Hybrid GACO approach for optimized floorplanning in partially reconfigurable FPGAs

This paper presents Hybrid GACO, a novel floorplanning framework for Partially Reconfigurable Field-Programmable Gate Arrays (PR-FPGAs) that integrates Genetic Algorithms (GA) and Ant Colony Optimization (ACO). The proposed approach addresses the multi-objective challenges of PR-FPGA floorplanning—including wirelength minimization, timing closure, resource utilization, and reconfiguration overhead—under architectural and runtime constraints. A formal FPGA grid model is introduced to account for heterogeneous resource types and forbidden regions. GA is employed for global exploration of the design space, while ACO provides fine-grained local refinement using pheromone-guided search. A weight calibration and adaptive feedback mechanism ensures cost function responsiveness to implementation metrics. The framework is evaluated on MCNC benchmarks and a real-time object detection system using the YOLOv3-Tiny model on a Xilinx Zynq UltraScale + MPSoC ZU9EG. Experimental results demonstrate that Hybrid GACO outperforms traditional MILP-based (FLORA), GA-based, and simulated annealing (SA)-based methods across all key metrics, including Half-Perimeter Wire Length (HPWL), resource efficiency, reconfiguration time, latency, and power consumption. These results confirm the framework's suitability for dynamic, resource-constrained, and latency-sensitive reconfigurable systems.