Encrypted Computing Architectures for AI Workloads

As AI systems are increasingly deployed on sensitive data, privacy-preserving computation has become a central architectural challenge. ARCH Lab develops hardware and system support for encrypted and secure AI workloads, enabling machine learning inference and computation without exposing private inputs, models, or intermediate data.

We are particularly interested in architectures that reduce the performance overheads of encrypted computation while preserving strong security guarantees. This includes support for homomorphic computation, secure execution mechanisms, and novel architectural techniques for balancing privacy, efficiency, and scalability in modern AI platforms.

Research themes

• Hardware acceleration for encrypted AI
• Architectural support for homomorphic and secure computation
• Privacy-preserving inference systems
• Efficient memory and data movement for secure workloads
• Secure architecture for collaborative and distributed AI