nRF Edge AI Library

The nRF Edge AI library is an optimized compute library for deploying Edge AI applications on Nordic Semiconductor devices.

Overview 

The library consists of several core modules:

Runtime - Compute engine for running Nordic Edge AI Lab generated models.
Digital Signal Processing (DSP) - Signal processing primitives for feature extraction (FFT, spectral analysis, statistics, and others).
Neural Network (NN) - Inference engine for neural network models.

See the following image to understand the structure and dependencies of the nRF Edge AI library:

The library provides the following key features:

It is written in portable C with no external dependencies beyond libc.
Memory footprint is fixed and predictable, making it ideal for embedded systems.
It is optimized for constrained microcontroller environments.
It is provided as precompiled static libraries for common Cortex-M architectures.

Core modules 

See the available modules for nRF Edge AI Library:

Building and configuring 

You can enable the library in your application through Kconfig options in the nRF Connect SDK build system. For details, see nRF Edge AI Library integration.

Once you enable the CONFIG_NRF_EDGEAI Kconfig option, the build system automatically builds and links the library with your application:

It locates the precompiled static library for your target architecture (for example, libnrf_edgeai_cortex-m4.a).
It includes the header files from the include/nrf_edgeai/ directory.
It compiles and links your Nordic Edge AI Lab-generated model sources with your application (for example, nrf_edgeai_user_model.c).
It compiles and links your firmware application with the library.

See also nRF Edge AI Observability Library for collecting runtime statistics from model output probabilities; it works with any inference engine, including this library, and can upload data to the Memfault cloud service.