Automated exploration of datapath in high level synthesis using temperature dependent bacterial foraging optimization algorithm
Anirban Sengupta, Saumya Bhadauria · 2014
This paper presents a novel methodology for automated exploration of datapath in high level synthesis using temperature dependent bacterial foraging optimization algorithm (BFOA) which has the ability to reach reach optimal solution in most cases. To the best of the authors' knowledge, this is the first work that proposes a direct mapping of BFO algorithm for design space exploration (DSE) problem in high level synthesis (HLS). The major contributions of the proposed methodology are as follows: a) Novel multi-dimensional bacterium encoding scheme to handle the DSE problem; b) Novel chemotaxis algorithm for imitating exploration drift during searching; c) A novel replication algorithm customized to the DSE problem; d) A novel elimination-dispersal (ED) algorithm to introduce diversity during exploration; e) A temperature dependent BFOA based exploration process to tradeoff between power-performance design metrics during HLS which mimics the actual Escherichia coli (E.coli) bacterium behaviour operating in its feasible temperature range; f) Adaptive mechanisms such as resource clamping and step size clamping to handle boundary outreach. Results indicated an average improvement in Quality of Result (QoR) of >27 % and reduction in runtime of > 44 % compared to recent genetic algorithm based approach which does not guarentee reaching optimal solution.