Edge AI Duty-Cycle Lab: TinyML Battery Budget (2D)

Deploying a neural network to a battery-powered microcontroller — the TinyML pattern behind Edge Impulse projects and Arduino ML boards — is an energy-budgeting problem as much as a machine-learning one. This 2D schematic draws a sensor node whose status LED pulses dim through sleep and flares amber through every sample-and-infer burst, scrolls a live accelerometer-style waveform into the classifier's input window, animates a real packet toward a gateway icon whenever an inference score clears the detection threshold, and renders a scrolling duty-cycle timeline strip you can drag to scrub back through recent history. Four controls — sampling rate, wake interval, detection threshold and model size (Tiny→Large) — drive a real energy model: inference time scales with the model's FLOP count on a fixed microcontroller clock, duty cycle falls out of how much of each wake period is spent actively sampling and computing versus asleep, and that duty cycle sets the average power draw and therefore the estimated battery life on a coin-cell budget. A second chart sweeps the wake interval directly, showing where battery life collapses once T approaches the unavoidable active-burst floor. Flipping to always-on mode removes sleep entirely, showing why an always-listening node needs mains power while a duty-cycled one can run for months on the same battery.