Autonomous Robotic Experiments Accelerate Discovery of Multi-Components Electrolyte for Rechargeable Lithium-Metal Batteries

Artificial intelligence (AI) driven approach for materials discovery, has attracted significant recent attention even in the field of rechargeable batteries. Instead of relying on the experience and intuition of researchers for exploring new materials, the approach employs data scientific techniques that can, in principle, reduce the time and cost of the discovery of new materials with superior battery performance and predict cycle life. Actually, an automated robotic experimental system was recently developed by our group for discovering new multi-component electrolytes for lithium-metal based rechargeable batteries, which exhibited superior energy density compared with conventional lithium-ion batteries (ref.1). Although superior searching throughput analyzing more than 1000 samples per day was achieved, adequate experimental design is essential to realize high-throughput exploration of electrolyte composition from large searching space. For example, when considering a combination of selecting 5 types from 20 types of chemicals, the candidates are over 10⁷. Thus, it is not realistic to comprehensively evaluate all the possible combination even through such robotic experiments. Therefore, a specific electrolyte composition that realizes a superior battery performance must be determined with only a limited number of experimental trials.<br/>In the present study, the effectiveness of a data-driven automated robotic experiments was investigated to discover multi-component electrolyte additives for lithium-metal based rechargeable batteries. Established machine-learning methodologies using Bayesian optimization were employed to solve optimization problems for analyzing datasets obtained from the automated robotic experiments, thereby minimizing the number of trials required to identify the ideal electrolyte composition. As results, we identified the specific electrolyte composition (1.5 M LiNO3, 0.1 M LiTFSI, 0.1 M LiBr, 0.5 mM LiCl, and 10 mM LiBOB in dimethylamide, with 5 vol.% 1,3-dioxolane) that enhanced the cycle life of the lithium-oxygen rechargeable batteries (ref.2). In additions, we also put our attention for the development of the orchestration system to realize a closed loop between AI searching algorithms and robotic experiments. Generally, different searching algorithms are utilized depending on the motivation of a materials exploration task and the procedure for controlling the devices largely depends on the characteristics of the robotic systems. Thus, the control software has far been developed on a case-by-case manner, limiting the widespread use of searching algorithms for robotic experiments. Based on these considerations, we recently developed the NIMS-OS to implement a closed loop of AI and robotic experiments for automated materials exploration (ref.3). Notably, NIMS-OS treats each AI algorithm and each robotic system as separate modules, resulting in the implementation of a closed loop with any combination of these modules. When modules for new AI algorithms or robotic systems are prepared, new closed-loop systems can be easily controlled via NIMS-OS. We believe that the such generic control software is the advantageous for NIMS-OS as standard platform for autonomous materials exploration with robotic experiments.<br/> <br/>References<br/>[1] <b><u>S. Matsuda</u></b> et al. Scientific Reports, 2019, 9, 6211<br/>[2] <b><u>S. Matsuda</u></b> et al. Cell Reports Physical Science, 2022, 3, 100832<br/>[3] R. Tamura, <b><u>S. Matsuda</u></b> et al. Science and Technology of Advanced Materials: Methods, doi.org/10.1080/27660400.2023.2232297