Secret-Key Generation from Physics Onur G unl u onur.gunlu@tum.de - PowerPoint PPT Presentation

Institute for Communications Engineering Technische Universit¨ at M¨ unchen Secret-Key Generation from Physics Onur G¨ unl¨ u onur.gunlu@tum.de Supervisor: Gerhard Kramer January, 2015 - SPCodingSchool, Campinas - Brazil Onur G¨ unl¨ u — Secrecy via Physical Unclonable Functions (PUFs) 1/4

Institute for Communications Engineering Technische Universit¨ at M¨ unchen Problem Definition Secret-key generation from Physical Unclonable Functions: They are cheap and safe alternatives to Non-Volatile Memory for storing the secret key. The source of randomness (i.e., unique fingerprints of Ring Oscillators [ROs] or SRAMs) is used to extract the key. Transmitter = Receiver = PUF. To reconstruct the key, side information (i.e., helper data) is stored to correct errors in the regenerated bits. The reconstructed secret key can be used in a one-time pad scenario to provide unconditional security or authentication. Onur G¨ unl¨ u — Secrecy via Physical Unclonable Functions (PUFs) 2/4

Institute for Communications Engineering Technische Universit¨ at M¨ unchen Discrete Cosine Transform (DCT)-based RO PUFs f C q Q c1 f1 q Gray(q ,L ) Concatenation 1 1 1 Type IV RO Array Gray B Quantization N x M - DCT N x M Coding Gray(q ,L ) qK K K c512 f 512 Figure 1: Secret-key Generation via DCT-based RO PUFs Main Benefits High uniqueness in RO PUF outputs by eliminating systematic and spatial variations; A trade-off between the number of bits generated and error probability of PUF outputs; Increased reliability by decreasing sensitivity to temperature and voltage variations. Onur G¨ unl¨ u — Secrecy via Physical Unclonable Functions (PUFs) 3/4

Institute for Communications Engineering Technische Universit¨ at M¨ unchen Reliability of the DCT Coefficients Define the average bit-error probability per extracted bit as the distortion measure for each DCT coefficient. Extract L bits from a coefficient if the distortion D ( L ) is less than the predefined error probability p b . 0.12 Mean Number of Errors D(L) 0.1 0.08 0.06 p b = 8/128 0.04 0.02 0 2 4 6 8 10 12 Number of Bits L Figure 2: Average bit-error probability per extracted bit Thank you! Onur G¨ unl¨ u — Secrecy via Physical Unclonable Functions (PUFs) 4/4