Neural network based power estimation on chip specification

This paper forwards a neural network based VLSI power estimation on VLSI chip specification. This paper used neural network to perform VLSI power estimation. Experiments were made on chip specification parameters extracted from the datasheet of TI series micro-controllers. Different net structure, training plans and vector organizations were applied. Based on limited number of test vector, experimental results showed the neural network based power estimation could give acceptable results on chip power with specific net structure. This method can achieve a much faster power estimation result on datasheet of the same kind of VLSI chips without simulation and analysis or simulations of detail structure and interconnections. Higher training/testing ratio leads to a more accurate power estimation value.