PrSi 2    1942   3.56  Ca11 Sb10    1310   3.5   HoGa 2    1583   3.57   PrSn 2    1437   3.55  CaSn   1374   3.25   HoP   2713   3.60   PrSn 3    1434   3.33  CeN   3025   2.97   HoS   2550   3.07   ScN   2747   3.44  CeP   2986   3.31   HoSb   2398   3.59   Sm11 Sn 10    1758   3.43  CeS   2652   2.65   La 2 H 5    1320   3.42   SmGa 2    1640   3.47  CeSb   2073   3.29   La 2 H 5    1320   3.58   SmGe   1790   3.56  CeSn 3    1433   3.33   La 2 Sn 3    1445   3.46   SmSi   2064   3.44  DyGa 2    1589   3.57   La 3 Sn 7    1436   3.58   SmSn 2    1373   3.58  DyGe   2138   3.36   LaGa 2    1725   3.34   SmSn 3    1381   3.40  DyP   2690   3.60   LaGe   1758   3.25   Sr 5 Sb 3    1309   3.27  DyS   2557   3.01   LaP   2727   3.27   SrAl 4    1327   3.51  DySb   2383   3.58   LaS   2429   2.47   SrGa 2    1283   3.27  ErGa 2    1578   3.61   LaSb   2051   3.29   SrGe   1426   3.58  ErS   2394   3.10   LaSn 3    1433   3.27   SrSn   1394   3.49  ErSb   2348   3.60   LuSb   2394   3.61   TbGa 2    1573   3.55  Eu16 Sb11    1811   3.49   Nd 2 Sn 3    1437   3.52   TbSi   2107   3.41  Eu 3 Ge 5    1330   3.61   NdGa 2    1732   3.39   TiF 3    1707   3.55  Eu 5 Sb 3    1791   3.24   NdGe   1797   3.50   TiN   3225   3.55  EuGa 2    1292   3.51   NdP   2842   3.36   Y 3 Ge 4    2146   3.60  EuGe   1423   3.56   NdS   2525   2.64   YGa 2    1630   3.56  EuO   2246   3.03   NdSb   2384   3.29   YP   2697   3.59  EuP   2256   3.38   NdSi   1941   3.59        Table  3.: List of candidate metallic materials for model (B): 4H-SiC contacting layer. The estimated mT  and mφ  are also summarized.     Formula   mT  (K)   mφ  (eV)   Formula   mT  (K)  mφ  (eV)   Formula   mT  (K)   mφ  (eV)  Ba 2 Ge   1431   2.75   Eu 5 Sb 3    1791   3.24   Pr 5 Sb 3    1999   3.11  Ba 2 Si   1409   3.02   EuO   2246   3.03   Pr 5 Sn 3    1772   3.21  Ba 2 Sn   1457   2.88   EuS   2534   2.85   PrAl   1714   3.17  Ba 5 Sb 3    1405   2.90   Gd 2 C   2451   3.20   PrS   2478   2.60  BaGa 2    1290   3.08   GdN   2994   3.23   PrSb   2384   3.30  BaSn   1416   3.29   LaS   2429   2.47   Sm   1352   2.91  ACCEPTED MANUSCRIPT1     CaSn   1374   3.25   LaSb   2051   3.29   Sm 2 C   1708   3.16  Ce 2 C 3    2547   3.29   LaSn 3    1433   3.27   Sm 5 Ge 3    1886   3.17  CeN   3025   2.97   Nd 3 Ga 2    1523   3.28   Sm 5 Sb 3    1997   3.31  CeO   2365   2.90   Nd 5 Ge 3    1911   3.27   Sm 5 Si 3    1892   3.32  CeP   2986   3.31   Nd 5 Si 4    1935   3.18   Sm 5 Sn 3    1774   3.27  Eu 2 Ge   1521   3.27   Nd 5 Sn 3    1816   3.22   Sr 5 Sb 3    1309   3.27  Eu 2 Si   1513   3.17   NdAl   1633   3.24   SrGa 2    1283   3.27  Eu 2 Sn   1623   3.13   NdSb   2384   3.29        Table  4.: List of candidate metallic materials for model (C): Sm 2 O 3  contacting layer.The estimated mT  and mφ  are also summarized.       