Integer linear programming based routing algorithm design for on-chip optical network

Abstract

This report explores the work done on optimising Optical Network on Chip (ONoC) routing by minimising signal contention while maintaining a thermal reliable route. Compared to electronic circuits, optical communication has a much larger overhead associated with communication contention and is hence more sensitive to communication contention. For large-scale ONoC, there would be a higher number of signals propagating simultaneously, there is a need to manage traffic flow to ensure signal contention is minimised to improve efficiency. Furthermore, optical components such as Micro-Ring Resonators (MR) widely used in ONoC are highly susceptible to thermal fluctuations and this can affect reliability of wavelength transmitted and hence, the quality of signal received. To maintain a thermal reliable route, thermal tuning is used to maintain a constant temperature. The mixed-integer linear programming (MILP) models developed aim to improve overall efficiency of ONoC while minimising signal contention and maintaining a thermal reliable route. The model is split into two parts: signal routing between nodes for inter-processor communication and thermal- aware task scheduling with task graphs. The models were implemented with AMPL and the cplex solver. The model developed was evaluated against python implementations of a continuous mapping algorithm and a greedy algorithm. There are clear improvements made in the implementation.