An analog neural network approach to global routing problem

Neural networks have been successfully applied to many combinatorial optimization problems. However, applying this technique in the integrated circuit routing problem has yet to be investigated. This paper proposes a modified Hopfield network to solve the global routing problem, which has been proven to be NP- complete. This network is constructed of two layers of neurons. One layer is used for reducing the interconnection wire length and the other layer is used for channel capacity enforcement. The operation and theory under this design are thoroughly discussed and a software simulator will be implemented to monitor the performance of this network. On the average, an approximate 20% total wire length reduction of randomly generated data is obtained.