Practical Entropy Accumulation for Random Number Generators with Image Sensor-Based Quantum Noise Sources.

The efficient generation of high-quality random numbers is essential in the operation of cryptographic modules. The quality of a random number generator is evaluated by the min-entropy of its entropy source. The typical method used to achieve high min-entropy of the output sequence is an entropy accumulation based on a hash function. This is grounded in the famous Leftover Hash Lemma, which guarantees a lower bound on the min-entropy of the output sequence. However, the hash function-based entropy accumulation has slow speed in general. For a practical perspective, we need a new efficient entropy accumulation with the theoretical background for the min-entropy of the output sequence. In this work, we obtain the theoretical bound for the min-entropy of the output random sequence through the very efficient entropy accumulation using only bitwise XOR operations, where the input sequences from the entropy source are independent. Moreover, we examine our theoretical results by applying them to the quantum random number generator that uses dark shot noise arising from image sensor pixels as its entropy source.

Accelerated implementation for testing IID assumption of NIST SP 800-90B using GPU.

In cryptosystems and cryptographic modules, insufficient entropy of the noise sources that serve as the input into random number generator (RNG) may cause serious damage, such as compromising private keys. Therefore, it is necessary to estimate the entropy of the noise source as precisely as possible. The National Institute of Standards and Technology (NIST) published a standard document known as Special Publication (SP) 800-90B, which describes the method for estimating the entropy of the noise source that is the input into an RNG. The NIST offers two programs for running the entropy estimation process of SP 800-90B, which are written in Python and C++. The running time for estimating the entropy is more than one hour for each noise source. An RNG tends to use several noise sources in each operating system supported, and the noise sources are affected by the environment. Therefore, the NIST program should be run several times to analyze the security of RNG. The NIST estimation runtimes are a burden for developers as well as evaluators working for the Cryptographic Module Validation Program. In this study, we propose a GPU-based parallel implementation of the most time-consuming part of the entropy estimation, namely the independent and identically distributed (IID) assumption testing process. To achieve maximal GPU performance, we propose a scalable method that adjusts the optimal size of the global memory allocations depending on GPU capability and balances the workload between streaming multiprocessors. Our GPU-based implementation excluded one statistical test, which is not suitable for GPU implementation. We propose a hybrid CPU/GPU implementation that consists of our GPU-based program and the excluded statistical test that runs using OpenMP. The experimental results demonstrate that our method is about 3 to 25 times faster than that of the NIST package.