Lightning Talks Session Chair: Mert Pese (Clemson University)-Secure on-sensor machine learning for automotive applications

No ratings

Presented at NDSS Symposium 2024 by

Mahesh Chowdhary

Swapnil Saha

To achieve ultra-low-power (uW level), low latency (uS-mS level), small footprint (< few mm), and private inference at the edge of the edge for time-critical automotive systems, sensor manufacturers now integrate custom processing cores directly within the sensor die. There are a large variety of automotive applications from simple application of detecting vehicle stationary condition to complex C-V2X application that require multi-sensor fusion, and can benefit from such compute capability on sensors. These ultra-low-power (uW level) processing capabilities enable moving some computation directly to the sensor from the different control units in automobiles, allowing on-chip sensor fusion, signal conditioning, and running machine learning models on-sensor. We present the latest generation of smart automotive MEMS sensors that can run decision trees and finite state machines within the sensor. We discuss the hardware architecture of these sensors and their integration in automotive systems. We also illustrate no-code automatic machine learning frameworks that can be used to generate smart and private automotive algorithms for these sensors from raw sensor data, as well as tools and techniques to train and deploy performant machine learning models for automotive applications.