A Software Framework for Rapid Application-Specific Hybrid Photonic Network-on-Chip Synthesis
| Autor: | Sudeep Pasricha, Shirish Bahirat |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2016 |
| Předmět: |
Engineering
synthesis algorithms Computer Networks and Communications lcsh:TK7800-8360 02 engineering and technology computer.software_genre 020210 optoelectronics & photonics Genetic algorithm 0202 electrical engineering electronic engineering information engineering network-on-chip photonic interconnects chip multiprocessors Electrical and Electronic Engineering business.industry Ant colony optimization algorithms lcsh:Electronics Particle swarm optimization 020202 computer hardware & architecture Software framework Network on a chip Hardware and Architecture Control and Systems Engineering Embedded system Signal Processing Simulated annealing Scalability Heuristics business computer |
| Zdroj: | Electronics, Vol 5, Iss 2, p 21 (2016) Electronics; Volume 5; Issue 2; Pages: 21 |
| ISSN: | 2079-9292 |
| Popis: | Network on Chip (NoC) architectures have emerged in recent years as scalable communication fabrics to enable high bandwidth data transfers in chip multiprocessors (CMPs). These interconnection architectures still need to conquer many challenges, e.g., significant power consumption and high data transfer latencies. Hybrid electro-photonic NoCs have been recently proposed as a solution to mitigate some of these challenges. However, with increasing application complexity, hardware dependencies, and performance variability, optimization of hybrid photonic NoCs requires traversing a massive design space. To date, prior work on software tools for rapid automated NoC synthesis have mainly focused on electrical NoCs. In this article, we propose a novel suite of software tools for effectively synthesizing hybrid photonic NoCs. We formulate and solve the synthesis problem using four search-based optimization heuristics: (1) Ant Colony Optimization (ACO); (2) Particle Swarm Optimization (PSO); (3) Genetic Algorithm (GA); and (4) Simulated Annealing (SA). Our experimental results show significant promise for the ACO and PSO based heuristics. Our novel implementation of PSO achieves an average of 64% energy-delay product improvements over GA and 53% improvement over SA; while our novel ACO implementation achieves 107% energy-delay product improvements over GA and 62% improvement over SA. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|