NEUROPT: neurocomputing for multiobjective design optimization for printed circuit board component placement

Abstract
The traveling salesman problem (TSP) can be mapped to the problem of optimal component placement for printed circuit boards (PCBs). An innovation that consists of adapting a neural network formulation of TSP to multiobjective component placement on the basis of wire-length criteria, as well as thermal reliability, is described. The author shows that the mathematical formulation of the Hopfield energy function for TSP is identical to the energy for the placement problem except for the cost (distance) function. The Hopfield cost function can be modified by introducing terms to model the wire-length and thermal reliability for alternative component placements. This approach was tested by coding a neural network simulator and comparing the quality of the resulting placement with standard methods. The positive results of that testing, the potential for a dramatic improvement in the time needed to calculate such optimal placements, and the natural way of extending the formulation to include more design criteria lead to a confidence that similar approaches will have a significant impact on multiobjective design.