Formula   mT  (K)   mφ  (eV)   Formula   mT  (K)  mφ  (eV)   Formula   mT  (K)   mφ  (eV)  Al 13 Fe 4    1434   4.28   GaNi 3    1658   4.63   NiSb   1437   4.36  Al 3 Cr   1762   4.32   GdB 4    2836   4.41   P 2 W   1965   4.76  Al 3 Ni 2    1885   4.32   GdH 2    1284   4.59   PW   2702   4.70  Al 3 Ni 5    1840   4.44   Ge 2 Mo   1815   4.51   PrB 4    2842   4.34  Al 4 Cu 9    1327   4.26   GeMo 3    2453   4.43   PrH 2    1339   4.50  Al 4 Ni 3    1899   4.40   Hf 3 N 2    3447   4.51   RbC 8    2915   5.04  Al 5 Co 2    1493   4.26   HfB 2    3489   4.43   Sc 39 N 34    2770   4.72  Al 5 W   1983   4.24   HfH 2    2487   4.41   Si 2 Mo   2263   4.44  AlCo   1899   4.36   HfP   2893   4.41   Si 2 Ni   1600   4.74  AlFe 3    1749   4.48   HfP 2    1935   4.61   Si 2 W   2435   4.55  AlMo 3    2077   4.25   HfS   2660   4.68   Si 3 Mo 5    2437   4.27  AlNi   1900   4.48   HoB 4    2737   4.45   SiMo 3    2464   4.52  AlNi 3    1627   4.58   HoH 2    1354   4.60   SiNi   1390   4.69  B13 C 2    2683   5.33   KC 8    3028   5.03   SiNi 2    1617   4.53  B 2 Mo   2707   4.65   LaB 4    2349   4.28   SiNi 3    1677   4.71  B 2 W   2738   5.06   LaH 2    1332   4.48   SmB 4    2800   4.39  BMo   2843   4.56   LuB 4    2694   4.52   SmH 2    1344   4.53  BW   3023   4.43   Mn   1537   4.29   SmN   2817   4.30  BW 2    3011   4.54   Mn 23 C 6    1841   4.32   SmP   2431   4.27  BaC 6    2696   4.91   Mn 29 H 2    1351   4.36   Ta 2 C   3577   4.82  Ce 2 O 3    2484   5.48   Mn 2 N   1703   4.50   Ta 2 N   3291   4.48  CeB 4    2368   4.43   Mn 2 P   1553   4.42   Ta 3 B 2    2899   4.25  CeB 6    2541   4.82   Mn 2 P   1553   4.39   Ta 3 B 4    3315   4.45  Co   1767   4.86   Mn 3 Co   1364   4.33   Ta 3 S 2    2491   4.26  Co 2 P   1654   4.57   Mn 3 Co   1364   4.38   Ta 5 S 8    2198   4.57  ACCEPTED MANUSCRIPT1     Co 2 Si   1648   4.50   Mn 4 N   1653   4.58   Ta 5 S 8    2198   4.61  Co 3 Ni   1467   4.53   MnAl   1474   4.35   Ta 7 S12    2182   4.37  Co 3 S 4    1561   5.02   MnAl 12    1354   4.25   TaB   3055   4.40  Co 9 S 8    1455   4.84   MnAl 6    1401   4.26   TaB 2    3360   4.61  CoB   1735   4.91   MnB   2157   4.62   TaCo 3    1837   4.45  CoGe   1490   4.48   MnB 4    2401   5.04   TaH 2    2125   5.08  CoH   1288   4.70   MnCo   1618   4.34   TaN   3282   4.62  CoP   1456   4.96   MnGa   1361   4.46   TaNi 3    1764   4.40  CoSi   1733   4.60   MnNi 3    1533   4.63   TaP   2282   4.25  CoSi 2    1661   4.74   MnP   1440   4.65   TaSi 2    2264   4.26  CoSn   1498   4.43   MnS 2    1441   5.05   TbB 4    2759   4.43  Cr 2 O 3    2521   5.05   MnV   1677   4.25   TbH 2    1427   4.58  Cr 3 C 2    2130   4.42   Mo 15 N16    2333   4.85   TbH 3    1286   5.22  Cr 3 Si   2030   4.47   Mo 15 N16    2333   4.75   TbP   2705   4.34  Cr 7 C 3    2065   4.35   Mo 2 C   2785   4.32   Ti 2 C   2514   4.26  CrB 4    2434   4.82   MoN   2329   4.99   Ti 2 O   2203   4.53  CrP   1882   4.48   MoP   2139   4.64   Ti 8 C 5    2974   4.39  CrSi 2    1756   4.42   MoP 2    1600   4.56   TiB   2598   4.33  CsC 8    2774   5.46   MoW   2880   4.40   TiB 2    3301   4.31  Cu   1340   4.52   Nb 2 B 3    3273   4.56   TiO   2136   4.50  DyB 4    2752   4.42   Nb 2 N   2307   4.51   TiP   2382   