ENERGY-EFFICIENT FEATURE EXTRACTION ENGINE AND SECURE CHIP IDENTIFICATION FOR UBIQUITOUS SURVEILLANCE

Abstract

One important part of surveillance is vision, and a critical step in computer vision is object detection. In terms of energy, the trend is towards pushing power consumption to sub-mW at tens of MHz frequency. An energy-quality scalable feature extraction accelerator is presented as the first chip demonstration of the Oriented FAST Rotated BRIEF (ORB) algorithm. In this accelerator, tuning knobs are introduced, allowing for adjustable balance between the energy consumption and quality of the feature extraction accelerator. Security issues are expected to arise in terms of data authenticity, integrity and confidentiality. To assure that the data and the sender are legitimate, chip identification using physically unclonable functions (PUFs) is done. A novel class of mono-stable static (PUFs) for secure key generation and chip identification. From a statistical quality viewpoint, the achieved reproducibility and uniqueness are the best-in-class, with lowest energy compared to state of the art in PUFs.