Introduction
Are you having trouble implementing AI/ML into your embedded devices?
(From here on, AI/ML will be abbreviated as AI)
Although there are ideas for incorporating AI into your company's products, there are likely to be some challenges, such as:
・Challenge 1: "I can't imagine to what extent AI/ML can be achieved with embedded devices."
In this article, we will introduce the benefits of NXP's AI solution, eIQ, in addressing these anticipated challenges !
What is eIQ?
Figure 1: eIQ Overview
As shown in Figure 1, the eIQ ® ML software development environment is comprised of the software libraries and AI model development environment required to run AI applications on NXP's MCUs (microcontrollers) and MPUs (processors), and is provided free of charge .
Issue 1: "I can't imagine how much AI/ML can be achieved with embedded devices."
Figure 2: AI hardware portfolio
NXP is a semiconductor vendor with a wide portfolio, ranging from extremely lightweight microcontrollers equipped with the Cortex-M series, a CPU suitable for AI, to processors equipped with the multi-core Cortex-A series, as well as an AI accelerator NPU.
Figure 3: Go point / Application Code Hub
Various AI-related demos for various MCUs/MPUs are available at the links below. By checking the details of these demos, you can get an idea of how much AI processing is possible with each grade of MCU/MPU.
・MCU demo
Application Code Hub | NXP Semiconductors
・MPU demo
GoPoint for i.MX Applications Processors | NXP Semiconductors
Additionally, the following page, modelzoo, has published various models converted for NXP devices along with evaluation data. By checking this page, you can get an idea of how much processing time will be required for each type of processing .
Figure 4: Model Zoo
Issue 2: "I don't know where to start" "I'm concerned about the cost of development tools and learning costs"
Figure 5: Model Convert
The free eIQ Toolkit allows you to develop AI models using GUI operations from time-series data/image data from sensors you have prepared, without any complicated setup.
By combining it with the various sample software mentioned above, you can start developing and evaluating embedded AI without incurring significant learning costs for AI tools.
・Time series model development tool (eIQ Time Series Studio)
Figure 6: eIQ Time Series Studio Overview
Figure 7: eIQ TSS training screen
Using eIQ Time Series Studio (TSS), you can develop machine learning models that perform anomaly detection, classification, and regression (prediction) using a GUI. End-to-end development functionality reduces the cost of developing time-series AI models for edge devices. Input values from various sensors can be used as training data, making it useful for introducing AI to a variety of devices, as shown below.
Figure 8: eIQ Time Series Studio usage example
Click here for "eIQ Time Series Studio (time series model development tool) Overview and Usage "
・Image processing model development tool (eIQ Portal)
Figure 9: eIQ image processing model development function
Using the image processing model development function of eIQ Toolkit, you can develop deep learning-based object detection/image classification models using a GUI.It offers a complete set of functions, including an augmentation function that augments the training data by processing images, training with specified hyperparameters, exporting in a state optimized for NXP MCU/MPU, and testing the developed model.
Figure 10: Overview of eIQ Model Water Marking Technology
Additionally, the eIQ model watermarking technology function helps protect your models by providing a way to prove whether your image processing models have been illegally copied.
For details, please refer to the link below and the eIQ_Toolkit_UG.pdf included in the eIQ Toolkit.
eIQ ® Model Watermark Technology | NXP Semiconductors
Issue 3: "I developed the AI components on a PC, but I'm worried about whether I can port it to embedded devices."
・eIQ Portal Model Convert function
Figure 11: eIQ Convert function overview
eIQ not only has a model development function, but also has a function to optimize users' own models for NXP MCUs/MPUs. During conversion, it is possible to specify options such as quantization format and per channel/per tensor, and it is configured so that assets built in a PC or rich SoC- based evaluation environment can be smoothly reused for NXP 's MCUs/MPUs suitable for embedded use in product development.
・Nvidia TAO Tool Extension
Figure 12: Nvidia Tao Toolkit Extension Overview
NXP is the first semiconductor vendor to directly integrate the NVIDIA TAO Toolkit API into its eIQ machine learning development environment, an AI enablement tool, enabling the deployment of NVIDIA pre-trained AI models on NXP edge processing devices.
For details, please refer to the article below and the eIQ_Toolkit_UG.pdf included in the eIQ Toolkit.
Issue 4: "I'm interested in the recently popular generative AI, LLM, but I'm not sure if it can be realized on embedded devices."
・LLM Solutions/GenAI Flow (Pre-release)
Figure 13: LLM Pipeline
NXP is one of the industry's first companies to offer an embedded generative AI/LLM solution called Gen AI Flow.
For detailed implementation instructions, please refer to the article " [Getting Started] i.MX 95: LLM_RAG Implementation Hands-on - eIQ Gen AI Flow - ".
-------
[Added 4/7/2025]
The first revision of eIQ genAI Flow was released in the BSP update for Q1 2025. Please see below for details.
------
This can be achieved by providing an environment incorporating RAG (Figure 14), providing an environment that combines sufficient practicality and security for embedded devices.
Figure 14: RAG Overview
*LLM: Large Language Model
*RAG: Retrieval Augmented Generation
LLM support enables embedded devices to provide more intuitive, conversational user experiences, from smart home devices that support voice commands to industrial equipment that can be controlled with natural language, to in-vehicle infotainment systems that allow users to command and operate in-vehicle functions through hands-free, two-way conversation.
Please also refer to the following white papers:
https://www.nxp.jp/webapp/Download?colCode=GEN-AI-RAG-WHITEPAPER
summary
NXP's eIQ provides the industry's highest level of functionality required for edge AI development, as shown below, reducing customers' AI implementation and development costs.
・We provide a free environment for developing time-series AI and image processing models using a GUI.
- Model conversion function and collaboration with NvidiaTool allow for smooth migration from evaluation environments such as PCs
・By utilizing GenAI Flow , you can respond to the latest trend , generative AI (LLM) .
-----
-This article is based on information current as of the time of writing (September 25, 2025).
=========================
We are currently unable to respond to comments in the "Comment" section of this post.
We apologize for the inconvenience, but when making inquiries, please refer to `` Technical Questions to NXP - How to Contact Us( Japanese Blog) ''.
(If you are already an NXP distributor or have a relationship with NXP, you may ask the person in charge directly.)
We will introduce the benefits of using NXP's AI/ML solution eIQ to address the challenges of introducing AI/ML to embedded devices.
- Plan to release LLM+RAG solution
-Developed time series AI and image processing AI models
-Convert function allows smooth migration of existing AI/ML assets