ABSTRACT

As an emerging communication infrastructure, wavelength-division multiplexing (WDM) based optical network-on-chip (ONoC) provides ultra-high bandwidth, low latency and low power dissipation for high-performance computing systems. However, there remain two major challenges in this architecture. One is the intrinsic thermal susceptibility. Under temperature variations, core functional optical devices suffer from significant thermal-induced optical power losses, which seriously threaten ONoCs’ reliability. The other is communication conflicts, which potentially counteracts the advantages of ONoCs in communication performance and energy efficiency. To co-optimize the communication performance and the thermal reliability of ONoCs, in this chapter, we propose a brand-new process variation (PV)-tolerant optical temperature sensor design, PV-OTS, for accurate and efficient thermal monitoring on ONoCs. Effective temperature monitoring, performing as the foundation of thermal-aware management techniques, is critical for ONoCs. On that basis, we further develop novel routing approaches to resolve both the communication conflict and thermal susceptibility challenges in ONoCs. Evaluation results show the high accuracy of the proposed PV-OTS, with an untrimmed inaccuracy of only 0.8650°C over a large operating temperature range from 25 to 105°C. Compared to state-of-the-art techniques, based on guaranteed thermal reliability, our routing approaches averagely improve communication performance by 159.64% and reduce energy overhead by 8.10%.