TREBUCHET: Fully Homomorphic Encryption Accelerator for Deep Computation¶
Authors: David Bruce Cousins; Yuriy Polyakov; Ahmad Al Badawi; Matthew French; Andrew Schmidt; Ajey Jacob; Benedict Reynwar; Kellie Canida; Akhilesh Jaiswal; Clynn Mathew; Homer Gamil; Negar Neda; Deepraj Soni; Michail Maniatakos; Brandon Reagen; Naifeng Zhang; Franz Franchetti; Patrick Brinich; Jeremy Johnson; Patrick Broderick; Mike Franusich; Bo Zhang; Zeming Cheng; Massoud Pedram
Year: 2023
Venue: arXiv 2023
type: accelerator collection: catalogue, sok-references status: auto
Links: arXiv:2304.05237
Summary¶
Auto-extracted summary (unreviewed)
Describes the TREBUCHET project under the DARPA DPRIVE program, which designs an FHE hardware coprocessor built from tiles of highly parallel ALUs for vectorized 128-bit modular arithmetic, targeting BGV, BFV, CKKS and FHEW at >=128-bit security and integrating with the PALISADE and OpenFHE libraries. This first-phase paper covers the LAWS ALU and ring processing unit (RPU) design, a FULCRUM architecture-generation flow, SPIRAL-generated NTT code mapped to the RPU ISA, and emulation of a single 11.83M-gate RPU on a Palladium system for NTT, ring multiply/add, modulus switching and basis extension kernels up to 64K points. A BGV logistic-regression training workload is modeled to estimate runtime, and chip I/O bandwidth of 62.5-250 GB/s is projected.
Abstract¶
Secure computation is of critical importance to not only the DoD, but across financial institutions, healthcare, and anywhere personally identifiable information (PII) is accessed. Traditional security techniques require data to be decrypted before performing any computation. When processed on untrusted systems the decrypted data is vulnerable to attacks to extract the sensitive information. To address these vulnerabilities Fully Homomorphic Encryption (FHE) keeps the data encrypted during computation and secures the results, even in these untrusted environments. However, FHE requires a significant amount of computation to perform equivalent unencrypted operations. To be useful, FHE must significantly close the computation gap (within 10x) to make encrypted processing practical. To accomplish this ambitious goal the TREBUCHET project is leading research and development in FHE processing hardware to accelerate deep computations on encrypted data, as part of the DARPA MTO Data Privacy for Virtual Environments (DPRIVE) program. We accelerate the major secure standardized FHE schemes (BGV, BFV, CKKS, FHEW, etc.) at >=128-bit security while integrating with the open-source PALISADE and OpenFHE libraries currently used in the DoD and in industry. We utilize a novel tile-based chip design with highly parallel ALUs optimized for vectorized 128b modulo arithmetic. The TREBUCHET coprocessor design provides a highly modular, flexible, and extensible FHE accelerator for easy reconfiguration, deployment, integration and application on other hardware form factors, such as System-on-Chip or alternate chip areas.
Source: arxiv
Extracted details¶
| Field | Value | Source | Confidence |
|---|---|---|---|
| contribution | tile-based FHE coprocessor design with parallel ALUs optimized for vectorized 128-bit modular arithmetic | llm | 0.75 |
| contribution | LAWS ALU and ring processing unit (RPU) with design-space exploration via the FULCRUM modeling and generation flow | llm | 0.75 |
| contribution | integration with the open-source PALISADE and OpenFHE libraries | llm | 0.75 |
| contribution | SPIRAL-generated NTT kernels translated to the RPU ISA | llm | 0.75 |
| contribution | Palladium emulation of a single RPU and workload modeling of BGV logistic regression training with functional bootstrapping | llm | 0.75 |
| contribution | chip I/O bandwidth modeling | llm | 0.75 |
| schemes | BGV | llm | 0.75 |
| schemes | BFV | llm | 0.75 |
| schemes | CKKS | llm | 0.75 |
