Intel demonstrated its Heracles accelerator featuring fully homomorphic encryption (FHE) at the International Solid-State Circuits Conference (ISSCC), unveiling what could be a watershed moment for secure cloud computing. Operating at 1.20 GHz, the new chip is roughly 1,074 to 5,547 times faster than a 24-core Intel Xeon W7-3455 when accelerating FHE mathematical operations.
For nearly two decades, fully homomorphic encryption has existed as a cryptographic ideal with a crippling drawback. FHE allows computing on encrypted data without decryption, but it's currently slow on standard CPUs and GPUs. The performance penalty has been severe enough to confine FHE use to narrow applications, despite its theoretical power to transform how organisations handle sensitive information.
The strategic implications are significant. When a voter queries an election database to verify her ballot was recorded correctly, under current systems the government must decrypt her data to process the request, creating a vulnerability. Heracles demonstrated this use case at ISSCC, simulating a request by a voter to verify her ballot had been registered correctly. The state maintained an encrypted database of voters and their votes, and to protect privacy, the voter encrypted her ID and vote before sending it to the government database. On an Intel Xeon server CPU, the process took 15 milliseconds; verifying 100 million voter ballots adds up to more than 17 days of CPU work versus a mere 23 minutes on Heracles.
Heracles departs from conventional CPUs and GPUs, which struggle with encrypted workloads. FHE math depends on extremely large integers, intensive polynomial calculations, and complex data transformations that quickly overwhelm general-purpose processors. The chip features a purpose-designed 8192-way SIMD compute engine composed of 64 tile-pairs, each containing 128 parallel arithmetic lanes arranged in an 8x8 mesh. The sheer amount of data involved meant linking 48-GB-worth of expensive high-bandwidth memory to the processor with 819 GB per second connections. Data moves through the array at 9.6 terabytes per second by hopping from tile-pair to tile-pair.
The capability gap that FHE acceleration aims to close has real-world consequences beyond voting systems. Potential applications include huge leaps in medical research and patient care without exposing patient data, more effective tools against money laundering without regulators actually seeing anyone's bank-account information, self-driving cars that can learn from each other without revealing driver details, and analytics about a business without examining customer information. No decryption occurs inside the processor, eliminating entire classes of attacks.
What remains challenging is moving from hardware breakthrough to widespread adoption. Software and algorithmic development has taken FHE far, but it's not nearly far enough for FHE to be practical in any but a small and narrow set of applications. Programming in FHE remains a major challenge because of the many parameters that must be correctly chosen for FHE, and the many auxiliary operations needed to manage FHE computations, which cannot be automatically handled by programming tools. Heracles is currently implemented as a PCIe accelerator card installed alongside standard servers and uses liquid cooling to manage its thermals.
Intel's timeline suggests accelerating maturation of the technology. The company developed Heracles under the DARPA DPRIVE program to design a fully homomorphic encryption accelerator for overcoming computation overhead barriers, indicating military and intelligence applications already shape development priorities. Future FHE applications will allow data sharing and collaboration in finance, insurance, and healthcare while maintaining compliance with privacy regulations.
Competitors are racing alongside Intel. Other research teams at universities and startups are developing their own FHE accelerators, recognising that hardware acceleration is essential before the technology can move beyond specialists. The standardisation of FHE schemes remains ongoing, with Intel's DARPA principal investigator presenting at the HomomorphicEncryption.org Standards Meeting in Seoul, and Intel and collaborators presenting homomorphic encryption research papers at industry conferences.
What distinguishes Heracles is not merely speed, but the demonstration that purpose-built architecture can overcome a fundamental mismatch between encrypted computation's mathematical demands and general processors' design. For organisations managing health records, financial data, or classified information, the ability to outsource processing to cloud services without trusting those services with plaintext represents a transformation in data governance. Whether Heracles itself becomes the platform or merely proves that FHE acceleration is practical, the bottleneck has been broken.