- 學位論文
動態可重組系統之階層式基因軟硬體分割演算法
Hierarchical Genetic-Based Hardware-Software Partitioning Algorithm for Dynamic Reconfigurable Systems
摘要
近年來在嵌入式系統 (Embedded System) 設計上導入軟硬體共同設計(Hardware-Software Co-design) 概念已為發展趨勢，嵌入式系統在符合系統限制條件下，如何分配各系統功能在軟體與硬體資源下執行以提高系統整體效能是一重要課題。本論文提出一階層式基因軟硬體分割演算法 (Hierarchical Genetic-Based Hardware-Software Partitioning Algorithm, HGA) 應用於動態可重組系統中，藉由提出的階層評估準則 (Hierarchy Estimator Norm, HEN) 將系統切割成數個階層，並應用基因軟硬體分割演算法 (Genetic Partitioning Algorithm) 求解系統軟硬體分割。階層評估準則將代表系統任務的染色體長度縮短，以更短的時間內得到系統軟硬體分割解，每個階層各自執行基因軟硬體分割演算法，藉由提出一新的適應性函數 (Fitness Function) 來篩選染色體，針對系統執行時間 (Execution time) 與功率消耗 (Power consumption) 為目標搜尋最適合的軟硬體分解。本論文使用TGFF (Task Graph for Free) 產生實驗範例，由實驗結果得知與基因演算法比較之下，階層式基因軟硬體分割演算法能減少22.6%的演算法執行時間。
並列摘要
In recent years, it’s a trend import the concept of hardware-software co-design to the design of embedded systems. How to allocate the system functions effectively to enhance the overall system performance with hardware and software resources becomes the primary topic. As mentioned, dynamic reconfigurable architecture based hardware is one of the ways. In this thesis, we propose a Hierarchical Genetic-based hardware-software partitioning Algorithm (HGA) for dynamic reconfigurable systems. The objective of the Hierarchy Estimator Norm (HEN) proposed to improve convergence procedure to achieve hardware-software partitioning. HEN made system tasks to several hierarchies that reduced the chromosome length of GA. Each hierarchy executed GA to obtain hardware-software partitioning solution respectively. We proposed a new fitness function partitioning the type of resource allocate to each task, and the target was optimization in system execution time and power consumption. Finally, the experiment sample used Task Graph for Free (TGFF). Experiment results show that the HGA can obtain a solution for Hardware Software Partitioning, and can reduce the 22.6% on execution time of algorithm compared to GA.