-
The emergence of large language models has significantly advanced scientific research. Representative models such as ChatGPT and DeepSeek R1 have brought notable transformations to the paradigm of scientific research. While these models are general-purpose, they have demonstrated strong generalization capabilities in the field of batteries, particularly in solid-state battery research. In this study, we systematically screened 5309268 articles from key journals up to 2024, accurately extracting 124021 relevant battery-related papers. Additionally, we comprehensively searched through 17559750 patent applications and granted patents from the European Patent Office and the United States Patent and Trademark Office up to 2024, from which we filtered out 125716 battery-related patents. Utilizing this extensive collection of literature and patents, we conducted numerous experiments to evaluate the knowledge base, in context learning, instruction-following, and structured output capabilities of language models. Through multi-dimensional model evaluations and analyses, we found the following: first, the model exhibited high accuracy in screening literature on inorganic solid-state electrolytes, equivalent to the level of a doctoral student in the relevant field. Based on 10604 data entries, the model demonstrated good recognition capabilities in identifying literature on in-situ polymerization/solidification technology. However, its understanding accuracy for this emerging technology was slightly lower than that for solid-state electrolytes, requiring further fine-tuning to improve accuracy. Second, through testing with 10604 data entries, the model achieved reliable accuracy in extracting inorganic ionic conductivity data. Third, based on solid-state lithium battery patents from four companies in South Korea and Japan over the past 20 years, the model proved effective in analyzing historical patent trends and conducting comparative analyses. Furthermore, the model-generated personalized literature reports based on the latest publications also showed high accuracy. Fourth, by leveraging the model's iteration strategies, we enabled DeepSeek to engage in self-thinking, thereby providing more comprehensive responses. The research results indicate that language models possess strong capabilities in content summarization and trend analysis. However, we also observed that the model may occasionally exhibit issues with numerical hallucinations. Additionally, while processing vast amounts of battery-related data, the model still has room for optimization in engineering applications. Based on the characteristics of the model and the above test results, we utilized the DeepSeek V3-0324 model to extract data on inorganic solid electrolyte materials, including 5970 entries of ionic conductivity, 387 entries of diffusion coefficients, and 3094 entries of migration barriers. Additionally, it includes over 1000 entries of data related to chemical, electrochemical, and mechanical properties, covering nearly all physical, chemical, and electrochemical properties associated with inorganic solid electrolytes. This also signifies that the application of large language models in scientific research has transitioned from assisting research to actively advancing its development. The datasets presented in this paper can be acess at the website:
https://cmpdc.iphy.ac.cn/literature/SSE.html (DOI:https://doi.org/10.57760/sciencedb.j00213.00172 ).-
Keywords:
- soild state battery /
- artificial intelligence /
- natural language process /
- materials database
-
图 1 本工作主要考核目标(左侧)与设计的实验(右侧)
Figure 1. The main assessment objectives (left) and the designed experiment (right) of this work.
图 2 DeepSeek R1, ChatGPT o3 min和Gemini 2.0对同样内容的总结比较
Figure 2. Comparison of summaries generated by DeepSeek R1, ChatGPT o3 min, and Gemini 2.0 on the same content.
图 3 DeepSeek R1分析日韩4家企业过去20年固态电池专利
Figure 3. Analysis of solid-state battery patents from four Japanese and South Korean companies over the past 20 years by DeepSeek R1.
图 4 DeepSeek R1分析每日电池文献(左图)和电池新闻(右图)
Figure 4. Analysis of solid-state battery literatures (left subfigure) and news (right subfigure) by DeepSeek R1.
图 5 DeepSeek R1 70B未挂载知识库(左图)和挂载知识库(右图)在回答同一问题的差异
Figure 5. DeepSeek R1 70B without knowledge base mounted (left) and with knowledge base mounted (right) showing differences in responses to the same question.
图 6 DeepSeek R1 70B Deep Research流程(左图)和优化后输出(右图)
Figure 6. The workflow of deep research of DeepSeek R1 70B (left) and answer by it (right).
图 7 无机固态电解质数据库构建流程
Figure 7. The workflow of constructing inorganic solid electrolyte database.
表 1 无机固态电解质文献分类准确性
Table 1. Accuracy of literature classification for inorganic solid electrolytes.
MatElab模型 DeepSeek R1 QwQ 32B 精确性 0.859 0.883 0.781 召回率 0.850 0.823 0.827 F1 0.854 0.854 0.801 
DownLoad: CSV
表 2 原位固化技术文献分类准确性
Table 2. Accuracy of literature classification for in-situ solidification technology.
