International Association
for Cryptologic Research

Discrete Gaussian sampling is an integral part of many lattice

based cryptosystems such as public-key encryption, digital signature schemes and homomorphic encryption schemes. In this paper we propose a compact and fast Knuth-Yao sampler for sampling from a narrow discrete Gaussian distribution with very high precision. The designed samplers have a maximum statistical distance of $2^{-90}$ to a true discrete Gaussian distribution. In this paper we investigate various optimization techniques to achieve minimum area and cycle requirement. For the standard deviation 3.33, the most area-optimal implementation of the bit-scan operation based Knuth-Yao sampler consumes 30 slices on the Xilinx Virtex 5 FPGAs, and requires on average 17 cycles to generate a sample. We improve the speed of the sampler by using a precomputed table that directly maps the initial random bits into samples with very high probability. The fast sampler consumes 35 slices and spends on average 2.5 cycles to generate a sample. However the sampler architectures are not secure against timing and power analysis based attacks. In this paper we propose a random shuffle method to protect the Gaussian distributed polynomial against such attacks. The side channel attack resistant sampler architecture consumes 52 slices and spends on average 420 cycles to

generate a polynomial of 256 coefficients.