Timing Variation-Aware Task Scheduling and Binding for MPSoC

This work addresses the new problem of timing variation-aware task scheduling and binding (TSB) for multiprocessor system-on-chip (MPSoC) architecture in the system-level design, where tasks have full flexibilities of resource (i.e., processor) sharing to meet the design constraints. With the timing variation of processors' clock speed, it has been observed that considering the effects of resource sharing on the resulting performance yield computation is critically important for accurate design space exploration and evaluation in the system-level design. Unfortunately previous statistical static timing analysis (SSTA) in the system-level has never considered resource sharing in computing the performance yield, or has overly simplified by employing the gate-level SSTAs. In this work, we overcome those limitations by proposing an effective SSTA technique called TSB-SSTA, which schedules and binds tasks to resources in the presence of resource sharing. We also propose a timing variation-aware (TV) framework, called TSB-TV, tightly integrating TSB-SSTA. We have tested the effectiveness of our approach through experimentation with benchmarks, which showed an average of 56.1% improvement in performance yield over conventional methods.