Using Neural Networks to Solve VLSI Design Problems

In this paper we summarise our study of the application of neural computation networks, as proposed by Hopfield and Tank, to several NP-complete problems in the domain of VLSI design. We have found that a number of important VLSI problems such as optimal module orientation and optimal assignment of pin positions can be easily mapped onto a neural network and solved in this way. The results of our simulations indicate that the solutions found using this technique compare favorably to those found by other optimization techniques such as simulated annealing. Furthermore, with the advent of neural network hardware, neural network algorithms should prove to be extremely fast.