Fuzzy biasless Simulated Evolution for multiobjective VLSI placement

In each iteration of Simulated Evolution (SE) algorithm for placement poorly placed cells axe selected probabilistically based on a measure known as 'goodness'. To compensate for the errors in 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 run-time and on the quality of 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 work, a biasless selection scheme for simulated evolution algorithm is proposed. This scheme eliminates the human interaction needed in the selection of bias value for each problem instance. Due to the imprecise nature of design information at placement stage, fuzzy logic is used in all stages of SE algorithm. The proposed scheme was compared with an adaptive bias scheme and was always able to achieve better solutions.