Neural Networks having the Time-variant Connection for Multi-layer Channel Routing Problem

摘要

It is necessary for designing VLSI to arrange wirings not to overlap each other in the wiring area of a layer. The problem to find such arrangement is called multi-layer channel routing problem. This article proposes the neural networks having the time-variant connection for such problem. In previous studies, a monotonously decreasing energy function is introduced into the system. Therefore, the system operates toward the minimum of the energy. Then, to find the optimum solution, this minimum corresponds to the minimum of the cost function. However, this method hardly finds the optimum solution if the energy function has many local minima. On the other hand, higher order connection system is scarcely influenced by local minima to find the solution. It takes, however, the system so long time to find the solution. Then, in this article, the time-variant connection system is proposed to find the solution quickly without the influence of local minima.