The choice of neighbourhood in self-organization scheme for VLSI

The self-organization process has been applied in many areas e.g. speech recognition, pattern recognition, placement of VLSI cells, regression analysis etc. The self-organizing process is a self-learning scheme in neural networks where weights of synapses between the output nodes and input nodes are adjusted to select the topological relationships at the input end to the output end. The common feature of this method is to find the most sensitive output node (called the winning node) to a given input pattern and to choose a neighbourhood around the winning node. The adjustments of the weights in this neighbourhood influences the outcome iteratively to other input patterns. Although it is known that the choice of neighbourhood affects the rate of convergence, the relationship between convergence and neighbourhood function is not known in quantitative terms. We examine a variety of neighbourhoods and study their effects in the placement of VLSI cells. The results are generalizable to other domains