Fully Homomorphic Encryption (FHE) enables computation on encrypted data, holding immense potential for enhancing data privacy and security in various applications. Presently, FHE adoption is hindered by slow computation times, caused by data being encrypted into large polynomials. Optimized FHE libraries and hardware acceleration are emerging to tackle this performance bottleneck. Often, these libraries implement the Number Theoretic Transform (NTT) algorithm for efficient polynomial multiplication. Existing implementations mostly focus on the case where the polynomials are defined over a power-of-two cyclotomic ring, allowing to make use of the simpler Cooley-Tukey NTT. However, generalized cyclotomics have several benefits in the BGV FHE scheme, including more SIMD plaintext slots and a simpler bootstrapping algorithm.We present a hardware architecture for the NTT targeting generalized cyclotomics within the context of the BGV FHE scheme. We explore different non-power-of-two NTT algorithms, including the Prime-Factor, Rader, and Bluestein NTTs. Our most efficient architecture targets the 21845-th cyclotomic polynomial — a practical parameter for BGV — with ideal properties for use with a combination of the Prime-Factor and Rader algorithms. The design achieves high throughput with optimized resource utilization, by leveraging parallel processing, pipelining, and reusing processing elements. Compared to Wu et al.’s VLSI architecture of the Bluestein NTT, our approach showcases 2× to 5× improved throughput and area efficiency. Simulation and implementation results on an AMD Alveo U250 FPGA demonstrate the feasibility of the proposed hardware design for FHE.

Fully Homomorphic Encryption (FHE) is a technique that allows computation on encrypted data. It has the potential to drastically change privacy considerations in the cloud, but high computational and memory overheads are preventing its broad adoption. TFHE is a promising Torus-based FHE scheme that heavily relies on bootstrapping, the noise-removal tool invoked after each encrypted logical/arithmetical operation.

The magic of Fully Homomorphic Encryption (FHE) is that it allows operations on encrypted data without decryption. Unfortunately, the slow computation time limits their adoption. The slow computation time results from the vast memory requirements (64Kbits per ciphertext), a bootstrapping key of 1.3 GB, and sizeable computational overhead (10240 NTTs, each NTT requiring 5120 32-bit multiplications). We accelerate the FHEW bootstrapping in hardware on a high-end U280 FPGA.To reduce the computational complexity, we propose a fast hardware NTT architecture modified from [5] with support for negatively wrapped convolution. The IP module includes large I/O ports to the NTT accelerator and an index bit-reversal block. The total architecture requires less than 225000 LUTs and 1280 DSPs.Assuming that a fast interface to the FHEW bootstrapping key is available, the execution speed of FHEW bootstrapping can increase by at least 7.5 times.

Homomorphic encryption (HE) enables calculating on encrypted data, which makes it possible to perform privacy-preserving neural network inference. One disadvantage of this technique is that it is several orders of magnitudes slower than calculation on unencrypted data. Neural networks are commonly trained using floating-point, while most homomorphic encryption libraries calculate on integers, thus requiring a quantisation of the neural network. A straightforward approach would be to quantise to large integer sizes (e.g. 32 bit) to avoid large quantisation errors. In this work, we reduce the integer sizes of the networks, using quantisation-aware training, to allow more efficient computations. For the targeted MNIST architecture proposed by Badawi et al. [1], we reduce the integer sizes by 33% without significant loss of accuracy, while for the CIFAR architecture, we can reduce the integer sizes by 43%. Implementing the resulting networks under the BFV homomorphic encryption scheme using SEAL, we could reduce the execution time of an MNIST neural network by 80% and by 40% for a CIFAR neural network.

Cryptocurrency mining is an energy-intensive process that presents a prime candidate for hardware acceleration. This work-in-progress presents the first coprocessor design for the ASIC-resistant CryptoNight-Haven Proof of Work (PoW) algorithm. We construct our hardware accelerator as a Xilinx Run Time (XRT) RTL kernel targeting the Xilinx Varium C1100 Blockchain Accelerator Card. The design employs deeply pipelined computation and High Bandwidth Memory (HBM) for the underlying scratchpad data. We aim to compare our accelerator to existing CPU and GPU miners to show increased throughput and energy efficiency of its hash computations.

Modern processors expose software to information leakage through shared microarchitectural state. One of the most severe leakage channels is cache contention, exploited by attacks referred to as PRIME+PROBE, which can infer fine-grained memory access patterns while placing only limited assumptions on attacker capabilities.

We present a domain-specific co-processor to speed up Saber, a post-quantum key encapsulation mechanism competing on the NIST Post-Quantum Cryptography standardization process. Contrary to most lattice-based schemes, Saber doesn’t use NTT-based polynomial multiplication. We follow a hardware-software co-design approach: the execution is performed on an ARM core and only the most computationally expensive operation, i.e., the polynomial multiplication, is offloaded to the co-processor to obtain a compact design. We exploit the idea of distributed computing at micro-architectural level together with novel algorithmic optimizations to achieve approximately a 6 times speedup with respect to optimized software at a small area cost, which we demonstrate on a Zynq-7000 ARM/FPGA SoC.

