Intel’s Fully Homomorphic Encryption Chip Could Revolutionize Privacy

Intel’s hardware-accelerated Fully-Homomorphic Encryption chip, Heracles, could bring fully E2EE server-side processing into viability.

Many of us send our data off to servers for processing everyday, whether it’s for directions in Google Maps or searching for a nearby restaurant. This data is processed by servers in the clear and all of these services have full visibility into everything we send them.

Some promise they don’t store info on us, but that’s only a promise and not backed up by any technical guarantees.

There have been attempts at reducing the trust needed, such as the rise of TEEs in servers.

A Trusted Execution Environment is a segregated area of memory and CPU that's protected from the rest of the CPU by using encryption. Any code outside that environment can't read or tamper with the data in the TEE. Authorized code can manipulate the data inside the TEE.

These are meant to reduce the “trusted computing base” to be as small as possible so there’s not much of a chance of data leaking out where it’s not meant to go.

However, the data is still decrypted and processed in the clear, and there have been vulnerabilities in the past that caused the data within the TEE to leak out to attackers with physical access to the hardware.

To ensure that data can’t be leaked even with physical access, you need proper E2EE similar to how messengers like Signal work where the data is never decrypted by any server.

Enter: Fully Homomorphic Encryption. FHE allows encrypted data to be processed server-side without ever decrypting it. Because of the type of encryption it uses, it's even quantum-resistant.

The catch? It’s thousands of times slower than processing the data normally.

This problem means that although FHE has existed for years and years, it hasn't been practical for most applications.

Intel's work began 5 years ago with the DARPA program Data Protection in Virtual Environments (DRIVE) to research hardware that could accelerate FHE.

The chips are optimized to perform computations in parallel, utilizing single instruction multiple data (SIMD) compute engines and high-bandwidth memory to quickly link the processors together. They compute synchronously, so the chips don't get stuck waiting for each other to finish.

It all adds up to a huge speed increase over traditional CPUs. According to the IEEE Spectrum article, "across seven key operations, Heracles was 1,074 to 5,547 times as fast."

Of course, for FHE to take off there needs to be support at all levels. Duality Technology focuses more on the software side of FHE.

Optalsys is a company looking to move away from the limits of traditional computers and utilize photonics, computing with light, to speed up FHE even more.

Some incredible progress has been made in the world of FHE. It could be the case that in just a few short years, it'll be the norm to make fully E2EE queries to Google or ask ChatGPT for dinner ideas in a fully E2EE manner.