4.32  DyH 2    1376   4.62   Nb 3 B 4    3183   4.33   TiS   2282   4.37  ErB 4    2679   4.43   Nb 3 S 4    1937   4.46   TmB 4    2666   4.47  ErH 2    1313   4.59   Nb 3 S 5    1949   4.39   TmH 2    1352   4.60  EuH 2    1292   4.32   Nb 5 N 6    2440   4.30   V 2 B 3    2767   4.57  Fe   1784   4.55   NbB 2    3293   4.49   V 2 N   2264   4.68  Fe 2 B   1576   4.77   NbCo 3    1726   4.25   V 2 O 3    2118   6.07  Fe 2 P   1642   4.54   NbH 2    1867   5.12   V 3 B 4    2718   4.38  Fe 2 W   1954   4.39   NbN   2439   5.62   V 3 S 4    2027   4.40  Fe 3 Co   1449   4.64   NbP 2    1839   4.41   V 3 Si   2198   4.31  Fe 3 Ge   1470   4.62   NdB 4    2850   4.37   V 5 B 6    2641   4.40  Fe 3 N   1861   4.92   NdH 2    1337   4.52   VB   2522   4.41  Fe 3 P   1646   4.51   NdH 3    1322   4.67   VB 2    2830   4.59  Fe 3 Si   1643   4.67   Ni   1752   4.79   VFe 3    1723   4.36  Fe 3 Sn 2    1489   4.41   Ni12 P 5    1466   4.62   VN   2481   4.79  Fe 9 Co 7    1699   4.47   Ni19 Ge12    1449   4.47   VNi 3    1643   4.36  FeB   1897   4.92   Ni 2 B   1678   4.81   VSi 2    1924   4.26  FeCo   1700   4.51   Ni 3 B   1471   4.50   VW   2931   4.28  FeGe   1385   4.56   Ni 3 Ge   1723   4.63   W   3526   4.48  ACCEPTED MANUSCRIPT1     FeH   1481   4.67   Ni 3 Mo   1684   4.39   W 2 N 3    2902   4.99  FeN   1848   5.80   Ni 3 P   1506   4.58   WC   2964   4.94  FeNi   1553   4.72   Ni 3 S 2    1359   4.95   YB 4    2951   4.45  FeNi 3    1498   4.83   Ni 3 Sn   1434   4.61   Zn 3 N 2    1579   4.41  FeP   1430   4.97   Ni 3 Sn 2    1500   4.39   ZnNi   1346   4.48  FeS   1452   4.77   Ni 3 Sn 4    1285   4.36   ZrB 2    3325   4.24  FeSn   1388   4.42   Ni 4 B 3    1499   4.82   ZrH 2    1910   4.28  Ga 31 Mo 6    1355   4.26   Ni 4 Sn 3    1501   4.41   ZrN   3252   4.44  Ga 3 Ni 2    1509   4.36   Ni 4 W   1766   4.31   ZrP   2537   4.30  Ga 3 Ni 5    1504   4.44   Ni 5 Ge 3    1450   4.48   ZrP 2    1764   4.51  Ga 9 Ni13    1505   4.35   Ni 5 P 4    1361   4.68   ZrS   2436   4.62  GaCo   1497   4.42   Ni 9 S 8    1331   4.71   ZrSi   2507   4.24  GaFe 3    1667   4.67   NiGe   1446   4.54   ZrSi 2    2200   4.33  GaMo 3    2461   4.30   NiH   1276   4.64        Table  5.: List of candidate metallic materials for model (D): BaS contacting layer.The estimated mT  and mφ  are also summarized.              SUPPLEMENTAL MATERIAL  ACCEPTED MANUSCRIPT1       Computational material screening for electrode materials of BaSi2 solar cells    Tomoaki Yazakia, Keisuke Arimotoa, Junji Yamanakaa, and Kosuke O. Harab aUniversity of Yamanashi, 7-32 Miyamae, Kofu, Yamanashi, Japan bNara Institute of Science and Technology, 8916-5 Takasyama, Ikoma, 630-0192, Nara, Japan   1. Data list of work functions Table S1 lists the work functions of elemental metals estimated by the density functional theory (DFT) calculations and the machine learning model [S1] together with reference values [S2].  