Model Export to Embedded AI Coder

ETAS Embedded AI Coder is a tool designed to convert trained neural networks into efficient C code optimized for embedded systems. It supports various neural network architectures, including fully connected networks, convolutional networks (e.g., ResNets), and LSTM networks. The tool generates code specifically tailored for ARM Cortex-M processors but can also produce generic code compatible with most microcontrollers, with customizable options for further hardware-specific optimizations. For more information, see the ETAS Embedded AI Coder Documentation.

When you export a model to Embedded AI Coder, ASCMO writes a platform-neutral JSON hand-off (*.ascmo.json) that the ETAS Embedded AI Coder can convert into C code for embedded targets.

One file is created per selected output using the pattern:

<base_name>_<output>.ascmo.json

Supported model types

• MLP (Static/Dynamic)

• MLP Classifier (Dynamic)

• CNN (Dynamic)

• RNN (Dynamic)

  Note
  In Embedded AI Coder V1.0, RNN models are supported only when Learn initial states is deactivated.

Support for additional model types will be introduced in the next version.