Monostable Multivibrator Networks - Extremely Low Power Inference at the Edge with Timer Neurons

Learn about Monostable Multivibrator Networks (MMVs) for edge computing in this technical presentation from imec researcher Lars Keuninckx. Explore how MMV networks can enable extremely low-power inference through timer neurons, with detailed coverage of MMV fundamentals and network timing conditions. Discover the training algorithm that optimizes excitatory/inhibitory connections and MMV periods using surrogate gradient techniques. Examine real-world applications through case studies including Google Soli radar gestures, Heidelberg keyword spotting, IBM DVS-128 gestures, and Yin-Yang symbol segmentation. Gain insights into a fresh perspective on neuromorphic engineering that focuses on leveraging fundamental electronic building blocks rather than strictly mimicking biological neurons. Understand how MMVs implemented through digital hardware counters offer a practical approach to efficient edge computing implementations.