Homomorphic Encryption makes privacy preserving computing possible in a third party owned cloud by enabling computation on the encrypted data of users. However, software implementations of homomorphic encryption are very slow on general purpose processors. With the emergence of `FPGAs as a service', hardware-acceleration of computationally heavy workloads in the cloud are getting popular. In this article we propose HEAWS, a domain-specific coprocessor architecture for accelerating homomorphic function evaluation on the encrypted data using high-performance FPGAs available in the Amazon AWS cloud. To the best of our knowledge, we are the first to report hardware acceleration of homomorphic encryption using Amazon AWS FPGAs. Utilizing the massive size of the AWS FPGAs, we design a high-performance and parallel coprocessor architecture for the FV homomorphic encryption scheme which has become popular for computing exact arithmetic on the encrypted data. We design parallel building blocks and apply pipeline processing at different levels of the implementation hierarchy, and on top of such optimizations we instantiate multiple parallel coprocessors in the FPGA to execute several homomorphic computations simultaneously. While the absolute computation time can be reduced by deploying more computational resources, efficiency of the HW/SW communication interface plays an important role in homomorphic encryption as it is computation as well as data intensive. Our implementation utilizes state of the art 512-bit XDMA feature of high bandwidth communication available in the AWS Shell to reduce the overhead of HW/SW data transfer. Moreover, we explore the design-space to identify optimal off-chip data transfer strategy for feeding the parallel coprocessors in a time-shared manner. As a result of these optimizations, our AWS-based accelerator can perform 613 homomorphic multiplications per second for a parameter set that enables homomorphic computations of depth 4. Finally, we benchmark an artificial neural network for privacy-preserving forecasting of energy consumption in a Smart Grid application and observe five times speed up.

Homomorphic encryption is a tool that enables computation on encrypted data and thus has applications in privacy-preserving cloud computing. Though conceptually amazing, implementation of homomorphic encryption is very challenging and typically software implementations on general purpose computers are extremely slow. In this paper we present our year long effort to design a domain specific architecture in a heterogeneous Arm+FPGA platform to accelerate homomorphic computing on encrypted data. We design a custom co-processor for the computationally expensive operations of the well-known Fan-Vercauteren (FV) homomorphic encryption scheme on the FPGA, and make the Arm processor a server for executing different homomorphic applications in the cloud, using this FPGA-based co-processor. We use the most recent arithmetic and algorithmic optimization techniques and perform designspace exploration on different levels of the implementation hierarchy. In particular we apply circuit-level and block-level pipeline strategies to boost the clock frequency and increase the throughput respectively. To reduce computation latency, we use parallel processing at all levels. Starting from the highly optimized building blocks, we gradually build our multi-core multi-processor architecture for computing. We implemented and tested our optimized domain specific programmable architecture on a single Xilinx Zynq UltraScale+ MPSoC ZCU102 Evaluation Kit. At 200 MHz FPGA-clock, our implementation achieves over 13x speedup with respect to a highly optimized software implementation of the FV homomorphic encryption scheme on an Intel i5 processor running at 1.8 GHz.

This article describes a field-programmable gate array (FPGA) cryptographic architecture, which combines the elliptic curve--based Ed25519 digital signature algorithm and the X25519 key establishment scheme in a single module. Cryptographically, these are high-security elliptic curve cryptography algorithms with short key sizes and impressive execution times in software. Our goal is to provide a lightweight FPGA module that enables them on resource-constrained devices, specifically for Internet of Things (IoT) applications. In addition, we aim at extensibility with customisable countermeasures against timing and differential power analysis side-channel attacks and fault-injection attacks. For the former, we offer a choice between time-optimised versus constant-time execution, with or without Z -coordinate randomisation and base-point blinding; and for the latter, we offer enabling or disabling default-case statements in the Finite State Machine (FSM) descriptions. To obtain compactness and at the same time fast execution times, we make maximum use of the Digital Signal Processing (DSP) slices on the FPGA. We designed a single arithmetic unit that is flexible to support operations with two moduli and non-modulus arithmetic. In addition, our design benefits in-place memory management and the local storage of inputs into DSP slices’ pipeline registers and takes advantage of distributed memory. These eliminate a memory access bottleneck. The flexibility is offered by a micro-code supported instruction-set architecture. Our design targets 7-Series Xilinx FPGAs and is prototyped on a Zynq System-on-Chip (SoC). The base design combining Ed25519 and X25519 in a single module, and its implementation requires only around 11.1K Lookup Tables (LUTs), 2.6K registers, and 16 DSP slices. Also, it achieves performance of 1.6ms for a signature generation and 3.6ms for a signature verification for a 1024-bit message with an 82MHz clock. Moreover, the design can be optimised only for X25519, which gives the most compact FPGA implementation compared to previously published X25519 implementations.

A Virtual Private Network (VPN) encrypts and decrypts the private traffic it tunnels over a public network. Maximizing the available bandwidth is an important requirement for network applications, but the cryptographic operations add significant computational load to VPN applications, limiting the network throughput. This work presents a coprocessor designed to offer hardware acceleration for these encryption and decryption operations. The open-source SigmaVPN application is used as the base solution, and a coprocessor is designed for the parts of Networking and Cryptography library (NaCl) which underlies the cryptographic operation of SigmaVPN. The hardware-software codesign of this work is implemented on a Xilinx Zynq-7000 SoC, showing a 93% reduction in the execution time of encrypting a 1024-byte frame, and this improved the TCP and UDP communication bandwidths by a factor of 4.36 and 5.36 respectively compared to pure software solution for a 1024-byte frame.