BTS: An Accelerator for Bootstrappable Fully Homomorphic Encryption¶
Authors: Sangpyo Kim; Jongmin Kim; Michael Jaemin Kim; Wonkyung Jung; John Kim; Minsoo Rhu; Jung Ho Ahn
Year: 2022
Venue: ISCA 2022
type: accelerator collection: catalogue status: auto
Links: DOI · arXiv:2112.15479
Summary¶
Auto-extracted summary (unreviewed)
BTS is presented as the first hardware accelerator for CKKS FHE that treats bootstrapping as a first-class operation; the authors analyze the off-chip bandwidth and compute demands of bootstrapping, identify parameter sets suited to acceleration under memory-technology limits, and design a grid of processing elements with a deterministic network-on-chip. BTS achieves 5,556x faster execution on ResNet-20 inference and 1,306x on logistic regression than a CPU, with a 373.6 mm^2 chip consuming up to 163.2 W.
Abstract¶
Homomorphic encryption (HE) enables the secure offloading of computations to the cloud by providing computation on encrypted data (ciphertexts). HE is based on noisy encryption schemes in which noise accumulates as more computations are applied to the data. The limited number of operations applicable to the data prevents practical applications from exploiting HE. Bootstrapping enables an unlimited number of operations or fully HE (FHE) by refreshing the ciphertext. Unfortunately, bootstrapping requires a significant amount of additional computation and memory bandwidth as well. Prior works have proposed hardware accelerators for computation primitives of FHE. However, to the best of our knowledge, this is the first to propose a hardware FHE accelerator that supports bootstrapping as a first-class citizen. In particular, we propose BTS - Bootstrappable, Technologydriven, Secure accelerator architecture for FHE. We identify the challenges of supporting bootstrapping in the accelerator and analyze the off-chip memory bandwidth and computation required. In particular, given the limitations of modern memory technology, we identify the HE parameter sets that are efficient for FHE acceleration. Based on the insights gained from our analysis, we propose BTS, which effectively exploits the parallelism innate in HE operations by arranging a massive number of processing elements in a grid. We present the design and microarchitecture of BTS, including a network-on-chip design that exploits a deterministic communication pattern. BTS shows 5,556x and 1,306x improved execution time on ResNet-20 and logistic regression over a CPU, with a chip area of 373.6mm^2 and up to 163.2W of power.
Source: arxiv
Extracted details¶
| Field | Value | Source | Confidence |
|---|---|---|---|
| contribution | first FHE accelerator supporting bootstrapping as a first-class citizen | llm | 0.85 |
| contribution | analysis of off-chip memory bandwidth and computation required by CKKS bootstrapping | llm | 0.85 |
| contribution | identification of HE parameter sets efficient for FHE acceleration under memory-technology limits | llm | 0.85 |
| contribution | BTS microarchitecture with a massive grid of processing elements | llm | 0.85 |
| contribution | network-on-chip design exploiting the deterministic communication pattern of HE operations | llm | 0.85 |
| schemes | CKKS | llm | 0.85 |
| models | ResNet-20 | llm | 0.85 |
| models | logistic regression | llm | 0.85 |
| tasks | FHE hardware acceleration | llm | 0.85 |
| tasks | CKKS bootstrapping | llm | 0.85 |
| tasks | private neural network inference | llm | 0.85 |
| tasks | private logistic regression | llm | 0.85 |
| hardware | BTS ASIC (373.6 mm^2, up to 163.2 W) | llm | 0.85 |
| hardware | CPU baseline | llm | 0.85 |
| security level | 128-bit | llm | 0.85 |
| interaction | non-interactive FHE | llm | 0.85 |
| code url | https://github.com/snucrypto/HEAAN (URL found in the full text) | regex | 0.50 |
| code url | https://github.com/ldsec/lattigo (URL found in the full text) | regex | 0.50 |
| code url | https://github.com/Yongyongha/SparseLWE-estimator (URL found in the full text) | regex | 0.50 |
| llm relevance | low | llm | 0.85 |
| key result | speed-up over CPU: 5,556x (ResNet-20 inference under CKKS FHE) | llm | 0.85 |
| key result | speed-up over CPU: 1,306x (logistic regression under CKKS FHE) | llm | 0.85 |
| key result | chip area: 373.6 mm^2 (BTS accelerator) | llm | 0.85 |
| key result | power: up to 163.2 W (BTS accelerator) | llm | 0.85 |
System card: BTS¶
| Group | 8. FHE hardware accelerators |
| Family | ASIC (CKKS) |
| Arithmetic | - |
| Interaction | Hardware |
| Model scale | - |
| Status | Foundational |
| Relation to the SoK | Accelerator designed around CKKS bootstrapping. |
| Note | Memory bandwidth centered design. |
| Systems page | BTS |
Keyword mentions¶
Counts of vocabulary terms in the full text (first page of occurrence in brackets).
- scheme: CKKS 66 [p.1], MPC 5 [p.2], Leveled / somewhat HE 3 [p.13], BFV 2 [p.2], BGV 2 [p.2]
- library: Lattigo 19 [p.2], PALISADE 2 [p.15], EVA / CHET 1 [p.14]
- model: ResNet 15 [p.1], CNN 4 [p.10], Logistic regression 3 [p.1]
- dataset: MNIST 2 [p.10], CIFAR-10 1 [p.10], ImageNet 1 [p.11]
- hardware: GPU 15 [p.1], FPGA 10 [p.1], ASIC 9 [p.1], NVIDIA A100 2 [p.5], NVIDIA V100 2 [p.10], Intel Xeon 1 [p.10], CUDA 1 [p.13], AVX-512 1 [p.14]
- technique: Bootstrapping 92 [p.1], NTT 57 [p.2], Key switching 18 [p.3], RNS 13 [p.3], Rescaling 6 [p.3], Convolution 6 [p.3], Rotations 4 [p.2], ReLU 3 [p.10], Compiler 3 [p.6], SIMD packing 2 [p.2]
- assumption: LWE 5 [p.1], Lattice-based 4 [p.14]
- topic: Homomorphic encryption 128 [p.1], Open source 3 [p.14]
Files¶
| Path (under papers/) | Collection | Pages | Size | SHA-256 | Note |
|---|---|---|---|---|---|
references/catalogue/8-114_bts-an-accelerator-for-bootstrappable-fully-homomorphic-encr.pdf |
catalogue | 15 | 1.5 MB | fa15c7c34fd9 |
Database key: bts-an-accelerator-for-bootstrappable-fully-homomorphic-encryption · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set bts-an-accelerator-for-bootstrappable-fully-homomorphic-encryption COLUMN VALUE