針對在單晶片網路系統中任務群組之研究

Abstract

單晶片網路系統已成為研究晶片上系統之通訊的一個新趨勢。 我們為單晶片網路系統發展了一個完整的設計流程，而本篇論文著重在單晶片網路設計流程中之任務群組部份。此任務群組問題需要考慮多重的限制，包括運算量，記憶體，輸出入以及整體通訊頻寬。在滿足所有的限制條件下，試圖減少整體群組之數量。我們導入一個演算法來解決這個問題。此演算法已被寫為一個自動化的程式，而我們也得到一些實驗結果。經由這個程式我們可以在數分鐘內減少群組數目並滿足所有條件限制。

The Network-on-Chip (NoC) has become a new trend in on-chip system communication design. We have developed a complete design flow for NoC system design. In this thesis we focus on the task clustering in our NoC design flow. The task clustering problem in the NoC system design needs to consider multiple constraints, which include computing power, memory, I/O and communication bandwidth. We introduce an algorithm to minimize the number of clusters in the result with all constraints being met. The algorithm has been written as an automatic program and some experimental results are presented. Using our program, we can reduce the number of clusters with all constraints satisfied in few minutes.