Adaptive bacterial foraging driven datapath optimization: Exploring power-performance tradeoff in high level synthesis

An automated exploration of datapath for power-delay tradeoff in high level synthesis (HLS) driven by bacterial foraging optimization algorithm (BFOA) is proposed in this paper. The proposed exploration approach is simulated to operate in the feasible temperature range of an actual Escherichia coli (E. coli) bacterium in order to mimic its biological lifecycle. The proposed work transforms a regular BFOA into an adaptive DSE framework that is capable to explore power-performance tradeoffs during HLS. The key sub-contributions of the proposed methodology are as follows: (a) Novel chemotaxis driven exploration drift algorithm; (b) Novel multi-dimensional bacterium encoding scheme to handle the DSE problem; (c) A novel replication algorithm customized to the DSE problem for manipulating the position of the bacterium by keeping the resource information constant (useful for inducing exploitative ability in the algorithm); (d) A novel elimination-dispersal (ED) algorithm to introduce diversity during the exploration process; (e) Adaptive mechanisms such as resource clamping and step size clamping to handle boundary outreach problem during exploration. Finally, results indicated an average improvement in QoR of > 35% and reduction in runtime of > 4% compared to recent approaches.