The interconnection network provides a processor connection structure for parallel computers. Typical interconnections include torus, hypercube, and star graph. Hypercube is the most widely used interconnection network with various advantages such as recursive structure, node symmetry, and simple routing algorithm. In this paper, we analyze what to consider when applying the devide-and-conquer method to improve the network cost of the hypercube. We also propose a new matrix hypercube based on the results of the study. The cluster properties of the proposed matrix hypercube are analyzed. In addition, we propose a routing algorithm, derive the diameter according to the routing, and compare the network cost with other networks. Matrix hypercube improved by 3/4 compared to hypercube with the same number of nodes. In terms of network cost, hypercube is, and matrix hypercube has. Therefore, the proposed matrix hypercube has better results than hypercube.