Named entity recognition (NER) is an important task for clinical knowledge discovery from electronic health records (EHRs). However, to protect the patient privacy, we want to minimizethe annotation needs for NER system development. Therefore, advanced semi-supervised learning algorithms are required. In this paper, we reviewed the recent advance in semi-supervised learning for named entity recognition and discussed the possibility of application in the clinical NER system. We hope our discussions help the development or improvementfor the clinical NER applications for other researchers.