Table S1 Work functions of elemental metals estimated by DFT calculations and the machine learning model [S1] together with reference values [S2]. Estimations are for (100) orientations. Reference values are for (100) if available; otherwise, for polycrystalline samples. Work function (reference [S2]) [eV] 4.64 4.2 5.47 2.52 4.98 2.87 1.95 5.1 4.32 3.9 Formula Work function (DFT) [eV]  Work function (machine learning) [eV]  Ag  4.39  4.22Al  4.38  4.25Au  5.23  4.90Ba  2.40  2.33Be  3.93  3.86Ca  2.76  2.79Cs  1.99  2.01Cu  4.65  4.61Ga  4.37  4.33Hf  3.69  3.57ACCEPTED MANUSCRIPT2  In 3.95 4.01 4.09 K 2.22 2.38 2.29 Li 3.06 3.01 2.93 Mg 3.69 3.51 3.66 Mo 3.98 4.00 4.53 Na 2.66 2.83 2.36 Nb 3.67 3.75 4.02 Os 4.89 4.81 5.93 Pb 3.85 3.94 4.25 Pd 5.23 5.03 5.22 Pt 5.81 5.59 5.64 Rb 2.17 2.21 2.261 Re 4.52 4.42 4.72 Rh 5.23 4.96 4.98 Ru 4.60 4.44 4.71 Sc 3.32 3.23 3.5 Sn 4.27 4.29 4.42 Ta 3.90 4.22 4.15 Ti 3.70 3.49 4.33 V 3.84 4.00 4.3 W 4.21 4.36 4.63 Zn 4.60 4.07 3.63 Zr 3.68 3.41 4.05  2. Validation of the screening workflow on pn-junction Si solar cells To validate our screening workflow, we searched for electrode materials for pn-junction crystalline Si solar cells. The device model for simulations consisted of an n-type Si layer (100 nm) on the top side and a p-type Si layer (200 μm) on the bottom side. The electron and hole concentrations in the n- and p-type layers were 1×1019 cm−3 and 1×1016 cm−3, respectively. The permittivity, bandgap, electron affinity, effective density of states of the conduction band, effective density of states of the valence band, electron mobility, and hole mobility were 11.9, 1.12 eV, 4.05 eV, 2.86×1019 cm−3, 2.66×1019 cm−3, 1450 cm2/V⋅s, and 505 cm2/V⋅s, respectively, according to ref. [S3]. The optical absorption coefficients were taken from ref. [S4]. Because wxAMPS does not accept Auger recombination coefficients, acceptor-type midgap defects ACCEPTED MANUSCRIPT2 were assumed to simulate realistic Si materials. We assumed the density, energy level, and capture cross section to be 1×1012 cm−3, 0.56 eV, and 1×10−16 cm2, respectively. This led to a carrier lifetime of approximately 1 ms. The simulation temperature was 300 K, and the effective surface recombination speed was 1×107 cm/s. The top and bottom reflectances were 0 and 1, respectively. Figure S1 shows the power conversion efficiency of the pn-junction Si solar cells as functions of the work functions of the bottom and top electrodes. These results indicate the range of work functions that yield high power conversion efficiency: Specifically, ≥ 5.45 eV for the bottom electrode and ≤ 3.65 eV for the top electrode. Then, computational material screening was performed. For the validation, the search space was confined to elements excluding noble gases, radioactive elements, and actinoids. The melting point threshold was set to 300 K. Table S2 summarizes the candidate materials in descending order of electrical conductivity (accurately, electrical conductivity divided by the relaxation time. Please see the main text for details). Work function criteria were not considered to prepare this table. If we apply the