AI at the edge

A curated list of hardware, software, frameworks and other resources for Artificial Intelligence at the edge. Inspired by awesome-dataviz.

Contents

• Hardware
• Software
• Frameworks
• Contributing
• License

Hardware

• OpenMV - A camera that runs with MicroPython on ARM Cortex M6/M7 and great support for computer vision algorithms. Now with support for Tensorflow Lite too.
• JeVois - A TensorFlow-enabled camera module.
• Edge TPU - Google’s purpose-built ASIC designed to run inference at the edge.
• Movidius - Intel's family of SoCs designed specifically for low power on-device computer vision and neural network applications.
  • UP AI Edge - Line of products based on Intel Movidius VPUs (including Myriad 2 and Myriad X) and Intel Cyclone FPGAs.
  • DepthAI - An embedded platform for combining Depth and AI, built around Myriad X
• NVIDIA Jetson - High-performance embedded system-on-module to unlock deep learning, computer vision, GPU computing, and graphics in network-constrained environments.
  • Jetson TX1
  • Jetson TX2
  • Jetson Nano
• Artificial Intelligence Radio - Transceiver (AIR-T) - High-performance SDR seamlessly integrated with state-of-the-art deep learning hardware.
• Kendryte K210 - Dual-core, RISC-V chip with convolutional neural network acceleration using 64 KLUs (Kendryte Arithmetic Logic Unit).
  • Sipeed M1 - Based on the Kendryte K210, the module adds WiFi connectivity and an external flash memory.
  • M5StickV - AIoT(AI+IoT) Camera powered by Kendryte K210
  • UNIT-V - AI Camera powered by Kendryte K210 (lower-end M5StickV)
• GreenWaves GAP8 - RISC-V-based chip with hardware acceleration for convolutional operations.
• Ultra96 - Embedded development platform featuring a Xilinx UltraScale+ MPSoC FPGA.
• Apollo3 Blue - SparkFun Edge Development Board powered by a Cortex M4 from Ambiq Micro.
• Google Coral - Platform of hardware components and software tools for local AI products based on Google Edge TPU coprocessor.
  • Dev boards
  • USB Accelerators
  • PCIe / M.2 modules
• Gyrfalcon Technology Lighspeeur - Family of chips optimized for edge computing.
• ARM microNPU - Processors designed to accelerate ML inference (being the first one the Ethos-U55).
• Espressif ESP32-S3 - SoC similar to the well-known ESP32 with support for AI acceleration (among many other interesting differences).
• Maxim MAX78000 - SoC based on a Cortex-M4 that includes a CNN accelerator.

Software

• TensorFlow Lite - Lightweight solution for mobile and embedded devices which enables on-device machine learning inference with low latency and a small binary size.
• TensorFlow Lite for Microcontrollers - Port of TF Lite for microcontrollers and other devices with only kilobytes of memory. Born from a merge with uTensor.
• Embedded Learning Library (ELL) - Microsoft's library to deploy intelligent machine-learned models onto resource constrained platforms and small single-board computers.
• uTensor - AI inference library based on mbed (an RTOS for ARM chipsets) and TensorFlow.
• CMSIS NN - A collection of efficient neural network kernels developed to maximize the performance and minimize the memory footprint of neural networks on Cortex-M processor cores.
• ARM Compute Library - Set of optimized functions for image processing, computer vision, and machine learning.
• Qualcomm Neural Processing SDK for AI - Libraries to developers run NN models on Snapdragon mobile platforms taking advantage of the CPU, GPU and/or DSP.
• X-CUBE-AI - Toolkit for generating NN optimiezed for STM32 MCUs.
• Neural Network on Microcontroller (NNoM) - Higher-level layer-based Neural Network library specifically for microcontrollers. Support for CMSIS-NN.
• nncase - Open deep learning compiler stack for Kendryte K210 AI accelerator.
• deepC - Deep learning compiler and inference framework targeted to embedded platform.
• uTVM - MicroTVM is an open source tool to optimize tensor programs.
• Edge Impulse - Interactive platform to generate models that can run in microcontrollers. They are also quite active on social netwoks talking about recent news on EdgeAI/TinyML.

Other interesting resources

• Benchmarking Edge Computing (May 2019)
• Hardware benchmark for edge AI on cubesats - Open Source Cubesat Workshop 2018
• Why Machine Learning on The Edge?
• Tutorial: Low Power Deep Learning on the OpenMV Cam
• TinyML: Machine Learning with TensorFlow on Arduino and Ultra-Low Power Micro-Controllers - O'Reilly book written by Pete Warden, Daniel Situnayake.
• tinyML Summit - Annual conference and monthly meetup celebrated in California, USA. Talks and slides are usually available from the website.

Contributing

• Please check for duplicates first.
• Keep descriptions short, simple and unbiased.
• Please make an individual commit for each suggestion.
• Add a new category if needed.

Thanks for your suggestions!

License

To the extent possible under law, Xabi Crespo has waived all copyright and related or neighboring rights to this work.