A Machine Learning Based Approach To Predict Power Efficiency Of S-Boxes

In the era of lightweight cryptography, designing cryptographically good and power efficient 4 x 4 S-boxes is a widely discussed problem. While the optimal cryptographic properties are easy to verify, it is not very straightforward to verify whether a S-box is power efficient or not. The traditional approach is to explicitly determine the dynamic power consumption using commercially available CAD tools and report accordingly based on a pre-defined threshold value. However, this procedure is highly time consuming, and the overhead becomes formidable while dealing with a set of S-boxes from a large space. This mandates development of an automation tool which should be able to quickly characterize the power efficiency from the Boolean function representation of an S-box. In this paper, we present a supervised machine learning (ML) assisted automated framework to resolve the problem for 4 x 4 S-boxes, which turns out to be approximately 14 times faster (using ANDOR-NOT gates) than the traditional approach. The key idea is to extrapolate the knowledge of the literal counts of various functional forms, AND-OR-NOT gate counts in the simplified SOP form of the underlying Boolean functions corresponding to the S-box to predict the dynamic power efficiency. We demonstrate the effectiveness of our framework by reporting a set of power efficient S-boxes from a large set of 4 x 4 optimal S-boxes. The experimental results and performance of our novel technique depicts its superiority with high efficiency and low time overhead.