FHERMA: Building a Practical Library for Fully Homomorphic Encryption

Fully Homomorphic Encryption (FHE) is a transformative cryptographic technique that allows computations to be performed directly on encrypted data—without ever needing to decrypt it. This opens the door to a wide range of privacy-preserving applications across domains like finance, machine learning, healthcare, and blockchain.

However, despite its promise, FHE remains largely inaccessible to most developers. The reason is simple: FHE is still too hard to use in practice.

The Problem

Today, building FHE applications typically requires deep expertise in cryptography, numerical methods, and low-level optimizations. Developers must work directly with encryption schemes, manage ciphertext noise growth, choose appropriate parameters, and often reinvent basic operations like comparison in encrypted form.

This complexity has resulted in three major barriers to broader adoption:

1. Lack of high-level FHE components — Developers must build everything from scratch.
2. Heavy computational overhead — FHE operations are often orders of magnitude slower than plaintext.
3. Lack of standardized tools or benchmarks — It's hard to know if your implementation is secure, efficient, or even correct.

The Need for a Component Layer

A critical step toward realizing the full potential of Fully Homomorphic Encryption lies in the introduction of a Сomponent Layer—a library of reusable, high-level operations that abstract away the complexity of underlying cryptographic mechanisms.

This layer serves a similar role to what TensorFlow and PyTorch have done for machine learning: offering developers ready-to-use, optimized building blocks that enable them to focus on application logic rather than the low-level details of encryption.

A well-designed FHE component library would allow developers to:

• Rely on trusted, community-reviewed building blocks (e.g., encrypted matrix multiplication, sorting, logistic regression)
• Concentrate on functional and application-level development instead of cryptographic engineering
• Share and reuse optimized solutions across projects and domains
• Leverage expert-driven performance improvements and security assurances

The FHERMA Approach: Community-Driven and Challenge-Based

To build this component library, a dedicated platform—FHERMA—was created by Fair Math and OpenFHE teams. FHERMA is an open infrastructure designed to crowdsource the development of FHE components through structured, merit-based challenges and transparent benchmarking.

Rather than developing the library within a single organization, FHERMA engages the broader global community of FHE practitioners—researchers, developers, students, and engineers—to contribute and evaluate solutions collaboratively.

The process is organized as follows:

• FHERMA publishes a set of well-defined challenges (e.g., “efficient encrypted matrix multiplication”)
• Participants submit implementations, which are automatically evaluated for correctness,security, and performance
• Top-performing solutions are released under the Apache 2.0 license and integrated into the open-source FHE component library Polycircuit
• To ensure full transparency and reproducibility, all test data and secret keys are published at the conclusion of each challenge

There are two challenges types :

• Black-box challenges, where only the encrypted outputs are submitted (preserving the confidentiality of the implementation)
• White-box challenges, where full source code is submitted to enable benchmarking and deeper analysis

Why This Model Is Effective

The competition-based, community-driven approach taken by FHERMA provides several distinct advantages:

• Efficiency: Through open participation, a broad range of implementation strategies is explored, often surpassing the quality of compiler-generated solutions
• Openness: All accepted contributions are released under permissive licensing, with no restrictive agreements or intellectual property transfer
• Ecosystem Development: The platform is designed to welcome both seasoned experts and new contributors, offering educational materials and an accessible evaluation process
• Scalability and Flexibility: The modular challenge system supports a variety of FHE schemes, including CKKS, BFV, BGV, and TFHE, and is designed to grow as new techniques and use cases emerge

The platform currently supports wide range of FHE libraries and additional tools, including OpenFHE, Lattigo, Apple Homorphic Encryption Library, HE Layers from IBM.

Toward a Practical FHE Ecosystem

FHERMA is more than just a competition platform. It serves as the foundation for building a sustainable, community-aligned FHE development ecosystem. By aligning academic openness with the structure of professional software engineering, the platform aims to:

• Lower the barrier to entry for FHE adoption
• Encourage best practices and promote reproducible research
• Accelerate the deployment of scalable, production-grade FHE applications in machine learning, blockchain, and financial services

While the project is still in its early stages, the initial results are promising. Winning solutions to early challenges—such as encrypted matrix multiplication and logistic function evaluation—have already been published and incorporated into the Polycircuit library. New challenges and use cases are continuously being added as the platform evolves.

FHERMA Cookbook: A Growing Library of Practical FHE Components

As part of the broader mission to make Fully Homomorphic Encryption (FHE) accessible and usable in real-world systems, the FHERMA initiative has published the first edition of the FHERMA Cookbook — a curated collection of reusable, high-level FHE components.

📄 Read the publication: https://eprint.iacr.org/2025/1302

The cookbook includes the initial series of components developed through formal challenges on the FHERMA platform. Each component solves a concrete task—such as encrypted comparison, sorting, or matrix operations—under clearly defined cryptographic and performance constraints.

What makes this effort unique is that the cookbook is not the product of a single organization. Instead, it brings together contributions from over 20 researchers across more than 10 leading international institutions, making it one of the first large-scale, community-driven resources of its kind in the FHE field.

The cookbook serves several key goals:

• Provide a reusable and extensible component base for FHE applications
• Offer detailed documentation of design strategies and trade-offs
• Establish a benchmarkable reference for evaluating performance and correctness
• Act as a foundation for standardization in component-level FHE development

This is just the beginning. The cookbook is a living document and will be continuously expanded as new challenges are introduced and solved through the FHERMA platform.

For developers and researchers looking to build privacy-preserving applications using FHE, the cookbook offers both a starting point and a growing knowledge base built by the community.

What’s Next

FHERMA continues to evolve as new challenges are launched and new contributors join. The platform is designed to be extensible: new schemes, new hardware backends, and new AI applications can be addressed through future iterations.

In parallel, Fair Math is building a broader infrastructure layer - this larger project includes advanced cryptographic infrastructure and privacy-native frameworks designed specifically to empower AI and decentralized systems at scale, bridging the gap between cutting-edge theory and practical application.

Together, these efforts will help unlock a future where privacy-preserving AI and decentralized technologies are the norm, not the exception.

Conclusion

For FHE to achieve widespread adoption, the field must move beyond theoretical breakthroughs toward practical usability. This requires a combination of well-engineered tools, reusable components, and an active, open community.

FHERMA represents a major step in that direction. By crowdsourcing both the development and benchmarking of reusable FHE components, the platform enables collective progress in building a foundational layer of infrastructure for privacy-preserving computation.

Those interested in contributing or following the ongoing challenges can visit the platform at: https://fherma.io