Ethos-u root repository

Clone this repo:
  1. 87c0cce 23.05 release by Jonny Svärd · 6 days ago master 23.05
  2. 7697b77 MLBEDSW-7505: Document limitation due to Runtime error on yolo_v2 on fpga_m55_h32 by Tim Hall · 6 days ago
  3. ad51f65 Update readme with 23.05 release notes by Måns Nilsson · 14 days ago
  4. bdf4122 23.02 release by Kristofer Jonsson · 3 months ago 23.02
  5. 3b10c9f Adding OpenAMP remotes by Kristofer Jonsson · 4 months ago

Arm(R) Ethos(TM)-U

This is the root repository for all Arm(R) Ethos(TM)-U software. It is provided to help users download required repositories and place them in a tree structure.

Fetching externals

The externals can be downloaded with a Python script. The default configuration is stored in externals.json which is a human readable JSON file.

$ ./ fetch

The default configuration can be overridden with the -c argument, for example.

$ ./ -c 22.11.json fetch

Directory structure

The script will build following directory structure.

+-- core_platform
+-- core_software
|   +-- applications
|   +-- cmsis
|   +-- cmsis-nn
|   +-- cmsis-view
|   +-- core_driver
|   +-- drivers
|   +-- lib
|   +-- rtos
|   +-- tflite_micro
+-- linux_driver_stack
+-- vela
.This is the root directory for all Arm Ethos-U software.
core_platformThis directory contains target specific files and is provided as an example how core software can be built for target platforms.
core_softwareThe software executing on Arm Cortex-M is referred to as Core Software. This folder provides a small build system that illustrates how to build the key components for the Arm Ethos-U core software.
core_driverThe Arm Ethos-U NPU driver.
cmsisCMSIS provides generic interfaces to boot and configure the Arm Cortex-M CPUs.
cmsis-nnCMSIS-NN provides optimized neural network kernels for Arm Cortex-M CPUs.
tflite_microThe TensorFlow Lite microcontroller framework is used to run inferences.
linux_driver_stackExample driver stack showing how Linux can dispatch inferences to an Arm Ethos-U subsystem.
velaThe Vela optimizer takes a TFLu file as input and replaces operators that are supported by the Arm Ethos-U NPU with custom operators designed to run on the NPU. Operators not supported by the NPU are executed in software.


The Arm Ethos-U is provided under an Apache-2.0 license. Please see LICENSE.txt for more information.


The Arm Ethos-U project welcomes contributions under the Apache-2.0 license.

Before we can accept your contribution, you need to certify its origin and give us your permission. For this process we use the Developer Certificate of Origin (DCO) V1.1 (

To indicate that you agree to the terms of the DCO, you "sign off" your contribution by adding a line with your name and e-mail address to every git commit message. You must use your real name, no pseudonyms or anonymous contributions are accepted. If there are more than one contributor, everyone adds their name and e-mail to the commit message.

Author: John Doe \<\>
Date:   Mon Feb 29 12:12:12 2016 +0000

Title of the commit

Short description of the change.

Signed-off-by: John Doe
Signed-off-by: Foo Bar

The contributions will be code reviewed by Arm before they can be accepted into the repository.


Please see Security.


Release 23.05

The 23.05 release has been tested against Tensorflow version 2.11 as reference, with the additional patches applied:

Without these patches there may be output diffs between TFL and TFLM for certain models.

There's a discrepancy in tensorflow/lite/micro/cortex_m_corstone_300/ fixed by

Known Issues

TensorFlow Lite for Microcontrollers Out of Memory Error during Runtime

During runtime the TensorFlow Lite for Microcontrollers framework might report the following fatal error:
Failed to resize buffer. Requested: X, available: Y, missing: Z, where X, Y and Z are numbers of bytes and X = Y + Z.
There can be several reasons for running out of memory during an inference but one cause is that too much memory was allocated to the Ethos-U during the offline compilation phase of the .tflite file using Vela. This can result in not enough memory being available at runtime for the other software components e.g. the application, the framework, or the reference kernels. The solution is to calculate the amount of memory required at runtime by all components and then update the amount allocated to the Ethos-U by using either the Vela CLI option --arena-cache-size or the arena_cache_size attribute in Vela's .ini configuration file. This calculation can be difficult to get right and so a pragmatic solution would be to start by reducing the amount allocated to the Ethos-U by the value ‘Z’ (from the error message) rounded up to the nearest multiple 16 (the default tensor alignment used in Vela). This may be an iterative process.

Trademark notice

Arm, Cortex and Ethos are registered trademarks of Arm Limited (or its subsidiaries) in the US and/or elsewhere.