ModusToolbox™ Machine Learning

Overview

ModusToolbox™ Machine Learning (ML) enables you to rapidly evaluate and deploy Machine Learning models on Infineon MCUs. ModusToolbox™ ML is designed to work seamlessly with the ModusToolbox™ ecosystem and can be dropped into existing projects to enable machine learning tasks on low-power edge devices.

Easy Integration and Flexible Tools

ModusToolbox™ ML tools enables you to:

  • Import models from popular training frameworks such as TensorFlow™
  • Optimize the model for embedded platforms to reduce the size and complexity
  • Validate the performance of the optimized model by checking performance against test data
  • Generate optimized model code and libraries which are integrated with the ModusToolbox™ flow

Optimized for Embedded Performance

ModusToolbox™ ML includes an embedded inference engine which supports optimized implementations of most popular Neural Network operators such as 1D/2D convolutions, a variety of activation functions as well as support for more complex operators for RNN networks such as GRU.

ModusToolbox™ ML adds configurators, tools, code examples and supporting libraries to help quickly get started with Machine Learning Model deployment on Infineon MCU’s. 

What’s New

The ModusToobox™ ML 1.2 release brings in a set of great new features and improvements to make the deployment and validation of Deep Learning models on the PSoC™ 6 platform even easier. The major updates to this release are:

  • Data streaming for on-device validation
  • Size optimization option during model conversion
  • Improved cycle, accuracy for 8-bit models
  • New kernels
    • LSTM, and Dilated CNN’s
  • Limited support for Lambda Layers - reduce_max, reduce_min, reduce_prod, reduce_mean

Please download the latest version of ModusToolbox™ and related patches to get access to ModusToolbox™ ML tooling.

  • Import from Keras(H5) type file models
  • Validation GUI for various quantization levels
  • Supported operators and kernels
Network Type Operators/Primitives Activation Format Weights Format
MLP Dense, Batch Normalization Float, 16-bit, 8-bit Float, 16-bit, 8-bit
Conv1D Conv1D, MaxPool1D, MaxPooling1D, Batch Normalization Float, 16-bit, 8-bit Float, 16-bit, 8-bit
Conv2D Conv2D, MaxPooling2D, SeparableConv2D, DepthwiseConv2D, Average Pooling 1D, Global Max Pooling 1D Float, 16-bit, 8-bit Float, 16-bit, 8-bit
Recurrent GRU, LSTM Float, 16-bit, 8-bit Float, 16-bit, 8-bit
Reshape Reshape 2D, Flatten, Input Layer Float, 16-bit, 8-bit Float, 16-bit, 8-bit
Merge Concatenate, Add Float, 16-bit, 8-bit Float, 16-bit, 8-bit
Activation Relu, tanh, Softmax, sigmoid Float, 16-bit, 8-bit Float, 16-bit, 8-bit
Auxiliary Dropout, Activation Float, 16-bit, 8-bit Float, 16-bit, 8-bit

To get started with ModusToolbox™ ML, please download the latest version of ModusToolbox™ software and patches and get the below hardware evaluation kits:

Please follow the steps in the ModusToolbox™ ML User Guide listed in the documentation which will guide you through the setup.

Support