Paper 2025/2284 Original: eprint.iacr.org/2025/2284 ↗

Meta-PBS: Compact High-Precision Programmable Bootstrapping

Shihe Ma
Tairong Huang
Anyu Wang
Changtong Xu
Tao Wei
Xiaoyun Wang
Abstract

Currently, most FHE schemes realize bootstrapping through the linear-decrypt-then-round paradigm. For the programmable bootstrapping (PBS) of TFHE, this means the lookup table (LUT) needs a redundancy of $O(\sqrt{N})$ to be able to remove the modulus switching noise, which limits the plaintext modulus of PBS to $O(\sqrt{N})$. We remove this requirement for redundancy by proposing the Meta-PBS framework, which allows us to start with under-redundant or non-redundant LUTs. Meta-PBS iteratively blind-rotates the LUT, during which the LUT redundancy gradually increases. The bootstrapping outputs the correct result when the redundancy eventually exceeds the noise bound. Asymptotically, Meta-PBS requires $O(1)$ blind rotations in dimension $N$ to evaluate a negacyclic function modulo $2N$, whereas PBS needs $O(\sqrt{N})$ blind rotations. Meta-PBS also enjoys an additive noise growth, allowing for more homomorphic arithmetic on bootstrapped ciphertext. We modified Meta-PBS to support the simultaneous evaluation of multiple LUTs on the same ciphertext and/or arbitrary LUTs. According to our implementation, when evaluating a 12-bit negacyclic function, Meta-PBS outperforms EBS (PKC'23) by 79 times. When evaluating an arbitrary function on an 8-bit LWE ciphertext, Meta-PBS reduces the running time of the Refined LHE (CCS'25) by half while allowing for a 27 times larger post-bootstrap linear combination.

Metadata
Available format(s)
PDF
Category
Public-key cryptography
Publication info
Preprint.
History
2026-08-12: last updated
2025-12-19: published
Short URL
https://ia.cr/2025/2284
License
https://creativecommons.org/licenses/by/4.0/

BibTeX

@misc{cryptoeprint:2025/2284,
      author = {Shihe Ma and Tairong Huang and Anyu Wang and Changtong Xu and Tao Wei and Xiaoyun Wang},
      title = {Meta-PBS: Compact High-Precision Programmable Bootstrapping},
      howpublished = {Cryptology {ePrint} Archive, Paper 2025/2284},
      year = {2025},
      url = {https://eprint.iacr.org/2025/2284}
}
Note: In order to protect the privacy of readers, this archive does not use cookies or embedded third party content.