Improving the efficiency of portrait segmentation is of great importance for the deployment on mobile devices. In this paper, we achieve the fast and light-weight portrait segmentation by introducing a new extremely light-weight backbone (ELB) architecture. The core element of ELB is a bottleneck-based factorized block (BFB), which can greatly reduce the number of parameters while keeping good learning capacity. Based on the proposed ELB architecture, we only use a single convolution layer as decoder to generate results. The ELB-based portrait segmentation method can run faster (263.2FPS) than existing methods yet retaining the competitive accuracy performance with state-of-the-arts. Experiments are conducted on two datasets, which demonstrates the efficacy of our method.