# Requirements

Start with an Ambiq firmware project that builds and runs on your target board.
heliaCORE supplies kernels; your board SDK supplies startup code, the linker
script, device support, and the flash/debug workflow.

## Before you begin

- A board application that you can build, flash, and debug.
- Your board SDK's compiler and build tools available in the shell or IDE.
- One heliaCORE release selected for headers and library sources or archives.
- For source integration, CMake 3.15.6 or newer where using this repository's top-level build; your SDK may require a newer version.

## Check your build

For an existing CMake build, run `cmake --build build --verbose` and inspect a
compiler command for a firmware source file. If nothing rebuilds, use your IDE's
build log or rebuild that file. Record the compiler executable/version and
`-mcpu`, `-mfpu`, and `-mfloat-abi` flags, including those supplied by your board
configuration.

Compare a prebuilt SDK's `manifest.json` with that command:

| Manifest field | Compare with your firmware |
|---|---|
| `toolchain.id` | Artifact toolchain name: `gcc`, `atfe`, or `armclang`. |
| `toolchain.compiler_id` and `toolchain.version` | Compiler family and version used to build the archive. |
| `abi.arch_flags` | CPU, FPU, and float ABI flags. |
| `features.f32` and `features.f16` | Whether the archive contains the floating-point kernels you need. |

If no release package matches your requirements, use a source build.

## Target and compiler

- Identify your target's Cortex-M core and the CPU, FPU, and float ABI flags used by your firmware.
- For a prebuilt library, select a release artifact matching your CPU and compiler. Check its `manifest.json` for the build configuration.
- For a source build, use your project's toolchain and target flags. DSP and Helium paths are selected at compile time where supported.

See [Cortex-M targets](https://ambiqai.github.io/ns-cmsis-nn/guide/architecture/cortex-m-targets/) for
architecture details and [Toolchains](https://ambiqai.github.io/ns-cmsis-nn/guide/architecture/toolchains/)
for compiler compatibility and floating-point requirements.

## Source or prebuilt?

| Choose | When it fits |
|---|---|
| **Prebuilt** | A release library matches your firmware's CPU, compiler, ABI, and required kernel features. |
| **Source** | You need custom operator selection, or no archive supplies your required features and compiler settings. |

FP16 and FP32 are opt-in APIs. Support varies by operator and
target. Check [data-type coverage](https://ambiqai.github.io/ns-cmsis-nn/guide/coverage/data-types-by-family/)
before enabling them. Changing a build option does not add kernels to an
already compiled archive; select a matching archive or rebuild from source.

## Prepare your integration

Use one heliaCORE release throughout your project. Download packages from
[GitHub Releases](https://github.com/AmbiqAI/ns-cmsis-nn/releases), verify the
provided checksums, and follow the guide for [CMake](https://ambiqai.github.io/ns-cmsis-nn/getting-started/cmake/),
[CMSIS-Pack](https://ambiqai.github.io/ns-cmsis-nn/getting-started/cmsis-pack/), [Zephyr](https://ambiqai.github.io/ns-cmsis-nn/getting-started/zephyr/), or
[neuralSPOT-X](https://ambiqai.github.io/ns-cmsis-nn/getting-started/neuralspot-x/).