Introduction
Check out Why FPGA/ADC Communication Prefers GPMC Interface in Industrial Fields? to learn about TinyML~
Today, I will introduce several open-source projects related to TinyML.
TinyML Cookbook
Overview
This book is about TinyML, a rapidly developing field at the unique intersection of machine learning and embedded systems, enabling AI applications on ultra-low-power devices like microcontrollers.
TinyML is an exciting field full of opportunities. With a minimal budget, we can bring objects that interact cleverly with the surrounding world to life and enhance our way of living. This book aims to eliminate barriers through practical examples, allowing developers with no embedded programming experience to get started with TinyML. Each chapter will be a standalone project that teaches how to use some core technologies of TinyML, interface with electronic components like sensors, and deploy ML models on memory-constrained devices.
License
MIT license
Lattice TinyVision & TingyML
https://www.latticesemi.com/Products/DevelopmentBoardsAndKits/HimaxHM01B0
Above are just reference links for Lattice’s related AI examples implemented on the low-power FPGA Up5k. You can search for more details on their official website, which has relevant introductions. Although the GitHub repository doesn’t seem official, there are many complete reference designs available.
Efinix TinyML
https://github.com/Efinix-Inc/tinyml/tree/661ae30f2bf5b083ab88c7a4e54f0185a859f9b8
Efinix offers a TinyML platform based on the open-source TensorFlow Lite for Microcontrollers (TFLite Micro) C++ library, which runs on RISC-V with the Efinix TinyML accelerator. This site provides an end-to-end design process to help deploy TinyML applications on Efinix FPGAs. It covers the design process from training AI models, post-training quantization, to running inference on RISC-V using the Efinix TinyML accelerator. Additionally, it showcases the deployment of TinyML on Efinix’s highly flexible domain-specific framework.
RISC-V SoC:
Design Process:
TinyAcc
This is a project implementing a neural network model with a drop function.
Conclusion
The TinyML projects introduced today are just a few. The current application scenarios are still relatively inclined towards embedded microprocessors, with only Lattice and Efinix FPGAs launching their own IP and example programs in this area. Lattice’s development leans more towards open-source promotion, so whether this application area is a “prospect” or a “dead end” is open to interpretation~
Finally, I would like to thank all the contributors of open-source projects that have greatly benefited us. If you have any projects of interest, feel free to leave a message in the background or add me on WeChat to leave a message. That’s all for today; I am your dedicated writer, looking forward to seeing you in the next article.