A generalized placement algorithm based on self-organization neural network

Kohonen's self-organization neural network is applied to VLSI cell placement. The algorithm proposed by A. Hemani and A. Postula (1990) adapted for two terminal nets is generalized to accommodate multiterminal nets. In contrast to the traditional gate-as-points model, which was proved improper for cell placement, the authors propose a net-as-points model in the new algorithm. Experiments showed that the convergence speed for the new algorithm was several times faster than for the gate-as-points model while the quality of solutions was better. Placements obtained by the new algorithm were superior to those achieved by the min-cut algorithm in terms of total wire length.<
>