Thesis (M.Sc.) - Cairo University - Faculty of Engineering - Department of Computer Engineering

In this study, we address the task mapping on Networks on Chip (NoC) based architectures. The study proposes a novel many-to-one task mapping methodology to replace the current one-to-one mapping methodology. Many- to-one task mapping is based on mapping more than one single task on a single processing unit in a NoC-based system. In this thesis, we implement many-to- one multi-objective mapping using Genetic Algorithm (GA). The objective function maximizes system reliability and minimizes both communication power consumption and average packet latency. The methodology is evaluated by comparing it with di{uFB00}erent one-to-one mapping techniques. We also developed a C++-based simulator to evaluate our methodology. Experimental results showed that our approach outperforms one-to-one NMap, Simulated Annealing(SA), Branch and Bound (BB) and GA mapping by 133%, 187%, 155%, and 158%, respectively

Issued also as CD