work function criteria, only Na remains as a candidate. However, Na is highly reactive with air and is therefore not used in practical solar cells. Additionally, a common approach is to heavily dope with impurities to avoid the effect of Fermi level pinning and form ohmic contacts on Si by taking advantage of its excellent dopability. Therefore, ignoring the work function criterion aligns with the design concept of practical Si solar cells. However, if we consider other semiconductors that do not significantly suffer from Fermi level pinning, the work function criteria would be useful for effective screening. Table S2 shows that Al, In, Ag, etc. are promising candidates. This result agrees with the electrode materials (Al and Ag) used in practical Si solar cells [S5]. In is avoided probably because of its low melting point, scarcity of resources, and high cost. Thus, our screening workflow has been shown to reach practical electrode materials for crystalline Si solar cells.  Figure S1 Power conversion efficiency of pn-junction solar cells simulated by wxAMPS as functions of work function of (a) bottom and (b) top electrodes.  ACCEPTED MANUSCRIPT    Table Sbandgalisted, conduc FormuAl In Ag Zn Pb Au Cd Tl Na Be Ga Sb Sn Ge    Referen[S1] Schhigh-thrS2 List of elemap, phase stabbut was nottivity dividedula mp-ID mp-134 mp-85 mp-8566mp-79 mp-2048mp-81 mp-94 mp-82 mp-1017mp-87 mp-142 mp-104 mp-117 mp-32 nces hindler P, Antroughput densments that pability, interfat used for sd by the relaxMelting6 83 72 toniuk ER, Chsity functionaassed computace reactivityscreening. Thxation time.g point (K) 9914041208647566127761756235015423148535031176heon G, et al.al theory and m2 tational matey, and meltinhe elementsWork function. Discovery omachine learnerial screeninng point. Thare listed in (eV) Elec4.194.054.194.033.964.873.933.812.804.464.294.704.294.74f stable surfacning. Adv Fung. The conside estimated in descendinctrical conductces with extrenct Mater. 202dered criteriawork functiong order of tivity (S/cm⋅s)3.13×101.99×101.66×101.65×101.64×101.56×101.23×101.04×107.29×107.22×105.40×102.82×102.71×101.76×10eme work fun24;34(19): 2  a included on is also electrical ) 06 06 06 06 06 06 06 06 05 05 05 04 04 03 nctions by 2401764. ACCEPTED MANUSCRIPT2 [S2] Lide DR, editor. CRC handbook of chemistry and physics, internet version 2005. CRC press; 2005. [S3] Sze SM., Semiconductor Devices: Physics and Technology, 2nd Edition. John Wiley & Sons. Inc.; 2002, translated by YNannichi Y, Kawabe M, and Hasegawa F, Japan UNI Agency, Inc.; 2004. [S4] Green MA and Keevers M, Optical properties of intrinsic silicon at 300 K, Prog Photovolt. 1995;3:189. [S5] Avrutin V, Izyumskaya N, and Morkoç H, Semiconductor solar cells: Recent progress in terrestrial applications. Superlattices Microstruct. 2011;49:337.   ACCEPTED MANUSCRIPT2    ACCEPTED MANUSCRIPT2    ACCEPTED MANUSCRIPT2    ACCEPTED MANUSCRIPT2    ACCEPTED MANUSCRIPT2    ACCEPTED MANUSCRIPT2    ACCEPTED MANUSCRIPT3    ACCEPTED MANUSCRIPT3    ACCEPTED MANUSCRIPT3