Fuzzy biasless simulated evolution for multiobjective VLSI placement

In each iteration of a simulated evolution (SE) algorithm for VLSI placement, poorly placed cells are selected probabilistically, based on a measure known as 'goodness'. To compensate for the error in the goodness calculation (and to maintain the number of selected cells within some limit), a parameter known as 'bias' is used, which has major impact on the algorithm's run-time and on the quality of the solution subspace searched. However, it is difficult to select the appropriate value of this selection bias because it varies for each problem instance. In this paper, a biasless selection scheme for the SE algorithm is proposed. This scheme eliminates the human interaction needed in the selection of the bias value for each problem instance. Due to the imprecise nature of the design information at the placement stage, fuzzy logic is used in all stages of the SE algorithm. The proposed scheme was compared with an adaptive bias scheme and was always able to achieve better solutions.