| schemes | FHEW | llm | 0.75 |
| libraries | OpenFHE | llm | 0.75 |
| libraries | PALISADE | llm | 0.75 |
| libraries | SPIRAL | llm | 0.75 |
| models | logistic regression | llm | 0.75 |
| tasks | FHE hardware acceleration | llm | 0.75 |
| tasks | logistic regression training (workload modeling) | llm | 0.75 |
| hardware | GF12LP ASIC multi-project wafer (150 mm2 target) | llm | 0.75 |
| hardware | Palladium emulation system | llm | 0.75 |
| security level | >=128-bit | llm | 0.75 |
| threat model | outsourced computation on untrusted systems | llm | 0.75 |
| interaction | non-interactive FHE | llm | 0.75 |
| code url | https://github.com/Microsoft/SEAL (URL found in the full text) | regex | 0.50 |
| limitations | results are from Palladium emulation of a single RPU; full-device performance is projected rather than measured | llm | 0.75 |
| limitations | full-device scaling, I/O and peripheral integration, and fabrication are part of an ongoing second phase | llm | 0.75 |
| limitations | FHE ciphertext expansion (e.g. 4 B to >20 KB per integer) makes data I/O transfer a bottleneck | llm | 0.75 |
| llm relevance | low | llm | 0.75 |
| key result | RPU gate count: 11.83M gates (ALU 113K gates, lane 137K gates) (single RPU emulated on Palladium) | llm | 0.75 |
| key result | kernel sizes verified: 1024, 16K and 64K points for Switch Modulus, Fast Basis Extension, RingMul, RingAdd, NTT (Palladium emulation) | llm | 0.75 |
| key result | projected chip I/O bandwidth: 62.5-250 GB/s (LVDS at SDR / DDR) | llm | 0.75 |
| key result | functional bootstraps per training epoch: 51 (33 sign/compare + 18 rescaling) (BGV logistic regression, ring size 2^16, 64 SIMD slots, depth 2) | llm | 0.75 |
| key result | target FHE-vs-plaintext computation gap: within 10x (project goal) | llm | 0.75 |
System card: Trebuchet¶
| Group | 8. FHE hardware accelerators |
| Family | ASIC (DARPA DPRIVE) |
| Arithmetic | - |
| Interaction | Hardware |
| Model scale | - |
| Status | Active |
| Relation to the SoK | DARPA DPRIVE accelerator for deep computation. |
| Note | Cited by the SoK as evidence for four to five orders of magnitude speedup from custom chips. |
| Systems page | Trebuchet |
SoK reference [104]¶
D. B. Cousins, Y. Polyakov, A. A. Badawi, M. French, A. Schmidt, A. Jacob, B. Reynwar, K. Canida, A. Jaiswal, C. Mathew et al., “Trebuchet: Fully homomorphic encryption accelerator for deep computation,” arXiv preprint arXiv:2304.05237, 2023.
Listed on the reference page of the site.
Keyword mentions¶
Counts of vocabulary terms in the full text (first page of occurrence in brackets).
- scheme: BGV 4 [p.1], CKKS 3 [p.1], BFV 2 [p.1], FHEW 2 [p.1]
- library: OpenFHE 15 [p.1], PALISADE 10 [p.1], Microsoft SEAL 3 [p.4], HElib 1 [p.4]
- model: Logistic regression 7 [p.5], CNN 2 [p.1]
- hardware: ASIC 5 [p.1], FPGA 3 [p.1], ARM 1 [p.1], GPU 1 [p.1]
- technique: NTT 20 [p.2], Rescaling 5 [p.5], Sigmoid 3 [p.5], Bootstrapping 2 [p.5], SIMD packing 2 [p.5], Convolution 2 [p.2], Compiler 2 [p.2]
- assumption: Lattice-based 7 [p.1], RLWE 2 [p.3]
- topic: Homomorphic encryption 31 [p.1], Open source 4 [p.1], Benchmark 3 [p.6]
Files¶
| Path (under papers/) | Collection | Pages | Size | SHA-256 | Note |
|---|---|---|---|---|---|
references/catalogue/8-117_trebuchet-fully-homomorphic-encryption-accelerator-for-deep.pdf |
catalogue | 6 | 963.8 KB | 98b624b28283 |
|
references/sok-references/104_trebuchet-fully-homomorphic-encryption-accelerator-for-deep.pdf |
sok-references | 6 | 963.8 KB | 98b624b28283 |
byte-identical to references/catalogue/8-117_trebuchet-fully-homomorphic-encryption-accelerator-for-deep.pdf |
Database key: trebuchet-fully-homomorphic-encryption-accelerator-for-deep-computation · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set trebuchet-fully-homomorphic-encryption-accelerator-for-deep-computation COLUMN VALUE