Current communication infrastructures for convolutional neural networks (CNNs) only focus on specic transmission patterns, not applicable to benet the whole system if the dataow changes or dierent dataows run in one system. To reduce data movement, various CNN dataows are presented. For these dataows, parameters and results are delivered using dierent trac patterns, i.e., multicast, unicast, and gather, preventing dataow-specic communication backbones from beneting the entire system if the dataow changes or dierent dataows run in the same system. Thus, in this paper, we propose MUG-NoC to support typical trac patterns and accelerate them, therefore boosting multiple dataows. Specically, (i) we for the rst time support multicast in 2D-mesh software congurable NoC by revising router conguration and proposing the ecient multicast routing; (ii) we decrease unicast latency by transmitting data through the dierent routes in parallel; (iii) we reduce output gather overheads by pipelining basic dataow units. Experiments show that at least our proposed design can reduce 39.2% total data transmission time compared with the state-of-the-art CNN communication backbone.