CHESS: Compiling Homomorphic Encryption with Scheme Switching

In recent years, cloud computing has become increasingly prevalent, with the number of services and applications that offload their computations to the cloud growing each year. Many of these services handle users' sensitive data, making security a critical concern. Fully Homomorphic Encryption (FHE) offers a promising solution by enabling computations on encrypted data. However, developing FHE applications remains complex and tedious, as it requires expertise in cryptography to select the correct parameters for secure and efficient evaluation. In addition, contemporary FHE schemes have distinct advantages and disadvantages, with each scheme being suitable for different types of data and operations. Moreover, current scheme switching techniques come with many limitations, which we address by proposing new methods that ensure ciphertext compatibility during mid-execution switching, and also align with state-of-theart methodologies for encrypted computation using the CGGI scheme with minimal overheads. To make scheme switching a viable solution, we introduce the first hybrid FHE compiler that efficiently translates high-level plaintext code into an intermediate representation (IR), facilitating scheme switching between different FHE schemes based on the IR. Unlike previous works that focus solely on single-scheme compilers, our approach stands out by leveraging the strengths of two FHE schemes. We evaluate the efficiency and usability of our scheme switching methodology and demonstrate a practical solution for end users.