Article ID Journal Published Year Pages File Type
6942114 Integration, the VLSI Journal 2018 22 Pages PDF
Abstract
In Network-on-Chip based multi-core systems, application mapping is a critical issue as it affects the overall system performance in terms of average packet delay (APD) and system reliability. As the number of Intellectual Property (IP) cores integrated into a single chip increase, heat dissipation becomes a major issue. Due to non-uniform heat spreading, thermal hotspots are formed which reduces the system reliability. In this work, we have proposed a reliability model which in turn considers thermal effect due to computation and communication entities present in the network. Based on the reliability model, we have proposed a Mixed Integer Linear Programming (MILP) formulation of the mapping problem for improving system reliability with a constraint on the APD of the network. To address large sized applications, we have used a Particle Swarm Optimization (PSO) based mapping method. The basic PSO has been augmented with a constructive heuristic to improve the system reliability further. Experiments have been conducted on applications belonging to SPLASH-2 and PARSEC benchmark suites, which shows significant improvement in system reliability, compared to other works mentioned in the literature.
Related Topics
Physical Sciences and Engineering Computer Science Hardware and Architecture
Authors
, , ,