Abstract: ‍ ‍With the rapid advancement of communication technologies and emerging scenarios， future wireless communication networks will face increasingly complex demands， including ubiquitous connectivity， high speed， high reliability， and intelligence. Simultaneously， the surge in data traffic and bandwidth requirements presents significant challenges to communication networks in new contexts. As a cutting-edge technology， semantic communication utilizes neural networks to extract and transmit the semantic information of data. This approach significantly reduces bandwidth requirements while enhancing transmission quality. Recently， foundation models have emerged， demonstrating powerful feature extraction and comprehension capabilities， which provide notable advantages in expressive power and predictive performance. Compared to traditional networks， foundation models possess broader application potential by effectively managing multimodal information and complex data. This paper reviews the current research and development trends of foundation model-enabled semantic communication， focusing on its applications in semantic encoding and decoding， as well as physical and network layer design， and practical deployment. First， we introduce the basic concepts of semantic communication， outlining its fundamental structure and distinguishing it from traditional communication schemes. An overview of semantic communication technologies based on foundation models follows. Next， we analyze the performance of foundation models and traditional neural networks in semantic communication from the perspectives of semantic encoding and decoding， physical layer design， and network layer implementation. This analysis highlights the potential of foundation models to improve communication accuracy， efficiency， and robustness. Finally， the paper summarizes future directions and challenges related to foundation models in communication， addressing issues such as model training， customization， and practical deployment. It also explores new strategies to tackle challenges including the acquisition and synchronization of training data， computational resource management， and hardware power consumption. These efforts aim to promote further application and development of foundation models in semantic communication.