Austin Rovinski

Today's most advanced cryptographic computing technologies — which enable privacy-preserving computation — are trapped in research labs by one critical barrier: they're thousands of times too slow for everyday use.

NYU Tandon, helming a research team that includes Stanford University and the City University of New York, just received funding from a $3.8 million grant from the National Science Foundation to build the missing infrastructure that could make those technologies practical, via a new design platform and library that allows researchers to develop and share chip designs.

The problem is stark. Running a simple AI model on encrypted data takes over 10 minutes instead of milliseconds, a four order of magnitude performance gap that impedes many real-world use cases.

Current approaches to speeding up cryptographic computing have hit a wall, however. "The normal tricks that we have to get over this performance bottleneck won’t scale much further, so we have to do something different," said Brandon Reagen, the project's lead investigator. Reagen is an NYU Tandon assistant professor with appointments in the Electrical and Computer Engineering (ECE) Department and in the Computer Science and Engineering (CSE) Department. He is also on the faculty of NYU's Center for Advanced Technology in Telecommunications (CATT) and the NYU Center for Cybersecurity (CCS).

The team's solution is a new platform called "Cryptolets.”

Currently, researchers working on privacy chips must build everything from scratch. Cryptolets will provide three things: a library where researchers can share and access pre-built, optimized hardware designs for privacy computing; tools that allow multiple smaller chips to work together as one powerful system; and automated testing to ensure contributed designs work correctly and securely.

This chiplet approach — using multiple small, specialized chips working together — is a departure from traditional single, monolithic chip optimization, potentially breaking through performance barriers.

For Reagen, this project represents the next stage of his research approach. "For years, most of our academic research has been working in simulation and modeling," he said. "I want to pivot to building. I’d like to see real-world encrypted data run through machine learning workloads in the cloud without the cloud ever seeing your data. You could, for example, prove you are who you say you are without actually revealing your driver's license, social security number, or birth certificate."

What sets this project apart is its community-building approach. The researchers are creating competitions where students and other researchers use Cryptolets to compete in designing the best chip components. The project plans to organize annual challenges at major cybersecurity and computer architecture conferences. The first workshop will take place in October 2025 at MICRO 2025, which focuses on hardware for zero-knowledge proofs.

"We want to build a community, too, so everyone's not working in their own silos," Reagen said. The project will support fabrication opportunities for competition winners, with plans to assist tapeouts of smaller designs initially and larger full-system tapeouts in the later phases, helping participants who lack chip fabrication resources at their home institutions

"With Cryptolets, we are not just funding a new hardware platform—we are enabling a community-wide leap in how privacy-preserving computation can move from theory to practice,” said Deep Medhi, program director in the Computer & Information Sciences & Engineering Directorate at the U.S. National Science Foundation. “By lowering barriers for researchers and students to design, share and test cryptographic chips, this project aligns with NSF’s mission to advance secure, trustworthy and accessible technologies that benefit society at large."

If the project succeeds, it could enable a future where strong digital privacy isn't just theoretically possible, but practically deployable at scale, from protecting personal health data to securing financial transactions to enabling private AI assistants that never see people's actual queries.

Along with Reagen, the team is led by NYU Tandon co-investigators Ramesh Karri, ECE Professor and Department Chair, and faculty member of CATT and CCS; Siddharth Garg, Professor in ECE and faculty member of NYU WIRELESS and CCS; Austin Rovinski, Assistant Professor in ECE; The City College of New York’s Rosario Gennaro and Tushar Jois; and Stanford's Thierry Tambe and Caroline Trippel, with Warren Savage serving as project manager. The team also includes industry advisors from companies working on cryptographic technologies.