Selection of SRAM Cells to improve Reliable PUF implementation using Cell Mismatch Metric

Physically Unclonable Functions (PUFs) are low-cost cryptographic primitives implemented in secret key generation and device authentication strategies. SRAM-PUFs are widely well-known as entropy source; however, they mainly experience a low reproducibility of the challenge-response pair because of non-deterministic noise conditions during the process of power-up. The reliability of SRAM-PUFs achieved these days comes by using complex error correcting codes (ECCs) combined with Fuzzy extractor structures introducing an increment in terms of power consumption, area cost and complexity of the design. In this paper we define an effective metric to classify the SRAM cells identifying the most reliable cells generating high reproductible responses for the PUF implementation and, identifying the most unpredictable ones for Random Number Generator (RNG). This metric is obtained from the mismatch between the cell's inverters and the start-up behavior. Also, the noise in the PUF is modeled to validate the classification results obtained by the proposed metric. The proposed metric can be used during SRAM PUF design to explore the impact on reliable cells significantly increasing the reproducibility of the PUF and minimizing the dependability on ECCs and Fuzzy extractor.