Arbitrarily sized cell placement by self-organizing neural networks

A new self-organizing neural network is described. It can solve arbitrarily sized cell placement problem with various constraints on their connection and dimension. The solution procedure modifies Kohonen´s self-organization algorithm to adapt to the subclass of self-organization problems in which the sample vectors are not easily available, as well as in the case of the cell placement problem. For arbitrarily sized cell placement, the overlap penalty function and the cell growing-up algorithm are introduced to the authors´ solution model where sizes of the cells are considered during the self-organization process in order to reduce overlaps among the cells. Their procedure is convergent in a reasonable number of iterations, and the resulting total wire lengths are at least the same as previous results