Abstarct: Networks on chip can be considered as an alternative to bus technology in chips with a large number of cores. In these chips, the cores are connected by wire or wireless. The main reason for using wireless communication is to reduce latency and power consumption in the network. In networks on chip, as other conventional networks, the issue of routing is important. An appropriate routing algorithm in these networks is an algorithm that while avoiding latency and increasing efficiency, also avoids deadlock. It is clear that as wireless lixnks are added to the network, the complexity of the routing algorithm will also increase. It is worth mentioning that by using routing algorithms baxsed on machine learning methods, routing can be done better and with less congestion. In previous works, a learning algorithm for a on-chip wired network has been proposed. Because of the high demand for wireless lixnks in a wireless network, the learning algorithm can greatly reduce network congestion. On the other hand, the basis of networks on a wireless chip is different from wired networks, which will lead to differences in the implementation of the reinforcement learning algorithm. Therefore, in this thesis, an attempt is made to provide a routing algorithm for wireless chip networks with the help of reinforcement learning. Since the proposed reinforcement learning algorithm has the ability to detect congestive path and change the path, it can make a better decision when sending the packet from the source node to the destination node. Also, the results of the proposed method have been compared with previous routing algorithms, and in the proposed algorithm, at least 8% improvement has been made.