Getting Started

Start with the integration path that matches the project you are building. Each path consumes the same heliaCORE kernel release, but packages it for a different tooling workflow.

Choose Your Path

Use your project shape as the decision point: pick the path that owns dependency resolution in your firmware today. Each path page gives you the exact download, build setting, and verification step for that workflow.

If you are integrating from scratch, start with the CMake package. If your tool already understands CMSIS components, start with the CMSIS-Pack. Zephyr and neuralSPOT-X users should follow their native integration pages.

What to Download

Every heliaCORE release publishes integration artifacts on the GitHub Release. Most users download one integration artifact plus its matching checksum.

Recommended for CMake ns-cmsis-nn-<cpu>-<toolchain>-<version>.tar.gz The full SDK package: headers, source snapshot, prebuilt archive, CMake config, and manifest. This is the right starting point for custom CMake firmware.
Recommended for CMSIS tooling Ambiq.NS-CMSIS-NN.<version>.pack The CMSIS-Pack for IDE and CMSIS-Toolbox workflows. Use it when your project resolves dependencies through CMSIS components.
Advanced use ns-cmsis-nn-staticlibs-<toolchain>-<version>.zip Bare static libraries grouped by CPU. Choose this when your build already owns include paths, CPU flags, and link wiring.
Verification *.sha256 Checksum files for validating downloads before you add them to a firmware build or internal package mirror.

Choose the CPU name in the artifact that matches your target:

  • cortex-m0 for Cortex-M0/M0+ baseline Apollo targets.

  • cortex-m4 for Cortex-M4 targets with FPv4 SP-D16.

  • cortex-m55 for Cortex-M55 targets with MVE.

The cortex-m0 artifact is the Cortex-M0/M0+ baseline package used for Apollo targets in that class.

Choose gcc, atfe, or armclang for the toolchain name. Source integration paths, such as the CMSIS-Pack Source Cvariant or Zephyr source mode, compile the kernels with your project toolchain instead.

For API and operator coverage, see Operator & Kernel Coverage and API.