Abstract

We review a true random number generator which internally exploits a pipeline analog-to-digital converter modified to operate in chaotic mode. Contrarily to other true random number generators it is not based on the explicit observation of noisy physical quantities, freeing the designer from subtle technology-dependent details. In ideal conditions the model can be formally proven to deliver perfectly uncorrelated output sequences by itself. Here we thoroughly investigate the effects of implementation errors and we introduce a data-postprocessing methodology suitable for their compensation and for assuring robust behavior.