MULTI-OBJECTIVE DESIGN AUTOMATION FOR RECONFIGURABLE MULTI-PROCESSOR SYSTEMS

Abstract

Reconfigurable Multi-Processor Systems (MPS) offer both high computational performance and low power consumption at the same time. However, the combination of software and hardware reconfigurability also drastically increases the complexity of the system and imposes new challenges for design automation tools.

With the aim of making the design flow for Reconfigurable MPS more productive and autonomous, this thesis provides algorithms and tools that accelerate the development process in both macro level and micro level synthesis through several contributions:

+ A throughput and energy-aware mapping approach. + A multi-stage leakage aware scheduling heuristic for partially reconfigurable platform. + A Machine Learning and Genetic Algorithm optimization framework that can address the time-consuming problem of DSE process for list-based heuristics. + A Design Space Exploration tool that utilizes the loop structure to shorten the development time of HLS tools. + An HLS tool that can auto-generate option pricing accelerators for reconfigurable platforms.