Channel Models for Physical Unclonable Functions based on DRAM Retention Measurements
Sven Müelich, Sebastian Bitzer, Chirag Sudarshan, Christian Weis, Norbert Wehn, Martin Bossert, Robert F. H. Fischer · 2019
Physical Unclonable Functions (PUFs) are hardware primitives, which exploit intrinsic randomness occurring from variations in manufacturing processes, to generate random binary sequences (responses) that can be used for secure generation and storage of cryptographic keys. In recent times, intrinsic randomness present in the behavior of memory cells is used for that purpose. While SRAM is the type of memory that has most often been chosen in literature, the use of DRAM has emerged during the last years. In this work, we propose two methods to remove the bias in binary sequences extracted from DRAM. For each method, we derive a channel model, which captures the instabilities that occur when re-extracting responses from DRAM. Although knowledge about the channel is beneficial when selecting or designing error-correcting codes, it is often neglected in the context of PUFs. To provide a proof of concept, we use data generated by exploiting the retention behavior of DRAM.