LLA: Enhancing Security and Privacy for Generative Models with Logic-Locked Accelerators

Authors: You Li, Guannan Zhao, Yuhao Ju, Yunqi He, Jie Gu, Hai Zhou

Venue: AAAI Conference on Artificial Intelligence (AAAI), 2025

As generative models are increasingly deployed on dedicated hardware, their parameters and behavior become exposed to end-users with physical access. LLA (Logic-Locked Accelerators) extends hardware logic-locking techniques from circuit IP protection to the AI setting, applying key-based locking to accelerators so that only authorized users obtain correct model behavior. The approach enhances the security and privacy of generative models by protecting their parameters at the hardware level, demonstrating a bidirectional synergy between traditional EDA security methods and modern machine-learning systems.