More Like this

1. OFHE: An Electro-Optical Accelerator for Discretized TFHE

   Zheng, Mengxin; Chu, Cheng; Lou, Qian; Youngblood, Nathan; Li, Mo; Moazeni, Sajjad; Jiang, Lei (May 2024, Proceedings International Symposium on Low Power Electronics and Design)

   This paper presents \textit{OFHE}, an electro-optical accelerator designed to process Discretized TFHE (DTFHE) operations, which encrypt multi-bit messages and support homomorphic multiplications, lookup table operations and full-domain functional bootstrappings. While DTFHE is more efficient and versatile than other fully homomorphic encryption schemes, it requires 32-, 64-, and 128-bit polynomial multiplications, which can be time-consuming. Existing TFHE accelerators are not easily upgradable to support DTFHE operations due to limited datapaths, a lack of datapath bit-width reconfigurability, and power inefficiencies when processing FFT and inverse FFT (IFFT) kernels. Compared to prior TFHE accelerators, OFHE addresses these challenges by improving the DTFHE operation latency by 8.7\%, the DTFHE operation throughput by $$57\%$$, and the DTFHE operation throughput per Watt by $$94\%$$. 

   more » « less
2. GPU Acceleration for FHEW/TFHE Bootstrapping

   https://doi.org/10.46586/tches.v2025.i1.314-339  

   Xiao, Yu; Liu, Feng-Hao; Ku, Yu-Te; Ho, Ming-Chien; Hsu, Chih-Fan; Chang, Ming-Ching; Hung, Shih-Hao; Chen, Wei-Chao (December 2024, IACR Transactions on Cryptographic Hardware and Embedded Systems)

   Fully Homomorphic Encryption (FHE) allows computations to be performed directly on encrypted data without decryption. Despite its great theoretical potential, the computational overhead remains a major obstacle for practical applications. To address this challenge, hardware acceleration has emerged as a promising approach, aiming to achieve real-time computation across a wider range of scenarios. In line with this, our research focuses on designing and implementing a Graphic Processing Unit (GPU)-based accelerator for the third generation FHEW/TFHE bootstrapping scheme, which features smaller parameters and bootstrapping keys particularly suitable for GPU architectures compared to the other generations.In summary, our accelerator offers improved efficiency, scalability, and flexibility for extensions, e.g., functional bootstrapping (Liu et al., Asiacrypt 2022), compared to current state-of-the-art solutions. We evaluate our implementation and demonstrate substantial speedup in the single-GPU setting, our bootstrapping achieves an 18x - 20x speedup compared to a 64-thread server-class CPU; by using 8 GPUs, the throughput can be further improved by 7x compared to the single-GPU implementation, confirming the scalability of our design. Furthermore, compared to the SoTA GPU solution TFHE-rs, we achieve a maximum speedup of 1.69x in AND gate evaluation. Finally, we benchmark several private machine learning applications, showing real-time solutions for (1) encrypted neural network inference for MNIST in 0.04 seconds per image, which is the fastest implementation to our knowledge.(2) private decision trees in 0.38 seconds for Iris dataset, where as prior 16 cores CPU implementation (Lu et al., IEEE S&P 2021) required 1.87 seconds; These results highlight the effectiveness and efficiency of our GPU-acceleration in real-world applications.As a technical highlight, we design a novel parallelization strategy tailored for FHEW/TFHE bootstrapping, allowing an automated optimization that partitions bootstrapping into multiple GPU thread blocks. This is necessary for FHEW/TFHE bootstrapping with scalable parameters, where the whole bootstrapping process may not fit into a single thread block. With this, our accelerator can support a broader range of parameters, making it ideal for upcoming privacy-preserving applications. 

   more » « less
3. An FPGA-based Programmable Vector Engine for Fast Fully Homomorphic Encryption over the Torus

   Gener, Serhan; Newton, Parker; Tan, Daniel; Richelson, Silas; Lemieux, Guy; Brisk, Philip (January 2021, SPSL: Secure and Private Systems for Machine Learning (ISCA Workshop))

   This paper describes an FPGA-based vector engine to accelerate the bootstrapping procedure of Fast Fully Homomorphic Encryption over the Torus (TFHE), a popular and high-performance fully homomorphic encryption scheme. Most TFHE bootstraping comprises many matrix-vector operations that are implemented using Torus polynomials, which are not efficiently implemented on today's standard arithmetic hardware. Our implementation achieves linear performance scaling with up to 16 vector lanes. Future work will switch to an FFT-based polynomial multiplication scheme and switch to larger FPGA parts to accommodate more vector lanes. 

   more » « less
4. LightBulb: A Photonic-Nonvolatile-Memory-based Accelerator for Binarized Convolutional Neural Networks

   https://doi.org/10.23919/DATE48585.2020.9116494  

   Zokaee, Farzaneh; Lou, Qian; Youngblood, Nathan; Liu, Weichen; Xie, Yiyuan; Jiang, Lei (March 2020, Design, Automation & Test in Europe Conference & Exhibition)

   Although Convolutional Neural Networks (CNNs) have demonstrated the state-of-the-art inference accuracy in various intelligent applications, each CNN inference involves millions of expensive floating point multiply-accumulate (MAC) operations. To energy-efficiently process CNN inferences, prior work proposes an electro-optical accelerator to process power-of-2 quantized CNNs by electro-optical ripple-carry adders and optical binary shifters. The electro-optical accelerator also uses SRAM registers to store intermediate data. However, electro-optical ripple-carry adders and SRAMs seriously limit the operating frequency and inference throughput of the electro-optical accelerator, due to the long critical path of the adder and the long access latency of SRAMs. In this paper, we propose a photonic nonvolatile memory (NVM)-based accelerator, Light-Bulb, to process binarized CNNs by high frequency photonic XNOR gates and popcount units. LightBulb also adopts photonic racetrack memory to serve as input/output registers to achieve high operating frequency. Compared to prior electro-optical accelerators, on average, LightBulb improves the CNN inference throughput by 17× ~ 173× and the inference throughput per Watt by 17.5 × ~ 660×. 

   more » « less
5. Glyph: Fast and Accurately Training Deep Neural Networks on Encrypted Data

   Qian, Lou; Bo, Feng; Geoffrey, Fox; Lei, Jiang (October 2020, Advances in neural information processing systems)

   null (Ed.)

   Because of the lack of expertise, to gain benefits from their data, average users have to upload their private data to cloud servers they may not trust. Due to legal or privacy constraints, most users are willing to contribute only their encrypted data, and lack interests or resources to join deep neural network (DNN) training in cloud. To train a DNN on encrypted data in a completely non-interactive way, a recent work proposes a fully homomorphic encryption (FHE)-based technique implementing all activations by \textit{Brakerski-Gentry-Vaikuntanathan} (BGV)-based lookup tables. However, such inefficient lookup-table-based activations significantly prolong private training latency of DNNs. In this paper, we propose, Glyph, an FHE-based technique to fast and accurately train DNNs on encrypted data by switching between TFHE (Fast Fully Homomorphic Encryption over the Torus) and BGV cryptosystems. Glyph uses logic-operation-friendly TFHE to implement nonlinear activations, while adopts vectorial-arithmetic-friendly BGV to perform multiply-accumulations (MACs). Glyph further applies transfer learning on DNN training to improve test accuracy and reduce the number of MACs between ciphertext and ciphertext in convolutional layers. Our experimental results show Glyph obtains state-of-the-art accuracy, and reduces training latency by 69%~99% over prior FHE-based privacy-preserving techniques on encrypted datasets. 

   more » « less