Vers des générateurs de nombres aléatoires uniformes et gaussiens à très haut débit

Renaud Santoro 1
1 CAIRN - Energy Efficient Computing ArchItectures with Embedded Reconfigurable Resources
Inria Rennes – Bretagne Atlantique , IRISA-D3 - ARCHITECTURE
Abstract : The purpose of a random number generator (RNG) is to produce a random binary sequence independent, unpredictable and uniformly distributed. A RNG is needed in all kinds of applications such as cryptography, telecommunications, computer simulations, testing of VLSI circuits or probabilistic algorithms. The number of embedded applications requiring hardware RNG is constantly increasing, especially in the reconfigurable circuits. In view of the physical design of a RNG, the solution is to use a hybrid generator composed of a truly random generator (TRNG) and a pseudo-random number generator (PRNG). Prove mathematically that a random generator is usually difficult or impossible. RNGs are often evaluated using statistical tests. Evaluation of a TRNG is much more complex than PRNG testing as the TRNG quality is a function of the noise source used in the external environment and the implementation quality within the target circuit. Fischer et al. have proposed one of the first TRNG totally embedded within an FPGA. Since, a large generator number has been presented in the literature. One objective of this thesis was to propose an objective methodology for analyzing the quality of recognized TRNGs in reconfigurable circuits. The evaluation method, fast and efficient is based on the use of hardware statistical tests proposed by the AIS 31 and the FIPS 140-2. The advantage of the method is to evaluate the TRNGs in the same FPGA chip to analyze them in similar conditions of use. The hardware version of AIS 31 allows the TRNG evaluation with a maximum throughput of 84.2 Mb/s. The material cost of it is 4042 LUTs, about 10 % of a Xilinx Virtex 5 FPGA SX50. Moreover, the method is used to find the optimal TRNG data rate for the targeted reconfigurable circuit, showing the TRNG pre-characterization interest. From the previous study, a TRNG is selected for its statistical properties. A high data rate hybrid generator is then proposed. However, despite the quality of the selected generator, the TRNG randomness can vary on time due to possible external attacks on the circuit, changes in external conditions or aging components. The quality of the seed sequences provided by a TRNG is important to ensure the unpredictability of a PRNG. To improve the robustness of a TRNG, online statistical tests must be implemented. Thus, a real-time evaluation of the generator behavior is performed and the TRNG weaknesses are filtered. This study proposes to use an efficient implementation of the FIPS 140-2 battery. The negligible area and the performance allow to control the TRNG quality in real time. A hybrid RNG implemented on a reconfigurable target and within a VLSI circuit is then proposed. The study conducted on the RNG hybrid resulted in an industry partnership with a company specialized in security. Then, in some applications, it is sometimes necessary to generate random numbers following a probability distribution different from a uniform distribution. From the use of the proposed hybrid RNG, a very high data rate gaussian random number generator is proposed. Furthermore, the circuit can generate random numbers according to a Rayleigh law. From previous studies, a hybrid RNG is proposed for FPGA implementation and is embedded into the quantum cryptography solution proposed by the Smart-Quantum company. The RNG generates uniform random bits and his innovating by the TRNG real time monitoring realized by embedded the FIPS 140-2 tests. Then, the realization of two ASIC in 130 nm technology are described. The chips will allow to evaluate the performance of ASIC TRNGs. Moreover, the second circuit is embedded high throughput gaussian and Rayleigh noise generators.
Complete list of metadatas
Contributor : Olivier Sentieys <>
Submitted on : Monday, January 11, 2010 - 10:07:22 AM
Last modification on : Friday, November 16, 2018 - 1:38:15 AM
Long-term archiving on: Wednesday, November 30, 2016 - 10:22:03 AM


  • HAL Id : tel-00438600, version 3


Renaud Santoro. Vers des générateurs de nombres aléatoires uniformes et gaussiens à très haut débit. Traitement du signal et de l'image [eess.SP]. Université Rennes 1, 2009. Français. ⟨tel-00438600v3⟩



Record views


Files downloads