On entropic aspects of VLSI designs

Abstract

In this paper, we revisit entropic aspects of modern VLSI designs. In particular, we discuss design guidelines for specific computation problems, considering issues associated with dynamic datapath multiplexing due to branching and sharing. We apply these design guidelines to parallel decoder designs of Reed-Solomon codes as an example and show that an appropriate choice of the decoding algorithm, without excess entropy generation in the datapath, results in a better implementation than algorithms that are more efficient just arithmetically. We expect that the present entropic considerations will become more important as the datapath becomes wider and more complex.