Abstract. We present both a hardware and a software implementation variant of the learning with errors (LWE) based cryptosystem presented by Lindner and Peikert. This work helps in assessing the practicality of lattice-based encryption. For the software implementation, we give a comparison between a matrix and polynomial based variant of the LWE scheme. This module includes multiplication in polynomial rings using Fast Fourier Transform (FFT). In order to implement lattice-based cryptography in an efficient way, it is crucial to apply the systems over polynomial rings. FFT speeds up multiplication in polynomial rings, which is the most critical operation in lattice-based cryptography, from quadratic to quasi-linear runtime. For the hardware variant, we show how this fundamental building block of lattice-based cryptography can be implemented and evaluated in terms of performance. A second important component for lattice-based cryptosystems is the sampling from discrete Gaussian distributions. We examine three different variants for sampling Gaussian distributed integers, namely rejection sampling, a rounding based approach, and a look-up table based approach in hardware.

Abstract. We present a GPU implementation of the Simple Sampling Reduction (SSR) algorithm that searches for short vectors in lattices. SSR makes use of the famous BKZ algorithm. It complements an exhaustive search in a suitable search region to insert random, short vectors to the lattice basis. The sampling of short vectors can be executed in parallel.Our GPU implementation increases the number of sampled vectors per second from 5200 to more than 120, 000. With this we are the first to present a parallel implementation of SSR and we make use of the computing capability of modern graphics cards to enhance the search for short vectors even more.