In　the　field　of　Natural　Language　Processing　(NLP),　traditional　Chinese　Named　Entity　Recognition　(NER)　tasks　often　only　involve　the　recognition　of　a　few　types　of　entities.　But　current　real-world　applications　require　more　fine-grained　types　of　entities　for　more　detailed　downstream　NLP　tasks.　Since　fine-grained　Chinese　NER　is　challenging　for　existing　models,　novel　models　need　to　be　introduced.　We　propose　a　model　with　RoBERTa　as　word　embedding,　a　convolutional　attention　layer　and　CRF　layer　for　outputting　the　entity　labels.　This　model　has　a　better　overall　performance　than　baseline　models　on　the　CLUENER2020　fine-grained　Chinese　NER　dataset.　©　2020　IEEE.