| With the continuous development of microelectronics technology,the coordination and realization of the rapid transfer of data between computing units within the chip has gradually become the need for further improvement of chip performance in the increasing scale of chip design.Network on chip is mainly used to transmit data,while the network routing algorithm is used to coordinate the transmission of data between computing units by controlling the network.Among the many network on chip architectures,Crossbar networks are a fast transmission speed and high parallelism network structure that is widely used for connecting multiple cores and interconnecting networks.However,Crossbar networks using synchronous circuits can cause problems such as clock skew and data loss when transmitting data across the clock domain as the area of chip and the clock frequency increase.To solve the above problems,this dissertation first introduces the process of the designing and building of an asynchronous Crossbar network.In addition,traditional network routing algorithms lack flexibility in coordinating computational units,making it difficult to achieve complex computations.Therefore,the author presents a kind of routing algorithm,which can map functions to a network.In order to verify the validity of the asynchronous Crossbar network and the routing algorithm,the author build up a computing array including a 2×2 asynchronous Crossbar network and a floating-point number computing element,which bases on the proposed asynchronous Crossbar network and routing algorithm,and map the convolution function in convolution neural network on it.The design and algorithm are implemented and simulated by Vivado design platform of Xilinx company.The experimental results show that asynchronous Crossbar network is more compatible with different data transfer speeds,and its single data path can operate at a data transfer frequency equivalent to 100 Mhz.Now that integrated circuit technology and industry are developing rapidly,the asynchronous Crossbar networks and routing algorithms proposed in this dissertation will be applied more widely and deeply. |