eIQ Getting Started with NXP Microcontrollers

[#1]

eIQ® is comprised of multiple pieces of hardware and software to enable users to run machine learning models on embedded devices. Some of the key pieces of eIQ enablement for NXP microcontrollers include:

• eIQ Time Series Studio - PC tool to create and deploy classical machine learning and neural network models for time series analysis
• eIQ Inference Engines - Included as part of MCUXpresso SDK or Yocto Linux, these are used to do inferencing of pre-trained models on embedded devices. Options for MCUs include TensorFlow Lite, and coming soon, ExecuTorch.
• eIQ Neutron NPU – Accelerator core architecture embedded inside specific NXP devices like MCX N and i.MX RT700 to accelerate the inference of neural network models
• eIQ Toolkit - Contains the Neutron Converter tool for enabling models to be accelerated with eIQ Neutron NPUs
• eIQ Model Zoo - browse models tested on NXP silicon
• eIQ Model Watermarking Extension - Enhance copyright protections on custom models
• eIQ Model Creator - Partnership with ModelCat for vision-based model development

There are two main paths for using AI/ML on NXP devices.

• Use eIQ Time Series Studio to generate and deploy a model for time series applications
• Use eIQ inference engines to deploy a pre-trained neural network model

This attached lab will cover the 2nd option by showing how to run the TensorFlow Lite for Microcontrollers (TFLM) inference engine examples found in the MCUXpresso SDK. TFLM support can be found for the following NXP devices in MCUXpresso SDK:

• MCX N
• i.MX RT500
• i.MX RT600
• i.MX RT700
• i.MX RT1050
• i.MX RT1060
• i.MX RT1064
• i.MX RT1160
• i.MX RT1170
• i.MX RT1180

Full details on how to download eIQ inference engine software libraries and run it with VS Code, MCUXpresso IDE, IAR, Keil MDK, or ARM GCC can be found in the attached Getting Started guide. 

For more information about eIQ and some hands-on labs for the i.MX RT family, see the following links:

eIQ Software for MCUs:

• eIQ Libraries for MCUs on MCUXpresso SDK Builder or Github
• eIQ Time Series Studio (TSS) for Time Series applications
• eIQ Toolkit (for Neutron Converter Tool)

MCU Hands-on Labs

• eIQ Time Series Studio Getting Started Lab
• i.MX RT700 NPU Getting Started Lab
• MCX N NPU Getting Started Lab
• TFLite for Microcontrollers Getting Started Lab

Resources:

• eIQ TSS Documentation
• eIQ MCUXpresso SDK Documentation
• eIQ FAQ
• Application Code Hub for AI/ML Examples
• GoPoint for i.MX Application Processors
• eIQ Model Creator by ModelCat.ai
• eIQ Model Zoo
• eIQ Neutron NPU Research Paper

eIQ App Notes

• i.MX RT700 eIQ Neutron NPU Enablement and Performance (AN14700)
• i.MX RT700 USB Camera Object Detection (AN14718)
• Anomaly Detection with eIQ using K-Means clustering in TF-Lite (AN12766)
• Transfer Learning and Datasets (AN12892)
• Handwritten Digit Recognition using TensorFlow Lite (AN12603)
• Caffe Model Development on MNIST Dataset with CMSIS-NN Library (AN12781)
• Gender Voice Recognition with TensorFlow Lite Inference (AN13065)

[nxf52904]

Hello! I tried running the cmsis_nn_cifar10 example from the SDK (configured to have eIQ) but it seems that the example has SDRAM and BOARD_SDRAM that overlaps (both starting at the same address and having the same size). When I try to adjust the memory to be the same as memory in example cmsis_nn_kws, the example is not writing in the right addresses. I also can't erase that memory. Can you please guide me to solve this issue?

[anthony_huereca]

Hi Ana-maria, 

   What board and what IDE are you using? Can you also share your modified linker file? 

-Anthony