Summary
This technical report discusses the advancements in event-based neural networks (NN) and their application in sensor fusion systems. It outlines the challenges faced by modern sensor systems, particularly in handling the increasing volume of data generated by high-performance sensors used in robotics, drones, and autonomous vehicles. The report details imec's research and development efforts to create flexible neural network architectures that leverage event-based processing to reduce latency and power consumption. The integration of event-based feature extraction with sensor fusion is emphasized, allowing only relevant features to be transmitted downstream. The report also presents the development of neuromorphic AI algorithms and hardware, designed to operate efficiently with minimal energy usage. A proof of concept is highlighted, demonstrating the feasibility of event-based processing with a digital spiking neural network chip. The report concludes with potential applications in various fields, including autonomous driving and medical sensors.
