New compiler brings fine-grained optimization to privacy-preserving encryption

Researchers have introduced Recifhe, a novel compiler designed to optimize programs built on fully homomorphic encryption (FHE)—a cryptographic technique that allows computations to run directly on encrypted data without decryption.

The optimization gap#

Current FHE compilers, including those targeting advanced schemes like RNS-CKKS, work at a coarse-grained level: they manage individual ciphertext operations, each of which internally runs a series of polynomial operations. This approach misses opportunities to eliminate redundant work that spans across multiple ciphertext operations, leaving performance on the table.

How Recifhe works#

Recifhe operates on two levels simultaneously. At the ciphertext level, it takes a standard (non-FHE) program and transforms it into an FHE-compatible version by inserting the necessary encryption management operations and applying global optimizations. At the polynomial level—the granular layer below—Recifhe identifies and removes duplicate polynomial computations that occur across different ciphertext operations.

Performance gains#

The dual-level optimization strategy delivers measurable improvements: Recifhe achieves approximately 1.25x speedup compared to systems that only optimize at the ciphertext level. This boost comes from eliminating the computational waste that conventional single-level compilers overlook.

Why it matters#

As FHE schemes become more practical for privacy-sensitive applications—ranging from secure cloud computing to confidential data analysis—improving their performance is crucial. Recifhe's approach of looking deeper into the computational pipeline demonstrates how compiler techniques can unlock efficiency gains at multiple abstraction layers simultaneously.