Abstract
Increasing on-chip integration with technology scaling in which more transistors are available on a chip than that can be simultaneously powered on will lead to the dark silicon era. Process variation gradually impacts the performance (power, speed, frequency etc.) of system over its usage life. In this paper, by adapting the application allocation algorithm at the software layer, reduction of the impact of process variation in multicore architecture is suggested and tested. Process variation causes core-to-core variations in power and frequency. To gain the maximum performance within the power limits, best subset of cores should be selected from the available cores to run an application. When the power constraints are used as cost parameter, simulation of adaptive allocation algorithm on Sniper multicore simulator promises twice the better performance compared to conventional core allocation strategy.
