Analysis of Image-Sensor Based Random Number Generators

Abstract

Random number generators (RNGs) generate output based on an entropy seed which can be generated by sampling natural processes. Quantum RNGs (QRNGs) generate true random numbers by extracting entropy from quantum systems that are inherently probabilistic. One way to do this is to collect quantum entropy from signals generated by complementary metal-oxide-semiconductor (CMOS) Image Sensors (CISs) while detecting photons emitted through radiative recombination in light-emitting diodes (LEDs). Here, we propose a framework for determining and setting up an RNG based on spontaneous emission and shot noise due to photon absorption using affordable commercial-off-the-shelf (COTS) CISs and LEDs. To verify the entropy of such RNG, we developed a performance analysis methodology based on the second-order correlation function, cross-correlation, and mutual information to study the spatial correlations on the CIS output. Our research makes implementing RNGs using COTS components easy, thereby increasing their adoption and use in various applications. We applied our methodology using COTS components and compared our results against the NIST SP 800-90B entropy estimation suite. Furthermore, we extracted the entropy using the Toeplitz-hashing function to generate truly random numbers. We tested more than 140 GB of random data using the Dieharder testing suite and passed all statistical tests.