Ultra-Fast Energy-Efficient Neuromorphic Edge Processing for Event-Based and Frame-Based Cameras

Explore a technical conference talk from ETH Zurich's Michele Magno discussing advanced neuromorphic edge processing systems for event-based and frame-based cameras. Learn about dynamic vision sensor (DVS) technology and its microsecond-level reaction capabilities that outperform traditional RGB cameras in perception tasks. Discover two innovative embedded platforms: ColibriUAV, designed for UAV applications with dual camera interfaces, and ColibriEYE. Understand the architecture of Kraken, a cutting-edge low-power RISC-V System on Chip featuring hardware accelerators for spiking and deep ternary neural networks. Examine benchmark results showing impressive performance metrics, including 7200 frames of events per second processing capability with only 10.7 mW power consumption - 6.6 times faster and 100 times more energy-efficient than USB interface approaches. Gain insights into practical applications for autonomous nano-drones and eye tracking, with overall system power consumption under 50 mW and millisecond-range latency performance.