English

• Solid-state chemistry is drawing increasing attention due to the rise of solid-state electrolytes (SSEs). SSEs enable the accelerated migration of multiple anions such as lithium (Li) and sodium at solid-state mode. Compared with routine organic liquid electrolytes in Li batteries, solid Li ionic conductors (SSLCs) can realize high thermal stability, high ionic conductivity, and wide electrochemical windows, which enables the application of Li metal anodes^[1]. Consequently, SSLCs have been treated as a promising solution to break through the anxiety of limited energy density of conventional Li-ion batteries. Tremendous efforts have been devoted to this field and have achieved significant progresses^[2]. However, the experimentally synthesized solid electrolytes cannot meet the multiple requirements in practical batteries, which is calling for emerging methodology and instructions to explore advanced solid-state electrolytes.

  Figure 1

  

  Figure 1.  Unsupervised clustering results for Li contained compounds. Most high ionic conductive materials are located in groups V and VI, which possess mild distortion anion structures. The materials in groups V and VI exhibit obvious talents of high ionic conductivity^[11], which are concentrated for the subsequent screening

  DownLoad: Full-Size Img PowerPoint

  The known SSLCs exhibit various crystalline structures, including layered patterns (Li₃N), garnet (Li₇La₃Zr₂O₁₂), NASICON (Li_1.5Al_0.5Ti_0.5(PO₄)₃), perovskite (Li_0.5La_0.5TiO₃), anti-perovskite (Li₃OX, X = Cl and Br), thio-LISICON (Li₁₀GeP₂S₁₂, LGPS) and argyrodite (Li₆PS₅X, X = Cl, Br, and I)^[3-7]. These typical ionic conductors possess disparate crystalline structure, physical/chemical properties, and ionic conductivities, whose internal connection is difficult to be completely analyzed and unraveled. Therefore, designing brand-new Li ionic conductive materials confronts tremendous challenges. Conventional explorations depending on experimental try and errors are not effective. Based on the already known materials, a rapid exploration and screening impel the development of material genome engineering^[8]. Mo and co-authors have conducted much beneficial work towards the prediction of (electro)chemical stability against Li metal and the electrochemical window of multiple solid-state electrolytes, affording an accurate guidance for interfacial engineering^{[9, 10]}. The supervised learning methods require abundant data to train the models. A large amount of involved data guarantees the accuracy of the trained model. However, the kinds of known SSLCs are insufficient to conduct supervised machine learning with a high accuracy. Most Li contained compounds do not possess high ionic conductivities, which cannot be treated as training materials. In order to figure out the disadvantages, Mo, Ling, and co-workers creatively conducted an unsupervised machine learning study, which is a new route to unravel the interior differences between SSLCs and thus can predict new potential SSLCs^[11]. The unsupervised machine learning model simultaneously divides the data into different groups according to data characteristics.

  In the unsupervised model, the authors designed a protocol as follows: digitalizing Li-contained compounds, clustering the targeted groups, and running ab initio molecular dynamics (AIMD) simulations to verify the predicted objects. The anion framework of Li-contained compounds is firstly digitalized by the modified XRD (mXRD) representation. In order to highlight the structural crystal features, only the anion framework is considered. The mXRD means that the characteristics of the anion frameworks are transformed into a group of lattice parameters shown in XRD patterns. Each compound will be simplified into a vector. Then the mXRD clustering classified compounds by their characteristics of their anion framework structures. The Li ions located in the highly symmetric lattices are constrained at the well-defined sites. The highly disordered framework will also locally trap ions and hinder possible percolations. It is concluded that anion frameworks with a mild distortion, which are located between the highly symmetric lattices and the highly disordered ones, possess high ionic conductivities. This fruitful insight affords a great reference for further screening of ionic conductors. The groups with mild distortion are further screened by AIMD simulation to choose objects with high ionic conductivities. More significantly, the unsupervised method dramatically shrinks the range for screening and increases screening efficiency compared with the conventional high throughput methods. The unsupervised machine learning is appropriate to estimate potential materials with high ionic conductivities. The results for screening also provide targets for experimental attempts. Because the new predicted materials exhibit disparate crystal structures, the results can also help to broaden the thoughts.

