Deploy a machine learning application to the Arm Ethos-U65 NPU on NXP FRDM i.MX 93 with Topo
Introduction
Understand the architecture of the machine learning application
Understand the toolchains used in the Topo Project
Build the Topo Project from scratch
Clone and deploy the application with Topo
Next Steps

  Log an issue

  Fork and edit

  Discuss on Discord

About this Learning Path

Skill level:  Introductory
Reading time:  1 hr
Last updated:  15 Jul 2026
Skill level:
  Introductory
Reading time:
  1 hr
Last updated:
  15 Jul 2026
Author:Tomas Agustin Gonzalez Orlando, Arm
Arm IP:Cortex-A Cortex-M Ethos-U
Tags: Containers and Virtualization Linux macOS Windows Topo Docker SSH ExecuTorch remoteproc-runtime
Author:
  • Tomas Agustin Gonzalez Orlando, Arm
Arm IP:
Cortex-A Cortex-M Ethos-U
Tags:
Containers and Virtualization Linux macOS Windows Topo Docker SSH ExecuTorch remoteproc-runtime

Who is this for?

This is an introductory topic for embedded/edge software developers who want to deploy machine learning workloads to heterogeneous Arm-based Linux targets using Topo, including leveraging Arm Ethos-U NPUs.

What will you learn?

Upon completion of this Learning Path, you will be able to:

  • Explain how Topo deploys an application that spans Cortex-A, Cortex-M, and Ethos-U
  • Deploy the topo-imx93-npu-deployment Topo Project, which operates across Cortex-A, Cortex-M, and Ethos-U, to perform image classification using an ExecuTorch MobileNetV2 model
  • Describe how the project is bootstrapped from Compose services, Remoteproc Runtime metadata, and Topo parameters and follow this process yourself
  • Understand how to take similar projects and create Topo Projects, including using Agent Skills

Prerequisites

Before starting, you will need the following:

  • A host machine (x86 or Arm) with Linux, macOS, or Windows
  • An NXP FRDM i.MX 93 target board with Linux setup, accessible over SSH with root access. To do this, see Use Linux on the NXP FRDM i.MX 93 board .
  • Docker installed on the host and target. For installation steps, see Install Docker .
  • At least 25 GB of free disk space on the host if you’re building without cache images.
  • The Device Tree Compiler (dtc) installed on the host.
  • lscpu installed on the target (pre-installed on most Linux distributions)
  • Topo installed on the host. For installation steps, see Deploy containerized workloads to Arm-based Linux targets with Topo .
  • Basic familiarity with containers, SSH, and CLI tools
  • (Optional) Access to an agent, such as Codex or Claude Code

Summary

AI-assisted

This summary was drafted with an approved AI-assisted workflow and reviewed by Arm contributors before publication. Human technical review remains part of the process so the final page reflects engineering rigor, accuracy, and Arm editorial standards.

Close
?
You’ll build and deploy a heterogeneous image classification application to an NXP FRDM i.MX 93 board using Topo. First, you’ll assemble a Topo Project from two base projects: a Cortex-A web application and a Cortex-M33 ExecuTorch runner, then convert the combined sources into a Compose project with Topo metadata and Remoteproc Runtime services. The application preprocesses images on Cortex-A, shares model and tensor data through shared memory, and issues RPMsg commands to Cortex-M33 firmware that delegates inference to the Arm Ethos-U65 NPU via the ExecuTorch backend. After validating target readiness with topo health and deploying, containers run on the board, and you’ll be able to access a browser-based MobileNetV2 classifier.

Frequently asked questions

AI-assisted

These FAQs were drafted with an approved AI-assisted workflow and reviewed by Arm contributors before publication. Human technical review remains part of the process so the final page reflects engineering rigor, accuracy, and Arm editorial standards.

Close
?
How do I confirm the FRDM i.MX 93 target is ready before deploying?
Run topo health --target <user>@<target-ip>. The host and target sections should show successful checks for SSH and the container engine, and the target should also report Remoteproc Runtime and Remoteproc as healthy.
Which components run on Cortex-A and Cortex-M33, and how do they communicate?
The Cortex-A side runs the web application that prepares images, writes model and tensor data into shared memory, and sends inference commands. The Cortex-M33 runs the ExecuTorch firmware and receives commands over RPMsg.
What should I expect after deploying the Topo Project?
Topo builds images on the host, transfers them to the target, and starts services on the board. Remoteproc Runtime starts the Cortex-M firmware, and the browser-based MobileNetV2 classifier becomes available.
What are the key steps to turn the two base projects into a Topo Project?
Combine the Cortex-A web app and the Cortex-M33 ExecuTorch firmware sources into a single repository, make it a normal Compose project, then add Topo metadata and Remoteproc Runtime services. The Topo Project is bootstrapped from Compose services, Remoteproc Runtime metadata, and Topo parameters.
What should I check if `topo health` reports a failure?
Resolve the specific errors shown by topo health before continuing. Confirm SSH access to the target, verify the container engine is available on both host and target, and ensure Remoteproc Runtime is present on the target.
Next