在網路晶片架構中,路由選擇演算法 (Routing Algorithm),流量控制(Flow Control)是兩個影響電路效能表現的因素。本文提出一個雙向網路晶片(BiNoC)的模型,根據這個模型,發展出雙向網路晶片中的雙向路由演算法以及一個應用時間分割多工的通道控制方法。雙向路由演算法使用雙向通道使得網路晶片中的封包可以選擇更多路由路徑;通道控制方法則可以根據晶片上封包的擁塞程度分配雙向通道的時間單位,這兩者均可以利用雙向網路晶片的特性來增進晶片中傳輸流量的平衡。本文利用一個精準時脈週期的測試環境進行模擬,對於在合成的交通型態傳輸情況下,提出的路由選擇演算法以及通道控制方法均可提升系統在雙向網路架構上的效能。
Network on Chip (NoC) is an emerging design implemented in an SoC for the communication of IPs since the advence of deep sub-micron technology in recent years. This Thesis proposes a three-dimensional model for a Bi-directional NoC (BiNoC). This three-dimensional model inspires us with a new routing algorithm for BiNoC, called bidirectional routing (BI-Routing) and a flow control method for BiNoC, called time division multiplexing BiNoC (TDM-BiNoC). Bi-Routing is a fully adaptive routing algorithm using different layers in the proposed three-dimensional model to avoid deadlock without prohibiting any paths. TDM-BiNoC adopts the time division multiplexing concept in the three-dimensional model to release congestion and thus avoid deadlock. TDM-BiNoC dynamically allocates the time slots for the bidirectional channels. In summary, both our proposed Bi-Routing and TDM-BiNoC can increase the load balance of an NoC and reduce its latency. Experimental simulation results showed its better performance with admissible area, power, and timing overhead.