The traditional router with virtual channel flow control alleviates the loss of link throughput and networks congestion by increasing virtual channels. However, it results in low utilization of buffers and excessive area cost of arbitration logic. And the router with dynamically allocated virtual channels can improve the buffer utilization and link throughput by organizing the buffers dynamically, but it can not avoid the rapid increase of the logic for flow control and arbitration. To improve the link throughput and buffer utilization, we propose a NoC router with dynamically allocated virtual-output-queues (DAVOQ) in order to obtain a good tradeoff between performance and overhead. In the router DAVOQ, the buffers are organized by rapid-linked lists and the logic area of arbitration, and pipeline is reduced and optimized based on lookahead routing. The simulation and synthesis results show that the router DAVOQ can reduce latency and improve throughput, yielding saving of standard cells area by 15. 1% and saving of leakage power by 12. 9% under 0. 13μm CMOS process in comparison with traditional router with virtual channels. And it can obtain considerable latency reduction with losing a little network throughput, and yield saving of standard cells area by 15. 6% and saving of leakage power by 20. 5% when compared with the router with dynamically allocated virtual channels.%传统虚通道流控技术的片上路由器通过增加虚通道缓解排头阻塞引起的链路吞吐率下降以及网络拥塞的同时,面临缓冲区低利用率、仲裁开销较大等问题.而动态虚通道流控的片上路由器虽可通过动态管理缓冲单元,提高缓冲区利用率与链路吞吐率,但却不可避免流控与仲裁逻辑复杂度与开销的快速增长.为了提高链路吞吐率与缓冲区利用率,获得较好的性能与开销折中,提出一种动态分配虚拟输出队列结构的片上路由器DAVOQ,该结构通过快速链表动态组织虚拟输出队列,同时使用超前路由机制以简化仲裁逻辑,优化流水线.模拟与综合的结果表明,相比传统虚通道路由器,DAVOQ路由器改善报文传输延迟与吞吐率的同时,在0.13 μm CMOS工艺下,节省了15.1％的标准单元面积与12.9％的漏电流功耗；而相比动态虚通道路由器,DAVOQ路由器能够以较小的吞吐率损失获得可观的延迟改善,同时节约15.6％的标准单元面积与20.5％的漏电流功耗.
展开▼
