FPGA Sensor Fusion for Edge AI Applications with Example MIPI Interfaces
(Room 303A)
06 Nov 25
11:45 AM
-
12:15 PM
Tracks:
Hardware - MIPI Embedded Vision
FPGAs have long been the preferred solution for bridging disparate interfaces in complex systems. With the rapid growth of sensor deployments in industrial and automotive domains—driven by increasing AI workloads—there is a critical need for scalable and reusable sensor fusion solutions at the edge. This paper presents a modular FPGA-based hardware design framework combined with parameterized firmware and host drivers to reduce development time and enhance system flexibility. We outline the key components required for implementing efficient sensor fusion using FPGAs in edge AI system architectures. The discussion includes a breakdown of essential hardware modules for interfacing with diverse sensors, including MIPI CSI-2, along with host interfaces for control and data streaming. We also detail firmware considerations for managing sensor data transport and preprocessing at the edge processor, as well as host SW drivers responsible for handling fused sensor data streams. Additional focus is given to protocols enabling dynamic discovery and enumeration of edge sensors, and APIs for sensor configuration, event management, and debugging. A practical example flow is provided, demonstrating I3C as the host control interface alongside 10G Ethernet for high-speed streaming and / or control. This structured approach enables scalable and efficient sensor fusion solutions for evolving edge AI applications.