FPGA based Parallel Neighborhood Search

An FPGA based generic parallel neighborhood search which exploits parallelism at both search and move levels is proposed. A neighborhood partitioning technique is employed to significantly increase parallelism at move level with minimum hardware resource increment. The proposed approach is applied to a tabu search and evaluated using the quadratic assignment problem. Experimental results show that the proposed technique can enhance the search speed by 13.3 times with a solution quality improvement of 11.9%. Compared with a GPU implementation, this work achieves a speedup of 20.2 times.