By 
Newark, NJ-based Duality Technologies announced on February 3 it has been awarded a $14.5M contract from the Defense Advanced Research Projects Agency (DARPA) DPRIVE program. Under the contract with DARPA, Duality will lead a team developing a novel ASIC (Application-specific Integrated Circuit) – code named ‘TREBUCHET’ – to accelerate computations using Fully Homomorphic Encryption (FHE), with the goal of making FHE cost-effective for even the most challenging high-value applications.
FHE is an advanced cryptographic technique, widely considered the ‘holy grail of encryption,’ which enables multiple users to process and glean insights on encrypted data while the data or models remain encrypted, preserving data privacy throughout the analytics process. This new hardware is crucial in helping data scientists cross the next frontier in FHE: the ability to train some of the most advanced machine learning (ML) models on encrypted data, enabling organizations to leverage greater amounts of diverse sensitive data for training.
Traditionally, encrypted data is not usable in computations. With increased privacy concerns and tightening data protection regulations, organizations across industries are looking for secure computing methods to train and execute AI models without exposing sensitive or confidential data or IP. For example, PETs based on FHE facilitate cross-border medical research projects, which rely on the interchange of data between healthcare providers and research institutes. Without privacy preservation, healthcare providers would not be able to share sensitive patient data of the sort necessary to yield important research breakthroughs – ultimately saving lives.
DARPA’s Data Protection in Virtual Environments (DPRIVE) program aims to design and implement a hardware accelerator for FHE computations that will dramatically reduce the runtime and therefore the cost of high-value FHE applications, such as privacy-preserving image recognition. The computing hardware to be developed by the Duality-led team will reduce run time overhead by many orders of magnitude compared to current software-based FHE computations on conventional CPUs. By accelerating the hardest and highest value FHE calculations to within one order of magnitude of current performance on unencrypted data, the new technology will overcome an obstacle which has hitherto limited widespread use of FHE.
“Duality team members have been supporting DARPA-funded innovation and application of FHE for over a decade. For the DARPA DPRIVE project, we’ve put together an expert team with knowhow covering the entire stack from encrypted data science applications, FHE software engineering, embedded design, ASIC architecture and layout, to state-of-the-art automated system optimization and design verification,” said Dr. David Bruce Cousins, director of Duality Labs and principal investigator for the project. “Some members of our team developed the first ever prototype HE hardware accelerators under the DARPA PROCEED program starting in 2010 and are lead developers for the PALISADE open source FHE library, first developed for the DARPA SAFEWARE program in 2015. We are excited to harness our extensive expertise, which spans the encrypted computing stack, to contribute to the DPRIVE program, making privacy-preserving analytics and machine learning a reality.”
“Duality has a track record of successfully transitioning DARPA research to commercial privacy-preserving data science software products, and DPRIVE’s final acceptance challenge of accelerating encrypted Neural Net training lines up perfectly with our product roadmap,” said Dr. Kurt Rohloff, CTO and co-founder, Duality Technologies. “No other company, either start-up or multinational, has been as successful as Duality in developing and delivering real-world solutions for practical application of advanced privacy technologies such as Fully Homomorphic Encryption.”
Source: Duality
