THAMON: Thermal-aware High-performance Application Mapping onto Opto-electrical network-on-chip

Abstract

Nanophotonics as a promising candidate for future on-chip networks has several benefits such as bandwidth transparency, light-speed data rate and low power consumption. Despite significant features of optical on-chip data transmission, the challenges of thermal and process variations (PVs) threaten the photonic network-on-chip reliability. These variations may cause a notable drift in the resonant wavelength of micro-ring resonators as the basic element of optical on-chip network and lead to data corruption, bandwidth wastage and eventually reliability reduction. In this paper, after discussing the effects of temperature variation (TV) on the reliability of optical on-chip network, a thermal-aware and high performance application mapping platform called as THAMON is proposed for manycore systems based-on hybrid opto-electrical network-on-chip. An integer linear programming (ILP) problem, as well as four run-time heuristic algorithms, are suggested to increase the network performance along with maintaining the network reliability through thermal management. The experimental results show that ILP and our proposed most effective heuristic approach improve the performance of the system by about 24% and 16% compared to the base-line approach, and also meet the thermal limitations of the photonic network.