2025/11/11 by Li Ni, Xijun Huang, Xinhui Chen +8 · 1 citation
Computer Science · Engineering · #Physical Unclonable Functions (PUFs) and Hardware Security #Advanced Memory and Neural Computing #Chaos-based Image/Signal Encryption
paper · doi:10.1109/ted.2025.3627172
Resistive random access memory (RRAM) is a promising nonvolatile memory (NVM) technology, yet its susceptibility to data breaches poses critical challenges for high-security applications such as AI acceleration, secure booting, and trusted execution environments (TEEs). In this work, we present a novel in-memory unified true random number generator (TRNG) and physical unclonable function (PUF) design to strengthen RRAM hardware security and attack resistance, leveraging the intrinsic randomness of RRAM switching time. By exploiting device-to-device (D2D) and cycle-to-cycle (C2C) variations, our design could synchronize PUF and TRNG responses via an RRAM array-based entropy source extraction circuit. Furthermore, a switching time-based recombination scheme, derived from the statistical distribution of switching time, is used to enhance PUF reliability. Experimental (simulation) results demonstrate an energy efficiency of 1.54 (3.6) pJ/bit at a throughput of 33.3 Mb/s, a native PUF bit error rate (BER) of 0.569% (0.307%), and zero errors over 2000 evaluations. Meanwhile, the TRNG passes all NIST statistical tests without postprocessing, achieving a minimum entropy of 0.98 under all conditions.