微调后的Qwen2-7B-Instruct DeepSeek R1 精确性 0.406 0.950 召回率 0.833 0.653 F1 0.576 0.774
DownLoad: CSV
-
[1] ChatGPT ChatGPT website
[2] DeepSeek-AI 2025 arXiv 2501.12948
[3] Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez A N, Kaiser L, Polosukhin I 2017 arXiv: 1706.03762
[4] Radford A, Narasimhan K, Salimans K, Sutskever I 2018 OpenAI Blog
[5] Devlin J, Chang M W, Lee K, Toutanova K 2018 arXiv 1810.04805
[6] Radford A, Wu J, Child R, Luan D, Amodei D, Sutskever I 2019 OpenAI Blog
[7] Brown T B, Mann B, Ryder N, Subbiah M, Kaplan J, Dhariwal P, Neelakantan A, Shyam P, Sastry G, Askell A, Agarwal S, Herbert-Voss A, Krueger G, Henighan T, Child R, Ramesh A, Ziegler D M, Wu J, Winter C, Hesse C, Chen M, Sigler E, Litwin M, Gray S, Chess B, Clark J, Berner C, McCandlish S, Radford A, Sutskever I, Amodei D 2020 arXiv: 2005.14165
[8] Wei J, Wang X Z, Schuurmans D, Bosma M, Ichter B, Xia F, Chi E, Le Q, Zhou D 2022 arXiv: 2201.11903
[9] o1模型 OpenAI o1 Hub | OpenAI
[10] Yang A, Yang B S, Hui B Y, Zheng B, Yu B W, Zhou C, Li C P, Li C Y, Liu D Y H, Huang F, Dong G T, Wei H R, Lin H, Tang J L, Wang J L, Yang J, Tu J H, Zhang J W, Ma J X, Xu J, Zhou J R, Bai J Z, He J Z, Lin J Y, Dang K, Lu K M, Chen K Q, Yang K X, Li M, Xue M F, Ni N, Zhang P, Wang P, Peng R, Men R, Gao R Z, Lin R J, Wang S J, Bai S, Tan S N, Zhu T H, Li T H, Liu T Y, Ge W B, Deng X D, Zhou X H, Ren X Z, Zhang X Y, Wei X P, Ren X C, Fan Y, Yao Y, Zhang Y C, Wan Y, Chu Y F, Cui Z Y, Zhang Z Y, Fan Z H 2024 arXiv: 2407.10671
[11] Zeng A H, Xu B, Wang B W, Zhang C H, Yin D, Zhang D, Rojas D, Feng G Y, Zhao H L, Lai H Y, Yu H, Wang H N, Sun J D, Zhang J J, Cheng J L, Gui J Y, Tang J, Zhang J, Sun J Y, Li J Z, Zhao L, Wu L D, Zhong L C, Liu M D, Huang M L, Zhang P, Zheng Q K, Lu R, Duan S Q, Zhang S D, Cao S L, Yang S X, Tam W L, Zhao W Y, Liu X, Xia X, Zhang X H, Gu X T, Lv X, Liu X H, Liu X Y, Yang X Y, Song X X, Zhang X K, An Y F, Xu Y F, Niu Y L, Yang Y T, Li Y Y, Bai Y S, Dong Y X, Qi Z H, Wang Z Y, Yang Z, Du Z X, Hou Z Y, Wang Z H 2024 arXiv: 2406.12793
[12] Gemini Gemini
[13] Chen J L, Xiao S T, Zhang P T, Luo K, Lian D F, Liu Z 2024 arXiv 2402.03216
[14] Lewis P, Perez E, Piktus A, Petroni F, Karpukhin V, Goyal N, Küttler H, Lewis M, Yih W T, Rocktäschel T, Riedel S, Kiela D 2000 arXiv: 2005.11401
[15] 吴思远, 王宇琦, 肖睿娟, 陈立泉, 2020 物理学报 69 226104
Google Scholar
Wu S Y, Wang Y Q, Xiao R J, Chen L Q 2020 Acta Phys. Sin. 69 226104
Google Scholar
[16] 离子输运数据库 http://e01.iphy.ac.cn/bmd
[17] Xiao R J, Li H, Chen L Q 2015 Sci. Rep. 5 14227
Google Scholar
[18] He B, Chi S T, Ye A J, Mi P H, Zhang L W, Pu B W, Zou Z Y, Ran Y B, Zhao Q, Wang D, Zhang W Q, Zhao J T, Adams S, Avdeev M, Shi S Q 2020 Sci. Data 7 151
Google Scholar
[19] Yang F L, dos Santos E C, Jia X, Sato R, Kisu K, Hashimoto Y, Orimo S, Li H 2024 Nano Mater. Sci. 6 256
Google Scholar
[20] Hargreaves C J, Gaultois M W, Daniels L M, Watts E J, Kurlin V A, Moran M, Dang Y, Morris R, Axandra M, Thompson K, Wright M A, Beluvalli-Eshwarappa P, Blanc F, Collins C M, Crawford C A, Duff B B, Evans J, Gamon J, Han G P, Leube B T, Niu H J, Perez A J, Robinson A, Rogan O, Sharp P A, Shoko E, Sonni M, Thomas W J, Vasylenko A, Wang L, Rosseinsky M J, Dyer M S 2023 npj Comput. Mater. 9 9
Google Scholar
[21] Wu S Y, Zhu T N, Tu S J, Xiao R J, Yuan J, Wu Q S, Li H, Weng H M 2024 Chin. Phys. B 33 050704
Google Scholar
[22] 无机固态电解质材料数据库https://cmpdc.iphy.ac.cn/literature/SSE.html
[23] Zhang Y, Xie M X, Zhang W, Yan J L, Shao G Q 2020 Mater. Lett. 266 127508
Google Scholar
[24] Li Y X, Daikuhara S, Hori S, Sun X Y, Suzuki K, Hirayama M, Kanno R 2020 Chem. Mater. 32 8860
Google Scholar
[25] Xu R C, Wu Z, Zhang S Z, Wang X L, Xia Y, Xia X H, Huang X H, Tu J P 2017 Chem. Eur. J. 23 13950
Google Scholar
[26] Li X L, Peng W X, Tian R Z, Song D W, Wang Z Y, Zhang H Z, Zhu L Y, Zhang L Q 2020 Electrochim. Acta 363 137185
Google Scholar
[27] Wu F, Fitzhugh W, Ye L H, Ning J X, Li X 2018 Nat. Commun. 9 4037
Google Scholar
[28] MatElab平台https://in.iphy.ac.cn/eln/#/recusertype
DownLoad:
Metrics
- Abstract views: 436
- PDF Downloads: 24
- Cited By: 0
