3 **CMSIS-NN** is an open-source software library that provides a collection of efficient neural network (NN) kernels developed to maximize the performance and minimize the memory footprint of neural networks running on Arm Cortex-M processors.
5 
7 CMSIS-NN functions are implemented in several variants and the optimal solution is automatically taken at compile time depending on the features available on the target processor architecture. Implementations for the following features are available:
8 - Single Instruction Multiple Data (SIMD) capability (e.g, Cortex-M0)
9 - DSP extension (e.g Cortex-M4)
10 - Arm M-Profile Vector Extension(MVE) (e.g Cortex-M55).
14 CMSIS-NN is actively maintained in a GitHub repository and is also released as a standalone package in CMSIS Pack format.
16 - [**CMSIS-NN GitHub Repo**](https://github.com/ARM-software/CMSIS-NN) provides the full source code as well as releases in CMSIS-Pack format.
17 - [**CMSIS-NN Documentation**](https://arm-software.github.io/CMSIS-NN/latest/) explains how to use the library and describes the implemented functions in details.
19 ## Key Features and Benefits
21 - CMSIS-NN provides a broad set of neural network kernels for Cortex-M devices.
22 - Optimized implementations for different Cortex-M capabilities (SIMD, FPU, MVE).
23 - Arm Compiler 6 and on Arm GNU Toolchain support.
24 - Follows int8 and int16 quantization specification of TensorFlow Lite for Microcontrollers.
25 - Widely adopted in the industry.