Robust entropy harvester for analogue noise sources in TRNG

Random numbers are important in diverse areas of science and technology. The requirements of randomness, however, are more stringent in cryptographic applications, where so-called true randomness is required. In particular, statistical randomness is not enough, and a degree of unpredictability stemmed in physical phenomenon is used to build a true random number generator (TRNG). A critical part in TRNG design is the entropy harvester, which extracts entropy from multiple entropy sources. Unfortunately, none of such sources provide true randomness in pure form, and the desired signal is heavily contaminated due to electrical coupling, electromagnetic interference, and other pseudo-random signals present in physical environment. Process variation and ageing affect the performance of noise sources. In this paper we propose an entropy extraction and estimation mechanism, which operates directly on raw randomness exhibited by analogue noise sources. The main strength of our design is its dynamic nature, i.e., it adapts as per the current state of source, and provides a robust mechanism to detect entropy failure. Our design can easily be implemented on commodity hardware in a cost effective manner.