A neural network approach to the placement problem

Abstract

In this paper, we introduce a new neural network approach to the placement of gate array designs. The network used is a Kohonen self-organising map. An abstract specification of the design is converted to a set of appropriate input vectors fed to the network at random. At the end of the process, the map shows a 2-dimensional plane of the design in which the modules with higher connectivity are placed adjacent to each other, hence minimising total connection length in the design. The approach can consider external connections and is able to place modules in a rectilinear boundary. These features makes the approach capable of being used in hierarchical floorplanning algorithms.