SHF: Small: A New Approach for Hardware Design of High-Precision Discrete Gaussian Sampling

North Carolina State University, Raleigh NC

Investigators

Abstract

New cryptographic systems can improve the security level and privacy assurance of everyday applications. These systems, however, often require new building blocks that did not exist earlier. High-precision sampling from Gaussian distributions is one such time-consuming and critical building block that has to be implemented efficiently with customized hardware solutions. The statistical accuracy of the resulting solutions must be guaranteed to abide by cryptographic security requirements. Moreover, the developed solutions must be flexible enough to support multiple current and potential future applications. Existing hardware designs cannot meet these conditions. This project will address this problem by a cross-cutting approach on hardware design, algorithms, statistics, and cryptography. Successful completion of the proposed activity will be a significant step towards automated solutions for provably-secure sampler designs. The project will help workforce development in the area of hardware security in collaboration with several other universities. The results of the project will be made open-source and will potentially be made available to other agencies, e.g., NIST. In this project, new sampler hardware designs will be developed to support lattice-based cryptographic systems such as those in post-quantum cryptography and homomorphic-encryption applications. Although sampling from uniform distributions has been thoroughly studied in the context of cryptography and other applications, there are significantly fewer works on sampling from non-uniform distributions such as the Gaussian distribution, which is unique to lattice-based cryptosystems. To that end, the project will first explore novel algorithmic approaches that simplify the random-search process used in Gaussian-sampling techniques. The project will then explore the approximation techniques that optimize the search while minimizing the impact on statistical deviations. Furthermore, the project will seek full design-automation solutions that can create optimized sampling hardware for a given set of parameters. The resulting solutions will be mapped to reconfigurable hardware and benchmarked against earlier proposals to compare the overheads and savings. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.