Application of Self Organizing Neural Network in Timing Driven System Partitioning on MCM

MCM is a new technology of ICs.In the design steps of MCM,partitioning is very important.Based on the improvement to Kohonen self organizing neural nerwork,the neural network approach for the timing driven system partitioning on MCM is presented.In the algorithm,the total routing cost between the chips and the circle time are both minimized,while satisfying area and timing constraints.The similarity of the directly connected cells is described by the connection weight between the cells and the internal delay of the combinatorial cell,and fuzzy similarity computing is applied to describing the similarity of indirectly connected cells.Each cell is mapped noto a two dimensional plane,and corresponds to one or a few neurons.The neural network is learned from a sequence of movements of cells,allows corresponding cells to cooperate during the learning process.By the self organizing learning process,the cells that have great similarity with each other will move as closely as possible.