11.6 Edge AI and TinyML
You can run a neural network on a microcontroller.
Before:06. Deep Learning
TinyML runs neural networks on microcontrollers under kilobyte-scale memory and milliwatt power budgets, using TFLite Micro and aggressive quantization — and embedded instincts are the differentiator here. It sits in the advanced bridge run as the destination the module points at. The design order is the rule: read the flash, RAM and latency budget first and design the model to fit, because a model designed first will not.
Work through these
TFLite and TFLite Micro workflow
The workflow for converting a trained model into a form a small device can run, including the reduced runtime intended for microcontrollers. This is the practical path from laptop to hardware.
Post-training quantization and QAT
Reducing numerical precision after training, or training with the reduction in mind. The second gives better results and costs more effort.
Memory, flash and latency budgets
Small devices have hard limits on working memory, storage and response time, and the model must fit all three. Designing against the budget from the start avoids a model that cannot be deployed.
Deploying to ESP32, Arduino Nano 33 or STM32
Putting the model onto one of the common development boards, which is where everything above becomes real. Choosing a board you can actually obtain is part of the exercise.
Sign in to keep your progress.
Free resources
We haven't checked most of these for screen reader use yet.
Links last checked 29 Aug 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.