International Association
for Cryptologic Research

<p>Numerous applications in homomorphic encryption require an operation that moves the slots of a ciphertext to the coefficients of a different ciphertext. For the BGV and BFV schemes, the only efficient algorithms to implement this slot-to-coefficient transformation were proposed in the setting of non-power-of-two cyclotomic rings. In this paper, we devise an FFT-like method to decompose the slot-to-coefficient transformation (and its inverse) for power-of-two cyclotomic rings. The proposed method can handle both fully and sparsely packed slots. Our algorithm brings down the computational complexity of the slot-to-coefficient transformation from a linear to a logarithmic number of FHE operations, which is shown via a detailed complexity analysis.</p><p>The new procedures are implemented in Microsoft SEAL for BFV. The experiments report a speedup of up to 44 times when packing 2^12 elements from GF(8191^8). We also study a fully packed bootstrapping operation that refreshes 2^15 elements from GF(65537) and obtain an amortized speedup of 12 times. </p>

In this paper, we perform a systematic study of functions $f: \mathbb{Z}_{p^e} \to \mathbb{Z}_{p^e}$ and categorize those functions that can be represented by a polynomial with integer coefficients. More specifically, we cover the following properties: necessary and sufficient conditions for the existence of an integer polynomial representation; computation of such a representation; and the complete set of equivalent polynomials that represent a given function. As an application, we use the newly developed theory to speed up bootstrapping for the BGV and BFV homomorphic encryption schemes. The crucial ingredient underlying our improvements is the existence of null polynomials, i.e. non-zero polynomials that evaluate to zero in every point. We exploit the rich algebraic structure of these null polynomials to find better representations of the digit extraction function, which is the main bottleneck in bootstrapping. As such, we obtain sparse polynomials that have 50% fewer coefficients than the original ones. In addition, we propose a new method to decompose digit extraction as a series of polynomial evaluations. This lowers the time complexity from $\mathcal{O}(\sqrt{pe})$ to $\mathcal{O}(\sqrt{p}\sqrt[^4]{e})$ for digit extraction modulo $p^e$, at the cost of a slight increase in multiplicative depth. Overall, our implementation in HElib shows a significant speedup of a factor up to 2.6 over the state-of-the-art.

We unify the state-of-the-art bootstrapping algorithms for BGV and BFV in a single framework and show that both schemes can be bootstrapped with identical complexity. This result corrects a claim by Chen and Han (Eurocrypt 2018) that BFV is more efficient to bootstrap than BGV. We also fix an error in their optimized procedure for power-of-two cyclotomics, which occurs for some parameter sets. Our analysis is simpler, yet more general than earlier work, in that it simultaneously covers both BGV and BFV. Furthermore, we also design and implement a high-level open source software library for bootstrapping in the Magma Computer Algebra System. It is the first library to support both BGV and BFV bootstrapping in full generality, with all recent techniques (including the above fixes) and trade-offs.

Fully Homomorphic Encryption (FHE) allows for secure computation on encrypted data. Unfortunately, huge memory size, computational cost and bandwidth requirements limit its practicality. We present BASALISC, an architecture family of hardware accelerators that aims to substantially accelerate FHE computations in the cloud. BASALISC is the first to implement the BGV scheme with fully-packed bootstrapping – the noise removal capability necessary for arbitrary-depth computation. It supports a customized version of bootstrapping that can be instantiated with hardware multipliers optimized for area and power.BASALISC is a three-abstraction-layer RISC architecture, designed for a 1 GHz ASIC implementation and underway toward 150mm2 die tape-out in a 12nm GF process. BASALISC’s four-layer memory hierarchy includes a two-dimensional conflict-free inner memory layer that enables 32 Tb/s radix-256 NTT computations without pipeline stalls. Its conflict-resolution permutation hardware is generalized and re-used to compute BGV automorphisms without throughput penalty. BASALISC also has a custom multiply-accumulate unit to accelerate BGV key switching.The BASALISC toolchain comprises a custom compiler and a joint performance and correctness simulator. To evaluate BASALISC, we study its physical realizability, emulate and formally verify its core functional units, and we study its performance on a set of benchmarks. Simulation results show a speedup of more than 5,000× over HElib – a popular software FHE library.