eIQ^™ for Arm^® CMSIS-NN

Overview

Arm CMSIS-NN is a collection of efficient neural network kernels used to maximize the performance and minimize the memory footprint of neural networks on Arm^® Cortex^®-M processor cores. For easy deployment, NXP eIQ integrates CMSIS-NN directly into MCUXpresso, along with all the other Arm CMSIS components. The use of CMSIS-NN requires more manual intervention than TensorFlow Lite, but yields faster performance and small memory footprint, at the expense of a limited set of neural network operators. Therefore, some models are not supported. Arm also provides a conversion script to parse the network topology and automatically determine the appropriate functions to be used.

MoreLess

Features

Delivered as middleware in the MCUXpresso SDK
Supports i.MX RT applications processors, LPC55S69 MCUs, and Cortex-M based devices
Developed by Arm to provide neural network support for Cortex-M4 and Cortex-M7 cores
Faster and smaller than TF Lite because CMSIS-NN development flow is entirely offline, creating a binary targeting M-class platform
Supports a basic set of operations to implement common model layers (e.g, convolution, fully-connected, pooling, activation)
Conversion scripts (from Arm) to convert Caffe models into CMSIS-NN API calls

MoreLess

Features

Delivered as middleware in the MCUXpresso SDK
Supports i.MX RT applications processors, LPC55S69 MCUs, and Cortex-M based devices
Developed by Arm to provide neural network support for Cortex-M4 and Cortex-M7 cores
Faster and smaller than TF Lite because CMSIS-NN development flow is entirely offline, creating a binary targeting M-class platform
Supports a basic set of operations to implement common model layers (e.g, convolution, fully-connected, pooling, activation)
Conversion scripts (from Arm) to convert Caffe models into CMSIS-NN API calls

Download