Genetic Algorithm based pipeline scheduling in high-level synthesis

In this work, we present a Genetic Algorithm (GA) based method for pipeline scheduling optimization. The objective is to minimize the circuit area under both data initiation interval and pipeline latency constraints. In the initialization, the scheduler generates a series of solutions between As Soon As Possible (ASAP) and As Late As Possible (ALAP) interval. Afterwards a Linear Programming (LP) algorithm is applied for transforming unfeasible solutions to feasible solutions, which are input to GA for searching the optimization result. In the experiments, our proposed algorithm achieves an average of 29.74% area improvement by comparing with ASAP and ALAP methods.