Quantization Schemes for PUFs: The Entropy-Area Trade-Off

Physical Unclonable Functions (PUFs) exploit manufacturing variations to derive chip-specific secrets that can be processed to cryptographic keys. For lots of PUFs, the actual PUF output can be understood as a sample drawn from a normal distribution that is then quantized to bits. This paper compares the most popular quantization methods for PUFs: The common sign-based, the two-metric helper data scheme, equiprobable quantization and the Lehmer-Gray encoding are evaluated regarding their implementation cost on ASIC and FPGA, the extracted entropy, and their reliability. This work also discusses the succeeding error correction necessitated by the respective quantization method. The results show the benefits of sign-based quantization in high-noise environments, of the twometric helper data scheme when mid-range errors coincide with large PUF area, and of equiprobable quantization in cases with very low noise. Additionally, the results reveal significant deficits in the Lehmer-Gray encoding regarding statistical properties and potential vulnerabilities.