A genetic algorithm assisted by a locally weighted regression surrogate model
ICCSA'12 Proceedings of the 12th international conference on Computational Science and Its Applications - Volume Part I
Parametric optimization of reconfigurable designs using machine learning
ARC'13 Proceedings of the 9th international conference on Reconfigurable Computing: architectures, tools, and applications
| Hi-index | 0.00 |
This paper presents a new methodology based on evolutionary multi-objective optimization (EMO) to synthesize multiple complex modules on reprogrammable devices. It starts from a behavioral description written in a common high-level language (for instance C) to automatically produce the register-transfer level (RTL) design in a hardware description language (e.g. Verilog). Since all high-level synthesis problems (scheduling, resource allocation and binding) are notoriously NP-complete and interdependent, these problems should be considered simultaneously. This drives to a wide design space, that needs to be thoroughly explored to obtain solutions able to satisfy the design constraints (e.g. area and performance). Since evolutionary algorithms are good candidates to tackle such complex explorations, in this paper we provide a solution based on the non-dominated sorting genetic algorithm (NSGA-II) to explore the design space and obtain the best solutions in terms of performance given the area constraints of a target reprogrammable device, for instance a Field Programmable Gate Array (FPGA). To further reduce the time needed for the exploration, that theoretically requires the complete logic synthesis of each visited point, the evaluation of the solutions have been speed-up by using two techniques: a good cost estimation model and a technique to exploit fitness inheritance by substituting the expensive actual evaluations with estimations based on closeness in an hypothetical design space. We show on the JPEG case study that the proposed approach provides good results in terms of trade-off between total area occupied and execution time. The results shows also that the Pareto-optimal set obtained by applying the proposed fitness inheritance model well approximates the set obtained without the proposed technique and reduces the overall execution time up to the 25% in average.