This paper proposed a light-weight single frame infrared small target detection network that combined cross-scale feature fusion and bottleneck attention module. Instead of bringing extra huge neurons, the network directly performs cross-scale feature interaction between the encoding and decoding sub-networks, maintain the response of small target in the deep CNN layers, and thus achieves the full fusion between the spatial structure features from shallow layers and high-level semantic features from deep layers. Based on cross-scale feature fusion module, a light-weight bottleneck attention module is introduced to further enhance the response the target feature in the deep layers of the network. Experimental results demonstrate that the network can effectively suppress the complex background clutter and achieve high performance of infrared small target detection with low amount of parameters.