A methodology for evaluation time approximation

This paper describes a feed-forward neural network model (FFNNM) for complexity prediction of path related objective functions, mainly average path length (APL) of an arbitrary Boolean function (BF). The proposed model is determined by neural training process of evaluation time derived from the Monte Carlo data of randomly generated BFs. Experimental results show a good correlation between the ISCAS benchmark circuits and those predicted by the FFNNM. This model is capable of providing an estimation of the performance of a circuit prior to its final implementation.