Fully homomorphic encryption is an immensely valuable tool for collaborating on the most sensitive data. Optalysys are developing hardware accelerators to overcome the technical limitations of FHE and enable the full potential of quantum-secure private computing.
One of the most significant avenues for accelerating FHE lies in targeting the efficient multiplication of encrypted data. As a fundamental part of how they work, all FHE implementations carry a need to repeatedly perform very large vector-polynomial multiplications.
To make this more efficient, all FHE schemes use a method in which these vectors are transformed into a form in which multiplication is easier and uses fewer operations. While this is more efficient than direct polynomial multiplication, this transformation is itself computationally expensive.
A major aspect of our work in making FHE faster therefore lies in accelerating this conversion beyond what can be achieved within the limits of conventional electronics.
However, different FHE schemes use different versions of this transform process. Some use a method called the Fourier transform, while others use a related process known as a Number-Theoretic Transform (NTT).
The NTT is also used extensively in the KYBER key encapsulation scheme selected by NIST to provide post-quantum internet security, which relies on the same fundamental cryptographic problem as modern FHE schemes.
Providing acceleration for all FHE schemes (and lattice-based cryptography in general) therefore means providing acceleration for both Fourier transforms and NTTs.
Our optical computing technology can be applied to both Fourier and Number-Theoretic transforms, allowing our hardware to provide acceleration for all fully homomorphic encryption schemes and next-generation PQC for internet traffic.
To allow developers to start working with the full range of optically-enabled transforms for lattice-based cryptography, we will be adding this capability to our ongoing beta programme to allow selection of both FT and NTT processing operations.
We will soon be launching our integrations with FHE libraries that make use of both types of transformation in the form of interfaces to the Concrete and PALISADE libraries.