  The solid Li ionic conductor is a complicated system, involving many chemical and physical parameters. This unsupervised machine learning system does not cover sufficient details in solid Li ionic conductors. Therefore, the accuracy of the machine learning model can be further improved. The unsupervised machine learning method is a new brand route for predictions, integrating the high throughput results towards screening solid-state lithium ion conductors for next-generation batteries.

  
  
• 1. [1]

     Cheng, X. B.; Zhang, R.; Zhao, C. Z.; Zhang, Q. Toward safe lithium metal anode in rechargeable batteries: a review. Chem. Rev. 2017, 117, 10403–10473. doi: 10.1021/acs.chemrev.7b00115

  2. [2]

     Takada, K. Progress in solid electrolytes toward realizing solid-state lithium batteries. J. Power Sources 2018, 394, 74–85. doi: 10.1016/j.jpowsour.2018.05.003

  3. [3]

     Bachman, J. C.; Muy, S.; Grimaud, A.; Chang, H. H.; Pour, N.; Lux, S. F.; Paschos, O.; Maglia, F.; Lupart, S.; Lamp, P.; Giordano, L.; Shao-Horn, Y. Inorganic solid-state electrolytes for lithium batteries: mechanisms and properties governing ion conduction. Chem. Rev. 2016, 116, 140–162. doi: 10.1021/acs.chemrev.5b00563

  4. [4]

     Awaka, J.; Takashima, A.; Kataoka, K.; Kijima, N.; Idemoto, Y.; Akimoto, J. Crystal structure of fast lithium-ion-conducting cubic Li₇La₃Zr₂O₁₂. Chem. Lett. 2011, 40, 60–62. doi: 10.1246/cl.2011.60

  5. [5]

     Lu, Y.; Huang, X.; Song, Z.; Rui, K.; Wang, Q.; Gu, S.; Yang, J.; Xiu, T.; Badding, M. E.; Wen, Z. Highly stable garnet solid electrolyte based Li–S battery with modified anodic and cathodic interfaces. Energy Storage Mater. 2018, 15, 282–290. doi: 10.1016/j.ensm.2018.05.018

  6. [6]

     Xiao, W.; Wang, J.; Fan, L.; Zhang, J.; Li, X. Recent advances in Li_1+xAl_xTi_2−x(PO₄)₃ solid-state electrolyte for safe lithium batteries. Energy Storage Mater. 2019, 19, 379–400. doi: 10.1016/j.ensm.2018.10.012

  7. [7]

     Zhao, Y.; Daemen, L. L. Superionic conductivity in lithium-rich anti-perovskites. J. Am. Chem. Soc. 2012, 134, 15042–15047. doi: 10.1021/ja305709z

  8. [8]

     Zheng, J.; Ye, Y.; Pan, F. 'Structure units' as materials genes in cathode materials for lithium-ion batteries. Natl. Sci. Rev. 2019, DOI: 10.1093/nsr/nwz178/5614552.

  9. [9]

     Zhu, Y.; He, X.; Mo, Y. First principles study on electrochemical and chemical stability of solid electrolyte-electrode interfaces in all-solid-state Li-ion batteries. J. Mater. Chem. A 2016, 4, 3253–3266. doi: 10.1039/C5TA08574H

  10. [10]

      Han, F.; Zhu, Y.; He, X.; Mo, Y.; Wang, C. Electrochemical stability of Li₁₀GeP₂S₁₂ and Li₇La₃Zr₂O₁₂ solid electrolytes. Adv. Energy Mater. 2016, 6, 1501590. doi: 10.1002/aenm.201501590

  11. [11]

      Zhang, Y.; He, X.; Chen, Z.; Bai, Q.; Nolan, A. M.; Roberts, C. A.; Banerjee, D.; Matsunaga, T.; Mo, Y.; Ling, C. Unsupervised discovery of solid-state lithium ion conductors. Nat. Commun. 2019, 10, 5260. doi: 10.1038/s41467-019-13214-1

• 

  Figure 1  Unsupervised clustering results for Li contained compounds. Most high ionic conductive materials are located in groups V and VI, which possess mild distortion anion structures. The materials in groups V and VI exhibit obvious talents of high ionic conductivity^[11], which are concentrated for the subsequent screening

  下载: 全尺寸图片 幻灯片
•