A stochastic neural network approach for circuit partitioning

We propose a stochastic neural network for solving combinatorial optimization problems, that can find global minima in theory. The network dynamic is based on integration of Langevin equation of neurons motion. The continuous network is applied to bipartitioning circuits represented by their hypergraphs. For this a new fuzzy net-cut model is used treating hypergraphs without splitting multi-pin-nets into two-pin-nets. The neural network has been tested with an industrial example and results will be given.